A major focus of current biological studies is to fill the knowledge gaps between cell, tissue and organism scales. To this end, a wide array of contemporary optical analytical tools enable multiparameter quantitative imaging of live and fixed cells, three-dimensional (3D) systems, tissues, organs and organisms in the context of their complex spatiotemporal biological and molecular features. In particular, the modalities of luminescence lifetime imaging, comprising fluorescence lifetime imaging (FLI) and phosphorescence lifetime imaging microscopy (PLIM), in synergy with Förster resonance energy transfer (FRET) assays, provide a wealth of information. On the application side, the luminescence lifetime of endogenous molecules inside cells and tissues, overexpressed fluorescent protein fusion biosensor constructs or probes delivered externally provide molecular insights at multiple scales into protein–protein interaction networks, cellular metabolism, dynamics of molecular oxygen and hypoxia, physiologically important ions, and other physical and physiological parameters. Luminescence lifetime imaging offers a unique window into the physiological and structural environment of cells and tissues, enabling a new level of functional and molecular analysis in addition to providing 3D spatially resolved and longitudinal measurements that can range from microscopic to macroscopic scale. We provide an overview of luminescence lifetime imaging and summarize key biological applications from cells and tissues to organisms.
In the past decade, there has been a surge of interest in imaging methodologies capable of measuring luminescence lifetime, which previously had been restricted to a small community of physicists and engineers. Luminescence lifetime is an intrinsic characteristic of any endogenous or exogenous luminescent biomolecule (luminophore), including endogenous molecules (autofluorescence), (in)organic dyes, fluorescent proteins, nanosensors and other types of biosensor probes (Fig. 1A; Box 1). Depending on its molecular structure, a luminophore can exhibit either fluorescence (with a lifetime of pico- to nano-seconds) or phosphorescence (with a lifetime of micro- to milli-seconds), and the methods to observe these two are designated as fluorescence lifetime imaging (FLI) and phosphorescence lifetime microscopy (PLIM), respectively (Fig. 1A). FLI methodologies can be integrated into traditional fluorescence imaging approaches, including widefield, confocal, multiphoton, super-resolution and light-sheet microscopy, as well as macro-imaging, fluorescence molecular tomography (FMT) and other optical imaging-based approaches (Ishikawa-Ankerhold et al., 2012; Meyer-Almes, 2017; Sarder et al., 2015; Denicke et al., 2007; Becker et al., 2017; Hirvonen and Suhling, 2020; Le Marois and Suhling, 2017; Datta et al., 2020; Poudel et al., 2020; Wang et al., 1992). FLI can also be combined with more specialized approaches such as Förster resonance energy transfer (FRET), which enables sensing of nanoscale interactions. (Fig. 1; Box 1) (Abe et al., 2013; McGhee et al., 2011; Piston and Kremers, 2007; Rajoria et al., 2014; Bunt and Wouters, 2017). In contrast to measurements of fluorescence intensity, lifetime measurements are independent of the fluorophore concentration (above the required threshold) and are minimally affected by the tissue optical properties. Thus, lifetime imaging provides an absolute value that is directly related to the luminophore structure and to how this structure is affected by environmental changes, such as temperature, pH, polarity and mechanical forces (Fig. 1B). Modern FLI and PLIM platforms cover an optical resolution ranging from sub-micrometer to several millimeters, a spectral range from 180 to 1700 nm and a time-domain resolution from sub-picoseconds to milliseconds, with measurement speeds that range from real-time for one-photon FLI and PLIM (Raspe et al., 2016; Trinh et al., 2019; Cheng et al., 2020) to several minutes per frame for two-photon PLIM (Rytelewski et al., 2019). By allowing quantitative, multiparameter and often non-invasive and non-destructive three-dimensional (3D) analyses of live cells and tissues to be performed, FLI and PLIM provide a broad support to the ongoing ‘3D molecular-imaging revolution’.
Luminescence is a physical phenomenon that encompasses both fluorescence and phosphorescence imaging approaches (Fig. 1A). There are a wide variety of microscopy and macroscopy imaging setups that enable the measurement of luminescence lifetimes of luminophores (i.e. a fluorophore or phosphor) in the visible and NIR wavelength range, with impact in many biomedical fields. Fluorescence lifetime imaging (FLI) includes microscopy-based approaches, described as FLIM (fluorescence lifetime imaging microscopy), and macroscopy-based optical imaging, described as MFLI (macroscopy fluorescence lifetime imaging). Phosphorescence lifetime imaging microscopy is referred to as PLIM.
While fluorescence intensity displays relative magnitude effects that are typically associated with the local concentration of the molecule, luminescence lifetime provides a ‘fingerprint’ of the fluorophore and/or phosphorescent dye structure in relationship to the external environment (sensing). The aim of luminescence lifetime measurements is to quantify the lifetime, τ (tau), which provides information about the immediate physical, chemical and biological environment of the luminophore, effectively making any luminophore a sensing probe. Thus, the measurement of lifetime changes using FLI or PLIM can detect the occurrence of molecular events, such as conformational changes of proteins induced by changes in the physical, chemical and biological environment surrounding the luminophore, which can be included in fluorescent proteins or conjugated to a wide variety of proteins (see figure). In the case of FLI, the lifetime values are within ∼0.1–20 ns, whereas phosphorescence lifetimes (measured using PLIM) are typically sensed in the ‘slower’ range of 1–1000 µs.
FLI and PLIM can be combined with FRET, a nanometer-range proximity assay that informs on protein–protein interactions occurring within 1–10 nm range, or used for other applications. Importantly, FLI and PLIM provide molecular information independently of the spatial resolution delivered by the imaging setup used to collect the data (see figure for a comparison of spatial resolution and depth of penetration for a variety of imaging modalities. CM, confocal microscopy; IVM, intravital microscopy; LISH, light sheet imaging; MFMT, mesoscopic FMT; MPM, multiphoton microscopy; SRM, super-resolution microscopy). For example, incorporating FLI–FRET assays into low-resolution optical imaging techniques, such as FMT, allows for the investigation of nanometer-range molecular interaction in intact animals, as shown for NIR MFLI deep-tissue imaging (Fig. 3).
A number of comprehensive reviews have described the technology underlying FLI and PLIM (Ishikawa-Ankerhold et al., 2012; Hirvonen and Suhling, 2020; Le Marois and Suhling, 2017; Levitt et al., 2009; Becker et al., 2017; Berezin and Achilefu, 2010; Dmitriev, 2017; Periasamy, 2001; Trinh et al., 2019), but they rarely address the benefits and applications of these powerful imaging approaches to researchers with training in biology and biomedical sciences. This Review aims to fill this gap by introducing readers to the FLI, PLIM and FLI–FRET-based methodologies that are important for multiparameter 3D imaging of cells, tissues and small animals, with a brief demonstration of the underlying biosensing technologies. Moreover, we will highlight the growing need for ‘3D FLI’, specifically in cancer and stem cell biology, developmental processes, intravital imaging, super-resolution microscopy, host–pathogen interactions and other areas of cell and molecular biology.
Luminescence lifetime imaging – a window into the molecular environment of 3D cells and tissues
Life inside the human body is seldom less than 3D. Subcellular organelles, cells, biofilms and tissues, all classically pictured as flat entities in textbooks and other scientific literature, overwhelmingly display, when imaged in vivo, significantly different 3D features, including shapes, dimensions and surface-to-volume ratios. Importantly, the 3D spatial (x,y,z) and temporal (t) organization and morphology of the nucleus, mitochondria, cytoskeleton, and the endo-lysosomal and biosynthetic membrane trafficking pathways, as well as microenvironmental parameters (viscosity, pH and others), are key features for the establishment of higher-complexity cellular processes, including cell migration, signaling, metabolism and proliferation in non-disease and pathological tissues (Wu et al., 2018; Pampaloni et al., 2007; Liu et al., 2018a; Ovečka et al., 2018; von Erlach et al., 2018; Rios and Clevers, 2018; Libanje et al., 2019; Wang et al., 2020). 3D organization influences the spatial distribution of surface receptors and confers biophysical and/or mechanical features that affect responses to different stimuli. The physics of cell–cell and cell–matrix interactions are highly dependent on the three-dimensionality of cells and tissues (Page et al., 2013; Teodori et al., 2017; Laurent et al., 2017). Three-dimensionality is also critical for native tissue organization, the function of a stem cell niche, developmental processes, and for applications in regenerative medicine (tissue engineering) and tumor biology, as well as for drug delivery, pharmacological research and other areas of biomedical science. Thus, the need to visualize cells and tissues in 3D has driven research and innovation to further develop quantitative advanced imaging technologies. In particular, luminescence lifetime (FLI and PLIM) imaging provides a unique molecular window into cells and tissues, bringing us closer to achieving the overall goal of comprehensive quantitative 3D imaging of live biological samples at the nanometer scale (Fig. 1; Box 1). Advances in manufacturing exogenous phosphorescent and fluorescent dyes, proteins, bioconjugates and biosensors, as well as improved characterization of endogenous fluorescence features, have allowed for the luminescence-based sensing of a plethora of physiological processes, structural components and biomarkers, ranging from protein–protein interactions to tissue oxygenation (Table 1) (Berezin and Achilefu, 2010; Blacker and Duchen, 2016; Quaranta et al., 2012; Sarder et al., 2015; Steinegger et al., 2020; Datta et al., 2020; Papkovsky and Dmitriev, 2013, 2018b; Conway et al., 2017; Rajoria et al., 2014).
Generally, ‘3D imaging’ of biological materials (e.g. ex vivo or in vivo systems, tissue engineered scaffolds, organoids and small model organisms) is performed by stacking data acquired in a two-dimensional (2D) mode (x,y) using optical sectioning or widefield detection. In this type of applications, the contrast acquired is confined to the plane imaged, and the sample is scanned sequentially plane by plane to obtain the third dimension. 3D stacking and reconstruction can then be achieved computationally for subsequent 3D analysis (Datta et al., 2020; Le Marois and Suhling, 2017) (Fig. 2). These approaches are mainly associated with microscopy techniques that are used on transparent and/or thin samples, allowing for good light penetration depth (Cheng et al., 2020). However, when considering intact biological tissues, scattering becomes a prevalent phenomenon, precluding the ability to confine the detected photon to a localized 2D plane. Hence, the 3D nature of light propagation needs to be considered. 3D imaging can be achieved via solving a mathematical inverse problem that is centered around a 3D light propagation model to directly generate 3D volumes, such as in optical projection tomography (OPT) or FMT (Ozturk et al., 2016; Venugopal et al., 2012; Ntziachristos, 2010; Yoon et al., 2020) (Fig. 3). Although these approaches require an increased level of processing and expertise, and lead to reduced resolution images, they offer the unique potential to image large volumes in deep tissues, enabling the monitoring of biological processes over organs and/or whole organisms.
Biological and biomedical applications of FLI, FRET and PLIM
Imaging metabolism and cell death in cancer and stem cells
Optical metabolic imaging was introduced over four decades ago in the form of autofluorescence imaging of the cell redox status, firstly using fluorescence intensity measurements (Barlow and Chance, 1976) and more recently leveraging lifetime contrast imaging (Lakowicz et al., 1992). The field has seen a significant revival owing to the wide availability of multiphoton fluorescence lifetime microscopy (FLIM) (Masters et al., 1998; Ranawat et al., 2019; König, 2020). On the application side, the balance between cell energy production pathways (i.e. oxidative phosphorylation and glycolysis) has been shown to reflect cancer and stem cell function (Blacker and Duchen, 2016; Brand and Nicholls, 2011; Vander Heiden et al., 2009; Rodríguez-Colman et al., 2017). The vast majority of optical metabolic imaging studies employ two-photon FLIM of endogenous NAD(P)H, sometimes combined with the detection of FAD and other co-factors (Blacker and Duchen, 2016; Ghukasyan and Kao, 2009; Liu et al., 2018b; Ranawat et al., 2019; Skala et al., 2007; Stringari et al., 2012; Digman et al., 2008; Alam et al., 2017). NAD(P)H in its free form displays a short fluorescence lifetime, indicative of glycolysis, whereas longer emission lifetimes that result from the binding of NAD(P)H to enzymes are indicative of an involvement in oxidative phosphorylation or other metabolic processes (Blacker et al., 2014). Depending on the subcellular localization, cell type, interactions with cellular enzymes and proteins, or presence of metabolites in medium, the observed lifetimes of NAD(P)H can be discriminated and even calibrated, thus allowing the distinction between NADH and NADPH (Blacker et al., 2014; Leben et al., 2019). The main disadvantages of autofluorescence imaging are the inability to always measure the absolute amounts of redox co-factors (especially in objects with multiple cell types) and a lack of correlation with other markers of cell metabolism, such as glycolytic and oxygen consumption rates (assessed by measuring oxygenation using the PLIM method) and membrane polarization (assayed by staining with membrane potential-sensitive dyes) (Table 1) (Brand and Nicholls, 2011; Foster et al., 2006; Okkelman et al., 2020b; Perottoni et al., 2020 preprint). Nevertheless, autofluorescence FLIM can discriminate between different metabolic states and has been recently applied to T cell classification using the ‘random forest’ machine learning approach (Walsh et al., 2020). Because it is based on two-photon excitation, the method is intrinsically useful for in vitro analysis of 3D multicellular systems, such as spheroids, tumoroids and organoids, as well as in vivo imaging of transparent specimens or exposed tissues (Wetzker and Reinhardt, 2019; Datta et al., 2020; Gómez et al., 2018). Recently, following the pioneering work of Szulczewski et al. (2016), metabolic autofluorescence FLIM measurements have been used to discriminate the spatial and temporal heterogeneity related to macrophage polarization in co-culture with primary human invasive ductal carcinoma cells (Heaster et al., 2020). So far, most of these studies have focused on the 2D analysis of 3D objects, which may result in inaccuracies owing to the rather arbitrary choice of a single (x,y) ‘equatorial’ plane in asymmetric tumor organoids or the analysis of averaged z projections.
Novel fluorescent and phosphorescent biosensor probes have been used in ‘multiparametric’ applications of autofluorescence-based redox imaging assays (Kalinina et al., 2016; Okkelman et al., 2020b). Thus, mitochondrial function and cell bioenergetics can be analyzed with complementary FLIM and PLIM approaches using genetically encoded NADH/NAD+ redox biosensors (Hung et al., 2011), glucose biosensors (Díaz-García et al., 2019), oxygen-sensing phosphors (Dmitriev et al., 2014, 2013; Okkelman et al., 2017a,b), pH-sensing proteins and mitochondrial membrane potential-sensitive probes (Okkelman et al., 2019). FLIM in 3D has been used to measure cellular NAD(P)H within live intestinal organoids expressing the stem cell marker Lgr5–GFP, demonstrating functional differences in redox status between oxidative phosphorylation and glycolysis in stem cells and non-stem cell types, and emphasizing the need to combine FLIM measurements of NAD(P)H with O2-sensing PLIM for the analysis of cell metabolism (Okkelman et al., 2020b).
Additional biomarkers and physical parameters that are important for cancer and stem cell biology (Nia et al., 2020), such as temperature (e.g. tumor-induced thermogenesis), cellular viscosity, polarity and tension, can be sensed using FLIM and various FRET biosensors. Temperature control is essential for cell viability, and growing evidence highlights the importance of the temperature gradients that are experienced by metabolically active biological tissues (Chrétien et al., 2018; Ogle et al., 2020; Zhou et al., 2020). Importantly, in breast and colon cancer cells, temperature gradients have been linked to the function of the uncoupling protein UCP1, mitochondrial function and cell oxygenation (Jenkins et al., 2016; Kawashima et al., 2020). Molecular tension FLIM has been used to map the cellular adhesive landscape, with biosensor probes revealing the tension forces experienced by cells (Glazier et al., 2019; Dumas et al., 2019; Ringer et al., 2017).
Cell death and proliferation are important hallmarks of cancer and developmental programs, and their detection is thus highly relevant. Notably, live-cell FLIM has been used to monitor cell proliferation via quenching of 5-bromo-2′-deoxyuridine (BrdU; Fig. 2B) and the co-expression of tagged cyclin kinase inhibitor nuclear proteins (Okkelman et al., 2016; Schreiber et al., 2012). DNA damage can be measured by FLIM-based visualization of the double-strand breaks (Lou et al., 2019), and activation of apoptosis in tumor spheroids has been assessed using FRET-based expressed caspase activity biosensors or by combining O2-sensing PLIM with fluorescent necrosis and dyes that report caspase-3 and/or caspase-7 activation (Dmitriev et al., 2015b). Interestingly, NADH FLIM has been used to discriminate between apoptosis and necrosis events (Wang et al., 2008).
Protein-based FLIM–FRET biosensors for intravital imaging of the therapeutic response
The area of intravital imaging using FLIM–FRET has been pioneered in the past two decades, spurred by advances in intramolecular fluorescent protein-based FRET biosensors, which are designed to change their 3D conformation in response to the targeted biological process (e.g. phosphorylation or interaction with other biomolecules) (Fig. 1B; Box 1). Altering the distance between the donor and acceptor fluorescent proteins in these biosensors can result in reversible changes to their donor fluorescence lifetime, hence directly reporting on the affected biological process (Table 1). FLIM–FRET has been increasingly employed in vivo using intravital imaging, owing to progress in the development of transgenic animals and imaging instrumentation (Nobis et al., 2013; Timpson et al., 2011a; Ellenbroek and van Rheenen, 2014). For example, Timpson and co-workers used in vivo FLIM–FRET to study cancer cell invasion in tumor xenografts expressing a RhoA GTPase biosensor, via implanted optical imaging windows (Timpson et al., 2011b). Currently available intramolecular FLIM–FRET biosensors enable the transient detection of different substrates (e.g. peptides, phosphorylated proteins, Ca2+-binding proteins and posttranslational modifications) by exploiting binding domains, cleavage sites, binding tags or other ligands (Conway et al., 2018, 2017; McGhee et al., 2011). For instance, FLIM–FRET biosensors have been used in 3D cancer models to analyze activation of apoptosis (Weber et al., 2015) and to visualize AMP-activated protein kinase (AMPK) gradients in response to drug stimulation (Chennell et al., 2016). FLIM–FRET redox biosensors represent a powerful alternative to autofluorescence-based metabolic imaging (Yellen and Mongeon, 2015). FLIM–FRET has enabled the spatial monitoring of the developing kidney using deep-tissue multiplexed 3D imaging of Ca2+ and cAMP in zebrafish embryos (Zhao et al., 2015). To advance the study of amyloid protein function in Alzheimer's disease, protein-based biosensors have been applied to monitor the development of aggresomes using FLIM and 3D structural illumination microscopy (Lu et al., 2019). As an alternative to biosensors, FLIM has been used to monitor both the kinetics of doxorubicin uptake via its quenching effect on histone H2B–eGFP (Bakker et al., 2012) and the release of doxorubicin from iron oxide nanoparticles in tumor spheroids (Basuki et al., 2013). A current limitation of FLIM is that, with some exceptions (pH-, O2-, temperature- and Ca2+-imaging), most of the research described above is based on semi-quantitative relative measurements of fluorescence lifetime. This is due to cell- and environment-specific effects on luminescence lifetime, which often preclude a precise calibration system that can be transferable to different subcellular compartments and cell types.
Intravital imaging of oxygenation, metabolism and Ca2+ concentration
Measuring O2 can be performed using two different imaging approaches: (1) ‘bottom-up’ one-photon imaging using dye- and nanosensor-based probes designed to image intracellular gradients; and (2) ‘top-to-bottom’ two-photon imaging using cell-impermeable dendrimer-protected porphyrin conjugates to image oxygenation in the bloodstream (Table 1). Intravital PLIM can be performed with both these two types of probes (Papkovsky and Dmitriev, 2018a,b), in addition to imaging of tissue-engineered constructs and studies of cellular O2 gradients (Dmitriev et al., 2012; Finikova et al., 2008; Mik et al., 2006; Yoshihara et al., 2017; Zhou et al., 2016; Napp et al., 2011; Mizukami et al., 2020). Two-photon imaging of vascular oxygenation (Sakadžić et al., 2010) has enabled 3D intravital PLIM combined with NADH and blood flow measurements (Devor et al., 2012; Gómez et al., 2018; Esipova et al., 2019; Finikova et al., 2008; Sakadžić et al., 2010). Successful design of vascular dendrimeric probes has also helped to map bone marrow oxygenation and to study O2-dependence of hematopoietic stem cell (HSC) migration (Christodoulou et al., 2020). The use of O2-sensing nanoparticles in vivo is still rare due to their suboptimal tissue distribution and quick clearance from the bloodstream (Dmitriev et al., 2015a). Nevertheless, a recent study successfully performed intravital time-lapse multiplexed FLIM–FRET using the AKT kinase activity of the ‘Eevee-Akt-mT2’ biosensor and O2-sensing PLIM to monitor hypoxic regions that are resistant to therapeutical targeting of the phosphoinositide 3-kinase (PI3K) pathway in pancreatic cancer (Conway et al., 2018). Similarly, modified two-photon excited PLIM (FaST-PLIM) has been used for intravital imaging of lymphocyte mobility in solid and hematological tumors in mice (Rytelewski et al., 2019).
Additional parameters that provide information on the function of excitable cells can also be sensed using FLIM. For example, FLIM can be used to monitor intracellular Ca2+ concentration via genetically encoded Ca2+ indicator (GECI) biosensors (Looger and Griesbeck, 2012; Tian et al., 2012) and small-molecule probes such as Ca2+-chelating Oregon Green BAPTA-1 (Kuchibhotla et al., 2009; Lattarulo et al., 2011) (Table 1). Traditionally, FLIM has been considered to be too slow for capturing fast Ca2+ changes. However, a parallelized time-correlated single-photon counting (TCSPC) two-photon imaging platform, capable of acquiring an image every 82 ms at 115 µm depth, has been used to monitor Ca2+ dynamics by time-lapse 3D FLIM (i.e. 4D FLIM) of a Cerulean–Citrine–troponin-C-based intramolecular FRET biosensor (CerTN L15) expressed in the brain stem of transgenic mice (Radbruch et al., 2015; Rakymzhan et al., 2017; Rinnenthal et al., 2013) (Fig. 2C).
Tissue engineering and organoids
FLI and PLIM in tissue engineering frequently deal with ‘top-to-bottom’ engineering approaches, such as organizing fluidic flow (tissues and organ-on-a-chip) (Perottoni et al., 2021; Zirath et al., 2018; Ahmed et al., 2019), scaffold materials (Appel et al., 2013; Sud et al., 2006; Neto et al., 2020) and, more recently, direct bioprinting of biologics (Ozturk et al., 2016, 2020; Kingsley et al., 2019). Autofluorescence FLIM has been used as a readout parameter to monitor the successful biofabrication of dermal matrix engineered scaffolds (Formigli et al., 2012) and scaffold composition during bioengineered cartilage growth (Fite et al., 2011). O2-sensitive ‘biosensor scaffolds’ (i.e. tissue engineering scaffolds containing phosphorescent O2-sensing dye) have been employed for the imaging of brain slices, scaffold-colonizing cancer cell aggregates and bone regeneration constructs (Jenkins et al., 2015; Schilling et al., 2019; Trampe et al., 2018; Xue et al., 2013, 2016; Yazgan et al., 2017). Moreover, multiplexed 3D imaging, combining O2 imaging with live-cell labeling, has enabled the tracing of hypoxia-dependent cell differentiation and anticancer drug action (Jenkins et al., 2015). A recent proof-of-concept study has demonstrated that O2-nanosensor-labeled scaffolds can be implanted into mice for long-term monitoring of O2 levels during regeneration and scaffold degradation (Elagin et al., 2020). In addition, such a biosensor scaffold approach allows the sensing of pH, Ca2+ and other parameters using 3D FLIM (O'Donnell et al., 2018; Okkelman et al., 2020a).
Organoids and other advanced ‘engineered tissue’ 3D models permit real-time, quantitative monitoring of genome organization (Lu et al., 2008; Hoppe et al., 2008; Wollman et al., 2020), self-organization of cellular processes (Lu et al., 2008; Hoppe et al., 2008; Wollman et al., 2020), metabolism, intracellular trafficking (Skruzny et al., 2020) and organelle–organelle interactions, as well as cancer cell migration, invasion and metastasis (Alexander et al., 2013; Steuwe et al., 2020), enabling an integrated understanding of the molecular biology of the cell in health and disease. Furthermore, metabolic FLIM and PLIM imaging are valuable tools in developmental biology and research into assisted reproductive technologies (Ma et al., 2019; Sosnik et al., 2016; Stringari et al., 2011, 2012), and hold promise for improving tissue engineering, recreating stem cell niches and tumor microenvironments and developing advanced tissues-on-a-chip systems (Okkelman et al., 2020b; Ozturk et al., 2020; Floudas et al., 2020; Barron et al., 2020).
Contrast enhancement for high-performance 3D imaging and super-resolution microscopy
Luminophores can be distinguished not only by spectra but also by their lifetime values. Thus, multiplexing is a clear advantage of luminescence lifetime imaging, effectively increasing the number of dyes and labels that can be sensed simultaneously in one sample (Fig. 2D). Because the same luminophore can display changes in lifetime via interaction with its immediate environment within the cell, an additional ‘contrast’ enhancement feature becomes available. In FLI-based approaches, lifetime signatures can be successfully discriminated from unwanted spectrally overlapping autofluorescence signals (Rudkouskaya et al., 2020b; del Rosal and Benayas, 2018; Okkelman et al., 2017b, 2020c; Rich et al., 2013). FLIM as a contrast enhancement methodology has become more popular with the advent of super-resolution microscopy (Auksorius et al., 2008; Bückers et al., 2011; Hell and Wichmann, 1994; Niehörster et al., 2016; Masullo et al., 2020; Thiele et al., 2020), providing a new way to multiplex fluorescence imaging. 3D FLIM for contrast enhancement has been demonstrated in proof-of-concept light-sheet microscopy with optically cleared and fixed samples (Hirvonen et al., 2020; Greger et al., 2011; Favreau et al., 2020; Funane et al., 2018; Mitchell et al., 2017; Weber et al., 2015; Li et al., 2020; Birch et al., 2016).
Imaging at the organ or organism level – meso- and macro-scopic luminescence lifetime imaging
FLI can be integrated into a wide range of imaging modalities while retaining the same fundamental physical principles and biological readout information across scales. This is particularly important in the imaging of larger volumes, including tissues, organs and tumors, or in whole-body imaging of small organisms. Here, we will examine FLI-based deep-tissue imaging with mesoscopic (depth of penetration of a few millimeters) and macroscopic (depth of penetration of a few centimeters) optical imaging platforms. In contrast to microscopy, macroscopy involves the visualization of entire volumes (∼mm3–cm3) with main applications in whole-body preclinical imaging. Macroscopy can be performed with 2D (reflectance) or 3D (tomography) imaging approaches, both in vivo and ex vivo (Sinsuebphon et al., 2018; McConnell and Amos, 2018; McConnell et al., 2016; Mizuno et al., 2021). Mesoscopic approaches refer to the scale between microscopic and macroscopic, and include imaging platforms that can visualize whole, intact samples in a size range of 50–100 µm to 2–3 mm. Current mesoscopic imaging applications focus on tissue engineering and/or subsurface preclinical imaging (below the intact skin and/or skull).
Confocal-based macro-FLI implementations collect point-by-point FLI data sets that are processed following the same approach as used in FLIM (Bloch et al., 2005). Macro-FLI confocal scanning platforms provide a larger field of view (of a few cm2, compared to a few hundred μm2 in microscopy) with near-cellular resolution FLIM images for the 2D analysis of, for instance, tumor metabolic heterogeneity and fluorescence FRET biosensors (Rinnenthal et al., 2013; Shcheslavskiy et al., 2018; Suhling et al., 2019; Zherdeva et al., 2018) (Fig. 3A). Because they do not account for the effect of light propagation on fluorescence decay, macro-FLI imagers are well suited for imaging of superficial tissues and/or transparent systems, as in both cases scattering is minimal. Macro-FLI imagers have also been used for ex vivo whole-organ imaging (Kantelhardt et al., 2016; Papour et al., 2013) and for imaging of the skin in studies of melanoma (Seidenari et al., 2013), pigmentation (Dancik et al., 2013) and wound healing (Rico-Jimenez et al., 2020), as well as to assess nanoparticulate zinc oxide sunscreen safety (Mohammed et al., 2020) and living dermal equivalents (Meleshina et al., 2018).
FLI-based sensing can be combined with other imaging methodological approaches, enabling improved tissue and organ visualization and quantitative analysis. For example, FLI has been integrated with optical coherence tomography (OCT) (Singh et al., 2016), a well-characterized non-invasive imaging approach that provides spatial resolution in the micrometer range by using low-coherence interferometry. Autofluorescence FLI has been merged with high-resolution OCT images to visualize tissue biochemistry and morphology of coronary artery sections (Sherlock et al., 2017; Shrestha et al., 2016) (Fig. 3B). OCT–FLIM has also been used in cancer diagnosis (Lee et al., 2018; Nam et al., 2018; Pande et al., 2016), as well as in the characterization of atherosclerotic plaques (Lee et al., 2018; Nam et al., 2018; Pande et al., 2016). In addition, FLI imaging has been enhanced through integration with OPT (Andrews et al., 2016; McGinty et al., 2011) (Fig. 3C), a form of tomography involving optical microscopy (Sharpe et al., 2002), ultrasound (Bec et al., 2014) or spatial frequency domain imaging (Smith et al., 2020a), a widefield diffuse optical imaging technique that can be used to obtain absorption and reduced scattering coefficients of turbid media (Gioux et al., 2019). Whereas these applications focus on shallow tissues and rely on the same principle as FLIM for image formation, for the imaging of deeper tissues, more complex mathematical formulations that take into account the optical tissue properties are required.
To perform imaging of deeper tissues, methodologies such as minimally invasive endoscopy or FMT need to be employed (Box 1). Deep-tissue imaging also requires the use of near-infrared (NIR) dyes, as tissue attenuation is reduced in the NIR optical window and autofluorescence is drastically decreased (Staudinger and Borisov, 2015). Moreover, macroscopic deeper imaging approaches are suitable for non-invasive imaging, allowing for imaging across the intact skin, which maintains the undisrupted native physiology and environment of organs and tissues, albeit at the cost of resolution. The flexibility of FLI has allowed its use in endoscopy for multiple in vivo applications via a fiber-optic bundle (Duran-Sierra et al., 2020; Hage et al., 2019; Jo et al., 2018; Won et al., 2016). FMT is a non-invasive and non-ionizing optical imaging modality that retrieves the biomarkers of interest in 3D volumes. It is based on acquiring optical data from spatially resolved surface measurements and performing mathematical computational tasks that involve the modeling of light propagation according to tissue scattering properties. Briefly, a forward model of light propagation is computed for every spatially resolved measurement acquired from the imaging system (Arridge, 1999) and then compared iteratively with the experimental data to yield 3D digitized maps of intensity contrasts (Chen et al., 2012). This mathematical task is referred to as an inverse problem and it is at the basis of image formation in numerous 3D imaging modalities, including computed tomography (CT) and nuclear imaging. In the case of FMT, the goal is to determine the 3D biodistribution of NIR dyes in deep organs, such as the liver, lungs and brain, or orthotopic tumor xenografts. Different tomographic imaging approaches have used FLI contrast to visualize and quantify gene function and protein expression in a variety of organs by using whole-body non-invasive imaging of genetically encoded NIR-fluorescent proteins or NIR-labeled nanoparticles (Cai et al., 2016; Kumar et al., 2018; Rice et al., 2015), as well as NIR-labeled antibody or ligand probes (Cai et al., 2016; Kumar et al., 2018; Nothdurft et al., 2009; Rice et al., 2015) (Fig. 3D). FLI–FRET tomographic reconstructions of deep-seated organs can provide nanoscale sensing in deep tissue, in the millimeter range, enabling the imaging of whole organs or even small animals (Box 1). To that end, volumetric imaging of genetically expressed fluorescent proteins (McGinty et al., 2011) or NIR-labeled probes (Venugopal et al., 2012) has already been obtained using FLI-based tomographic reconstruction of FRET measurements (Fig. 3E,F). One very promising application for these emerging FLI–FRET whole-body imaging systems is the monitoring of the intracellular delivery of targeted probes in live animals by quantifying the ligand–receptor efficiency across organs. For example, FMT has been used in conjunction with structured light illumination to achieve quantitative 3D FRET tomographic non-invasive in vivo macroscopic FLI (MFLI) imaging of subcutaneous tumor xenografts to measure drug delivery (Rudkouskaya et al., 2018, 2020b). Subsequently, in vivo multiplexed MFLI–FRET imaging of receptor–ligand target engagement and glucose metabolism in tumor xenografts in live intact animals has been demonstrated (Rudkouskaya et al., 2020a). However, tomographic image reconstructions are computationally expensive and require expertise in light propagation modeling, limiting their use in biomedical applications (Leblond et al., 2010). Hence, there is a growing interest in developing fit-free FLI analysis, such as phasor (Chen et al., 2019; Ma et al., 2019) and deep learning (Smith et al., 2019; Ochoa et al., 2020) methodologies. Considering the plethora of FLIM-based approaches for imaging 2D oncologic samples, and their many applications (Ardeshirpour et al., 2018; Basuki et al., 2013; Mehidine et al., 2019; Pal et al., 2019; Sparks et al., 2018; Weitsman et al., 2017), further advancing the ability to perform macroscopic FLI and tomographic reconstructions of deep-seated tumors in their native and undisrupted environment has the potential to accelerate cancer drug screening and efficacy assessment, as well as tumor biology research.
FLIM for live subcellular imaging in 3D
Advances in structured plane illumination and super-resolution microscopy have increased the appreciation for the complexity of subcellular organelle morphology and dynamics in 3D (Liu et al., 2018a). Thus, by enabling a functional in situ and live readout from these complex structures, FLIM is particularly suited to the study of chromatin compaction, subnuclear organization (Angelier et al., 2005; Wollman et al., 2020; Cremazy et al., 2005; Rino and Carmo-Fonseca, 2009), cell membrane-anchored processes, organelle biogenesis and other subcellular processes. For instance, FLIM–FRET has been used to monitor the heterochromatin organization in live Caenorhabditis elegans (Llères et al., 2017), and 3D FLIM was employed to study chromatin compaction in response to DNA damage using histone H2B–fluorescent protein fusion probes (Sherrard et al., 2018). In addition, a phasor-based approach to FLIM–FRET analysis of chromatin compaction has been described (Lou et al., 2019). DNA-staining dyes and fluorescent protein reporters have also been used for FLIM analysis of events in the nucleus (Abdollahi et al., 2018; Audugé et al., 2019; Estandarte et al., 2016; Kawanabe et al., 2015; Okkelman et al., 2016; Sparks et al., 2018).
The interactions of luminescent dyes with biomolecules, resulting in changes in their aggregation state, molecular rotation and lifetime values, have been extensively employed to study membrane composition by FLIM using probes, such as C6-NBD-PC and Laurdan, and viscosity-sensitive fluorescent molecular rotors (Kuimova, 2012; Levitt et al., 2015; Ringer et al., 2017; Stöckl et al., 2012). Use of these probes has enabled a deeper understanding of cell cycle-dependent changes in membrane composition (Denz et al., 2017) and of lipid bilayer organization with regard to rafts and microdomains (Carquin et al., 2016; Ma et al., 2018; Malacrida and Gratton, 2018). Recently, FLIM–FRET has been used to provide mechanistic insight into lipid and protein interaction sites between the endoplasmic reticulum and the trans-Golgi network (Venditti et al., 2019). Furthermore, 3D FLIM–FRET has allowed for the analysis of viscosity in tumor spheroids (Shirmanova et al., 2017) and lipid droplets in intestinal organoids (Okkelman et al., 2020a).
Emerging developments and application areas
The main current bottlenecks of FLI are associated with software optimization, standardization of functional assays and overall cost. For example, one of the classical limitations of FLIM is in the ability to quantify fluorescence lifetimes in low photon budget conditions, such as those encountered in biomedical microscopy applications. This has been typically addressed by the use of iterative algorithms that necessitate user inputs and can be computationally expensive. Recent innovations in phasor plot analysis (Zhang et al., 2019; Vallmitjana et al., 2020) represent attractive alternatives given their speed, user friendliness and potential for standardization. To highlight this, we provide below two examples of emerging 3D FLI developments.
Machine learning for FLI
To date, leveraging of new developments in artificial intelligence has mainly been performed for classification and segmentation of raw FLI data (Mannam et al., 2020) using classical supervised machine learning algorithmic techniques, such as support vector machines (SVM) (Zhang et al., 2019), k-means clustering (Brodwolf et al., 2020; Zhang et al., 2019), or ‘random forest’ machine learning algorithms (Walsh et al., 2020). However, deep learning, another class of artificial intelligence algorithms, is well known to outperform such ‘shallow learning’ classifiers by, for example, forming the FLI image directly from raw data, and deep learning methods are expected to become the norm for such image-processing tasks. Moreover, deep learning approaches are well suited to tackle the mathematical inverse problem associated with lifetime quantification. This has been demonstrated in a recent study that proposed a deep learning model, named FLI-Net, that is user-input free, processes whole images almost in real time and, of great importance to biological applications, is more accurate in low photon budget conditions than current ubiquitous techniques (Smith et al., 2019). Moreover, deep learning methodologies have been used to enhance unmixing approaches for increased specificity (Smith et al., 2020b). Lastly, they can facilitate the implementation of the next generation of high-end FLI instruments, especially hyperspectral macroscopic FLI systems (Pian et al., 2017; Yao et al., 2019; Ochoa et al., 2020). Hence, deep learning methods are expected to play a crucial role in the increased utilization of 3D, spectral and time-lapse reconstruction of live multiparameter sets of FLI, FLIM and PLIM data in biomedical imaging applications.
FLI for viral and pathogen research
Viral and pathogen infection of mammalian cells result in changes to membrane composition. In addition, virus replication and capsid assembly are highly demanding for cellular bioenergetics and metabolism. Therefore, viral research is an attractive application for FLI, FLIM and PLIM approaches. Unsurprisingly, FLIM has been extensively used to characterize human immunodeficiency virus 1 (HIV-1) infection, including capsid assembly and viral particle trafficking to the plasma membrane, as well as its dependence on cell metabolism and cholesterol content (Jones and Padilla-Parra, 2015; Coomer et al., 2020; de Rocquigny et al., 2014; El Meshri et al., 2015; Greiner et al., 2011). Other FLIM studies have investigated the assembly of influenza virus capsids (Thaa et al., 2010), adenovirus type-5 entry and disassembly (Martin-Fernandez et al., 2004), enterovirus 71 entry into cells (Ghukasyan et al., 2007), the effects of enterohemorrhagic Escherichia coli infection (Buryakina et al., 2012), and leaf chlorosis due to plant virus infection (Lei et al., 2017). These studies have mostly made use of single-cell FLIM imaging approaches (Witte et al., 2018). However, they could clearly benefit from being combined with more physiologically relevant 3D tissue organoid models. Moreover, the use of highly efficient tools to track viruses in live and physiological settings (Clevers, 2020) would improve and further streamline antiviral research and the development of therapeutics. Thus, FLI and PLIM methodologies are not restricted to animal cell and tissue imaging and can be used more widely, beyond biomedical applications, to include imaging-based research of other classes and kingdoms of life (Box 2).
The fields of aquatic and plant biology research, among others, also benefit from 3D FLIM (Moßhammer et al., 2019). For instance, O2 imaging using various nanoparticles is useful for the study of corals and plant-based inter-species systems such as those that exist in the rhizosphere (Fabricius-Dyg et al., 2012; Brodersen et al., 2020). Similar to NAD(P)H sensing in biomedical imaging, chlorophyll autofluorescence FLIM can be used to analyse the metabolic function of coral symbionts. Unfortunately, only a few research studies have attempted to investigate this (Cox et al., 2007; Petrášek et al., 2009; Shapiro et al., 2016).
Although the leaves of terrestrial plants and their photosynthetic pigments were among the first biological systems to be imaged using autofluorescence FLIM (Donaldson, 2020; García-Plazaola et al., 2015; Holub et al., 2000; Oxborough and Baker, 1997), the technique is rarely used in plant sciences and photosynthesis research (Cisek et al., 2009; Emiliani et al., 2003; Iermak et al., 2016). Plant seeds (including barley embryos) have been studied using autofluorescence imaging, with two-photon FLIM multiplexed with magnetic resonance imaging (Stark et al., 2007) being applied to distinguish between functionally different parts of the seed, specifically the seed coat, embryo and starch. FLIM–FRET has also been used to monitor protein interactions using overexpressed biosensors in another kingdom of life, the fungi (Altenbach et al., 2009).
Bacterial biofilms, which play highly important barrier and transport functions in microbial communities, represent another interesting application for 3D FLIM (Schlafer and Meyer, 2017). Biofilms often grow as 3D structures, and understanding their function is of high importance for basic biology, environmental science and biomedicine, for example in host–microbe interactions and in dental and human skin microbiome research (Brandwein et al., 2016; Hatzenpichler et al., 2020; Soll and Daniels, 2016). Until now, FLIM of biofilms has been focused on introducing staining dyes to map extracellular pH and FRET-based measurements (Schlafer and Meyer, 2017), as well as labeling and visualizing DNA and RNA using two-photon FLIM (Neu et al., 2004). In situ FLIM imaging using the dye SYTO 13 to measure DNA:RNA ratio has allowed the visualization of different bacterial communities within the same biofilm (Walczysko et al., 2008). Among the important components of the biofilm matrixome, the dynamics of O2 has been assessed using spinning-disk FLIM (Kühl et al., 2007).
Luminescence lifetime imaging, as it becomes more available to researchers of varying expertise, enables a redefinition of classic fluorescence imaging, thus opening a new avenue for the use of dyes and biosensor probes (making virtually any dye a biosensor) and producing a new, ‘molecular’ dimension in a wide variety of imaging platforms. Although it can easily provide additional benefits to imaging modalities that utilize dead, fixed or optically cleared tissues (for example by improving quality, resolution and the effective number of used fluorophores in super-resolution microscopy), its main power and potential to be harnessed in the coming decade lies in advanced live imaging of complex 3D objects, including studying virus infection in engineered ‘niches’, tissues, tissue organoids and small organisms.
We would like to thank the members of the Barroso, Intes and Dmitriev laboratories for comments and discussions that are at the basis of this Review. In particular, we would like to thank Dr Jonathan Barra for his careful and detailed review of the manuscript.
Our research in this area is funded by grants from the National Institutes of Health (R01 CA250636, R01 CA207725 and R01 CA237267 to M.M.B. and X.I.) and Universiteit Gent (BOF/STA/202009/003 to R.I.D.). Deposited in PMC for release after 12 months.
The authors declare no competing or financial interests.