Altered substrate metabolism in neurodegenerative disease: new insights from metabolic imaging

Despite being only 2% of the body’s mass, the brain consumes 20% of the body’s oxygen intake and 25% of its glucose intake [1]. This energy utilization is not uniform, with 70–80% of total energy consumed via neurons, and microglia, astrocytes and oligodendrocytes using the remaining portion [2‐4]. Although the brain primarily consumes glucose to meet its energy demands, in disease states or starvation energy utilization can shift to other substrates [5‐9]. During aging, glucose metabolism in the brain is impaired, a phenomenon that is exacerbated in neurodegenerative (ND) and neuroinflammatory diseases (NID) such as Alzheimer’s disease (AD), Parkinson’s disease (PD) and multiple sclerosis (MS). Specifically in AD, several changes to overall brain metabolism have been observed, including dysregulated oxidative metabolism, amino acid metabolism and creatine degradation [10], and impaired glucose transport [11]. In particular, the pathogenesis of AD is complex, and despite major research efforts, successful therapeutics that can prevent, or delay AD are lacking. Since perturbed brain metabolism has been repeatedly implicated in AD, understanding changes in substrate utilization prior to and during the development and progression of AD may highlight critical mechanisms that can be exploited to diagnose or treat AD.

Despite changes in brain metabolism being a potential mechanism driving the development of NDs, the field has been stymied by the lack of techniques that can accurately measure brain metabolism with sufficient temporal, spatial, and cell-specific resolution. On one hand, it is becoming widely appreciated that both neurons and glial cells, which have very different metabolic needs, contribute to disease pathology. Recent methodological advances have facilitated the identification of microglial subpopulations with defined metabolic phenotypes that may protect against or precede AD [12, 13]. However, these measurements have been made in isolated cells that have been subjected to mechanical or enzymatic stress, which likely modifies their metabolic state, underscoring the need for methodologies that can determine cell-specific metabolic changes in situ.

Specific brain regions have different cellular composition, functions, and therefore metabolic needs. Hence metabolic perturbations in certain regions may have a greater impact on disease state and severity. Nonetheless, existing “metabolic imaging” modalities, such as magnetic resonance imaging (MRI), magnetic resonance spectroscopy (MRS), and positron emission tomography (PET), have provided important insights regarding regional changes in metabolic processing that occur preceding and during NDs, such as AD [14]. Although these techniques are becoming increasingly advanced, they cannot discriminate between regional and cell-specific metabolic aberrations. In contrast, the development of two-photon imaging and fluorescence lifetime imaging microscopy (FLIM), enables direct measurement of the endogenous metabolic fluorophores, such as reduced nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide (FAD), which have very distinct lifetimes in free and bound-to-enzyme forms [15]. The ratio of free and bound-to-enzyme NADH or FAD can be determined within an individual cell or on a cell cluster in tissue, and therefore has the potential to measure cell and region-specific changes in brain metabolism that may precede or drive AD onset and progression. Here, we revisit our understanding of brain metabolism and review recent findings suggesting that changes in brain metabolism may be a potential target for the treatment of ND. We also highlight recent advances in metabolic imaging that enable precise monitoring of brain metabolism, guiding the design of interventions that may prevent or improve outcomes for individuals with NDs such as AD.

Neurodegenerative diseases affect millions of people worldwide and can be devastating for the afflicted and their families [63]. Of these diseases AD is the most common [2, 64], and will become more prevalent as the world’s population gets older [65]. Importantly AD, PD and MS share metabolic characteristics such as neuroinflammation and mitochondrial dysfunction. For example, during neuropathogenesis metabolic homeostasis becomes lost, which is associated with glial activation, neuronal damage, and cell death [66, 67]. Existing treatments for these diseases are not curative and at best slow the progression of the disease, highlighting the need to further understand metabolic aberrations underlying disease onset to develop new metabolism-focused intervention strategies with the potential to reverse or halt disease progression. Here, we will review our current understanding of the metabolic changes involved in AD, PD and MS.

It is clear by now that neurodegenerative diseases are accompanied by metabolic alterations in one or more cell types of the brain. Therefore, examining these alterations is a crucial step in understanding the etiology and progression of these diseases. There are many ways to study metabolism in cells and tissues such as mass spectrometry, activity and colorimetric assays, seahorse XF analyzers, and imaging [128]. Specifically, metabolic imaging is advantageous in a sense that it provides the spatial and temporal information that other techniques lack. Such spatiotemporal measurements are essential to resolve the metabolic changes in the different regions of the brain, as well as at different time points for longitudinal studies of disease progression.

Metabolic imaging modalities can be primarily divided into either ionizing radiation methods or non-ionizing radiation methods. Ionizing radiation methods include nuclear imaging techniques such as positron emission tomography (PET) and single-photon emission computed tomography (SPECT). Whereas non-ionizing radiation methods include magnetic resonance imaging (MRI) techniques such as magnetic resonance spectroscopy (MRS), hyperpolarized magnetic resonance, chemical exchange saturation transfer (CEST), optoacoustic (photoacoustic) imaging, as well as fluorescence microscopy imaging methods [129]. Coherent Raman scattering (CRS) microscopy [130] and mass spectrometry imaging (MSI) [131, 132] are other non-ionizing radiation methods which have been used to image metabolism in cells and tissues; although the latter does involve desorption/ionization of analytes by means of absorbing short ultraviolet (UV) wavelengths that are non-ionizing per se. To be clinically relevant, metabolic imaging techniques need to be noninvasive or minimally invasive (i.e., no biopsies and tissue sampling are required), and preferably without the use of exogenous tracers. Moreover, they need to provide enough resolution to differentiate between different cell types at least, if not at a subcellular level.

In vivo changes in brain metabolism have been assessed using the following noninvasive or minimally invasive neuroimaging techniques: MRI, functional magnetic resonance imaging (fMRI), MRS, diffusion weighted imaging (DWI), diffusion tensor imaging (DTI), PET and SPECT [133], in addition to functional photoacoustic microscopy (PAM) [134] (a focused review on advanced MRI and MRS methods used in brain metabolic imaging can be found in [135]). On the other hand, postmortem ex vivo examination of these changes has been generally performed using microscopy techniques [133] such as immunohistochemistry (IHC) [136], fluorescence microscopy with a variety of biosensors [137] (although such biosensors have also been used in in vivo brain metabolic imaging [138]), stimulated Raman scattering (SRS) [139]—which is a CRS technique, MSI techniques [140‐142], and more recently fluorescence lifetime imaging microscopy (FLIM) [143‐145] (though FLIM has also been used in in vivo brain metabolic imaging [53]). Autoradiography is another technique that has been used to observe changes in brain cell receptor expression or architecture [146‐148]. Here, a radioactive tracer is injected into the sample prior to brain isolation and exposure of a radiation sensitive imager plate [149]. This technique provides good data on the regional localization of the target of a given radioactive tracer. Depending on the tracer, inferences can then be made on changes in metabolism. This technique has been applied to analyze changes in PD, AD, and TBI [146‐149]. This technique is particularly useful to determine drug or ligand occupancy of specific-receptors and is therefore a gold-standard technique in pharmacological studies evaluating novel drugs, and their targets [150, 151]. However, the use of radiation, and the ex vivo nature of this approach is not always desirable.

Additionally, brain imaging methods can be broadly divided into three categories based on the information they provide: structural imaging, functional imaging, and molecular imaging. Structural imaging modalities include techniques such as computed tomography (CT) and MRI. CT images are created by exposing a patient to a rotating beam of X-ray radiation, whose signal is then collected and reconstructed to form a 3D image [152]. This can be used in conjunction with intravenously administered contrast agents, e.g., CT angiography, to allow for better imaging of vasculature [152, 153]. In the clinic, CT is useful for its quick acquisition time and is often used to rule out acute pathology such as hemorrhage, ischemia, or lesions. However, magnetic resonance images can provide better structural resolution especially when it comes to ND-associated structural changes. These images are created by exciting hydrogen ions and detecting energy release to construct an image [153]. Variations of MRI such as structural MRI and DTI have also been developed to study brain volume and tissue characteristics, and the integrity of white matter tracts [154].

Many MRI-based modalities have been developed to provide functional and molecular information. The most prevailing functional neuroimaging modality is blood oxygen level-dependent (BOLD) fMRI. This method relies on differences in the magnetic resonance signal created by changes in blood flow to brain regions, and differences in the magnetic properties of oxygen-rich and oxygen-poor blood. This provides an indirect measure of neural activity by associating metabolic activity with hemodynamic parameters that imply increased delivery or availability of nutrients to a specific brain region [155]. BOLD fMRI can be used for mapping brain networks associated with tasks or stimuli [156]. A similar technique is arterial spin labeling (ASL) wherein arterial blood is magnetically labeled and acts as a tracer to then measure differences in cerebral blood flow [157]. Rather than directly measure metabolism, these techniques provide good information on cerebral blood flow that can then be used to make inferences on regional cerebral metabolism.

PET is a well-established method of imaging that can provide metabolic information; the metabolic activity of brain regions can be imaged and analyzed using radioactive forms of glucose, dopamine, or radioligands for the 5-HT2 receptor, serotonin transporter, or D2 receptor [158, 159]. ¹⁸F-fluorodeoxyglucose (FDG) is one such tracer that has been developed to study regions of altered glucose metabolism in the brain. This is now widely used to study regions of infection, inflammation, or tumor growth [158, 160]. It should be noted, however, that localization of FDG is a measure of glucose uptake and does not provide a precise measure of glucose metabolism [96]. As previously mentioned, FDG will bind hexokinase but not be metabolized.

Each of the aforementioned methodologies has its merits. Nonetheless, there exist limitations to their usability and suitability for the metabolic research in question. PET technique, which is one of the most commonly used metabolic imaging methods for the brain- in addition to MRS—is limited by the high cost accompanying the production of its radiopharmaceuticals as well as the inability to distinguish anatomic details [161]. MRS limitations include potential tissue heating from the applied radio frequency (RF) energy, lengthy infusion and sampling schemes of the isotopically labeled substrates, and more significantly the metabolic composition retrieved from a given voxel may belong to different tissue types [161], and thereby depreciating the importance of brain tissue-dependent metabolite heterogeneity. Additionally, some MRS techniques (carbon spectroscopy) require either long acquisition time or infusion of ¹³C-labeled substrates, while others (hyperpolarization-based) are limited by their hyperpolarized metabolic probe’s short relaxation time [162]. Since many of these also rely on cerebral vasculature to either determine differences in brain regions or to transport a substrate, making these measurements susceptible to confounding by anatomical variation, as well as pathology that alters pericytes, endothelial cells, and smooth muscle cells which play a large role in controlling cerebral blood flow [163, 164]. Admittedly, many diseases also affect cerebral microvasculature, but this complicates separating metabolic differences from vascular differences [165].

The previously mentioned methods provide good structural, functional, and molecular information especially when used in conjunction with one another. However, they only provide regional resolution and are dependent on the detectable metabolites. On the other hand, optical imaging modalities provide cellular resolution. In light of these differences, an optical imaging modality such as FLIM, may be a better tool to study metabolic changes in the brain [15].

Reduced nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide (FAD) are endogenous molecules in cells and tissues that have weak natural fluorescence [166]. Using specific wavelength of a light source, they can be transformed into an excited state with subsequent relaxation back to the ground state followed by simultaneous emission of a light signal of a different wavelength. The first NADH fluorescence imaging on live cells was performed in 1957 by Duysens and Amesz [167]. Since then, it has been widely used to study metabolic state of cells and tissues. NADH and FAD have divergent emission spectra with a little spectral overlap, which allows them to be imaged simultaneously. Fluorescence microscopy of NADH and FAD molecules helps to visualize where metabolic alterations are taking place in biological samples, including cells and tissue sections.

While the fluorescence intensity signal of NADH and FAD is highly informative and represents the activity of metabolic pathways, it is proportional to their concentration in the imaged samples. This can become problematic when reproducing measurements or when a fluorophore is only present at small concentrations in a sample. One way to circumvent this reliance on fluorophore concentration is to look at fluorescence lifetime instead of fluorescence intensity. With fluorescence lifetime the amount of time a fluorophore spends in the excited state is measured. This property of a fluorophore is dependent only on the molecule’s microenvironment and not on its concentration [168‐170]. This can be particularly advantageous in the field of biology as some traditional fluorophores are cytotoxic or not native to a cell’s environment. As previously mentioned, FLIM allows for the use of endogenous fluorophores that are already present in biological systems [167, 168, 171]. Looking at a fluorophore’s fluorescence lifetime as opposed to its fluorescence signal intensity also eliminates the requirement that a fluorophore displays spectral shift in response to a change in environment [168]. This also allows for the use of fluorophores with large wavelength absorption spectra, potentially lowering the cost of instrumentation [168].

Development of laser scanning microscopy methods in conjunction with fluorescent molecules lifetime measurement marked the advent of FLIM. With FLIM, contrast in an image is generated by differences in fluorescence lifetimes [168]. This new method allows for the study of a fluorophore’s microenvironment, shedding light on the different environments in a cell or tissue. Examples of factors that can change a fluorophore’s fluorescence lifetime include the presence of oxygen, calcium, pH, energy transfer, as well as many other factors like nearby proteins or quenchers [172‐184]. As the field grows, more fluorophores will enter the field. NADH and FAD, which have intrinsic fluorescence and their lifetime is tied to the state of the molecule [185, 186]. Given the subcellular resolution fluorescence microcopy provides, examination of these fluorophores allows for a direct analysis of the metabolic profile of individual cells.

Advances in the field of microscopy over the last century have been key in the development of FLIM. Two-photon excitation microscopy allowed for deeper imaging of tissues and reduced photodamage even with longer imaging times and allowed for the interrogation of femtoliter volumes in a sample [185, 187‐191]. The ability to capture spatial and temporal resolution simultaneously in microscopy was also key in the development of FLIM as this allowed for fluorescence lifetime data acquisition in volumes small enough to analyze individual subcellular compartments in real time [189, 191].

The fluorescence lifetime of NADH changes depending on whether it is free or in a bound-to-enzyme form [186, 192]. For free NADH, the lifetime is around 0.4 ns versus around 3.4 ns when it is in a bound-to-enzyme form [15]. Other factors that affect the fluorescence lifetime include pH, temperature, and redox state. Since NADH is intricately involved as a cofactor in multiple metabolic pathways, differences in its lifetime can be used to determine which pathways a cell is using to meet its metabolic demands. FAD is another cofactor whose autofluorescence can and has been used to investigate metabolic processes in situ [193‐196]. Specifically, FLIM has found a niche in analyzing tissue for markers of cancer, since a hallmark of cancerous cells is the reprogramming of metabolic pathways [197‐203]. It should be noted that there are many other molecules in cells and tissues that exhibit autofluorescence and whose fluorescence spectra overlap with these probes, and techniques have been developed to bypass this problem [186, 192, 204, 205].

The phasor approach for FLIM data analysis helps to avoid multiexponential analysis used in typical fluorescence decay fitting [187, 189]. In phasor analysis each pixel of an image has its fluorescence decay plotted as a vector on a universal semicircle plot with vertices at (0,0) and (0,1). These points correspond to an infinite lifetime and a lifetime of zero, respectively. Pixels with contribution of multiple different lifetimes lie inside the circle while pixels with more homogenous lifetime composition lie closer to the periphery of the circle [187, 189]. The resultant phasor plot provides insight on the state of various fluorophores within a sample, which is not intended to determine exact fluorescence lifetimes (Fig. 2). Notably, there are caveats to this fit free analysis of FLIM data. Phasor analysis is not sensitive to low photon counts [206], and cannot reliably distinguish small changes in the redox state in the sample [186]. Moreover, the metabolic analysis does not have a single pathway resolution and response is reported by the ratio of free-to-bound fractions, where both numerator and denominator can change, while maintaining similar ratio [207]. Further, if the calibration decay is not acquired properly, this can lead to improper conversions of data to phasors [208]. Additionally, if the phasor transformation is not done with data acquired in entire period of the laser repetition rate, advantages of phasor plot analysis can be diminished [209]. The consequences of such errors as well as ways to avoid them are outlined elsewhere [185, 209].

As previously discussed, many neurodegenerative diseases involve metabolic reprogramming. By using FLIM, these changes in brain metabolism can be characterized with the possibility of identifying metabolic derangements prior to the onset of symptoms. For example, we have recently shown that neurons lacking the lipid processing enzyme lipoprotein lipase (LPL) show a shift in the NADH phasor plot indicative (Fig. 2) of increased free NADH compared to wild-type neurons, suggesting that LPL depletion results in metabolic shifts towards glucose utilization and away from oxidative metabolism (Fig. 2) [144]. These findings were the first to combine electrophysiological measurements of neuronal function with metabolic changes and highlight how FLIM may be used to measure metabolism alterations in fresh-frozen brain sections [144]. Recent studies have also employed FLIM to successfully determine metabolic changes in primary cultures of neurons and astrocytes in response to toxins that mimic the effects of PD [210]. Specifically, manganese (Mn) treatment leads to increased bound NAD(P)H in neurons consistent with enhanced apoptosis, but leads to decreased bound NADH in astrocytes, possibly due to a shift towards glycolytic metabolism and impaired respiration [210]. FLIM has also been used to study chronic neuroinflammation in the brains of mouse models of MS and in venous blood of human subjects [211‐213]. Here, NAD(P)H was used as a fluorophore to detect active NADPH oxidase and as an analog for oxidative stress [211‐213]. NAD(P)H can be used as an analog for oxidative stress because it serves as a source of electrons to generate ROS via NADPH oxidase [214‐216]. In the brain, overactivation of NADPH oxidase has been linked to neurodegenerative disease [214‐216]. This technique has also been applied to studying the pathophysiology of cerebral amyloid angiopathy and glioblastoma [217, 218].

With enough characterization of a given fluorophore, and its possible intracellular bound states and respective fluorescence lifetimes, one can generate a database that allows for near complete analysis of a cell’s metabolic state in various compartments [219]. Leben et al. undertook such an endeavor, generating a database of the various fluorescence lifetimes that NAD(P)H has when bound to various enzymes. With the help of RNA-seq data, they identified metabolic enzymes like malate dehydrogenase or lactate dehydrogenase, inducible enzymes like nitric oxide synthase and NAD(P)H oxidases, as well as other abundant small enzymes like sorbitol dehydrogenase that were most likely to bind to NAD(P)H and change its fluorescence lifetime [219]. By performing measurements of homogenous mixtures of NAD(P)H and single enzymes they were able to generate a database listing nearly all possible fluorescence lifetimes for this fluorophore [219]. This demonstrates that with enough fluorophore characterization, FLIM has the power to identify the state that a given fluorophore is in, what enzymes it is bound to, as well as what its microenvironment looks like in various intracellular or extracellular compartments. Although FLIM characterization is somewhat new, and often require validation studies to support the findings, it is clear even from the limited number of published studies that FLIM analysis can resolve cell-specific changes in brain metabolism that would otherwise be missed by alternative imaging modalities.

Microglia are of particular interest since altered microglial metabolism coincides with changes to phenotype and function that may contribute to, or protect against, the pathogenesis of many neurodegenerative diseases. However, existing methods for studying them rely on removing microglia from their native environment to study their gene expression and phenotype.

Recently, several publications have begun to utilize FLIM to understand inherent characteristics of microglial metabolism that may differentiate them from other cells in the brain. Sagar et al., measured the lifetime of NADH in both primary microglial cultures and in 100-μm-thick coronal sections from mice that had been treated with increasing doses of LPS in order to modulate the inflammatory status of the brain and “activate” the microglia [145]. Interestingly, they found that microglia in their activated state surveillant or inactivated state had a shorter NADH mean lifetime compared to other glial cells, and that they had a greater fractional contribution from free NADH [145]. In contrast, LPS treatment increased their mean NADH lifetime and decreased the fractional contribution of free NADH (compared to vehicle treatment) [145]. In support, Bernier et al., 2020 recently reported that resting microglia in situ have a much shorter NAD(P)H lifetime than surrounding non-microglial cells [53], and therefore a more glycolytic state. Importantly this study performed glycolytic inhibition with iodoacetate to confirm that glycolytic activity correlates with measurements of free short lifetime NADH [53]. Together, these studies suggest that microglia are much more glycolytic than previously thought, particularly in comparison to non-microglial cells and that activated microglia in fact may metabolically shift away towards oxidative metabolism to support the greater bioenergetic needs of activation-associated functions such as phagocytosis. Since these findings are somewhat in contrast to previous literature, and in particular findings from (albeit peripheral) inflammatory and macrophage models, it is imperative to further validate these findings using models of neurodegenerative disease such as AD, PD and MS. These discrepancies also highlight the importance of further work to validate our understanding of cell-and-disease-state specific brain metabolism, and to determine whether our findings corroborate with those from existing metabolic imaging approaches. In other words, will our findings using FLIM challenge our previous notions? For example, FDG-PET, described above, is predominantly used to measure brain glucose metabolism, which is interpreted as neuronal metabolism and neuronal activity. Since we know that non-neuronal cells majorly contribute to glucose uptake and metabolism, could glial cells also contribute to FDG-PET readouts? Although further method development is required, FLIM analysis of brain tissue from various metabolic and disease state paradigms, in conjunction with cell-specific markers, may eventually help answer these questions. Nonetheless, it is clear from the few studies employing FLIM, that this technique will enable major advancements in our understanding of microglial metabolism in situ. With more widespread use of FLIM and better metabolic characterization of microglia, FLIM could be used to identify mechanisms that promote or protect against the development of ND.

A striking similarity between ND is the immunometabolic switching of glial cells. Specifically, in AD, PD and MS oxidative-to-glycolytic, and shifts are often observed. It is plausible to suggest that targeting glycolytic shifts may be a potential therapeutic strategy for the treatment of ND. This is particularly promising in the case of AD, since disease-modifying treatments are mostly lacking. Here we suggest that imaging modalities such as FLIM, that can discriminate between subcellular, cell-type and regional differences in substrate utilization may be instrumental in the development of strategies that target metabolic processes to modify cellular function and improve outcomes in patients with AD and beyond.