Using stable isotope labeling to advance our understanding of Alzheimer’s disease etiology and pathology

Abstract

Stable isotope labeling with mass spectrometry (MS)-based proteomic analysis has become a powerful strategy to assess protein steady-state levels, protein turnover, and protein localization. Applying these analyses platforms to neurodegenerative disorders may uncover new aspects of the etiology of these devastating diseases. Recently, stable isotopes-MS has been used to investigate early pathological mechanisms of Alzheimer’s disease (AD) with mouse models of AD-like pathology. In this review, we summarize these stable isotope-MS experimental designs and the recent application in the context of AD pathology. We also describe our current efforts aimed at using nuclear magnetic resonance (NMR) analysis of stable isotope labeled amyloid fibrils from AD mouse model brains. Collectively, these methodologies offer new opportunities to study proteome changes in AD and other neurodegenerative diseases by elucidating mechanisms to target for treatment and prevention.

1 |. BACKGROUND

Alzheimer’s disease (AD) is the most common neurodegenerative disorder in the world as well as the most common form of dementia, accounting for 60–80% of dementia cases. Currently, an estimated 5.8 million people in the United States are living with AD, and this prevalence is projected to nearly triple in the coming decades (Alzheimer’s Association, 2020). In addition to the increasing health concerns, the economic burden is already estimated at over $300 billion and is expected to balloon as the population ages (Takizawa et al., 2015). Thus, there is a dire urge for more insightful pre-clinical diagnosis and effective treatments. AD is most commonly characterized behaviorally as a deterioration of short-term memory. However, as the disease progresses and neurodegeneration spreads, AD can subsequently cause more severe cognitive impairments (Bondi et al., 2017). The pathological characteristics that define AD and differentiate it from other dementias are the presence of neurofibrillary tangles and amyloid plaques (DeTure and Dickson, 2019). Tangles are primarily found inside neurons and are mainly composed of hyperphosphorylated tau proteins. Amyloid plaques are primarily found extracellularly are mainly composed of misfolded amyloid beta (Aβ) peptides. These neuropathologies have traditionally been determined during post mortem analysis. Recently, earlier diagnosis has been aided by technological advancements including positron emission tomography-based imaging as well as mass spectrometry (MS) for proteomic assessment of cerebrospinal fluid (CSF) or blood plasma (Bader et al., 2020; Higginbotham et al., 2020; Lowe et al., 2019; Morris et al., 2009). These technologies have also advanced the understanding of how the accumulation of misfolded proteins begins decades before the onset of memory loss or other behavioral symptoms.

1.1 |. AD Etiology

A small percentage of AD cases (<1%) are caused by genetic mutations, specifically in the presenilin genes, PSEN1 and PSEN2, as well as in the amyloid precursor protein (APP) gene (Karch et al., 2014). These mutations cause familial AD (FAD) and result in an early and aggressive clinical onset (Tanzi, 2012). Interestingly, mutations in the PSEN1 gene account for ~70% of these cases, suggesting that the particular species of Aβ may be critically important for pathology (Guerreiro et al., 2012). With APP, causative mutations have been identified throughout the gene, and most mutations lead to an increase in the production or an increase in the aggregation propensity of Aβ (Cruts et al., 2012). In all cases, these causative FAD mutations push APP processing toward the amyloidogenic pathway and toward the production of more toxic species of Aβ. As a consequence of increased accumulation of Aβ, cellular homeostasis likely becomes skewed, eventually overwhelming protein degradation or clearance machinery, promoting buildup of Aβ oligomers and plaques, and causing cell dysfunction and degeneration. The vast majority of AD cases are considered sporadic AD (SAD), or late-onset AD. The biggest risk factor for SAD is aging, but pathology likely arises due to many factors (Selkoe and Hardy, 2016). Over the past several years, genome-wide association studies and whole exome DNA sequencing studies have identified over 30 genes associated with increased risk of AD (Kunkle et al., 2019). Discovering and understanding these genetic risk factors has helped point to a number of important disease pathways, including lipid trafficking, neuroinflammation, synaptic function, and endocytosis (De Strooper and Karran, 2016). In most of these cases, SAD risk factors ultimately influence the processing and clearance of Aβ oftentimes by microglia.

Importantly, while genetics point to Aβ as a critical factor in AD pathogenesis, research has shown the disease to be far more complex. Aβ plaques or even tau tangles, while necessary, are not sufficient on their own to cause the cognitive impairment that defines AD. In fact, several studies focusing on individuals at advanced ages (80 or 90 years and older), such as the 90+ study or the SuperAger study, have identified cognitively normal individuals who still harbor significant amyloid burden (Harrison et al., 2018; Kawas, 2008; Sakr et al., 2019; Villemagne et al., 2008). The amyloid burden identified in these older individuals would normally be associated with AD, especially in consideration of age, however, these individuals show no loss of cognitive function. These studies suggest that AD pathogenesis is much more complex than the presence of plaques or tangles and sophisticated methods investigating the interplay of aberrant protein accumulation and impairment of protein networks is required to advance our understanding of this disease.

1.2 |. APP and Aβ Processing

One of the major questions in the field is how Aβ peptides confer toxicity to and ultimately kill neurons. Aβ is generated from sequential processing of APP, a ubiquitously expressed, type-I transmembrane protein with a large extracellular domain (Heber et al., 2000; Vassar et al., 1999; Wasco et al., 1993). Based on localization, interactome, and knock out studies, full length APP has been suggested to have roles in axon growth and guidance, maintenance of dendritic spine density and dynamics, as well as synaptic plasticity and learning (Muller et al., 2017; Rice et al., 2019). In addition to its many functions, APP undergoes proteolytic processing, resulting in several smaller fragments that have their own roles in the nervous system. APP processing occurs through several mechanisms, most commonly the non-amyloidogenic or amyloidogenic pathways (Selkoe and Hardy, 2016). In the non-amyloidogenic pathway, α-secretase cleaves APP within the Aβ region, producing the soluble APP fragment and the α-carboxyl terminal fragment. In the amyloidogenic pathway, which is favored when APP is internalized within clathrin-coated vesicles, β-secretase cleaves APP at the amino terminus of the Aβ region (Cirrito et al., 2008; Cirrito et al., 2005). In both pathways, the C-terminal fragments are subsequently cleaved by γ-secretase, producing the APP intracellular domain, and either the p3 fragment or the Aβ peptide, respective of the non-amyloidogenic or amyloidogenic pathways. The non-amyloidogenic pathway physiologically dominates, which helps limit production of Aβ and protects neurons against Aβ toxicity (Postina et al., 2004; Yuan et al., 2017). The amyloidogenic pathway, on the other, hand ultimately produces the Aβ peptide, which is subsequently released outside of the cell and can form oligomers, protofibrils, fibrils and plaques. Longer species of Aβ, particular Aβ₄₂, are more prone to misfold and aggregate into oligomers and plaques with oligomers representing the most toxic species often leading to synaptic dysfunction and eventually neuronal loss (Cline et al., 2018; Fontana et al., 2020; Huang and Liu, 2020).

1.3 |. Modeling AD like pathology in mice

Traditionally, researchers have relied on mice that overexpress APP with or without FAD mutations. These transgenic overexpression models include the PDAPP, Tg2576, APP23, and TgCRND8 models (Chishti et al., 2001; Games et al., 1995; Hsiao et al., 1996; Mucke et al., 2000; Sturchler-Pierrat et al., 1997). These mice recapitulate extracellular Aβ deposits in the brain and develop cognitive dysfunction, however, they do not develop the neurofibrillary tangles or neurodegeneration that is seen in humans. In an effort to better model more aspects of AD pathology, researchers generated mice with additional FAD mutations, often in the PSEN1 gene. These models include the 5XFAD mice, which carry five mutations in the APP and PSEN1 transgenes and aggressively develop amyloid plaques, gliosis, synaptic degeneration, neuronal loss, and cognitive deficits (Oakley et al., 2006). However, these mice still do not develop neurofibrillary tangles. Thus, to recapitulate tangle pathology, transgenic APP mice were crossbreed with mice carrying tau transgenes, such as the 3xTg-AD mice (Oddo et al., 2003). However, the tau mutation in these mice are not linked to AD, rather frontotemporal dementia with parkinsonism, and do not accurately recapitulate AD pathogenic mechanisms. Furthermore, one of the biggest limitations with these transgenic models is that APP overexpression not only massively overproduces APP and Aβ, but it also overproduces the other various APP cleavage produces. As a result, some phenotypes seen in these mice may be a result of the elevated levels of several APP fragments and may not be relevant to human AD.

Since APP overexpression may lead to non-relevant phenotypes, researchers developed an App knock-in (App KI) mouse that overproduces pathogenic Aβ without any overexpression (Saito et al., 2014). In order to produce toxic Aβ these App KI mice express a humanized Aβ sequence along with FAD mutations in the APP gene that cause APP to be cleaved preferentially along the amyloidogenic pathway (Sasaguri et al., 2017). These mice carry different combinations of FAD mutations leading to different degrees of pathology. The first line of App KI mice expresses the Swedish KM670/671NL (App^NL/NL) mutation. While the Swedish mutation increases production of Aβ, it is almost entirely Aβ₄₀, which is efficiently degraded. As a result, these mice have normal physiology and behavior (Salas et al., 2018). The second model expresses the Beyreuther/Iberian (I716F) and Swedish (App^NL-F/NL-F) mutations. The Iberian mutation increases the Aβ₄₂/Aβ₄₀ ratio by a factor of 30, leading to Aβ deposition, enhanced neuroinflammation and later cognitive dysfunction. The third model expresses the Arctic (E693G), Iberian, and Swedish (App^{NL-G-F/NL-G-F}) mutations. The additional Arctic mutation makes Aβ more prone to oligomerization/fibrillization (Basun et al., 2008; Nilsberth et al., 2001; Sahlin et al., 2007). As a result, these mice display accelerated pathology with Aβ deposition and cognitive decline occurring nearly three times as fast as the App^NL-F/NL-F mice. App KI mice have been used to show that some of the phenotypes in transgenic mice are not reproduced and may be non-relevant artifacts of transgenic overexpression (Saito et al., 2016). While these models eliminate the confounding variable of APP overexpression, they are not without their own caveats. For instance, the combination of FAD mutations is not observed in the human population and the interaction of these mutations may also lead to irrelevant phenotypes. Finally, similar to transgenic mice, these App KI mice do not develop neurofibrillary tangles (Sasaguri et al., 2017). Thus, these models represent a very early preclinical stage of AD and are effective for analyzing Aβ pathogenesis, preventative drug development, and identification of early biomarkers. App KI mice represent a leading rodent model system to investigate amyloid pathology as it relates to AD. They are particularly suitable because the different strains of mice allow for comparative investigations into different stages of pathology while controlling for age and APP expression.

2 |. Proteome wide measures of relative protein abundance using isotopically labelled cells and tissues

A couple decades ago, stable isotopes, such as Nitrogen-15 (¹⁵N) were utilized to quantitate changes in proteome expression and modifications between different pools of cells (Conrads et al., 2001; Oda et al., 1999). Shortly after, Stable Isotope Labeling with Amino acids (e.g. Arginine and Lysine) in Cell culture (SILAC) also became a mature and widely used approach to obtain robust measures of relative protein abundance between two or more biological conditions (Andersen et al., 2005; Ong and Mann, 2006; Schwanhausser et al., 2011). Heavy isotopically labeled proteins are biochemically identical, and proteins that incorporate the amino acids differ only in mass, enabling identification and quantification of both heavy and light peptides via shotgun proteomics. One drawback of using labeled amino acids is that only peptides that contain that particular amino acid can become labeled, leading to some peptides evading analysis.

Around the same time, several labs developed analogous methods for metabolic labeling of whole organisms with non-radioactive stable isotopes in the food or water (Doherty et al., 2005; Larance et al., 2011; Sury et al., 2010; Wu et al., 2004). These studies predominantly utilize rodent models and have been employed to assess changes in protein abundance. Stable Isotope Labeling in Mammals (SILAM) usually utilizes a Spirulina based chow that is enriched with heavy isotopes and has been shown to label rodents ≥ 95% with ¹⁵N. Importantly, even after multiple generations of labeling with this chow, no adverse health effects have been seen (McClatchy et al., 2007). SILAC and SILAM, when coupled to tandem MS, provide robust analytical workflows capable of determining changes in the relative protein abundance based on direct comparisons or ratio-of-ratio strategies (i.e. spiked internal standard) (Figure 1A–B).

Figure 1: Using stable isotope labeling to measure relative protein abundance in mouse models of AD-like pathology.

(A) Direct comparison of relative protein levels between two mouse models. A mouse model of AD-like pathology remains unlabeled (¹⁴N) and a relevant control mouse model is labeled with stable isotopes (¹⁵N). Brains are dissected and extracts are mixed 1:1 before proceeding for analysis by LC-MS/MS. (B) Comparison of relative protein levels using an internal standard. Two different mouse models of AD-like pathology remain unlabeled (¹⁴N). One control mouse model is labeled with stable isotopes (¹⁵N). Brains are dissected from all three mouse models. Control brain extracts are mixed 1:1 with AD model 1 as well as AD model 2. The two mixtures are then processed separately for LC-MS/MS. Analysis enables comparison of the two unlabeled AD models. (C) Multiscale analysis using stable isotopes. A mouse model of AD-like pathology remains unlabeled (¹⁴N) and a relevant control mouse model is labeled with stable isotopes (¹⁵N). Brains are dissected and extracts are mixed 1:1. Before proceeding for MS analysis, mixed extracts are processed further, including but not limited to, isolation of specific cell types, purification of organelles, or fractionation for synaptosomes. After the processing steps, samples are analyzed with LC-MS/MS.

2.1 |. Investigating Changes in Relative Protein Abundance Using Stable Isotopes with in vitro culture models of Aβ toxicity

SILAC is traditionally used to label two or three samples, labeling one with “heavy” isotopes and the other with “medium” and/or “light” isotopes. One advantage of the in vitro SILAC approach is that the role of defined cell types in AD can be directly interrogated, as opposed to in vivo experiments which provide composite measurements from multiple cell types. Another advantage of in vitro SILAC experiments is that they can provide well-controlled platforms that have clearly defined temporal resolution (i.e. expression of transgenes with viruses or spiking in toxic polypeptides). In fact, this approach has been applied to AD culture models to compare Aβ treated cells or Aβ producing cell to control cells in order to compare protein levels or proteome changes associated with Aβ induced pathology (Andrew et al., 2019; Correani et al., 2017). However, there are limitations to the SILAC approach such as variability between cultures and the fact that in vitro experiments fail to recapitulate the complex environment of the intact and living brain.

2.2 |. Determining changes in global steady state levels and enriched biochemical fractions in brains from AD mouse models.

Stable isotope labeling of whole mice followed by MS-based proteomic analysis is used to investigate proteins’ relative abundance even if none of the experimental samples is labeled. The most powerful strategy is to mix light and heavy extracts 1:1 to directly compare relative changes in protein abundance between control and experimental brain extracts (Figure 1C). Mixed samples are then analyzed by liquid chromatography-tandem MS (LC-MS/MS). Alternatively, isotopically labeled tissues extracts, usually from a wild type mouse brain, can be used as a spiked-in internal standard (Figure 1C) to compare relative changes in protein abundance among unlabeled experimental or control extracts. Using an internal standard can help eliminate any variation that would arise due to technical procedures or sample processing (Butko et al., 2013; MacCoss et al., 2005; Savas et al., 2015; Savas et al., 2017). For example, wildtype (WT) ¹⁵N labeled whole brain homogenates can be mixed 1:1 with unlabeled ¹⁴N cortical or hippocampal homogenates from multiple experimental groups followed by crude synaptic fraction isolation from the mixed ¹⁴N and ¹⁵N samples. Mixing the samples as early as possible in the experimental workflow limits potential technical variations that may occur during the isolation of synaptosomes. This approach is strategic compared to isobaric labeling approaches, such as TMT or iTRAQ because with those techniques, samples are labelled and mixed at the peptide level after fractions are isolated.

Recently, we used this approach in the context of AD in order to assess how AD-like pathology affects synaptic protein abundance. We compared App KI pathogenic genotypes to the relevant control (App^NL-F/NL-F / App^NL/NL or App^{NL-G-F/NL-G-F} / App^NL/NL) (Hark et al., 2020). Hark et al. found that at six months of age, most proteins that were significantly altered in App^NL-F/NL-F or App^{NL-G-F/NL-G-F} cortical extracts had increased levels compared to age matched App^NL/NL mice, suggesting synaptic proteins accumulate early in pathology. In contrast, at twelve months of age, most proteins with significantly altered levels in App^NL-F/NL-F or App^{NL-G-F/NL-G-F} cortical extracts had decreased abundance compared to App^NL/NL controls, consistent with the abundant evidence that synapses eventually degenerate in AD (Masliah et al., 1994).

3 |. Investigating Changes in Protein Turnover Using Dynamic Stable Isotopic labeling

Efficient protein turnover underlies critical biological processes by eliminating old or damaged proteins, replacing them with newly synthesized versions. A dysregulation of protein homeostasis (proteostasis) is considered a “hallmark” of aging (Balch et al., 2008; Lopez-Otin et al., 2013). This loss of proteostasis also clearly plays a factor in neurodegenerative diseases, as they are all share the common feature that misfolded proteins accumulate in the brain (Beal et al., 2006). In AD for example, impaired protein degradation machinery is prevalent. Chaperones and co-chaperones have been linked to APP metabolism, Aβ production and Aβ clearance; the ubiquitin proteasome system (UPS) has been shown to be involved in Aβ degradation, and impairments of the UPS have been shown to increase Aβ levels; and autophagic vacuoles are accumulated while genes encoding autophagy regulators have been shown to be altered in the brains of AD patients (Bustamante et al., 2018; Oddo, 2008; Zare-Shahabadi et al., 2015). Consequently, analyzing protein turnover in neurodegeneration models could provide critical insight into the mechanisms underlying pathology.

Historically, measuring protein turnover was accomplished by using radioisotope analogs of molecules, such as ³H, ¹⁴C, or ³⁵S labeled amino acids, along with scintillation counters. Toward the end of the 20^th century, spurred by advancements in MS instrumentation and data analysis software, there was a shift toward using stable isotope labels using ²H, ¹⁵N, ¹⁸O, and ¹³C. The advantages of stable isotope systems included safety, bio-orthogonality, and ability to quantify thousands of labeled proteins in each analysis. Dynamic pulse-chase SILAC has been applied to several cell models including bacteria, yeast, and mammalian cell lines and has helped uncover many basic principles of protein turnover (Maier et al., 2011; Pratt et al., 2002; Schwanhausser et al., 2009). However, SILAC based protein turnover studies have not been used to investigate in vitro models of AD-like pathology.

3.1 |. Investigating Protein Degradation in AD Mouse Models

Dynamic pulse-chase SILAM (pcSILAM) has been utilized to analyze the differences in protein turnover rates between various tissues such as liver, heart, and brain (Cambridge et al., 2011; Guan et al., 2012; Price et al., 2010). In addition, pcSILAM has been used to analyze and discover extremely long-lived proteins that do not turn over for many months (Savas et al., 2012; Toyama et al., 2013). Finally, this approach has also be used to analyze differences in protein turnover in the context of aging and various treatments such as rapamycin or calorie restriction (Basisty et al., 2018a; Basisty et al., 2018b). However, one area that has largely been left unexplored with this technology is long-lived proteins and protein degradation impairments in disease models, especially proteinopathies such as AD.

Pulse-chase stable isotope metabolic labeling with LC-MS/MS-based proteomic analysis can be effectively used to better understand AD by identifying protein networks with impaired degradation in mouse models of AD (Hark et al., 2020). Using the App KI mice, in particular, enables a unique and powerful opportunity to investigate how the degradation of individual proteins is altered at different stages of AD-like pathology. More specifically, protein degradation impairments can be assessed before and after measurable amyloid pathology. We focused on an age when the App^{NL-G-F/NL-G-F} brains have elevated levels of Aβ₄₂ and amyloid deposits but when the App^NL-F/NL-F brains do not have detectable increases in Aβ₄₂ levels or amyloid deposits. Based on previous pathological characterizations, and our own detailed confirmation of the App KI mice, six to seven months of age fit that criteria. In fact, App^NL-F/NL-F mice do not have significantly elevated Aβ₄₂ levels before ten months of age nor do they display thioflavin S positive amyloid plaques at six months of age. Notably, our analysis of the SDS-insoluble fraction revealed elevated levels of insoluble Aβ at six months, suggesting that there is a small pool of Aβ fibrils in App^NL-F/NL-F brains. App^{NL-G-F/NL-G-F} mice display elevated Aβ₄₂ levels as early as two months of age and harbor many amyloid plaques by six months of age. App^NL/NL brains, on the other hand, do not accumulate detectable levels of Aβ₄₂ peptides or amyloid plaques in the cortex, hippocampus, or cerebellum at any age (Saito et al., 2014; Salas et al., 2018).

To set up the pcSILAM analysis platform, a small cohort of female App^NL/NL, App^NL-F/NL-F, and App^{NL-G-F/NL-G-F} mice were labeled with ¹⁵N (Hark et al., 2020). These App KI mice were then bred with males of the same genotype and the homozygous progeny continued to be metabolically labeled with ¹⁵N until weaning. After weaning, the second-generation pups were chased with a standard ¹⁴N chow for months. This two-generation experimental design generates mirrored “label swapped” datasets (i.e. moms and pups) that provide independent confirmatory opportunities to broaden the analysis. The ¹⁴N labeled peptides in the first generation and the ¹⁵N labeled peptides in the second generation can be measured to track protein turnover both proteome-wide and for individual proteins. Extended labeling and chase periods (e.g. six months) are beneficial in part because at shorter labeling periods chimeric proteins composed of both ¹⁵N and ¹⁴N atoms dominate the brain proteome (Savas et al., 2016). Consequently, the ability to monitor global protein lifetimes across short periods is severely hindered, as chimeric proteins cannot be reliably identified by MS / MS and MS1 isotopic envelops broaden. Indeed, in previous in vivo ¹⁵N pulse-chase studies using WT mice, the number of measured proteins after one- or two-month chase was three times fewer than at very short or long chase periods (Savas et al., 2016).

Following the pulse-chase experimental design, the brain or other tissues can be dissected, and global protein turnover is analyzed by measuring the levels of each protein’s ¹⁴N remaining [¹⁴N / (¹⁵N + ¹⁴N)] in the first generation or ¹⁵N remaining [¹⁵N / (¹⁵N + ¹⁴N)] in the second generation. In order to assess how AD-like pathology affects protein turnover in the App KI mice, the pathogenic genotypes, App^NL-F/NL-F or App^NL-G-F/NL-G- are compared to the App^NL/NL controls. Assessment of global protein degradation across the genotypes revealed no systematic protein turnover impairments. However, by assessing each individual protein’s turnover, we were able to home in on specific alterations. Ratios > 1 for ¹⁵N remaining in the second generation or > 1 for ¹⁴N remaining in the first-generation brains indicated proteins with impaired turnover presumably due to AD-like pathology. Afterwards, gene ontology (GO) overrepresentation analyses provided a way to investigate if the impaired proteins are enriched for particular functions or cell compartments.

Notably, we found that the most significantly enriched GO cell component terms from the cortical and hippocampal datasets related to axons terminals or the presynaptic compartment. In contrast, proteins with impaired turnover in the cerebellum, a brain region that typically exhibits pathology later in disease progression (Xu et al., 2019), were not associated with GO terms related to axon terminals. Among the proteins that were found impaired were many soluble N-ethylmaleimide sensitive factor attachment protein receptors (SNAREs) including, Syntaxin 1B (Stx1b), Synaptobrevin 1 and 2 (Vamp1 and Vamp2), synaptosomal nerve-associated protein 25 (Snap25), and the calcium sensor Synaptotagmin 1 (Syt1). In addition, several key synaptic vesicle (SV) endocytosis factors, some of which have previously been genetically associated with SAD (Seshadri et al., 2010), had hampered turnover in the cortex and hippocampus of App^NL-F/NL-F and App^{NL-G-F/NL-G-F} mice. These proteins included clathrin coat assembly protein AP180 (Snap91), myc box-dependent interacting protein 1 (Bin1), and amphiphysin (Amph). Interestingly, many of these SV cycle proteins had slower turnover in App^NL-F/NL-F mice, despite these mice just beginning to exhibit increases in misfolded Aβ peptides at this age.

3.2 |. Investigating Protein Degradation in human AD

Stable isotopes have also been used to analyze toxic proteins and biomarker kinetics in humans in the context of neurodegenerative disorders. These studies, known as Stable Isotope Labeling Kinetic (SILK) studies, use short infusions (hours long) of ¹³C and ¹⁵N stable isotopes before measuring the labeled proteins in blood, CSF or tissue samples (Bateman et al., 2006; Paterson et al., 2019). Due to the short labeling duration and low degree of labeling, these experiments mainly provide dynamic measurements of protein synthesis, protein clearance from tissues, or protein release into body fluids. Specifically, in the context of AD, SILK has been used to analyze Aβ in FAD, SAD, and unaffected patients. These studies have found that Aβ₄₂ production and clearance rates positively correlated with amyloid plaque load and age, suggesting that faster turnover of Aβ₄₂ indicates increased deposition of Aβ₄₂ in plaques (Bateman et al., 2009; Wildsmith et al., 2012). Furthermore, SILK studies have investigated how pharmacotherapies or sleep affects Aβ clearance rates (Dobrowolska et al., 2014; Lucey et al., 2018). In addition, SILK has been used to investigate the kinetics of tau proteins both in vitro using induced pluripotent stem cell derived neurons and in vivo in humans. These studies have found that disease associated isoforms or posttranslational modifications of tau affect the half-life and kinetics, and that amyloidosis increased tau production rate (Sato et al., 2018). These studies have demonstrated the power and utility of using stable isotopes to investigate changes in humans and have opened up a new avenue to study AD in the most physiologically relevant system.

SILK studies have also been combined with Nanoscale secondary ion mass spectrometry (NanoSIMS) imaging to investigate the dynamics of Aβ plaque growth and distribution (Wildburger et al., 2018). These SILK-NanoSIMS studies have the power to visualize and localize the isotopic tracer relative to amyloid plaques in AD brain helping determine where the labeled proteins reside. This strategy has been applied in both mouse model and human AD brains to better understand that amyloid plaques are highly variable and dynamic structures. While these SILK studies provide very useful information, especially in humans, they are limited by poor labeling efficiency as the infusion period needs to be balanced by the cost and the burden on study participants. Furthermore, human studies have limitations related to the challenges/impossibilities of sampling frequency and genetic or pharmacological modulations.

4 |. Exploring the possibility of solving amyloid structures with NMR using isotopically labelled amyloid fibrils from AD mouse model brains

Amyloid structures in the brain are found in a variety of compositions and confirmations (DeTure and Dickson, 2019). It has been suggested that structurally dissimilar aggregates (i.e. strains) of the Aβ peptide may potentially explain differences observed in the progression and severity of AD. Synthetic Aβ peptides have been used and manipulated to demonstrate some of these differences (Tycko, 2016). However, structures generated in vitro from synthetic peptides are unlikely to fully recapitulate the complex and diverse morphology observed in human AD brain. We recently embarked on a project aimed to explore the possibility of using purified amyloid material from isotopically labeled AD mouse models alongside nuclear magnetic resonance (NMR) spectroscopy to analyze Aβ morphology. The first idea is to compare and contrast how different FAD mutations (i.e. APP and PSEN1/2) influence amyloid structure. Specifically, we are exploring how amyloid structures differ based on mouse genotypes such as App^NL-F/NL-F, App^{NL-G-F/NL-G-F}, and 5XFAD at different degrees of pathology (i.e. early versus late). The second idea is to determine how amyloid seeds from different sources (i.e. human AD brain or other mouse models) influence the structure of amyloid in TG AD and App KI brains. Previous research using amyloid seeds from various human AD brains has found that seeds from difference sources induce divergent amyloid structures in synthetic Aβ preparations in vitro (Kodali et al., 2010; Lu et al., 2013; Paravastu et al., 2008; Petkova et al., 2005; Qiang et al., 2017). The power of using stable isotopes is that one can, in theory, perform parallel experiments aimed to determine amyloid structure diversity in the living animal. Amyloid aggregates isolated from isotopically labeled AD mouse model brains are analyzed by solid-state NMR spectroscopy. The NMR fingerprint spectra subsequently illuminates structural homogeneity, strain variety, and distribution. These experiments could provide useful information into the diversity of Aβ morphology and can help inform how different pathological pathways emerge from the same peptide.

5.0 |. Conclusion

Stable isotope labeling followed by discovery-based LC-MS/MS analysis is a powerful technique that can be used to investigate relative protein levels between multiple experimental groups both proteome wide and within particular cellular fractions. In the context of AD, as referenced here, this technique has been used to analyze how amyloid pathology alters the proteome using in vitro cultures as well as in vivo mouse models of AD. Determining relative changes to the proteome can aid in the understanding of how the cell responds to Aβ accumulation, or which cellular compartments are uniquely vulnerable. Similarly, stable isotopes are used in dynamic pulse-chase experimental designs followed by LC-MS/MS in order to investigate protein turnover. Only very recently has this technique been used to study models of neurodegeneration, specifically using mouse models of AD-like pathology (Hark et al., 2020). The timing of the pulse-chase labeling can be modified to investigate different stages of pathology, which can help elucidate the timeline or order of pathological mechanisms. Overall, these experiments have helped reveal early and vulnerable substrates of impaired protein degradation during amyloid pathology, and may point targets for preventing the synaptic dysfunction that precedes neuron loss. Finally, stable isotope labeling can be used in combination with NMR spectroscopy to investigate the structure of purified Aβ, which can help elucidate the consequences of the diverse morphology of Aβ in an in vivo context.

It is widely appreciated that Aβ accumulation occurs decades before cognitive impairment manifests in AD. As a result, targeting Aβ has been the main therapeutic strategy to treat or prevent AD. However, nearly all clinical trials have failed, likely due to treatments beginning too late into disease progression. Thus, it is critically important to better understand the early mechanisms of AD pathogenesis and the proteomic consequences of Aβ accumulation. Stable isotopes and preclinical mouse models of AD provide powerful tools to help investigate these early mechanisms of AD and can help reveal the proteins or protein networks that should be targeted to treat or prevent the disease before irreversible neurodegeneration takes place. Finally, while discussed here with respect to AD, stable isotope labeling can be applied to many different neurodegenerative disease models beyond AD to help elucidate critical mechanisms for a variety of diseases.

Figure 2: Pulse-chase stable isotope labeling to measure protein turnover in mouse models of AD-like pathology.

(A) Single generation labeling with stable isotopes. Mice from a model of AD-like pathology are put on a diet of ¹⁵N enriched Spirulina chow for months. After several months, brains are dissected and extracted for LC-MS/MS analysis in order to identify and quantify the proteins that remain labeled with ¹⁴N. (B) Two generation labeling with stable isotopes. Female mice from a mouse model of AD-like pathology are put on a diet of ¹⁵N enriched Spirulina chow for months. Mice of the same genotype are introduced for breeding. Homozygous pups are born labeled with ¹⁵N. These second-generation mice are subsequently chased with a chow containing the natural ¹⁴N for several months. Brains are dissected and extracted for LC-MS/MS analysis in order to identify and quantify the proteins that remain labeled with ¹⁵N. (C) Pulse chase experimental designs can be affected by altering the timing of the labeling or chase. Based on the two-generational labeling, different labeling or chase periods allows for investigation into different stages of disease progression, including but not limited to Aβ accumulation, plaque deposition and Aβ turnover, or advanced aging.

Figure 3: Using stable isotopes and NMR to determine structures of amyloid in vivo.

(A) Assessing how amyloid structures differ amongst various mouse models of AD-like pathology. Several mouse models of AD-like pathology, including but not limited to, App KI mice, App transgenic mice, App/PS1 transgenic mice, and 5XFAD mice are labeled with ¹⁵N. Amyloid fibrils are then purified from each of these mouse models and amyloid structure is assessed with solid state nuclear magnetic resonance imaging (NMR). (B) Assessing how Aβ seeds from different sources influence Aβ structure. Amyloid fibrils are purified from human AD brains with different degrees of pathology, and seeds are injected into an isotopically labeled mouse model of AD-like pathology. After seeding, ¹⁵N labeled amyloid is purified from the mouse and the amyloid structure changes as a result of the injected seeds is assessed with solid state NMR.

Abbreviations:

Aβ

Amyloid Beta

AD

Alzheimer’s Disease

Amph

Amphiphysin

APP

Amyloid Precursor Protein

App KI

Amyloid Precursor Protein Knock In

Bin1

Myc box-dependent interacting protein 1

CSF

Cerebrospinal Fluid

FAD

Familial Alzheimer’s Disease

GO

Gene Ontology

iTRAQ

Isobaric Tag for Relative and Absolute Quantitation

LC-MS/MS

Liquid Chromatography Tandem Mass Spectrometry

MS

Mass Spectrometry

NanoSIMS

Nanoscale Secondary Ion Mass Spectrometry

NMR

Nuclear Magnetic Resonance

pcSILAM

Pulse-chase Stable Isotope Labeling in Mammals

PSEN1

Presenilin 1

PSEN2

Presenilin 2

SAD

Sporadic Alzheimer’s Disease

SILAC

Stable Isotope Labeling with Amino acids in Cell Culture

SILAM

Stable Isotope Labeling in Mammals

SILK

Stable Isotope Labeling Kinetics

Snap25

Synaptosomal nerve-associated protein 25

Snap91

Clathrin coat assembly protein AP180

SNARE

Soluble N-ethylmaleimide sensitive factor Attachment protein REceptors

Stx1b

Syntaxin 1B

SV

Synaptic vesicle

Syt1

Synaptotagmin 1

Tg

Transgenic

TMT

Tandem Mass Tag

UPS

Ubiquitin Proteasome System

Vamp1

Synaptobrevin 1

Vamp2

Synaptobrevin 2

WT

Wildtype