Insights into neurodegeneration in Parkinson’s disease from single-cell and spatial genomics

Abstract

Parkinson’s disease is pathologically defined by the death of dopaminergic neurons within the pars compacta of the substantia nigra. To date, the etiology of this multifaceted disease remains largely unclear, which may contribute in part to a current lack of disease-modifying therapies. Recent advances in single-cell and spatial genomic profiling tools have provided powerful new ways to measure cellular state changes in brain diseases. Here, we describe how these tools have offered insight into these complex disorders and highlight a recently performed comprehensive study of dopaminergic neuron susceptibility in Parkinson’s. The data generated by this recent work provide evidence for the role of specific pathways and common genetic variants in giving rise to the loss of a critical dopaminergic subtype in Parkinson’s. We conclude by outlining a set of basic and translational opportunities that arise from those data and insights gathered from this work.

Background

Parkinson’s disease (PD) is a common movement disorder whose most consistent pathological correlate is the death of dopaminergic (DA) neurons residing within the substantia nigra pars compacta. The loss of these cells is responsible for the pathognomonic motor symptoms of PD¹, and is observed in both the 10% of PD cases that are familial–which are largely caused by highly-penetrant mutations in a handful of genes²–and the 90% of sporadic PD cases of unknown etiology. Numerous cellular mechanisms have been identified that may contribute to PD-associated DA neuron death such as dysfunctional autophagy³, loss of calcium homeostasis, mitochondrial dysfunction, and protein misfolding^4,5. Beyond the substantia nigra, other subcortical brain regions, contain neurons susceptible to PD-associated degeneration, including the dorsal motor nucleus of the vagus nerve (DMNV)^6–10, the nucleus basalis of Meynert^11–13, the medial/dorsal raphe^14–17, the locus coeruleus^12,18–20, and the pedunculopontine nucleus (PPN)^{14,15,21–24} (Figure 1), which could additionally explain some aspects of movement issues (especially in the case of the PPN^25–27), as well as non-movement related symptomatology of PD, such as autonomic and olfactory dysfunction²⁸. The pathogenic reasons for the vulnerability of these specific populations–compared with the hundreds of thousands of other neuronal cell types in the brain–remains a fundamental question in PD biology whose answer could unlock new potential treatment targets.

Figure 1. Single-cell and spatial profiling of Parkinson’s disease-susceptible brain regions.

(Left) Schematic of sagittal cut of human brain with colored regions of brain with evidence of neurodegeneration in Parkinson’s disease. (Upper right) Axial cross-section of midbrain with substantia nigra pars compacta colored in shades of red. Darker shades indicate location of more severely-affected neurons. (Bottom right) Schematic for droplet-based single-nucleus RNA-sequencing (snRNA-seq) and Slide-seq/spatial transcriptomics. (Bubble inset) Cartoon of the most susceptible (SOX6/AGTR1) neuron subtype identified in Kamath et al. SNc = substantia nigra pars compacta, DMNV = dorsal motor nucleus of the vagus nerve, PPN = peduncolopontine nucleus, LC = locus coeruleus.

Even within the nigra, molecularly- and spatially-defined DA neuronal populations show variable rates of loss, evidence of which dates back to stereotactic studies from the early 20th century²⁹. Those neurons more resistant to PD-associated degeneration are found more frequently in the dorsal, as opposed to ventral, tier of the A9 catecholaminergic region of the substantia nigra³⁰. Functionally, these neurons have been shown to project to the ventral striatum in rodent studies, suggesting an alternate function more consistent with modulation of reward behavior³¹ similar to their tegmental area counterparts³². In addition, early histological studies of PD brains suggested that the disease-resistant DA neurons exhibit molecular differences compared to their more susceptible counterparts, including a higher baseline expression of the gene CALB1, first described by Yamada and colleagues^{33–35,33,34} and further corroborated by more recent work³⁶. The pattern of resistance of the calbindin-expressing DA neurons was recapitulated in toxin-based rodent and non-human primate models of PD which mirrored the movement phenomenon of PD^35,37, undergirding the primacy of the loss of non-calbindin expressing neurons in PD pathogenesis.

To clarify and expand upon these molecular differences, later efforts capitalized on laser capture microdissection and RNA-sequencing based techniques to extensively profile midbrain DA neuron populations in healthy rodent and human brain^38–41. With laser-capture microdissection, these profiles could be further separated spatially, thereby revealing key molecular differences between putatively susceptible neurons. Further efforts aimed to catalog the molecular pathways which were dysregulated in those dopaminergic neurons in PD brains^42,43. These early studies laid evidence for a number of translational theories about the molecular basis of susceptibility and particularly pathways which were amenable to therapeutic interventions^44,45. For example, the consistent identification of CALB1 as a marker for neurons resistant to PD has spawned a number of intriguing hypotheses about the role of calcium in disease^4,46. One such theory maintains that calcium itself catalyzes the aggregation of alpha-synuclein which may in turn lead to oxidative stress and neuronal damage^47,48. Yet another maintains that the slow oscillatory nature of L-type calcium channels, unique to DA neurons, may lead to an excess of mitochondrial oxidant stress and subsequent dysfunction. The dysfunction of mitochondria in DA neurons, a consistent molecular pathology^49,50, ultimately leads to the demise of these cells when calcium-buffering elements are unavailable.

The importance of calcium regulation to PD pathogenesis was ultimately put to the test in a set of clinical trials measuring the safety profiles^51,52 and efficacy⁵³ of isradipine, an often-used dihydropyridine calcium-channel blocker (CCB) agent, in reducing PD symptoms. While these trials did not meet their primary endpoint, post-hoc analyses suggested that isradipine significantly delayed the need for levodopa therapy in those PD patients receiving adequate pharmacological dosing of calcium blockade⁵⁴. Such results are congruent with observational work^55,56 suggesting that CCBs are in fact most effective at reducing the incidence of PD and, as such, may continue to warrant investigation especially in prodromal PD patient cohorts.

On the flip side of the same coin, efforts have been made to understand the valence of the specific molecules that define susceptible DA neuronal populations. The gene ALDH1A1, for example, encodes an aldehyde dehydrogenase enzyme, and marks neurons that are preferentially lost in PD⁵⁷. Interestingly, impaired metabolism of dopamine by aldehyde dehydrogenase has been shown to produce neurotoxic byproducts and the inhibition of this molecule has been shown to bring about dopaminergic cell loss⁵⁸. This would suggest that certain intrinsic capacities of dopamine neurons to metabolize the molecule dopamine itself are related to the pathogenesis of PD via the buildup of toxic substrates over the course of years^59,60. The enzyme has also been shown to participate in the production and metabolism of GABA in midbrain DA neurons, adding further complexity to its functional roles in vivo⁶¹. It remains to be understood what other specific molecules are responsible or protective of DA neurons in PD-associated degeneration.

Finally, one major pitfall of prior bulk sequencing technologies in understanding brain cell type function and dysfunction is a lack of precision. The human brain is composed of a remarkable diversity of cell types, even with anatomically- and functionally-defined regions such as the nigra. These bulk sequencing techniques required pre-defined aggregation of samples (e.g. by location in the form of laser capture microscopy), preventing a bottom-up ascertainment of cellular diversity, and thereby an unbiased assessment of specific neuron types that are lost in neurodegeneration.

Single-cell genomic profiling identifies a uniquely susceptible population to Parkinson’s disease

The advent of single-cell sequencing has allowed for the rapid development of high-dimensional datasets that enable systematic definitions of cell types within complex tissues^62,63. Application of these tools to the human brain has focused on uncovering processes and cell types that underlie disease⁶⁴. Recent efforts have begun to use these high-dimensional analyses to offer insights into the molecular basis of neurodegenerative disease. Single-cell sequencing studies, for example, have nominated specific neuron types that are vulnerable to degeneration in Alzheimer’s disease⁶⁵ and multiple sclerosis⁶⁶. In addition, these data have helped determine specific AD-associated angiogenic vascular cells⁶⁷, identified common signatures associated with tau tangles⁶⁸, a hallmark of AD, and determined specific molecular profiles of reactive glial cells in association with amyloid and tau aggregates⁶⁹. These studies have continued to expand in size and scope, with additional modalities such as measurements of open chromatin⁷⁰, offering additional insights into epigenetic and protein abundance changes in association with complex degenerative diseases. The vast majority of these efforts have focused on cortex, owing in part due to the intrinsically smaller length scale of variability (compared to less stereotyped areas such as the midbrain) and the existence of highly detailed cortical cell atlases in several mammalian species^64,71,72.

Apart from traditional single-cell sequencing methods, the advent of spatial molecular profiling^73–78 has been employed in the identification of patterns and cell types in complex tissue offering exciting opportunities for understanding disease biology. New ‘-omic’ spatial technologies broadly fall within two categories: capture-based methods that rely on a spatially-defined indexed surface to locally capture mRNA from tissue, and probe-based technologies in which molecules are tagged with fluorescent probes based on hybridization in situ⁷⁸. To date, both approaches have revealed a census of cells in complex regions within the rodent and human brain, thereby revealing previously known and, as-of-yet, unexplored gradients and molecular patterns^79–82. More recent work has aimed at defining spatially-variable changes that might result from a tissue-wide perturbation, such as in response to traumatic brain injury⁷⁷ and aging⁸³ or in the vicinity of local amyloid plaque buildup⁸⁴. The additional advances made in computational tools to investigate and understand these data have allowed for the better assignment of cell-cell interactions, tissue organization, and differential expression within and between cell types across experimental conditions⁸⁵. Finally, spatial profiling methods have recently begun to expand outside of ascertainment of RNA molecules, with the introduction of spatial DNA⁸⁶ and antibody profiling⁸⁷. Undoubtedly, these newer approaches will begin to reveal more hidden information about the molecular basis of brain diseases and fundamental biology about cell types in the brain.

The application of these spatial and single-cell methodologies to PD-relevant tissues has been especially revelatory with the single-cell characterization of mouse midbrain DA neurons^88–90. These studies to date have identified numerous molecularly-defined populations that reside within specific areas of the rodent midbrain, corroborating prior sequencing and histological work that separate broad populations into calbindin-positive and aldehyde dehydrogenase-expressing subtypes⁸⁸. Further comparisons between rodent and ESC-derived cell populations have identified early developmental differences in DA populations that might underlie the vast diversity of their mature successors⁸⁸. In addition, profiling studies of the substantia nigra, both in health^91,92 and disease⁹³, have defined mature DA neuron subtypes and attempted to understand what specific populations are gained or lost in the process of disease. These studies have also pointed to specific processes that occur within cell types, such as the reactivity of glial populations⁹³. One key limitation, however, to these studies is the under-representation of midbrain dopaminergic neurons in human tissue samples–they are surrounded by large white matter tracts and therefore make up a tiny fraction (<1%) of the surrounding cells–making it challenging to comprehensively define DA subpopulations in the human, or to make robust claims about affected molecular pathways in PD subjects.

The dataset provided by Kamath et al⁹⁴–utilizing a novel sorting procedure to enrich DA neurons in human tissue– sampled a sufficient number of DA neurons across subjects to enable a comprehensive annotation classifying them into 10 molecularly-defined subpopulations. Computational integration of these data with those generated from similar regions in other mammalian species, such as those in the rodentia and scandentia orders, identified one population that was solely present in humans and non-human primates, suggesting intrinsic differences between the nigral cytoarchitecture of humans and that of model organisms. Further, spatial profiling of primate brains revealed that these 10 subpopulations fell along a dorsal-ventral gradient, consistent with prior studies suggesting calbindin-expressing DA neurons were more restricted to the dorsal tier of the A9 catecholaminergic region. These subpopulations broadly fell within a gradient of SOX6 and CALB1 expression, with more specific markers including TRHR, GEM, PART1, and DDT, to name a few. Interestingly, these other subpopulations also showed a gradation of susceptibility that matched well with increasing expression of CALB1 and a stronger dorsal tier preference for more resistant populations. Finally, while future work is needed to better characterize the precise surviving fraction in these populations in a large sample of PD cases, further studies have already corroborated the identification of these 10 subpopulations, in addition to identifying new types across the midbrain⁹⁵.

Importantly, comparisons of the relative composition of these 10 subpopulations between neurotypical controls and individuals that died of PD or LBD identified one population that was significantly lost in the degenerative disease process, marked by the expression of an angiotensin type 1 receptor, AGTR1 (Figure 1). This same population was also marked by ALDH1A1, a previously-identified marker of susceptibility, but was non-specific to the specifically degenerating population in this study. In addition, other dopaminergic populations were relatively increased in abundance in association with disease, suggesting these types are preferentially preserved. Indeed, those preserved cells not only exist along the dorsal tier, but also express the marker, CALB1, as demarcating resistance³³. Transcription factor-based pathway analysis identified a number of significantly induced transcription factor programs, driven by genes known to play roles in PD such as TP53 and NR2F2, among others. Importantly, these programs were only activated in the DA neuron population most susceptible to loss, providing an internal control on the valence of these pathways in the process of neurodegeneration. Finally, the same population with the most significant degree of loss additionally showed the strongest enrichment for common variant risk of genes expressed in this cell type, suggesting cell-intrinsic processes underlie PD-associated degeneration.

Basic and translational impact

The identification of a specific population of dopaminergic neurons and their markers provides a launching point for a range of basic and translational avenues. These areas are ripe for insight and offer therapeutic and biomarker potential. First, the substantial contrast in the heritable risk of PD versus that of Alzheimer’s disease (AD) suggests that there are markedly different processes that underlie these diseases. While AD common variant risk primarily points to an immune process, evidence suggests the PD genetic risk seems to converge on DA neurons^{91,94,96–99}. It is important to note that the heritability enrichment does not mean that any individual allele acts in DA neurons; indeed, several elegant studies have described activities of particular risk loci in non-neuronal populations^100–102. Rather, the result suggests that human genetic risk has a significant component that acts in DA neurons. Importantly, this convergence of risk on the most susceptible neuron populations suggests an ability to further test epistatic models of PD. While PD is pathognomically defined by the loss of these neurons, some histological studies suggest that other neurons die at similar rates over the course of this illness^12,19. A brain-wide single-cell sequencing survey of PD brains will help uncover the particular cell populations that are similarly most vulnerable, and assess their underlying enrichment of genetic risk. Comparison of all of these vulnerable populations–including the AGTR1-expressing ventral tier dopaminergic neurons–should help to identify molecular similarities that are likely drivers of cell-intrinsic processes associated with PD-induced degeneration.

Second, the identification of particular neuron types that uniquely decline in PD versus other neurodegenerative conditions offers a springboard for determining diagnostic biomarkers more specific to the underlying neurobiology of PD. Indeed, antibodies against the protein AGTR1–the gene we find to be among the most enriched in the degenerating population–have been shown to be elevated in the serum of 117 individuals with PD as compared to a group of 106 age-matched controls ¹⁰³. Interestingly, this same study also showed a strong correlation between autoantibodies against the angiotensin II receptor type 1 and serum inflammatory cytokines in PD patients. Such findings are consistent with the study authors’ proposed mechanism that the death of these AGTR1-expressing DA neurons causes a release of intracellular antigens and subsequent antibody formation by infiltrating B cells¹⁰³. Interestingly, recent work would suggest a role for the adaptive immune system, via T cell brain parenchymal entry and recognition of alpha synuclein, in neuronal degeneration and the pathogenesis of PD and LBD^104,105. Future approaches could be aimed at mining transcriptional data gathered from various brain regions to identify candidate biomolecules based on marker genes defining the most susceptible neuron types in PD and other degenerative conditions. Additional paired postmortem brain and CSF or serum samples would allow for matched data to validate these computational approaches to identifying disease biomarkers.

Third, comparisons of the most susceptible neuronal populations between PD-affected brains and age-matched controls allows for the identification of molecular pathways altered in the course of neuronal decline. Two specific transcription factors with prior evidence for relevance to PD pathogenesis, encoded by the genes TP53 and NR2F2, rose to significance in regards to their programmatic activity in disease. Specifically, knockout of the protein Tp53 in dopaminergic neurons prevented progression of neurodegeneration in a mitochondrial toxin-based model of PD¹⁰⁶. Further, nuclear p53 has been shown to play a critical role in regulating autophagy, a key cellular pathway which mediates neuronal health, via the PD-associated protein PINK1¹⁰⁷. Additionally, the transcription factor Nr2f2 has been shown to be induced in publicly available transcriptomic datasets of nigral DA neurons from PD patients¹⁰⁸. Over-expression of this key transcription factor accelerated mitochondrial dysfunction and neurodegeneration in a mitochondrial pathology-based model of PD¹⁰⁸.

Finally, the identification of specific markers of the dying population may offer insights into druggable pathways. Increasing evidence points to counter-regulatory functions of the dopamine production pathway and the renin-angiotensin-aldosterone system (RAAS)¹⁰⁹. Within the nigra, activation of the RAA system has been implicated in production of oxidative stress and the knockdown of the angiotensin II type 1 receptor within a zebrafish model of PD ameliorated nigral degeneration¹¹⁰. From a clinical standpoint, observational studies of individuals with PD suggest a disease-modifying therapeutic effect of commonly-prescribed angiotensin receptor blockers (ARBs)¹¹¹. Given that these medications are quite well tolerated in large swaths of the population, such drugs could be more easily evaluated for therapeutic efficacy in PD populations with randomized controlled trials. Additional mining of pathways specifically defining other dying neuron populations may offer similarly fruitful repurposed drug candidates.

Conclusion

Parkinson’s disease is a multifaceted and devastating degenerative disorder with no disease-modifying therapies available, in large part due to our incomplete understanding of the key molecular events that occur in vulnerable neuronal populations. Nigral dopaminergic neurons are one of the most clearly affected cell types in the PD brain and have manifest clinical relevance. Single-cell and spatial sequencing technologies provide an unprecedented opportunity to directly study, in detail, the molecular events that occur within human brain cells affected by disease. These methods have increasingly begun to be applied to PD with success in identifying vulnerable cell populations and intrinsic signaling processes that are likely to be pathogenically important. Additional profiling across other modalities, including a combination of spatial, epigenetic, and protein measurements, and in other affected brain regions will no doubt offer a clear picture of the key molecular events in PD. In turn, we see such data as integral to the development of therapeutic interventions and diagnostic metrics.