Selective vulnerability in neurodegenerative diseases

Abstract

Neurodegenerative diseases have two general characteristics that are so fundamental we usually take them for granted. The first is that the pathology associated with the disease only affects particular neurons (‘selective neuronal vulnerability’); the second is that the pathology worsens with time and impacts more regions in a stereotypical and predictable fashion. The mechanisms underpinning selective neuronal and regional vulnerability have been difficult to dissect, but the recent application of whole-genome technologies, the development of mouse models that reproduce spatial and temporal features of the pathology, and the identification of intrinsic morphological, electrophysiological, and biochemical properties of vulnerable neurons are beginning to shed some light on these fundamental features of neurodegenerative diseases. Here we detail our emerging understanding of the underlying biology of selective neuronal vulnerability and outline some of the areas in which our understanding is incomplete.

The clinical manifestation of a particular neurodegenerative disease reflects the region of the brain and the specific population of cells within it that are affected. Why proteins that usually show widespread expression should accumulate in one set of cells but not in apparently similar neighboring cells is a fundamental question for the field. Addressing the basis of this selective neuronal vulnerability will not only help explain the molecular underpinnings of neurodegenerative diseases, but will also inform on the basic biology and complexity of neuronal sub-types that we currently have little appreciation for. In this review, we describe the neuropathology of several neurodegenerative diseases, focusing on cell types impacted, and we relate what is known about the cell types and pathways that may underlie their vulnerability. We propose that vulnerable neurons have a higher propensity to accumulate disease-related proteins and organelles due to their intrinsic anatomy and biochemistry. Because of this, vulnerable neurons can be considered to teeter on the brink of a ‘catastrophic cliff’. For many of the protein aggregation disorders, whether a neuron falls off the cliff is largely dependent on the solubility of the aggregation-prone protein and the efficiency of clearance mechanisms that keep misfolded or aggregated proteins in check. Correspondingly, genetic variability in the expression level of the deposited protein is important in pathogenesis, and risk factors for disease are often related to intrinsic or extrinsic protein-clearance pathways. Interacting with these variations in expression level and clearance mechanisms are the propagation properties of the deposited proteins, which contribute to the spread of induced neuronal failure. Concepts related to the biochemistry and anatomy of different cell types are discussed here in terms of why a particular cell may be predisposed to accumulate toxic proteins and degenerate, coupled with insight from genetics.

Types of neurons vulnerable to proteinopathy and degeneration

The major neurodegenerative diseases differ from each other not only in the type of pathological protein that accumulates but also in the regions impacted and the types of neurons that are vulnerable. Vulnerability usually refers to vulnerability to pathology, and in most cases, cells that accumulate pathology-associated proteins are also the cells that are lost as the result of cytotoxic events (for example, synaptic toxicity, cell-death-related signaling pathways, or neuroinflammation). For the purpose of this review, we refer to vulnerability as selective, but in many cases it is differential, with cells showing varying degrees of susceptibility.

Alzheimer’s disease.

The brain of patients with Alzheimer’s disease (AD) is characterized by the presence of amyloid plaques composed of β-amyloid (Aβ) and the presence of neurofibrillary tangles composed of misfolded, hyperphosphorylated tau protein. Amyloid deposition typically starts in the terminal fields of neurons expressing high levels of the gene encoding amyloid precursor protein (APP)¹. However, selective loss of vulnerable neurons in early AD is more closely linked to tau pathology. Neurons that are vulnerable to the accumulation of pathological forms of tau and that are lost early in the disease mainly include large pyramidal neurons in layer II of the entorhinal cortex (EC), the subiculum, the CA1 region of the hippocampus^2–5, corticopetal cholinergic neurons in the basal forebrain^6,7, and noradrenergic neurons in the locus coeruleus^8,9. Moreover, reelin-immunoreactive excitatory neurons in layer II of the EC^5,10 and pyramidal neurons immunoreactive to SMI32 (nonphosphorylated medium and heavy neurofilament proteins)^11,12 are particularly vulnerable in AD, whereas inhibitory neurons expressing calcium-binding proteins (parvalbumin, somatostatin, calbindin-D28k, and calretinin) are less vulnerable in AD or animal models of AD^13–15. Dentate gyrus granule neurons, deeper parts of layer III and layers V and VI of EC, and cortical interneurons are relatively spared in early AD^2,16,17. Interestingly, a comparison between rostral neurons (which express neuronal markers with forebrain cortical fates that are vulnerable in AD) and caudal neurons (which express classical markers of hindbrain and spinal cord and are relatively spared in AD) derived from induced pluripotent stem cells (iPSCs) of patients harboring an APP mutation revealed that both the generation of Aβ and the responsiveness of tau to Aβ were affected by neuronal cell type. Rostral neurons, which represent a mix of excitatory and inhibitory cells, were more sensitive than caudal (predominantly inhibitory) neurons, suggesting that cell-autonomous factors may, in part, dictate the pattern of selective regional vulnerability in human neurons in AD¹⁸.

Parkinson’s disease.

The motor manifestations of Parkinson’s disease (PD) are primarily linked to the selective and progressive loss of substantia nigra pars compacta (SNpc) dopaminergic neurons¹⁹. Interestingly, even within the SNpc there is a major loss (~90%) of dopaminergic neurons in the ventral tier, while dopaminergic neuron loss in the dorsal tier may be as little as 25%^20,21. In contrast, the very similar dopaminergic neurons in the ventral tegmental area (VTA) demonstrate a much lower degree of degeneration¹⁹. Neuronal loss is apparent in a handful of other regions. For example, cholinergic neurons in the basal forebrain²² and in the pedunculopontine nucleus are lost, but not glutamatergic or GABAergic pedunculopontine nucleus neurons²³. There is also a modest loss of glutamatergic neurons in the intralaminar nuclei of the thalamus and the basolateral amygdala^24,25. Of note, GABAergic neurons, regardless of where they are, appear to be resistant to Lewy pathology^20,26. Although nigral dopaminergic neurons have been the primary focus of PD research, they are not the first affected. Instead, neurons in the dorsal motor nuclei of the medulla oblongata, raphe nucleus, and locus coeruleus of the brainstem, as well as in the anterior olfactory nucleus, succumb first^27,28. The loss of nigral dopaminergic neurons is followed by degeneration of neurons in the transentorhinal region, motor and sensory cortex, and prefrontal cortex¹⁷.

Amyotrophic lateral sclerosis.

Motor neurons (MNs) in spinal cord and in brainstem, as well as upper MNs in the motor cortex, are selectively vulnerable in amyotrophic lateral sclerosis (ALS). Among MNs, fast-fatigable MNs are particularly vulnerable in ALS, while slow MNs are least vulnerable^19,29–31. Furthermore, the motor neurons of Onuf’s nucleus, as well as the oculomotor, trochlear, and abducens nerves, remain largely unaffected by cell loss even at late disease stages^19,32–34. The degeneration of MNs in ALS is often initiated in, and progresses from, the lower to upper spinal cord, followed by loss of upper MNs in the cerebral cortex, although there is considerable variability among patients^17,35.

Frontotemporal lobar degeneration.

Frontotemporal lobar degeneration (FTLD) affects the pregenual anterior cingulate cortex (ACC), extending back to midcingulate cortex, whereas AD involves posterior cingulate regions and spares the ACC. ACC and frontal insula deficits best differentiate behavioral variant (bv) FTLD from AD^36,37. Large bipolar spindle-shaped von Economo neurons (VENs) in layer Vb of those regions have been shown to represent an early target in bvFTLD but not in AD. In bvFTLD, a 69% reduction in VENs was found after controlling for neighboring layer 5 neuron loss³⁸. This VEN selectivity was seen even in patients with early-stage disease. In contrast to bvFTLD, late stage AD (Braak stage VI) showed no selective loss of ACC VENs. Tangles have yet to be seen in VENs³⁹. Pick’s disease is a rare cause of FTLD in which pyramidal neurons in the hippocampus and granular neurons in the dentate fascia are particularly vulnerable⁴⁰.

Huntington’s disease.

Huntington’s disease (HD) selectively affects medium spiny GABAergic neurons (MSNs) in the striatum, whereas large aspiny cholinergic interneurons and other striatal interneurons that express parvalbumin, calretinin, or nitric oxide synthase are relatively spared from degeneration in HD^31,41,42. Even within MSNs, those neurons expressing D2-type dopamine receptors, metenkephalin, or neurotensin are particularly vulnerable in HD, whereas MSNs expressing predominantly D1-receptors, substance P, and dynorphin are relatively spared in the early stages of the disease^31,41,43,44.

Primary and secondary selective neuronal vulnerability

There are potentially two types of selectively vulnerable cell types: those affected in the initial stage of the disease (primary vulnerable cells) and those affected later, in regions where the pathology has spread to (secondary vulnerable cells). For AD, PD, and ALS, the distribution of pathology at different stages of the disease has been mapped out using cross-sectional, postmortem immunohistochemistry analysis^27,45–48. Pathology-mapping studies suggest that disease proteins accumulate in regions of primary vulnerability and spread (propagate) to regions of secondary vulnerability along anatomical connections^48,49. Numerous studies suggest that the propagation of pathology, especially for AD and PD, occurs through a protein-templating mechanism whereby conformational ‘seeds’ of a pathological protein are transmitted from cell to cell in a prion-like manner^50–54. Whether prion-like proteins propagate in the same way in the human brain has been difficult to validate, but the fact that affected regions are connected supports the idea. It will be interesting to determine whether the physiological factors that make a cell vulnerable to accumulating pathological proteins in primary areas are the same as those in cells in secondary areas that develop pathology later as a result of propagation.

conformational strains and selective vulnerability

One of the intriguing features of neurodegenerative diseases that are caused by the same protein is the observation of clinical diversity. For example, the ‘4R tauopathies’ (which include progressive supranuclear palsy, corticobasal degeneration, and argyrophilic grain disease) are all caused by the accumulation of a form of tau protein that contains four microtubule binding domain repeats (4R tau). The tau protein in 4R tauopathies can take on different conformations, impact different neuronal and non-neuronal cell types, accumulate in different areas of the brain, and cause different clinical manifestations^55–57. Thus, the conformation of a given protein has been postulated to dictate the patterns of cell pathology, progression rate, and regional and neuronal vulnerability, but the basis of this structure-driven cellular vulnerability remains unknown. Recently, sophisticated techniques such as cryo-electron microscopy have been applied to elucidate the ultrastructure of both Aβ₁₋₄₂ and tau filaments^58–60, which, when combined with cellular vulnerability studies, may help explain why diseases such as the 4R tauopathies are so clinically diverse. Identifying whether the formation of particular strains is determined by genetic variants associated with particular cell types may help explain how conformational strains originate in the first place.

Potential mechanisms of selective neuronal vulnerability

Insight from genomics.

With whole-genome expression studies and the development of tissue-specific and cell-type-specific expression databases, the pathogenic loci identified by genetic analysis can be used as seeds to identify other genes that show similar expression patterns, and these can then be cross referenced with gene ontology databases to pull out networks of genes with related functions and expression profiles^61,62. These networks can be systematically investigated to identify other loci genetically involved in disease pathogenesis. As examples, the TREM2 module of co-expression in response to Aβ deposition in transgenic mice includes many other AD genetic risk loci that are largely expressed by microglia⁶³. Similarly, PD Mendelian and risk loci are involved in mitophagy (Table 1)⁶⁴. For several neurodegenerative syndromes, many loci map to particular biochemical pathways and are expressed in particular cell types (Tables 1 and 2). Identifying genetic loci can lead to identification of critical and intrinsic pathways that are close to failure in the normal brain and that fail in the diseased brain, for example, the endosome–lysosome system in diseases in which pyramidal neurons are lost or RNA metabolism in motor neurons (Table 1).

Table 1 |.

Risk loci and pathway implicated for the major neurodegenerative diseases

Disease	Gene associated with risk or disease-causing mutation	Pathways
AD	APP, PSEN1, PSEN2, APOE, CR1, CLU, BIN1, ABCA7, INPP50, CD2AP, EPHA1, MS4A6A, PICALM, CD33, HLA, PTK2B, SORL1, SLC24A4, DSG2, MEF2C, NME8, ZCWPW1, SP11, FERMT2, CASS4, TREM2, ABI3, PLCG2	Lipid metabolism Innate immunity Endosome-lysosome Ubiquitin proteasome
PD	SNCA, PKRN, PINK1, DJ-1, LRRK2, ATP13A2, PLA2G6, FBX07, VPS35, DNAJC6, SYNJ1, DNAJC13, VPS13C, RAB39B, GBA, NUCKS1, ITPKB, SIPA1L2, ILR2, TMEM163, SCNA3A, STK39, SATB1, NCKIPSD, ALS1, CHMP2B, MCCC1, TMEM175, FAM200B, FAM47E, ANK2, ELOVL7, ZNF184, HLA, KLHL7, CTSB, MICU3, SORBS3, SH3GL2, FAM171A1, BAG3, DLG2, MIR4697, OGFOD2, GCH1, TMEM229B, GALLC, CCQ7, ZNF846 TOX3, ATP6VOA1, MAPT, SYT4, LSMT, DDGK1, COMT	Endosome-lysosome Inflammation (adaptive immunity) Mitophagy Dopamine metabolism Vesicle fusion
ALS	C9orf72, TARDBP (TDP-43), SOD1, FUS, KIF5A, DCTN1, MATR3, TIA1, CHCHD10, VCP, SQSTM1 (p62), OPTN, UBQLN2, TBK1, CCNF, MOBP, SCFD1, SARM, UNC13A, C21orf2	Axonal transport Mitophagy DNA and RNA metabolism Autophagy and ubiquitin proteasome Toxic aggregation
FTD	C9orf72, GRN, MAPT, CHMP2B, CHCHD10, VCP, SQSTM1 (p62), OPTN, UBQLN2, TBK1, CCNF, HLA, TMEM106B, CTSC	Endosome and lysosome Autophagy and lysosomal pathway Mitochondrial damage Toxic aggregation Inflammation (adaptive immunity)
HD	HTT, MSH3, MTRNR2L2, DHFR	DNA mismatch repair

Bold font indicates Mendelian genes, italics indicates risk loci, and bold italics indicates that the locus appears in both categories. FTD, frontotemporal dementia.

Table 2 |.

Regions and neurons vulnerable in neurodegenerative diseases

Disease	Protein aggregates	Early-affected regions	Early vulnerable neurons
AD	Aβ42, Tau	LC, TEC, EC, BF, HP	Pyramidal neurons in EC-II & HP-CA1; cholinergic neurons in BF, noradrenergic neurons in LC
PD	α-synuclein	OB, DMV, SNpc	Dopaminergic neurons
ALS	TDP-43, SOD1, FUS, DPRs	MNC, SC, BS	Fast-fatigable motor neurons
bvFTLD	Tau, TDP-43, FUS	ACC, FI	VENs
PiD	Tau	HP, DG	Pyramidal in HP, granular neurons in DG
HD	Huntingtin	ST	MSNs

PiD, Pick’s disease; Aβ₄₂, Aβ peptide (1–42); Tau, microtubule-associated protein tau; TDP-43, TAR DNA-binding protein 43; FUS, RNA-binding protein fused in sarcoma; DPRs, dipeptide repeat proteins related to C9orf72; TEC, transentorhinal cortex; BF, basal forebrain; EC-II, entorhinal cortex layer II; CA1, Cornu Ammonis area 1 of hippocampus.

Mechanisms that have been linked to vulnerable cell populations.

In the aging brain and in the context of external stressors (for example, peripherally produced cytokines), several pathways can fail, leading to neurodegeneration⁶⁵. Why specific cells (Table 2 and Fig. 1) are selectively vulnerable to the breakdown of pathways that are critical in both vulnerable and resistant cells is not known. Explanation as to why will only be possible once neuronal subtypes have been better defined at a molecular level and we have a deeper understanding of their physiology.

Fig. 1 |. Regions and neurons that are vulnerable in neurodegenerative diseases.

Early-affected regions in different neurodegenerative diseases are indicated by different colors. LC, locus coeruleus; HP, hippocampus; OB, olfactory bulb; DMV, dorsal motor nucleus of the vagus; MNC, motor neocortex; SP, spinal cord; BS, brainstem; FI, frontal insula; DG, fascia dentata of the dentate gyrus; ST, striatum.

Protein supersaturation/metastable subproteome.

Cells go to great lengths to maintain proteins in a soluble state, as protein aggregation is associated with a wide variety of human diseases. The proteins most prone to aggregation are those whose cellular concentrations are high relative to their solubilities, i.e., proteins that are supersaturated. During stress, aging, and neurodegenerative disorders, cellular homeostasis of intrinsically supersaturated proteins becomes dysfunctional^66–70. Proteins impacted in AD, PD, ALS, and HD are collectively known as the ‘metastable subproteome’, and they aggregate as plaques, tangles, Lewy bodies, and intracellular inclusions. Metastable proteins are inherently supersaturated, especially in neurons compared to in astrocytes and microglia, which may contribute to neuronal-specific vulnerability. Consistent with this, gene duplication events and haplotypes that lead to increased gene expression are associated with increased risk of disease⁷¹. While it is not known whether particular cell types within a population express different levels of a metastable protein, expression levels do differ between primary and secondary affected regions of the AD brain, suggesting a causal link between vulnerable cell populations and protein supersaturation⁷⁰. It is notable that many risk alleles include genes involved in either lysosomal or ubiquitin proteasome system (UPS) function, which can directly affect the levels of supersaturated proteins. Thus, genetic risk data is consistent with the view that protein concentrations are critical, and being close to saturation puts a neuron at risk.

Protein homeostasis.

As mentioned previously, it seems likely that for each protein prone to misfolding, certain types of neurons are more affected by how that protein disrupts cellular protein homeostasis networks, and this may contribute to their vulnerability to a particular neurodegenerative disease²⁹. Protein homeostasis genes are altered in pretangle-bearing neurons as neurofibrillary tangle (NFT) pathology spreads through the brain^72,73, indicating disrupted protein homeostasis before clinical symptoms in AD. A gene co-expression analysis revealed that protein trafficking and clearance mechanisms, including specific branches of the endosomal–lysosomal systems and UPS, play a particular role in maintaining the homeostasis of the metastable subproteome associated with AD⁷⁴. Moreover, a transcriptome-wide microarray analysis across more than 500 healthy brain tissues from the Allen Brain Atlas revealed a quantitative correlation between the histopatho-logical staging of the disease and the expression patterns of the proteins that co-aggregate in amyloid plaques and neurofibrillary tangles, together with those of the protein homeostasis components that regulate Aβ and tau. Because this expression signature was evident in healthy brains, the analysis provided an explanatory link between a tissue-specific environmental risk of protein aggregation and a corresponding vulnerability to AD⁷⁰. Although these human-tissue-based findings are inherently correlative and cannot address causality, they can point the way for more detailed mechanistic approaches for understanding selective neuronal vulnerability in preclinical models.

In ALS and frontotemporal dementia, motor neurons and pyramidal neurons are vulnerable to overload of the UPS. Pyramidal neurons appear to be vulnerable to lysosomal failure in dementia with Lewy bodies as well as in frontotemporal dementia. It is interesting that cortical pyramidal neurons are susceptible to failure in either the lysosome system or the UPS. Both of these very general pathways are protein homeostasis mechanisms. Pyramidal cell death is associated with amyloid deposition and either tau tangle pathology (in AD) or with Lewy body (α-synuclein) pathology (in dementia with Lewy bodies). It is tempting to speculate that tangle pathology occurs when the UPS fails and that Lewy body pathology occurs when the lysosome system fails in the context of Aβ accumulation⁶⁵. In HD, cell-type-specific differences in the ability to maintain proteostasis (for example, lower autophagic degradation capacity in vulnerable striatal neurons than in resistant cortical and even striatal interneurons) may contribute to selective vulnerability to toxic huntingtin protein^75–78. Vulnerability may also involve neuron-specific combinations of dysfunction in cellular stress and proteostasis pathways, aggravated by advancing age, gene predisposition, and environmental factors.

Calcium homeostasis.

The lack of calcium-buffering proteins (for example, parvalbumin and calbindin D-28k) and disturbed cellular Ca²⁺ regulation is thought to play an important role in the vulnerability of neurons in AD, PD, and ALS. Neurons expressing calcium-buffering proteins are resistant, or less vulnerable, in the neocortex of AD patients^3,13,14, the spinal cord and brainstem of ALS patients, the spinal cord of a mouse model of ALS³, and the striatum of HD patients⁴². Gene expression profiling studies in rats and mice demonstrate that calbindin transcripts are enriched in VTA dopaminergic neurons as compared to in SNpc dopaminergic neurons^79,80. A similar expression pattern of calbindin was observed in human midbrain^19,81. The lack of calcium-buffering proteins parvalbumin and calbindin D28k may render human motor neurons particularly vulnerable to calcium toxicity following glutamate receptor activation⁸².

CA1 neurons are selectively vulnerable to degeneration in AD, whereas CA3 neurons are less vulnerable, and dentate granule neurons do not degenerate^17,83. Studies in transgenic mice (line 3xTg) suggest that excessive Ca²⁺ influx through L-type voltage-gated calcium channels (L-VGCC) may be one possible explanation for the selective vulnerability of CA1 neurons⁸⁴.

SNpc dopaminergic neurons, selectively affected in PD, share a set of traits that may underlie their vulnerability. They have long, highly branched axons with an extraordinary number of transmitter release sites. This combination of features—broad spikes, pace-making, low intrinsic Ca²⁺ buffering, and cytosolic Ca²⁺ oscillations—is what appears to distinguish vulnerable neurons, and these features lead to Ca²⁺ overloading and mitochondrial oxidant stress and damage^20,85. L-type Cav1.3 calcium channels, Kir6.2, K-ATP channels, and SK3 (the small-conductance calcium-activated potassium channel member) are molecular determinants for the electrophysiological differences between SNpc and VTA dopaminergic neurons, resulting in the selective vulnerability of SNpc neurons in PD¹⁹.

Among MNs, fast-fatigable MNs are particularly vulnerable in ALS. They exhibit the highest thresholds for excitation, firing rarely and in bursts²⁹. Vulnerable MNs express low levels of cytosolic calcium-buffering proteins and are subjected to large and fast calcium fluxes across intracellular organelles such as mitochondria and smooth endoplasmic reticulum⁸⁶. Large intracellular calcium fluxes and large numbers of neuromuscular junctions lead to particularly high energetic demands for function in these neurons²⁹. Consistent with this notion, calcium dyshomeostasis and/or mitochondrial dysfunction have been implicated in many pathogenic processes in neurodegenerative diseases^31,87–89.

Mitochondria and energy demand.

Large and long projection neurons in the EC and hippocampal CA1 are characterized by particularly high energy consumption and are vulnerable to decreased glucose and oxygen delivery through the vasculature and thus to energy deprivation⁹⁰. Similarly, SNpc dopaminergic neurons are estimated to form much larger axonal arbors and a higher number of synapses than VTA dopaminergic neurons, which may result in the pronounced redistribution of mitochondria to their axonal terminals and the tremendous elevation of their energy demand, as well as their susceptibility to insults jeopardizing the neuronal energy supply^91,92. Also, SNpc dopaminergic neurons were found to have a lower mitochondrial mass (i.e., higher level of mitochondrial DNA deletions) than VTA dopaminergic neurons, which might contribute to the selective vulnerability of SNpc over VTA^93–95. In addition, SNpc neurons in PD appear to be close to a catastrophic failure in mitochondrial function. Why this should be so is not clear, but one possibility is that dopamine synthesis and metabolism lead to oxidative damage to mitochondria; this may contribute to their sensitivity and may be the reason that genes involved in mitophagy lead to their selective loss^65,96.

Moreover, it has been suggested that projection neurons with sparsely myelinated axons would require prodigious energy expenditure to maintain axonal function and transport⁵¹. Such high energy demands would result in continuously high levels of oxidative stress and mitochondrial dysfunction that could increase neuron vulnerability to α-synuclein aggregation in PD. Consistent with this, projection neurons with long, thin axons that were only sparsely myelinated or unmyelinated were vulnerable to α-synuclein aggregates, whereas neurons with long but thickly myelinated axons with large diameters were resistant to the formation of such aggregates⁵¹. Intriguingly, a highly similar pattern can be noted in AD: sparsely myelinated temporal mesocortex exhibits tau aggregates first, whereas such pathology is last to appear in the heavily myelinated primary cortical fields⁹⁷.

ALS-vulnerable motor neurons are large cells with long axonal processes, which lead to requirements for a high level of mitochondrial activity compared to other neuronal groups⁹⁸. Motor neurons also have high perisomatic expression of the glutamate transporter protein (EAAT2) and very high expression of the cytosolic free-radical scavenging enzyme Cu/Zn superoxide dismutase (SOD1), which may render this cell group vulnerable in the face of genetic or post-translational alterations interfering with the function of these proteins⁸².

Striatal MSNs require high amounts of energy to maintain a hyperpolarized state, i.e., to remain electrophysiologically silent. This unique energy requirement of MSNs may contribute to their susceptibility to mitochondrial damage⁴². Another possibility for why MSNs are vulnerable to mitochondrial dysfunction may be associated with their low expression of the superoxide free-radical scavengers SOD1 and SOD2, which are enriched in resistant cholinergic interneurons⁴².

Neurotransmitters and neurotransmitter receptors.

The difference in several types of neurotransmitters and neurotransmitter receptors between subtypes of neurons has been proposed to explain selective neuronal vulnerability in neurodegenerative diseases. The expression profile and subunit composition of ionotropic glutamate receptors, especially NMDA receptors, may confer vulnerability in AD^9,99,100. Previous studies demonstrate that GluN1 and GluN2B subunits of NMDA receptors are more susceptible to the effects of aging¹⁰¹ and the progression of AD pathology¹⁰², suggesting that there might be intrinsic and differential expression profiles of NMDA receptor subunits between vulnerable and resistant neurons. Further characterization of the expression profile of ionotropic glutamate receptor subunits will help us better understand the selective neuronal vulnerability in AD. The expression of metabotropic glutamate receptor 2 (mGluR2) is increased in a similar pattern to both the regional and cellular subtype of neuronal vulnerability to degeneration and neurofibrillary alterations¹⁰³, while the differential expression of mGluRs (groups I and II) on specific neuronal populations might be responsible for selective neuronal degeneration in ALS^104,105. SNpc neurons may be more vulnerable due to their production of reactive oxygen species, from dopamine and its metabolites, that eventually kill neurons¹⁰⁶. Vulnerable neurons in the ventral SNpc exhibit increased expression of factors such as D2 dopamine autoreceptors, GIRK-2, lactotransferrin, and the dopamine transporter, coupled with a relative lack of neuroprotective elements, such as the dopamine vesicle transport protein and a number of trophic and growth factors²¹.

Differential expression of GABA_A and glycine receptors^29,107, as well as GluR2⁸², in ALS-resistant versus ALS-vulnerable motor neurons may render human motor neurons particularly vulnerable to hyperexcitation. The function of VENs in FTLD are as yet unknown, but the finding that VENs express dopamine (D3), serotonin (1b/2b), and vasopressin receptors is of interest because the neurotransmitters involved with these receptors are related to the behavior problems observed in patients with bvFTLD^108,109, suggesting that those neurotransmitter receptors might be responsible for the selective vulnerability of VENs in FTLD.

Similarly, glutamate excitotoxicity induced by activation of NMDA and AMPA receptors in the striatum is considered as one cause of the selective vulnerability of MSNs in HD^42,110. The NR2B subunits of NMDA receptors are highly expressed in vulnerable MSNs, while the NR2D subunits are highly expressed in resistant striatal interneurons^111,112. This differential expression pattern may explain the increased vulnerability of MSNs¹¹³. Moreover, increased expression of NR2B-subunit-containing extrasynaptic NMDA receptors in HD mouse striatum alters the balance between synaptic and extrasynaptic NMDA receptors activity, which may determine the vulnerability in HD^42,114,115.

Synaptic transmission happens when a neurotransmitter activates its receptors on the postsynaptic neuron. Synapse-related markers (for example, synaptophysin and synaptotagmin-1) are substantially reduced in NFT-bearing hippocampal CA1 pyramidal neurons and basal forebrain cholinergic neurons in AD patients, compared to normal non-NFT-bearing neurons in aged-matched controls¹¹⁶. Altered expression of genes related to synaptic transmission and synaptic vesicle transport is found in the entorhinal cortex and hippocampus of AD brains, which suggest that there are dramatic synaptic changes in vulnerable neurons affected early in AD¹¹⁷.

Aging.

Age is the major risk factor for both AD and PD, the two most prevalent neurodegenerative diseases. It is also a major inter-actor with and the second most important determinant (after the CAG-repeat length) of HD onset. Selective neuronal vulnerability may, in part, be a consequence of mature or aged neurons being close to different catastrophic cliffs, depending on their function, history of stress exposure, and genetic predisposition, and this may explain why certain inclusions and aggregates preferentially injure certain types of neurons⁹⁰. Aging has been shown to be associated with metabolic impairment, oxidative stress, perturbed neuronal calcium handling, and proteostasis dysfunction^17,118–122, which could trigger and accelerate Aβ, tau, and α-synuclein pathologies, thereby setting the stage for disease-specific neuronal vulnerabilities and transneuronal propagation of the proteinopathies.

Several other cellular and molecular changes that occur during normal aging could render neurons vulnerable to degeneration. Age-related reduction in calbindin expression has been implicated in the selective vulnerability of basal forebrain cholinergic neurons and entorhinal cortex layer II neurons in AD, dopaminergic neurons in PD, and MSNs in HD¹⁷, suggesting that alterations in cellular Ca²⁺ homeostasis, especially the expression of calcium-binding proteins, might play an important role in selective vulnerability during aging. Alterations in numerous neurotransmitter (for example, dopamine) and neurotrophic factor (for example, brain-derived neurotrophic factor or BDNF) signaling pathways occur during normal aging, and many such changes are amplified in neurodegenerative disease¹⁷.

It is increasingly appreciated that synapses are the most vulnerable compartments of neurons. Differences among synapses in terms of structure, metabolism, and signaling mechanisms might therefore be determinants of selective vulnerability. Analysis of cell-type-specific aging-altered genes reveals enrichment of pathways associated with synaptosomes in downregulated neuron-specific genes¹²³. Synaptic changes have also been identified in a study that examined the initial cell-type-specific transcriptional changes in a mouse model of ALS¹²⁴.

Other factors.

Several other factors have been implicated in selective neuronal vulnerability, including neurotrophins and neurotrophin receptors, protein kinases, and phosphatases. For example, a major downregulation of neurotrophin receptors (for example, TrkA, TrkB, and TrkC) and neurotrophin genes (for example, Gdnf, Ngfb, and Ntf4) in CA1 pyramidal neurons and basal forebrain cholinergic neurons associated with tau pathology has been reported in mild cognitive impairment or early-stage AD compared to controls^116,125. Moreover, the expression of pan-neurotrophin receptor p75(NTR) in basal forebrain cholinergic neurons has also been implicated as a vulnerability risk factor, probably due to the interaction with Aβ and with proapoptotic ligands that induce neuronal cytoskeletal abnormalities, leading to NFT formation in the basal forebrain cholinergic neurons of AD^126,127. NFTs are also impacted by the activity of a wide range of protein kinases and phosphatases that are differentially expressed. For example, PRKCB and MAPK1 levels have been shown to be higher in highly vulnerable CA1 pyramidal neurons¹²⁸, while DAPK1 is selectively activated in excitatory pyramidal neurons in the entorhinal cortical layer II of AD mice¹²⁹. In addition, downregulation of protein phosphatase 1 and protein phosphatase 2 subunit mRNAs was observed in AD compared with normal control basal forebrain cholinergic neurons and CA1 neurons¹¹⁶. Although the exact role of these factors in selective vulnerability is unclear, tau aggregation and downstream death-related signaling pathways would be likely targets.

Developmental factors have also been implicated in selective vulnerability. For example, some of the transcription factors expressed during neurogenesis continue to be expressed in dopaminergic neurons during adulthood, but the expression pattern is different between SNpc and VTA dopaminergic neurons, which may mediate the relative vulnerability of the former in PD. Other transcription factors that seem to be differentially regulated between SNpc and VTA neurons include PITX3^130,131, homeobox protein OTX2¹³⁰, and DCC¹³². In addition, ASCL1 and LMX1B seem to be required for specification of many brainstem regions that are susceptible to degeneration in early PD¹³¹. The role of these transcription factors in selective vulnerability, however, needs to be validated in the future.

Taken together, determinants of selective neuronal vulnerability might include intrinsic morphological, electrophysiological, and biochemical properties, as well as exposure to stressors such as aging. It should be noted that these determinants are not likely to be independent, and they may play synergistic roles in the selective loss of particular neurons in vulnerable regions of the brain.

Future directions

The majority of studies that define selective vulnerability are based on end-stage pathology, and their assessments are based on neuronal loss, structural or visible changes from immunostaining, or histochemistry in case series. While critically important, this is a crude assessment of cell function and vulnerability, as remaining neurons that appear normal could have vastly altered electrical properties, synaptic connections, gene expression patterns, or other changes that are poorly measured using histology-based techniques. The recent development of novel techniques such as single-cell profiling and the development of cell-type-specific human iPSC-derived neurons will help elucidate the molecular mechanisms underlying selective cellular vulnerability, including at early stages of the disease.

Single cell profiling.

Our ability to discriminate between neuronal subtypes is currently rudimentary, and therefore defining the gene signature of different neurons (especially at the single-cell level) will be critical for understanding why particular subtypes of neurons are vulnerable in neurodegenerative diseases. Techniques such as laser-capture microdissection and bacterial artificial chromosome-translating ribosome-affinity purification (bacTRAP) methodology, combined with microarray^19,116 or bulk RNA-seq^124,133,134, have been employed to create gene expression profiles of different types of neurons in various brain regions that are vulnerable to or resistant in AD and other neurodegenerative diseases. Moreover, single-cell RNA-seq^135,136 and single-nuclei RNA-seq techniques^137,138 have been successfully applied to identify microglia and to distinguish between different subtypes of excitatory and inhibitory neurons in human postmortem brain tissues based on their unique transcriptomics. One caveat of using transcriptomics in postmortem tissue to identify key vulnerability factors is that changes in mid- or late-stage disease may be an attempt at compensation and may not reflect the signaling milieu that underlies the initial cell-type-specific vulnerability in early-stage disease. Thus, comparison to early-affected regions with much less pathology would be beneficial.

Of note, a novel single-cell RNA-seq method, fluorescent in situ RNA-sequencing (FISSEQ)¹³⁹, has been developed to consider the cellular and spatial context of each RNA within a biological sample, which is absent in other single-cell or single-nucleus RNA-seq methods. If this method can be optimized for human fresh-frozen brain tissue—or ideally, fixed brain tissue—it will be a very powerful tool to identify the gene expression signature of vulnerable and resistant neurons in neurodegenerative diseases and will provide important insights for elucidating the mechanisms underlying selective vulnerability.

Exploring selective vulnerability using cell-type-specific human iPSCs-derived models.

Human iPSC-derived neurons have been extensively used to examine disease-relevant cellular and molecular phenotypes in a physiologically and genetically relevant context, potentially recapitulating the early stages of disease and allowing researchers to test therapeutic targets and small molecules in a human neuronal environment^18,140–142. With the advent of approaches to differentiate human iPSCs into specific cell types, such as dopaminergic^143–145, glutamatergic¹⁴⁶, and GABAergic neurons^147,148, we will be better able to explore which subtypes of neurons are vulnerable and why they are vulnerable to different proteins or different forms of the same protein, especially when these studies are combined with single-cell RNA-seq analysis and systems-biology approaches. However, a critical issue for these human iPSCs-derived models is the fact that without further manipulation (for example, overexpressing the progerin gene)¹⁴⁹ they do not represent the mature or aged state in which neurodegeneration takes place. New approaches, such as direct conversion¹⁵⁰, can recapitulate disease phenotype, cell type, and age, and they may have more utility for cell-based studies of selective vulnerability.

In conclusion, recent efforts to use genetics to identify cell-type-specific genetic risk factors and the cell populations affected, coupled with molecular approaches that can better define afflicted cells in the diseased brain, have advanced our understanding of the pathogenic processes associated with selective cellular and regional vulnerability. Considerable effort, however, is now needed to develop new techniques that can identify cell signatures in postmortem human brain tissue and identify relevant disease-associated pathways in complex biological systems.