Transcriptomic and epigenomic landscapes of Alzheimer's disease evidence mitochondrial-related pathways

Alzheimers disease (AD) is the main cause of dementia and it is defined by cognitive decline coupled to extra-cellular deposit of amyloid-beta protein and intracellular hyperphosphorylation of tau protein. Historically, efforts to target such hallmarks have failed in numerous clinical trials. In addition to these hallmark-targeted approaches, several clinical trials focus on other AD pathological processes, such as inflammation, mitochondrial dysfunction, and oxidative stress. Mitochondria and mitochondrial-related mechanisms have become an attractive target for disease-modifying strategies, as mitochondrial dysfunction prior to clinical onset has been widely described in AD patients and AD animal models. Mitochondrial function relies on both the nuclear and mitochondrial genome. Findings from omics technologies have shed light on AD pathophysiology at different levels (e.g., epigenome, transcriptome and proteome). Most of these studies have focused on the nuclear-encoded components. The first part of this review provides an updated overview of the mechanisms that regulate mitochondrial gene expression and function. The second part of this review focuses on evidence of mitochondrial dysfunction in AD. We have focused on published findings and datasets that study AD. We analyzed published data and provide examples for mitochondrial-related pathways. These pathways are strikingly dysregulated in AD neurons and glia in sex, cell-and disease stage-specific manners. Analysis of mitochondrial omics data highlights the involvement of mitochondria in AD, providing a rationale for further disease modeling and drug targeting.


Introduction
Mitochondria are double-membrane organelles with a specialized machinery required for many cell functions, including cell respiration and ATP production.To perform such functions, mitochondria carry a specialized genome that codes for 13 proteins of the electron transport chain as well as their own translation machinery.Mitochondria can also act as energy-sensing organelles, driving cellular processes such as free radical production and detoxification, calcium signaling, and apoptosis.These processes require specialized proteins whose genes are not included in the mitochondrial DNA (mtDNA).Hence, the mitochondrial proteome contains not only mitochondrial-encoded but also nuclearencoded post-translationally imported proteins [1].
Alzheimer's disease (AD) is the most common cause of dementia.The main risk factors for developing AD are age, genetic risk factors such as Apolipoprotein E (APOE) and Triggering Receptor Expressed on Myeloid Cells 2 (TREM2), and acquired risk factors such as metabolic comorbidities [2].AD has two main genetic presentations: familial AD (FAD, which accounts for 5 % of all cases) and sporadic AD (SAD, which accounts for nearly 95 % of all cases).Neuropathological hallmarks of AD are amyloid-beta aggregates and intracellular hyper-phosphorylated tau protein fibrils that lead to progressive neuronal loss.Many efforts to target such hallmarks have not had robust beneficial effects.Recently, mitochondria have been strongly implicated not only in AD pathogenesis but in aging.Understanding how mitochondrial function and dysfunction impact AD remains an effort in the field.
Mitochondrial-encoded gene expression has been shown to be differentially regulated in a tissue-specific manner [3].In energydemanding cells and tissues, such as neurons and brain tissues, mitochondria are essential for cellular maintenance and metabolic processes regulating cell fitness and survival.These post-mitotic cells without marked regenerative potential require intact mitochondrial structure and function for their activity and viability [4].During neurodegeneration, mitochondrial contribution to oxidative stress has been proposed as one of the processes involved in the onset and progression of the pathology.During aging, mitochondrial production of high-energy phosphates is attenuated, and the cellular oxidative stress is increased due to an imbalance between antioxidant components and the freeradical accumulation from mitochondria; therefore, mitochondria have been the focus of many aging theories [5][6][7].Oxidative stress has been investigated as a target to prevent cell death, neurodegeneration and inflammation [8][9][10][11] and understanding expression patterns and regulation of mitochondrial genes, transcripts and proteins is crucial for elucidating mechanisms of aging, and neurodegenerative diseases.
In this review, we outline current knowledge of the mitochondrialand nuclear-encoded genes that impact mitochondria in Alzheimer's disease (AD) and highlight a trend towards the development of integrative analysis and disease modeling in AD.

Mitochondrial genome
The mammalian mitochondrial genome is a compact, specialized circular double-stranded DNA of 16,569 base pairs (bp) in size and contains 37 genes that encode 13 protein subunits of the oxidative phosphorylation (OXPHOS) system, 22 tRNAs, and 2 rRNAs.It contains a noncoding region, the D-loop, (1124 bp in size) that acts as a promoter for both strands of the mtDNA.mtDNA strands include heavy (H) or light (L) strands, which have distinct composition, with H strand containing an increased abundance of purines (which have a higher molecular weight than pyrimidines), while the L strands contain more pyrimidines.mtDNA replication originates in the D-loop and is performed by different proteins than those for nuclear DNA replication.In human cells, polymerase γ (POLγ) is the replicative and reparative polymerase in mitochondria, first identified in 1970 [12], and it is encoded by the POLG2 gene on chromosome 15q26.Due to the highly oxidative environment, the absence of histones, and no evidence of nucleotide excision repair [13], mtDNA repair is limited compared to the nuclear DNA.Mutations in the mtDNA have been reported to occur in human diseases, such as mutations in the D-loop in cancer [14] and mutations in OXPHOS subunits increasing the susceptibility to metabolic diseases [15].Most mitochondrial diseases have strong neurological phenotypes and have been reviewed elsewhere [16].
There are hundreds of mtDNA copies in every eukaryotic cell, and mutations can be present in either all copies (homoplasmy) or in some of them (heteroplasmy).When tissues accumulate a certain ratio of mutant versus wild-type mtDNA, cells produce less ATP, reactive oxygen species (ROS) levels increase, and apoptotic cell death could occur [17].The symptoms of mitochondrial disorders vary from person to person, hampering clinical diagnosis.Mitochondrial diseases caused by mtDNA abnormalities are numerous and clinically diverse, and their progression is accelerated with aging as OXPHOS also becomes impaired [18].Many of these diseases have a late-onset and are characterized by progressive cell death.Interestingly, age-related diseases, such as AD and Parkinson's disease (PD), diabetes mellitus type 2 and cardiovascular disease have been associated with mitochondrial dysfunction and mtDNA abnormalities [19].
Mitochondrial single nucleotide polymorphisms (mtSNPs) have been investigated for potential associations with a variety of chronic conditions, such as neurodegenerative and cardiovascular diseases.mtDNA haplogroups are defined by the frequencies of specific mtSNPs (letter names of all haplogroups run from A to Z). Certain mtDNA haplogroups may affect OXPHOS, either predisposing or protecting from disease [20].Studies on mitochondrial haplogroups are essential for lineagetracing of human populations due to the maternal inheritance of mitochondria [21].Interestingly, the mtDNA haplogroups K and U confer protection against AD pathology in patients carrying Apolipoprotein E4 (ApoE4) alleles [22].The ApoE4 allele is the major risk factor for decreased longevity, cardiovascular disease, and sporadic AD (SAD).The study by Carrieri et al. suggested that the penetrance and effect of ApoE4 alleles depends on the interaction with specific mtDNA fingerprints.For example, multiple hot spots for base exchanges within the MT-ATP6, MT-ATP8, and MT-ND4 mitochondrial genes [23] were identified in a human population correlating to its haplogroups.It has been proposed that mitochondrial haplotype and APOE genotype should be taken into account to advance precision medicine and risk assessment in AD [24].

Mitochondrial epigenome
Nuclear DNA (nDNA) is organized in nucleosomes containing histone octamers and further compacted into chromosomes.In contrast, mtDNA is packed in mitochondrial nucleoids [25], which are the basic organizational unit of mtDNA and contain around 1.0-1.5 mtDNA molecules [26].The mitochondrial transcription factor A (TFAM) protein is the major structural protein in mammalian mitochondrial nucleoids and is responsible for the mtDNA packaging [27].In a mitochondrion, nucleoids that are highly compacted are believed to be related to mtDNA storage, while nucleoids with intermediate compaction (and therefore larger) are thought to be involved with active replication and translation [27].Indeed, mtDNA gene expression and replication are blocked by TFAM compaction in vitro [28].The TFAM-to-mtDNA ratio is considered to be essential in the regulation of mtDNA replication and mitochondrial gene expression in vivo [29].Furthermore, post-translational TFAM modifications, such as acetylation [30] and phosphorylation [31] within the high mobility group protein (HMGA) domains, can affect TFAM capacity to bind and compact DNA, which suggest that these modifications might be related to mtDNA transcription in vivo [29].It is reasonable to think of TFAM function as an epigenetic regulator of mtDNA.However, it is unclear whether the compaction of mtDNA by TFAM has an analogous role in gene expression as chromatin compaction in the nDNA.To further investigate the regulation of mtDNA replication and expression, DNA accessibility or protection assays (DNase I, MNase, ATAC) [32,33] and immunoprecipitation (ChIP) of mtDNA followed by sequencing [34] can be useful since they map the protein-DNA interaction landscape in human cells.These data can be correlated with gene expression and can provide insights into the effect of mtDNA accessibility on mtRNA abundance.To our knowledge, up to this date, such study has not been reported.
Nuclear epigenetic mechanisms include (hydroxy)methylation of cytosine residues, resulting in 5-(hydroxy)methylcytosine (5mC and 5hmC) on CpG sites, which regulate gene expression.While (hydroxy) methylation of nDNA is well described, (hydroxy)methylation of mtDNA has been controversial, mainly due to methodological limitations in accurately determining 5mC and 5hmC levels in mtDNA.The technological advances for the study of nDNA methylation, such as bisulfite sequencing, adapted to mtDNA, allowed for new methods to investigate mtDNA (hydroxy)methylation [35,36].The majority of studies support the evidence for (hydroxy)methylation of cytosine residues in mtDNA [37][38][39][40][41].Other studies also support the occurrence of mtDNA methylation and hydroxymethylation by showing the mitochondrial presence of DNA methyltransferases (DNMT1, DNMT3A, and DNMT3B) [41][42][43][44] and ten-eleven-translocation (TET1 and TET2) enzymes [45], respectively.A recent study used mass spectrometry to estimate 5mC in the mammalian mtDNA and revealed that the levels of 5mC makeup only 0.3-0.5 % of total mtDNA and that they do not seem to localize to any specific region, suggesting that mtDNA does not play a universal role in mtDNA gene expression or mitochondrial metabolism [36].Nonetheless, other studies agree that the regulatory D-loop region is one of the most methylated sites in the mtDNA [35,46].
In general, there is a consensus that mtDNA (hydroxy)methylation occurs at much lower levels than nDNA (hydroxy)methylation.Despite such low levels of mtDNA (hydroxy)methylation, many studies indicate that mtDNA methylation may influence mtDNA replication and transcription [47][48][49], and that exposure to environmental agents can affect the patterns of mtDNA (hydroxy)methylation [50].It has been shown that aging decreases mtDNA 5hmC in the mouse frontal cortex, accompanied by upregulation of mitochondrial genes, suggesting an epigenetic regulation of mtDNA transcription during aging [45].Moreover, patterns in mtDNA (hydroxy)methylation have been associated with various diseases, including different neurodegenerative diseases (reviewed in [51]).Despite considerable evidence implicating low (hydroxy)methylation of mtDNA at the D-loop region in both humans and animal models of neurodegeneration [42,44,[52][53][54], further investigation is warranted to elucidate the biological significance and causal relationship between mtDNA methylation and neurodegeneration.

Mitochondrial transcriptome
In the mitochondrial matrix, mitochondrial RNA (mtRNA) is present as polycistronic [3,55] precursors from both heavy and light chains [55] where 22 intercalated tRNAs are excised and liberate individual rRNAs and mRNAs.These individual RNAs undergo maturation, which comprises polyadenylation of the 3 ′ ends of mRNAs and rRNAs, and addition of the CCA trinucleotides to the 3 ′ ends of tRNAs.Mitochondrial transcription systems involve not only the mitochondrial-encoded genes, but nuclear-encoded RNA polymerases, transcription factors, endonucleases, aminoacyl-tRNA synthetases, RNA-modifying enzymes, structural components and biogenesis factors for the mitochondrial ribosome, among other auxiliary factors, as they have been reported in mitochondrial proteomic screenings [1,56,57].The small size of the mtDNA affords a depth of coverage to sample total mtRNA and characterize subtle events or differences between samples [3].
Studies of whole cell-transcriptomes have allowed the initial profiling of mitochondrial transcriptomes in different tissues through the Illumina Body Tissue Atlas [3], and the RNA Atlas [58].The abundance of some mitochondrial transcripts is higher in tissues with highenergy demand such as brain, liver, and heart.Moreover, the proportion of mitochondrial transcripts in the total pool of mRNAs in these tissues is up to 30 %.These data suggest that specialized cells and tissues exhibit a distinct mtRNA pattern, in terms of transcript abundance.Initially, the mitochondrial purification protocol coupled to RNA extraction was the approach for mitochondrial RNA sequencing.However, the presence of many nuclear-encoded transcripts indicated that these transcripts were co-purifying contaminants present in ER membranes and ribosomes.RNase A treatment of purified mitochondria leads to effective depletion of cytosolic transcripts, obtaining a "cleaner" mitochondrial fraction.RNA sequencing of RNase A-treated, purified mitochondria has led the identification of nuclear-encoded noncoding RNAs (ncRNAs) 5S rRNA, MRP, and RNase P in multiple studies [3,55,59].The presence of these transcripts in mitochondrial fractions depends on polynucleotide phosphorylase (PNPASE), a 3 ′ to 5 ′ exoribonuclease and poly-A polymerase, in the mitochondrial intermembrane space.It was shown that RNase P liberates and maturates tRNAs [59], and this process is crucial for mitochondrial function; whether RNase P and these bona fide transcripts are translated in mitochondria remains to be investigated.

Post-transcriptional changes of mitochondrial RNA
Mitochondrial transcriptome sequencing shows differential expression of mitochondrial-encoded genes and indicates that posttranscriptional mechanisms are key players in this regulation [60][61][62].To assess mitochondrial gene expression, it is important to characterize the mtRNA-binding elements that regulate it.Methods to sequence RNase-accessible regions of mtRNA have uncovered a landscape of posttranscriptional regulation of mitochondrial-encoded gene expression [63].
The mitochondrial translation machinery has been mostly preserved over the course of evolution and it is specialized for the translation of the OXPHOS components [64].The mammalian mitochondrial ribosome (mitoribosome) consists of a 28S small subunit and a 39S large subunit, which contain a catalytic 12S and 16S mt-RNA, respectively.Events such as aminoacylation of tRNAs [65], translation initiation, elongation and termination require several nuclear-encoded protein factors [64], such as TEFM, tRNA methyltransferases, RNases, mitochondrial ribosomal proteins (MRPs), among others.Mutations in these proteins cause strong mitochondrial phenotypes with characteristic clinical features, such as growth defects, muscle weakness (myopathy), and neuropathy [64].
More recent, deep sequencing studies of mitochondrial transcriptomes have demonstrated a role for the transcription elongation factor (TEFM) [66][67][68] not only for transcription elongation but for RNA processing in mitochondria.Constitutive Tefm knockouts exhibit embryonic lethality.Studies on conditional Tefm knockout mitochondria of heart and skeletal muscle indicated the presence of long unprocessed transcripts where mt-tRNAs were not excised properly coupled to mitochondrial dysfunction.The absence of TEFM leads to impaired RNA processing.Additionally, TEFM was demonstrated by proximity labeling (BioID) to bind to several other RNA processing factors, such as G-rich RNA sequence binding factor 1 (GRSF1), Fas-activated serine/threonine kinase (FASTK) protein family members FASTK, FASTK2 and FASTKD5, the mitochondrial poly(A)-polymerase (mtPAP), methyltransferases, RNA helicases, and the degradosome (SUPV3L1-PNPase) complex [66].The mitochondrial transcriptome has revealed complexity in its composition and regulation, even though transcription is initialized as polycistronic transcripts.The differential abundances of mature transcripts showcase the role of post-transcriptional processing, maturation, and degradation as regulatory steps in mitochondrial gene expression.The proposed interactions between TEFM and these RNA processing factors remain to be further characterized.

Mitochondrial and nuclear crosstalk
In eukaryotic cells, the nuclear and mitochondrial transcriptomes need to be coordinated in order to respond to environmental cues, cellular states and to meet metabolic demands.This mitochondrialnuclear crosstalk relies on the interplay between metabolite availability and epigenetic modifiers (reviewed on [69]).Moreover, mitochondrial genomes affect nuclear gene expression and epigenetics [43,70].The epigenetic modifications of nuclear-encoded mitochondrial genes are reported to be tissue-specific, indicating that the cross-talk between the nuclear and mitochondrial genomes is also tissue-specific [71,72].
The nuclear epigenome is affected by mitochondrial metabolism and metabolite availability.While mtDNA does not contain histones and the mtDNA epigenome is not subject to histone tail modifications, nuclear epigenetic mechanisms are well described and include DNA methylation and histone modifications, such as methylation, acetylation, phosphorylation, ubiquitination and sumoylation.These chromatin modifications are mediated by enzymes such as DNA methylases and TETs, histone methyltransferases and demethylases, histone acetyltransferases/ deacetylases, sirtuins, kinases and phosphatases.The majority of these histone-modifying enzymes require metabolic intermediates produced directly or indirectly by mitochondrial metabolism pathways.Examples of such cofactors include 2-hydroxyglutarate, α-ketoglutarate, acylcoenzyme A (Acyl-CoA), adenosine diphosphate (ADP), and adenosine triphosphate (ATP), flavin adenine dinucleotide (FAD), nicotinamide adenine dinucleotide (NAD+), S-adenosyl-homocysteine (SAH), S-adenosyl methionine (SAM) [69].In addition to chromatin modifications, RNA molecules of all classes, such as miRNAs, tRNAs, rRNAs, lncRNAs, and mRNAs, can also be covalently modified, in what is called the epitranscriptome [73] and are responsible for the regulation of mitochondrial transcription and function [65].
As previously mentioned, the nuclear-encoded DNA modifiers DNMT1 and TET1/2 translocate to mitochondria and are thought to induce mtDNA (hydroxy)methylation.Other nuclear-encoded epigenetic enzymes are also known to translocate to mitochondria, such as histone deacetylases 1 and 7 (HDAC 1 and 7) [74], lysine acetyltransferase 8 (KAT8) [75], lysine acetyltransferase 2A (GCN5) [76], and sirtuin deacetylase 3 (SIRT3) [77].While the nuclear function of these enzymes is well described, their mitochondrial function is still unresolved.An exception is SIRT3, known to translocate to mitochondria in response to stress, mediating deacetylation of metabolic enzymes [78,79].Mitochondrial SIRT3 is a tumor-suppressor factor in mammary tumors [80], and its deficiency has been shown to contribute to the metabolic syndrome in mice [81].Mitochondrial SIRT3 is known to mediate adaptive neuronal responses to exercise and glutamatergic signaling and to prevent neuronal death in models of epilepsy and Huntington's disease [82].In general, studies indicate that SIRT3 is important for mitochondrial health and quality control in response to stress [83].Neuronal SIRT3 also regulates mitochondrial biogenesis, the balance between mitochondrial fusion and fission, and plays a role in mitophagy, all of which are part of the mitochondrial quality control machinery and are relevant processes involved in the pathology of neurodegenerative diseases [84].

Mitochondrial dynamics
Mitochondria are highly dynamic organelles, which undergo coordinated cycles of fusion and fission to maintain their shape, cell distribution, and size.Fission is a dynamin-related protein 1 (DRP1)dependent process [85] that allows division of one mitochondrion into daughter mitochondria and has been classically linked to cell survival pathways such as autophagy-linked cytoprotection [86].A hallmark of apoptotic cell death is mitochondrial fragmentation, which may result from excessive fission or inefficient fusion.
Disrupted mitochondrial dynamics have been strongly implicated in neurodegenerative diseases, especially in AD [87,88].Several processes have been linked to mitochondrial function and dynamics, and the majority of these proteins are transported into mitochondria, having their genes encoded in the nucleus.This section will focus on processes that regulate mitochondrial and cellular functions, including oxidative phosphorylation, mitochondrial calcium signaling and autophagy, that are shortly presented below.

Oxidative phosphorylation
The mammalian oxidative phosphorylation (OXPHOS) system comprises five protein complexes.During respiration, electrons are transferred from the citric acid cycle products NADH and succinate through complexes I and II, respectively, to ubiquinone.Then, electrons pass through complex III and cytochrome c, terminating at complex IV.In this process, complex IV reduces O 2 to H 2 O, and the resulting electrochemical gradient generates a motive force driving ATP synthesis by the fifth complex, ATP synthase complex V.
The assembly process of OXPHOS complexes is tremendously intricate as mitochondrial complexes are multimeric and have subunits encoded both in the mitochondrial and nuclear genome (except complex II) [89].Exhaustive research on human complex I in the last years has identified assembly factors that, when mutated, are associated with pathological processes.
The interaction between OXPHOS complexes and their organization is still largely unclear.The field of proteomics has led to the discovery and characterization of supercomplex assemblies, composed of several complexes functioning together [90].Supercomplex assembly mediates the rescue of mitochondrial phenotypes associated with mutations of mitochondrial encoded subunits [91].Mutations in genes involved in supercomplex assembly, such as OPA1, COX7RP, HIGD2A, PHB1, SLP2 among others, have been described in skeletal muscle and brain tissue [92], and their potential role in the aging brain remains to be investigated.

Mitochondrial calcium signaling
Calcium is a signaling messenger involved in many cellular processes such as electrical activity, neurotransmitter release, cell adhesion and metabolism and has been extensively studied in neurodegeneration [93,94].Upon electrical or receptor-mediated stimulation, calcium concentrations rise to micromolar concentrations facilitated by opening/activation of specific ion channels such as voltage-gated calcium channels, glutamate and acetylcholine receptors [95][96][97][98], and by calcium release from the intracellular store.Disruption or the so-called "remodeling "of cytosolic calcium signaling has been described in AD [99]; it has been attributed to amyloid-enhanced calcium entry to neurons via the calcium homeostasis modulator 1 (CALHM1), to increased expression of ryanodine receptor (RYR) which leads to an increased release of intracellular calcium from the ER.According to the calcium hypothesis of AD, this general upregulation of calcium signaling stimulates mitochondrial calcium uptake and ultimately leads to apoptosis.Cytosolic calcium dysregulation AD has been reviewed elsewhere [96] and it is not within the scope of this review.Interestingly, in a recently-described form of cell death in neurons, ferroptosis, mitochondrial calcium uptake is increased [100], promoting an oxidant environment that can induce lipid peroxidation and membrane damage.
Following an increase in the cytosolic Ca 2+ , to prevent an overload with Ca 2+ , mitochondria have the capacity to rapidly take up the excess of cytosolic Ca 2+ .During increased neuronal activity of excitotoxicity, mitochondrial Ca 2+ overload can lead to several detrimental processes, including increased ROS production, decreased ATP production, and release of cytochrome c and apoptosis inducing factor (AIF).In AD, oligomeric forms of Aβ were shown to induce an increase in mitochondrial Ca 2+ uptake and Ca 2+ transfer from the ER stores, leading to neuronal cell death [101,102].Many key gatekeepers in the mitochondrial calcium signaling are pore-forming proteins that regulate calcium flux in a voltage-or ligand-dependent manner, such as voltagedependent anion channel 1 (VDAC 1), mitochondrial calcium uniporter (MCU) and the mitochondrial sodium/calcium exchanger protein (NLCX).The biological relevance of these channels depends on their function, which is dynamic and much of its fine-tuning occurs in a matter of milliseconds.For such fast and acute mechanisms, transcriptomics does not provide an adequate strategy to study calcium signaling.However, in chronic disease, calcium channels expression is affected at the extent of being detectable in transcriptomic data of AD patients and animal AD models.Dysregulation of intracellular calcium signaling has been extensively reviewed and proposed as a potential therapeutic opportunity [96,103,104].
VDAC1 regulates the transport of calcium between the outer mitochondrial membrane (OMM) and intermembrane space (IMS).It has been shown that Aβ binds to VDAC1 [105,106] and that disease progression correlates with increased VDAC1 expression and decreased mitochondrial complex V activity [107].Reduced VDAC1 expression is proposed to be protective against AD by increasing complex V activity and restoring ATP production [108].These data indicate that the VDAC family may represent a potential therapeutic target to preserve mitochondrial calcium transport in AD.
MCU is located in the inner mitochondrial membrane (IMM) [109] and mediates calcium influx to the mitochondrial matrix from the IMS [110,111].MCU opening is regulated by the calcium-sensitive subunits MICU1 and MICU2 [112][113][114].Mitochondrial calcium overload drives ROS production, as observed in neurodegeneration, making it an attractive target for potential therapies in AD [115][116][117][118].The NCLX is the main efflux mechanism for mitoCa 2+ [119] and a recent study has demonstrated that NCLX expression decreases during aging, in AD experimental animal models and in AD patients [120].These findings strongly suggest that calcium efflux dysregulation precedes neuropathological and clinical hallmarks in AD.These findings strongly suggest that mitochondrial calcium influx and efflux dysregulation precedes neuropathological and clinical hallmarks in AD, whether these processes are potential therapeutic targets in AD, remains to be investigated.

Autophagy regulates mitochondrial function
The highly oxidative environment inside the mitochondrial matrix is efficiently neutralized by mitochondrial enzymes, such as superoxide dismutase 2 (SOD2) or peroxiredoxins.However, in the presence of redox-active metal ions like Fe 2+ , formation of extremely reactive hydroxyl radicals is capable of modifying mtDNA, mtRNA, proteins and lipids [121].House-keeping processes, such as the unfolded protein response (UPR) and DNA repair, are involved in the maintenance of mitochondrial components.In case of excessive damage, mitochondria can undergo mitophagy, a selective degradation process based on autophagy.Autophagy is an energy-dependent process in which the cellular material is degraded by lysosomes and recycled.It involves a complex signaling network for which its components have been studied in neurodegenerative diseases [122].Autophagy is relevant for mitochondrial biology because: i) it is an energy-sensitive process, as it is dependent on the energetic sensor AMPK, coupling starvation and bioenergetic deficit with initiation of autophagy [123], ii) defective autophagy leads to accumulation of misfolded proteins [124], iii) mitophagy is involved in mitochondrial quality control.Deficiency of some autophagy-related genes (Atgs), such as ATG7 causes neurodevelopmental disease [125], and in some cases neonatal lethality [126].Moreover, components of autophagy and mitophagy pathways are involved in neurodegeneration (reviewed in [127]).Studies have shown a link between Parkin and PINK1 and mitochondria-induced inflammation in PD [128], and evidence of dysregulated Parkinmediated mitophagy in AD models and AD brains [129].

Mitochondrial damage-associated molecular pattern (DAMPs) release
Mitochondrial dysfunction and neuroinflammation are linked to neurodegeneration, and mitochondrial dysfunction can be an inflammatory trigger.Mitochondrial and cellular stresses can induce the release of mitochondrial components such as cytochrome c (CytC), mitochondrial transcription factor A (TFAM), cardiolipin, and mtDNA into the cytosol or the extracellular space, where they act as a damageassociated molecular pattern (DAMP), triggering inflammation [130,131].Once outside of the mitochondria, different sensors can recognize the mitochondria-derived DAMPs, depending on their location (cytosolic, lysosomal, extracellular) and the effector cell (neuron or glia).Danger sensors include the endosomal Toll-like receptor 9 (TLR9), the cytosolic NLRP3 inflammasome, AIM2 inflammasome, the cytosolic cyclic GMP-AMP synthase (cGAS)-stimulator of interferon genes (STING) pathway, and the extracellular TLR4 and receptor for advanced glycation endproducts (RAGE) [mtDNA reviewed on [132] and other DAMPs reviewed on [133].
During apoptosis, BAX/BAK oligomers form pores in the mitochondrial outer membrane (MOM) through which pro-apoptotic proteins, such as Cyt c or the apoptosis-inducing factor (AIF), are released into the cytosol which in turn activate caspases (9 and 3) and further downstream apoptotic cell death.The apoptosis-related BAX/BAK were classically thought to be the only oligomers to form large macropores in the mitochondrial outer membrane (MOM) and mediate mtDNA release [134], which would suggest that mtDNA would only be released during apoptosis.Nonetheless, VDAC oligomers have recently been shown to form pores in the MOM during oxidative stress, from which mtDNA can escape in live cells [135].As mentioned in Section 3.3, VDAC1 are gatekeepers in the mitochondrial calcium signaling.Moreover, VDAC1 mediate the transfer of nucleotides, metabolites, ions, and ROS between the mitochondrion and the cytoplasm, can mediate the release of CytC and are involved in the transport of cholesterol, fatty acid, and hemes [136].VDAC1 has also been shown to be important for neurite maintenance [137].The central role of VDAC in regulating mitochondrial function and serving as a gatekeeper for cell death, metabolism, survival, and inflammation makes it an attractive potential therapeutic target [138].
While chronic neuroinflammation is a known driver of neurodegeneration, direct evidence of the link between mtDNA release and other mitochondrial-DAMPs and AD is still lacking.mtDNA can trigger type I interferon (IFN) responses [132], and IFN is a driver of neuroinflammation and synapse loss [139].However, mtDNA in blood plasma declines with age [140], and studies on the presence of mtDNA in the CSF of AD patients show contrasting results [141,142], refuting mtDNA as a suitable biomarker for AD [142] (while being a potential biomarker for PD [143,144]).The presence of (free or vesicle-wrapped) extracellular mitochondria and their derivates in the CSF has been extensively reviewed elsewhere [145].The missing link between mitochondrial dysfunction, mtDNA release and neuroinflammation might lie in mitophagic processes, as discussed in Section 3.4.

The mitochondrial-cascade hypothesis in AD
AD is a multifactorial neurodegenerative disease that impacts learning and memory.AD has two main genetic presentations: familial AD (FAD, which account for 5 % of all cases) and sporadic AD (SAD, which accounts for nearly 95 % of all cases).Neuropathological hallmarks of AD are amyloid-beta aggregates and intracellular hyperphosphorylated tau protein fibrils, that lead to progressive neuronal loss.For years, the amyloid cascade hypothesis dominated the field, which focused on the strong genetic components of FAD, and on the prediction that Aβ causes AD, while removing Aβ or inhibiting its formation/aggregation could prevent AD.Clinical efforts targeting Aβ and aiming at reduced AD pathology showed little beneficial effects, e.g.aducanumab, while the majority of these clinical trials largely failed to present improvement in AD pathology.Possible reasons for these failures could be explained by an incomplete understanding of the complex pathophysiology of AD.As an effort to integrate and reconcile previously proposed hypotheses with the pathophysiology of SAD, the mitochondrial cascade hypothesis has been proposed, with mitochondria as an essential target to prevent the cascade of events that lead to AD [146].
The mitochondrial cascade hypothesis of AD encompasses mitochondrial involvement in both SAD and FAD and states that: 1) genetic inheritance defines baseline mitochondrial function, 2) inherited and environmental factors could contribute to age-associated mitochondrial changes, 3) when mitochondrial functional changes reach a critical threshold will influence AD pathology and symptoms.This hypothesis assumes that SAD and FAD share important mechanisms, while key differences prevail.Although, both SAD and FAD models exhibit mitochondrial dysfunction, the pathways in which mitochondria become dysfunctional may differ.For instance, in SAD, mtDNA are associated with age-related mitochondrial changes, while in FAD, changes in APP processing are linked to mitochondrial dysfunction.APP and Aβ were reported to disturb mitochondrial function and affect mitochondrial membranes.Moreover, Aβ excessive accumulation and aggregation are not only pathognomonic signs of AD, but also a consequence of brain aging.At the etiological level, the mitochondrial cascade hypothesis delineates FAD from SAD pathology and proposes that delaying brain aging will limit both the development and progression of AD.In addition, attempts to understand and target aging and neurodegenerative pathologies, led to several mitochondrial theories, such as mitochondrial free radical theory [147] and gradual ROS response hypothesis [148].All in all, the mitochondrial cascade hypothesis integrates aging, mitochondrial function and cognition in AD and supports mitochondrial-targeted therapeutics.

Transcriptomic and epigenomic evidence of mitochondrial dysfunction in AD 4.2.1. Bulk-RNA-seq
Transcriptomic studies of brain cells of AD patients and age-matched controls have found a strong association between AD pathology and mitochondrial and mitochondrial-related pathways.In a recent metaanalysis of 22 human brain transcriptomic datasets [149], it was proposed that there are seven consistently differentially expressed genes in all regions of AD brains: early response gene ZFP36L1 (ZFP36 Ring Finger Protein Like 1), RERE (Arginine-Glutamic Acid Dipeptide Repeats), PURA (Purine Rich Element Binding Protein A), OGT (O-Linked N-Acetyl Glucosamine Transferase), SPCS1, SOD1 (Superoxide dismutase 1) and NDUFS5 (NADH Dehydrogenase Ubiquinone Fe-S Protein 5).[149], Patel et al. also analyzed other neurological and mental health disorders such as schizophrenia, Huntington's disease, bipolar disorder, major depressive disorder and Parkinson's disease (PD), and propose that these seven genes are specific to AD.Three out of these seven reported AD-specific genes, NDUFS5, SOD1 and OGT are nuclear-encoded mitochondrial proteins.
The NDUFS5 gene is a subunit of mitochondrial complex I, and its expression is decreased in AD, indicating that ATP production could be affected in AD [150].SOD1 protein is also present in the mitochondria and contributes to the antioxidant mechanisms detoxifying ROS, which are mainly generated during the ETC process.The expression of SOD1 genes is downregulated in AD, indicating a potential cellular inability to prevent lipid peroxidation mediated by high mitochondrial ROS levels.The OGT gene codes for glycosyltransferase enzymes that posttranslationally modifies neuronal tau and amyloid β precursor protein (AβPP) [151].OGT expression is increased in AD and its activity has been associated with protection against tau and Aβ peptide toxicity, while its genetic ablation causes an increase in tau phosphorylation [152].Pharmacological inhibition of the O-GlcNAcase (OGA), a process which increases the pool of OGT product, prevents cognitive decline in AD mice harboring tau and app transgenes [153].These findings support the notion that preserving mitochondrial function and targeting these proteins might be beneficial to prevent the pathological aspects of AD.
A recent meta-analysis study investigating the gene expression profiles of 2114 postmortem samples of AD patients and age-matched controls [154] identified specific gene patterns associated with age-, sex-and pathology status.This meta-analysis demonstrated differences in the AD-associated gene expression changes between female and male subjects.We performed gene-set enrichment analysis (GSEA) on upregulated genes in males and females with Metascape and focused on genes that affect mitochondrial function [155].GSEA yielded as top hits an enrichment for mitochondrial-related pathways, such as electron transport, TCA cycle and mitochondrion organization in both sexes (Fig. 1A), while some other pathways such as synaptic transmission, neuron projection, morphogenesis and synapse organization were strongly enriched only in females.All in all, the disparity of differentially expressed genes (DEGs) between sexes is in line with the current view of AD, which disproportionally affects female subjects with stronger disease phenotypes and earlier age-of-onset [156].However, mitochondrial-related pathways are enriched in both sexes and may present a joint avenue for investigating disease mechanisms.A recent meta-analysis study investigating the gene expression profiles of 2114 postmortem samples of AD patients and age-matched controls [154] identified specific gene patterns associated with age-, sex-and pathology status.This meta-analysis demonstrated differences in the AD-associated gene expression changes between female and male subjects.We performed gene-set enrichment analysis (GSEA) on upregulated genes in males and females with Metascape and focused on genes that affect mitochondrial function [155].GSEA yielded as top hits an enrichment for mitochondrial-related pathways, such as electron transport, TCA cycle and mitochondrion organization in both sexes (Fig. 1A), while some other pathways such as synaptic transmission, neuron projection, morphogenesis and synapse organization were strongly enriched only in females.All in all, the disparity of differentially expressed genes (DEGs) between sexes is in line with the current view of AD, which disproportionally affects female subjects with stronger disease phenotypes and earlier age-of-onset [156].However, mitochondrial-related pathways are enriched in both sexes and may present a joint avenue for investigating disease mechanisms.
Transcriptomic analysis of AD brains and other neurodegenerative proteinopathies, such as progressive supranuclear palsy (PSP; a tauopathy) [157] reported a gene expression module for myelination in AD brains and not in PSP.Genes involved in AD such as PSEN1 (familial AD), or sporadic AD-associated risk genes, including BIN1 (Bridging Integrator-1), and CR1 reside within said module.BIN1 was upregulated in AD compared with other proteinopathies.BIN1 has been suggested to be a risk factor for AD from GWAS studies [158].Interestingly, BIN1 expression negatively correlates with oxidative phosphorylation and synaptic function [159].The relationship between BIN1 isoforms and mitochondria is that BIN1 is essential for organization of mitochondria and nucleus positioning [160], which may impair mitochondrial transport in the axon, promoting degeneration in AD.A recent study suggests that BIN1 alters Tau clearance and promotes the release of Tau-enriched extracellular vesicles by microglia [161].Another study has shown BIN1 functions in regulating neuronal activity and implicates a potential molecular mechanism in calcium signaling [162].Interestingly, it was shown that BIN1 isoforms are differentially expressed in neurons, astrocytes and microglia, and that neuronal and astrocytic isoforms are implicated in tauopathy [163].Whether specific BIN1 isoforms affect microglial phenotypes that are relevant for AD remains to be further dissected.All in all, the strict role of BIN1 in mitochondrial function in health and disease has not been properly interrogated and remains to be investigated.
The study by Patrick et al. [164] identified five modules of coexpressed genes that are related to microglia in two AD cohort studies.Two of such modules relate to β-amyloid, and another relates to tau pathology.Inside the tau pathology-related module genes such as vasodilator-stimulated phosphoprotein (VASP), MAP3K3 and TBC1 domain family member (TBC1D1) were enriched.It was previously shown that VASP is involved in focal adhesions and filopodia formation and interacts with mitochondrial proteins to regulate formation of such processes [165].Interestingly, MAP3K3 was shown to be present in a panel of microglial Aβ response proteins (MARPs) for advanced stages of AD in a time-resolved proteomic characterization of microglial proteome in AD models [166].TBC1D1 is a gene associated with obesity and regulated by Tat activating regulatory DNA-binding protein (Tardbp or TDP-43) [167].Microglial-specific knockout of TDP-43 promotes clearance of Aβ in various mouse models of AD [168] and AD patients.Whether the TDP-34-TBC1D1 axis regulates mitochondrial function or lipid metabolism in microglia has not been investigated.All in all, the study of Patrick et al. proposes new genes to be dysregulated in specific Fig. 1.Transcriptomes of Alzheimer's disease evidence dysregulation in mitochondrial-related pathways in a sex-, cell-, and disease stage-specific manner.(A) Gene set enrichment analysis (GSEA) with metascape: briefly, statistically enriched GO/KEGG terms and canonical pathways, p-values and enrichment factors were calculated and used to analyze upregulated genes of male and females from an Alzheimer's disease cohort (Wan et al., 2020).The heatmap is colored by p-value, gray cells indicate the lack of enrichment for that term in the corresponding gene list.(B) GSEA for selected pathways on different brain cell types in AD (Mathys et al., 2019).In brief, differentially expressed genes (DEGs) between single cell RNAsequencing of samples with no brain pathology of AD and varying degrees of pathology were analyzed with metascape.Selected p-values for pathways of interest were extracted and hierarchically clustered according to cell-type and pathology extent.First column: pooled pathology versus no pathology; second column, early pathology versus no pathology; third column: late pathology versus early pathology.The heatmap is colored by p-value, gray cells indicate the lack of enrichment for that term in the corresponding gene list.(C, D) Venn diagrams and circos plot that illustrate the overlap of DEGs between (C) early pathology versus no pathology comparison in excitatory neurons, oligodendrocytes and astrocytes, and DEGs between (D) late pathology versus early pathology.On the outside of the circos plot, each arc represents the identity of each gene list.On the inside, each arc represents a gene list, where each gene has a spot on the arc.Dark orange color represents the genes that appear in multiple lists and light orange color genes that are unique to that gene list.Purple lines link the same gene that are shared by multiple gene lists.The greater the number of purple links and the longer the dark orange arcs implies greater overlap among the input gene lists.(For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)pathology states of AD [168].
The group of Zhang et al. has constructed gene-regulatory networks from 1647 postmortem brain tissues from LOAD patients and nondemented subjects.These molecular networks host immune-and microglia-specific gene modules dominated by TYROBP, which is upregulated in LOAD [169].In a follow-up study with 1053 postmortem brain tissue samples of 125 individuals, specific trait-associated genes (TCGs) were identified.These TCGs are associated with the severity spectrum of dementia and the neuropathology of AD.The gene-trait correlation analysis identified TCGs that negatively correlated with dementia severity.These enriched pathways included oxidative phosphorylation, TCA cycle, respiratory electron transport, mitochondrial morphology and dynamics [170].In a recent study, it was proposed that ATPase H + Transporting V1 Subunit A (ATP6V1A) drives LOAD and that it could be a potential novel target for disease-modifying therapies.These findings were observed in human induced pluripotent stem cellderived neurons, transgenic AD Drosophila models and 364 LOAD donors [171].ATP6V1A dysregulation has been previously reported in a smaller cohort of AD, demonstrating overall downregulation and implicating oxidative phosphorylation in AD [172].
Bulk-RNA-seq is a powerful tool to infer region-, disease-stage, and tissue-specific gene dysregulations in AD, indicating that many mitochondrial-related processes could be targetable and studied further to prevent AD pathology.Integration of these large studies and their comparison with other cohorts of proteinopathies will enrich our view on mechanisms that drive neurodegeneration and AD in particular.

Spatial transcriptomics
The potential of bulk-RNA-seq to determine region-specific signatures remains limited due to its resolution (e.g.capturing whole-sample transcriptome per sample).For comprehensive understanding of tissue pathology, spatial transcriptomics methods aim to correlate highresolution tissue imaging coupled with RNA-seq employing spots of ~100 μm on tissue sections.Ortiz et al. [173] have generated an atlas of the adult mouse brain that supports the spatial identity of neurons from their single-cell RNA profile.Ortiz et al. [173] have generated an atlas of the adult mouse brain that supports the spatial identity of neurons from their single-cell RNA profile.This analysis yielded transcriptomic fingerprints that allow further neuroanatomical sub-region identification.These types of studies will provide more insights on gene expression and its correlation with specific cell types and brain area in disease models.
The study of the group of Bohr et al. yielded in a novel dataset that identifies mitochondrial protein Bok (Bcl-2 related ovarian killer) in both AD mice and patient brains [174].Navarro et al. compared their results with the single-cell atlas generated by Grubman et al. and observed significant downregulation of Bok in AD [175].Bok localizes in the endoplasmic reticulum (ER) and binds to inositol 1,4,5-triphosphate (IP3) receptors (IP3Rs), mitofusin 2 (MFN2) and dynamin-related protein 1 (DRP-1), which regulate fusion of the OMM.Bok deletion has been shown to result in specific mitochondrial phenotypes, such as fragmentation, without compromising cell sensitivity to apoptotic stimuli.
Employing spatial transcriptomics on brains of APP knock-in mice, coupled to in situ sequencing of AD brains in advanced Braak stages (V and VI, where there is extensive neocortical damage), Chen et al. demonstrated that microglia and astrocytes that are near amyloid plaques contain 57 inflammatory and lysosomal upregulated genes (Plaque induced genes; PIGs) and downregulated myelination genes in oligodendrocytes [176].
While spatial transcriptomic techniques provide spatial context to the molecular signatures throughout the brain, the resolution of most of these approaches does not reach single-cell level.Some fluorescent in situ sequencing methods (e.g., FISSEQ, STARmap) [177,178] have subcellular resolution, although the number of detectable genes is limited and require prior identification and selection of potential target genes.New technologies are improving the resolution of spatial transcriptomics (Slide-seq: spot size ~10 μm [179]; HDST: 2 μm beads [180]) and can provide multiomic information by combining spatial transcriptomics with protein profiling (DBiT-seq [181]).However, the difficult scalability of spatial transcriptomic technologies is currently a major limitation.

Single cell RNA-seq
Brain samples exhibit cellular heterogeneity, and the composition of cell populations in the brain may be altered in neurodegenerative disease.Single-cell RNA sequencing provides increased resolution to the brain transcriptome addressing such challenges.Conventional singlecell sequencing requires fresh tissue for the isolation of cells.Recently, single nucleus RNA-seq (snRNA-seq) was developed as a technology which can successfully capture single cell transcriptomes form frozen tissues.In snRNA-seq, nuclei are isolated from frozen samples and serve as a proxy to the whole-cell transcriptome of fresh samples of neurons [182] and microglia [183,184].Mathys et al. performed a single nuclei RNA sequencing study of 80,660 isolated nuclei from post-mortem brain of 48 individuals.This dataset clusters different cell types and subtypes such as excitatory neurons, inhibitory neurons, astrocytes, oligodendrocytes, and other cells.Additionally, these data follow the progression of AD in various cell types based on the pathological and clinical evidences of the severity of the disease and it is depicted as early pathology and late pathology.Comparisons between transcriptomes of control individuals with no pathology and age-matched individuals with varying degrees of pathological hallmarks (early-and late-pathology) yielded in multiple lists of DEGs between conditions.To exhibit the mitochondrial involvement in the development of AD we performed GSEA on different brain cell type-specific DEGs (Fig. 1B) [185].
Membership and enrichment analyses for excitatory neurons and for inhibitory neurons in the no-pathology versus pathology comparison demonstrated dysregulation of mitochondrial-related pathways, such as mitochondrial gene expression, oxidative phosphorylation and mitochondrial transport.Some of the dysregulated genes corresponded to the mitochondrial ribosomal proteins, mitochondrial complex I subunits such as NDUFA1, NDUFA5, and complex V components such as COX6C.It is noted by the authors that further comparison of subjects without pathology and those with early-pathology, as well as between those with early and late-pathology would provide insights in disease progression.Such comparisons yielded in enrichment of electron transport (e.g., NDUFAB1), autophagy (HSBP1, HSPA8, PINK1, DYNC1H1), inner mitochondrial membrane organization (mitochondrial contact site and cristae organizing system; MICO system), detoxification of reactive oxygen species (SOD1) for both excitatory and inhibitory neurons in early AD stages.Both types of neurons exhibited a stronger enrichment of pathways such as OXPHOS, autophagy and mitochondrion organization in late stages of disease when compared to early AD pathology.
Astrocytes and oligodendrocytes have important metabolic functions on which neurons depend on, such as glucose, lactate and acetate metabolism [186].In early stages of the disease, there was no detected dysregulation of mitochondrial-related pathways, while in late stages of pathology, pathways such as ATP metabolic process (OXPHOS subunits), detoxification (SOD and APOE), mitochondrion organization (complex I assembly subunits) and chaperone-mediated autophagy and aggrephagy (HSPA8, PARK7, DYNLL1) were altered and enriched (Fig. 1B and C).In the study of Mathys et al., glial cells such as astrocytes and oligodendrocytes were suggested to withstand metabolic insult better than neurons in early stages of disease than in late stages.The data set from Mathys et al. provides a snapshot of mitochondrial-related processes in the progression of AD in neurons and oligodendrocytes.
Microglia play important roles as the resident innate immune cells in the central nervous system (CNS) [187,188] and can adopt a wide range of phenotypes, depending on the stimuli and local microenvironment [189][190][191].Given these plastic characteristics of microglia, many activation states have been associated with disease progression, for instance disease-associated microglia (DAM), characterized by the dysregulation of the genes Apoe, Trem2, Tyrobp, Tmem119 and P2ry12 [192,193].The study of Mathys et al. led to the identification of 28 genes that overlap with reported DAM genes [192,194], however the number of profiled microglia is low compared to the other cell types, showcasing the limitation of single nuclei sequencing to study microglia.
A recent study by Gerrits et al. [195] employed a strategy where neuronal and oligodendrocyte-lineage nuclei were depleted to enrich for nuclei of less abundant cell types.In the microglial population, some subpopulations associated with amyloid-β, and others with tau pathology, suggesting that these microglia have different phenotypes.For the amyloid-β-associated populations, ITGAX, APRP1, MSR1 and APOE were enriched; for the phospho-tau associated microglia, CX3CR1, ADGRB3 and GRID2 were enriched.Some other subpopulations were characterized, such as inflammatory, proliferative and homeostatic microglia.The transcriptomic signature of metabolic pathways of the amyloid-β-associated microglia is similar to the inflammatory subpopulation (Fig. 2).On the other hand, these data suggest that the taurelated microglial population exhibit a metabolic signature more similar to homeostatic microglia.These data showcase that microglia are a plastic population that provide stimuli-specific responses.For instance, it was reported by Steffen et al. that only one third of IBA1positive microglia reported CX3CR1 expression in an APP/PS1 transgenic mouse model of AD with mitochondrial dysfunction.Furthermore, these CX3CR1-positive cells are recruited in close proximity to amyloid depositions [196], but mitochondrial dysfunction in these cells did not alter plaque size in mice at 100 days of age.Although studies like these illustrate the role of mitochondrial dysfunction of microglia in AD, further investigation in the metabolic changes of the different pathological microglial phenotypes (e.g.amyloid-associated and phospho-tau associated) will offer new targets for microglial state-specific AD therapeutic strategies.
snRNA-seq data has relevance since many mitochondrial proteins have a nuclear origin.Notably, mitochondrial-encoded transcripts are not present in these data.In bioenergetically active organs, mtRNA can contribute to up to a third of the total polyadenylated mRNA [3].However, mtRNA expression in brain has not been properly investigated in the context of neurodegenerative diseases yet, and whether mtRNA expression is differentially regulated in AD is not yet known.On a different experimental setup, mitochondrial-encoded RNAs have been found to be enriched in circulating extracellular vesicles (EVs) in AD individuals [197].Additionally, these EVs exhibit increase abundance of MT-ND1-6 mRNAs in AD.The authors further demonstrated that both Aβ aggregates and H 2 O 2 , promote the release of EVs in vitro, which are enriched in mitochondria and correspond to the status of the cell.The insights from this study have supported the use of circulating mtRNA as a diagnostic biomarker.In summary, the field of mitochondrial transcriptomics in AD has advanced during the last years (Table 1), and it represents a promising avenue to dissect mechanisms of disease in AD.

Epigenomics and multi-omics
Understanding the regulation of gene expression in AD is important to provide fundamental insights that link genetic variants with transcriptional and metabolic changes associated with AD.A recent multiomics study integrated epigenomic, transcriptomic, and proteomic techniques to generate comprehensive analyses of molecular pathways involved in AD [198].Their transcriptomic analysis revealed downregulation of genes related to cellular respiration and oxidation and upregulation of genes related to transcription and chromatin in AD.The proteomic analysis identified increased global levels of acetylated histone H3 lysine 27 (H3K27ac) and H3K9ac in AD, two histone marks related to active transcription.Chromatin-immunoprecipitation sequencing (ChIP-seq) analysis confirmed a higher number of H3K27ac and H3K9ac peaks with significant gains in acetylation in AD.The functional pathways associated with these disease-specific gains were GO terms related to transcription (for H3K27ac) and nucleic acid metabolism (for H3K9ac).Notably, DNA motif enrichment analysis of sites with H3K27ac or H3K9ac disease-specific gains showed enrichment for NRF1 and CTCF transcription factors.NRF1 is involved in the regulation of mitochondrial genes, cell cycle genes, and DNA damage response [199,200].Altogether, their analyses support a role for aberrant epigenetic activation in AD, with H3K27ac and H3K9ac as potential epigenetic drivers of AD, which is corroborated by other studies [201,202].
Genome-wide association studies (GWAS) have revealed several variants associated with AD [203][204][205][206].The majority of functional noncoding SNPs are predicted to influence disease progression via changes in transcription factor binding and regulatory element function [207], which are highly cell type specific.A recent study by Corces et al. employed bulk as well as single-cell chromatin accessibility assays to reveal brain-regional epigenomic and cell type specific heterogeneity [208].Single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq) clustering analysis from four brain regions of cognitively healthy individuals identified 24 distinct clusters, including different types of neurons (9 clusters with 30 subclusters) and glial cells (15 clusters).Linkage disequilibrium (LD) score regression using two relevant GWAS revealed a significant increase in per-SNP heritability for AD in the microglia peak set, but no significant enrichment for AD SNPs in the neuronal subclasses, which confirms the specific significance of AD SNPs to microglia and corroborates previous studies [203,209,210].
To identify the target genes of each GWAS locus, Corces et al. mapped the enhancer-centric three-dimensional chromosome architecture Fig. 2. Mitochondrial-related pathways are differentially regulated in microglia associated with Alzheimer's disease hallmarks.Heatmap depicting average expression of selected metabolic pathways in microglial subclusters in AD (Gerrits et al., 2021).The heatmap is colored by Log Fold change (LogFC).AD1 refers to amyloid-β-associated microglia.AD2 refers to tau-associated microglia.
using Hi-C library preparation followed by chromatin immunoprecipitation (HiChIP) [211] for H3K27ac.Using a catalog of putative diseaserelevant noncoding polymorphisms based on recent GWAS [203,204,212], these studies combined a tiered multi-omics approach, implementing a machine-learning framework, to annotate the functional effects of GWAS polymorphisms.Their models predicted a potential functional variant (rs1237999) that disrupts a putative oligodendrocyte-specific regulatory element upstream of PICALM, a gene previously implicated in AD by GWAS [213].Another variant, rs1237999, showed a 3D interaction with both PICALM and EED genes.A functional study has linked PICALM to autophagy and tau accumulation in vitro and in vivo [214].Consistent with the role of RIN3 in the early endocytic pathway, crucial for microglial function and of particular disease relevance in AD [215], the multi-omics approach identified a single SNP (rs10130373) that disrupts an SPI1 motif and communicates specifically with the promoter of the RIN3 gene in microglia.A

Table 1
Main mitochondrial-related findings in selected transcriptomic and epigenomic studies of AD models and cohorts.
Increase in per-SNP heritability for AD in microglia, but none for neurons.rs1237999: putative oligodendrocytespecific regulatory element upstream of PICALM.rs10130373: microglia-specific, SPI1 motif, communicates specifically with RIN3 promoter rs6733839 and rs13025717: predicted to two BIN1 microglia-specific regulatory elements.
recent functional study further corroborates Ras And Rab Interactor 3 (RIN3) role in regulating endosomal signaling and trafficking in AD [216].Furthermore, RIN3 is a guanine nucleotide exchange factor (GEF) for Rab5 subfamily, and the Rab GTPase network is involved in the regulation of autophagy [217,218].Their study further implicates two SNPs (rs6733839 and rs13025717) that are predicted to disrupt two microglia-specific regulatory elements upstream of BIN1, a previously studied gene for its implication in AD [219,220].A recent functional study suggests that BIN1 alters Tau clearance and promotes the release of Tau-enriched extracellular vesicles by microglia [161].Another functional study has shown BIN1 functions in regulating neuronal activity and implicates a potential molecular mechanism in calcium signaling [162].
Overall, GWAS combined with multi-omics approaches continues to confirm the significant role of microglia in AD.Although the work of Corces et al. annotated possible functions of noncoding variants in AD, further research is necessary to confirm and validate the role of such variants in a cell-specific manner in the context of neurodegeneration.Moreover, their single-cell chromatin accessibility analysis was based on cells isolated from cognitively healthy individuals.Performing singlecell multi-omics analysis in cells isolated from patients with dementia might provide further insight into the functional role of noncoding SNPs and possible dysregulated pathways.

Metabolomics
Strong indications of mitochondrial involvement in AD have been pointed out by metabolomic studies in animal models and AD patients [221].Metabolomics encompasses the measurement of 100's to 1000's of small molecules (<1.5 kDa) such as amino acids, fatty acids, lipids, carbohydrates, or other products of cellular metabolic functions.Metabolic signatures and their alterations will help provide a snapshot of disease state and underlying mechanisms.Recently, metabolomics has become widely used for different sample types such as brain tissue, blood, plasma, serum, saliva, urine, and cerebrospinal fluid (CSF) [222].Therefore, metabolomics is helpful to determine the systemic nature of AD.Metabolomic studies have shown metabolic perturbations in preclinical AD, emphasizing the potential to advance diagnosis and target nomination in early stages of AD.Numerous studies have shown consistent dysregulation of lipid metabolism and mitochondrial bioenergetic pathways in a plethora of AD sample types.One example is the study of Paglia et al., where mitochondrial aspartate metabolism was dysregulated in AD [223].This finding led the authors to suggest that crucial Acetyl-CoA shuttles entering mitochondria are defective in AD.Additionally, whether metabolome changes are secondary to defective mitochondrial function is a possibility and most likely both situations are intertwined.This topic remains to be further explored to obtain more mechanistic insights.All in all, metabolomics is a strong technology supporting the biomarker discovery platform in AD.

Lessons learned from preclinical and clinical studies on mitochondria-targeted AD therapeutics
In accordance with the mitochondrial cascade hypothesis, in this section we aim to outline the main mitochondrial studies that have been proven successful to improve cognition, neuropathological hallmarks or mitochondrial/bioenergetic health in AD models and to discuss the potential targets described in the previous section in the light of current literature.The drug discovery pipeline for AD is a long process that takes around 15 years from the drug discovery phase until the regulatory approval.Since most amyloid-targeting strategies did not manage to pass the clinical evaluation, rational design of compounds acting on the targets proposed in this study have the potential to accelerate AD drug development pipelines.

Preclinical studies
Several in vivo AD studies showed that mitochondrially-localized targets elicit beneficial responses on cognition, neuropathology, and mitochondrial health.Some examples of mitochondrially-localized targets and processes presented in Table 2 include: the TAR DNA-binding Protein 43 (TDP-43), beta-amyloid peptide, amyloid binding alcohol dehydrogenase (ABAD), NADPH Oxidase 2 (NOX2), Sigma non-opioid intracellular receptor 1, Dynamin-related protein 1 (Drp1), and mitochondrial respiratory complex I (MCI) and NAD + -consuming enzymes.Interestingly, amyloid-targeting studies reported recovery of mitochondrial health showcasing mitochondrial dysfunction secondary to amyloid-deposition.One should be careful when interpreting these data, that these studies were carried out in animal models expressing human mutant genes that drive FAD.All in all, the mitochondrial cascade hypothesis and the proposal of the so-called "mitochondrial medicines" are supported by current literature.

Nominated therapeutic targets
In order to provide an overview of current proposed mitochondrial targets for AD pathology, we surveyed the Agora database (https:// agora.adknowledgeportal.org/genes)for mitochondrially-localized protein products and compared them with the Human MitoCarta 3.0 [234].This search enabled us to find 80 mitochondrial targets with potential role in AD pathology (Supplementary Table 1).The Agora database contains a list of over 500 nascent drug targets for AD that were nominated by AD researchers.The list of nominated targets was collected by researchers from the National Institute on Aging's Accelerating Medicines Partnership in Alzheimer's disease (AMP-AD) consortium as well as other international research teams (Supplementary Table 2).
Mitochondrially-targeted antioxidant approaches have shown to improve AD behavioral and mitochondrial deficits [226,235], supporting the role of SOD1 as a target in AD as described previously by Patel and Mathys.MnTBAP is a small-molecule SOD1-mimetic.Up to this date, no preclinical or clinical study with this compound has been described.
Nuclear-encoded components of the respiratory complex I such as NDUFS5 and NDUFAB1, among others, were also proposed by Patel, Wan and Mathys.These specific targets have not been proposed in the Agora database (Supplementary Table 1).However, other subunits such as NDUFA10, NDUFA11, NDUFA13, NDUFA9, NDUFS1, NDUFS2, NDUFV1 are present in the Agora list.The function of respiratory complex I is NADH:ubiquinone oxidoreductase and its activity can be modulated by the NADH/NAD+ ratio and several small molecules.For instance, targeting the NDUFA1 subunit with the small molecule CP2 induces a stress response (caloric restriction-and exercise-mimetic) that shows to be a promising approach to restore mitochondrial health and prevent AD.

Clinical trials
In recent decades, an increasing number of clinical studies have focused on the translational effect of in vitro and in vivo animal AD model targeting mitochondrial functions.Several mitochondrialtargeted potential therapies are currently under investigation (Table 3), including mitochondrially-targeted antioxidants, Sirtuin-NAD activators and Nicotinamide riboside and cofactors, that will undoubtedly provide evidence to build upon the mitochondrial cascade hypothesis of AD.

Considerations
Each of the studies described in this manuscript have different cohort compositions and sample characteristics, which allow for distinct analyses.We have taken this into account when selecting study data to analyze.For instance, the meta-analysis from Wan et al., in which we performed GSEA for downregulated genes, employed a linear regression model where diagnosis, sex and other identified donor covariates were considered, making sure that the transcriptional changes are sex-and diagnosis-dependent.However, in that study, the type of sample (bulk-RNAseq) does not allow to obtain a cell type-specific transcriptomic signature.Several co-expression modules from Wan et al. are strongly enriched for cell-type-specific signatures, which raises the question of whether these changes are due to changes in cell proportions, such as neuronal cell death or gliosis (which occur during AD development and are increased in women).Sex-specific differences were also shown in the study from Mathys et al.The overrepresentation of distinct sub-clusters of excitatory neurons, astrocytes and oligodendrocytes for males and females was not due to disproportional cell contribution of particular individuals.All in all, two different strategies support the sex-specific transcriptomic differences in AD.
Black populations have a lower dementia prevalence, increased risk factors and greater cognitive impairment than white populations [236].It is important to mention that none of the studies we addressed in the review studied race-dependent effects on AD transcriptomes.The metaanalysis of Wan, the study of Mathys and the Gerrits paper focused on predominantly white donors, not allowing them to obtain insights into the effects of race in AD.Additionally, an approach to improve the insights on age, gender, race will be to perform further single-cell-level analyses with larger sample sizes and specific cohort compositions.snRNA-seq by definition does not contain mitochondrial-encoded transcripts.Targets proposed by the reviewed studies are relevant for mitochondrial function given that some of their protein products are post-translationally imported to mitochondria and others impact mitochondrial health by altering mitochondrial transport, organization, quality control, turnover, among others.However, mtRNA expression remains to be further investigated at the single cell level.Mitochondrial transcriptomics in the field of AD has not been thoroughly explored.Efforts to perform sub-cellular resolution transcriptomic analysis, such as single-soma RNA-seq [237] may prove to be capable to evidence changes in mitochondrial transcriptomes in AD.

Outlook
Multiple omics analyses identified mitochondrial-related genes and proteins involved in mitochondrial pathways as contributors to disease progression [238].Several studies [239,240] have pointed to mitochondrial dysfunction as a process that leads to neuronal loss, microglial activation, and the onset of the pathological hallmarks of AD.Whether these genes and processes are not only biomarkers but whether they represent therapeutic targets able to modulate AD etiology remains to be further investigated.Despite all progress made in the field of cellspecific mitochondrial gene expression and function, many areas and topics require further study: integration of multiple mitochondrial omics data sets.The access to rapidly evolving omics technologies and tools for analysis represents a novel avenue to delineate disease mechanisms in AD and in other neurodegenerative diseases as well.

Gene set enrichment analysis for differentially expressed genes in AD
Lists of DEGs corresponding to downregulated genes in males and females in AD reported by Wan et al. [154] were input into Metascape [155].Briefly, statistically enriched GO/KEGG terms and canonical pathways, p-values and enrichment factors were calculated and used to analyze upregulated genes.Hierarchical clustering was based on Kappastatistical similarities among their gene memberships.The heatmap is colored by p-value, gray cells indicate the lack of enrichment for that term in the corresponding gene list.Lists of DEGs corresponding to downregulated genes in males and females in AD reported by Wan et al. [154] were input into Metascape [155].Briefly, statistically enriched GO/KEGG terms and canonical pathways, p-values and enrichment factors were calculated and used to analyze upregulated genes.Hierarchical clustering was based on Kappa-statistical similarities among their gene memberships.The heatmap is colored by p-value, gray cells indicate the lack of enrichment for that term in the corresponding gene list.
Differentially expressed genes (DEGs) between single cell RNAsequencing of samples with no brain pathology of AD and varying degrees of pathology were analyzed with Metascape.Selected p-values for pathways of interest were extracted and hierarchically clustered according to cell-type and pathology.The values are depicted in the first column as pooled pathology versus no pathology; in the second column, as early pathology versus no pathology; and in the third column, as late pathology versus early pathology.The heatmap is colored by p-value, dark blue cells indicating the lack of enrichment for the term shown in the corresponding gene list, and increased enrichments appearing as white and further enrichment in red.
Hierarchical clustering of gene expression changes in microglia.The average expression of selected metabolic pathways of interest in microglial subclusters in AD [184] is represented by a hierarchically clustered heatmap with annotated subclusters based on kappa similarities.Briefly, genes were extracted if they were enriched in at least one subcluster.The heatmap was generated using the Complex Heatmap package in R, and is colored by log fold change (LogFC).Scripts and data used to generate the plots in this manuscript are available upon request.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table 2
Main mitochondrial-targeted preclinical studies and their effect in cognition and mitochondrial health.

Table 3
Clinical trials with a focus on mitochondria on AD.