Background
The inheritance of
APOEε4 allele is the major genetic risk factor for late-onset AD [
1]. The 3 alleles of human
APOE‐
APOEε2,
APOEε3, and
APOEε4—translate to 3 protein isoforms, APOE2, APOE3, and APOE4, which differ only in 2 amino acid residues at positions 112 and 158 [
2]. APOE is a 299-amino acid-long protein and a major component of low-density (LDL) and very low-density (VLDL) lipoproteins circulating in the blood. APOE is highly expressed in the brain, is secreted primarily by astrocytes, and its major role is to transport cholesterol and phospholipids as HDL-like particles in the interstitial fluid [
3]. The first and major regulatory step in the formation of brain HDL is the lipidation of APOE by ATP-binding cassette transporter A1 (ABCA1) [
4]. Properly lipidated APOE containing lipid particles in the brain affect synaptogenesis, play an important role in binding Aβ and lipid species, and facilitate their clearance through the blood-brain barrier and by microglia (reviewed in [
4,
5]).
An increased risk for AD in
APOEε4 carriers is undisputable: it is materialized in the earlier age of AD onset (approximately half of ε4-homozygotes will develop AD before age of 85, compared to only 10% of non-carriers), accelerated course of the disease, and more pronounced brain pathology [
6‐
8]. The molecular mechanisms mediated by
APOEε4 expression remain poorly understood, but a role for APOE4 in greater Aβ aggregation/deposition and neuronal toxicity, reduced clearance, and isoform-specific effects on neuroinflammation and neurogenesis have been demonstrated [
5,
9]. The protective effect associated with
APOEε2 is far from understood and ironically, compared to
APOEε3/4 or
APOEε4/4,
APOEε2/2, and
APOEε2/3 genotypes,
APOEε2/c are less represented in experimental and clinical studies. It is well established, however, that, excluding “oldest-old,” in the presence of clinical dementia and neuropathological criteria for AD, the effect of
APOEε2 is unaffected by age, it is independently associated with lower Braak neurofibrillary tangle stages, possibly fewer neuritic plaques, milder AD pathology, and less severe antemortem cognitive impairment [
10‐
13].
During the last decade, advanced sequencing technologies, improved mass-spectrometry platforms, and “omics” approaches have been constantly providing massive datasets comprising tens of thousands of genes, metabolites, and lipid molecular species with enormous potential to address questions relevant to disease pathogenesis and development, and possibly, drug discovery for neurodegenerative disorders [
14‐
20]. In this regard, the established association between lipid metabolism, Aβ generation, and its clearance from the brain [
21], as well as recent reports on the changes in transcriptomic profiles in the brain of AD patients and AD model mice [
22], has prompted further research using “multi-omics” assays. Their application is also motivated by the increasing evidence that changes in cholesterol and bilayer- and non-bilayer-forming phospholipids’ content play a role in the pathogenesis and progression of AD [
5]. The “multi-omics” approaches become particularly relevant considering the inheritance of the
APOEε4 allele as a major genetic risk factor of AD, earlier onset, and aggravated AD phenotype, as well as the protective effect of inherited
APOEε2 allele. Studies of brain lipidomes in AD model mice revealed alterations in phospholipid composition of the synaptic mitochondrial membranes, with cardiolipin (CL) content diminished during the early stages of pathology, connecting specific lipid changes to AD-like neurodegenerative process [
22]. Changes in the intracellular content of phosphatidylethanolamine (PE) as well as changes in its synthesis and metabolism have been associated with AD and other neurodegenerative disorders [
23,
24]. While the vast majority of lipidomics studies have compared lipidomes of AD brains to non-demented healthy controls, there have been no reports correlating changes in brain transcriptomics profiles to changes in lipid profiles, particularly in the context of
APOE genotype [
25‐
27]. Here, we present “multi-omics” profiling of postmortem AD brain samples from the inferior parietal lobule.
The inferior parietal lobule was chosen for two reasons: (1) neurofibrillary tangle formation occurs in a well-defined order, starting in the medial temporal lobe early in the disease and subsequently progressing towards the lateral temporal, parietal, prefrontal cortices, and finally the motor and sensory areas [
28,
29]. By contrast, in the earlier stages of the disease, amyloid deposits first affect the posterior association cortices and inferior parietal cortex; the areas of the medial temporal lobe might then be affected, but it is not very common in the early stages [
28,
30,
31]. Thus, the goal was to reveal differential changes in brain transcriptomes and lipidomes possibly associated with
APOE genotype that favors a delayed neurofibrillary tangle formation and slower amyloid deposition; (2) morphological and histochemical studies have shown that the initiation and progression of AD-related destruction inversely recapitulate primarily the progress of cortical myelination [
28]. In humans, myelination of axons in the prefrontal association areas and temporal and parietal lobes has the most protracted myelination which continues until the end of the sixth decade of human life. Late-myelinating neocortical areas at the same time are the most vulnerable to developing the pathognomonic lesions of AD consisting of neuritic plaques and neurofibrillary tangles [
32‐
34] (for a detailed review and extensive list of references, see Bartzokis [
35]). Longitudinal MRI data and high-throughput analysis studies, however, have provided evidence that initial, early signs of mild cognitive impairment (MCI), based on Clinical Dementia Rating, are associated with a similar rate of atrophy across all medial temporal lobe regions and inferior parietal lobule [
36,
37]. Moreover, comparing individuals without a diagnosis of MCI or AD but with cognitive complaints or cognitive decline, studies demonstrated involvement—detectable atrophy of posterior parietal lobule, more specifically the angular gyrus [
38,
39]. Very recently, a study examining the distribution and severity of tau-PET binding in cognitively normal adults with preclinical AD, as determined by positive β-amyloid PET, found that the precuneus and inferior parietal cortex were among the eight regions with the highest tau-PET binding. The findings were interpreted as consistent with preclinical involvement of the medial temporal lobe (MTL) and parietal lobe in AD [
40]. It is not known, however, if there are
APOE genotype-associated differences in transcriptional profiles in the inferior parietal lobule at those very early—almost impossible to investigate—or very late stages of the disease, brain samples available at the time of death, and if they can explain the differences in disease progression.
The results of our study demonstrate APOE allele-associated gene expression and lipid patterns at advanced stages of the disease. Weighted gene co-expression network analysis (WGCNA) revealed 14 co-expression network modules with a significant correlation to the APOE genotype. Utilizing Gene Ontology (GO) analysis with highly connected hub genes and lists of differentially expressed genes, we identified enriched GO terms associated with myelination, macroautophagy, regulation of macroautophagy, protein ubiquitination, and phosphatidylethanolamine biosynthetic process. The correlation between significantly changed lipid molecular species and differentially expressed genes indicated that differences in intracellular catabolic processes that deliver cytoplasmic components to lysosomes, as well as polyubiquitylation—implicated in proteasomal and lysosomal protein degradation—are among those underlying APOE allele-associated differences in AD pathology.
Discussion
The goal of this study was to reveal and analyze a differential association of APOE genotype with transcriptomic and lipidomic profiles in postmortem AD brain samples and to determine correlations. Since
APOEε2 allele is significantly related to a reduced disease risk, especially in people under the age of 85 years [
6‐
8], in groups with no statistical difference by age at death, we would expect
APOEε2/c postmortem brains at lower Braak stages and not as severe brain pathology. Thus, transcriptomic profiling of
APOEε2/c,
APOEε3/3, and
APOEε4/c postmortem brains would likely reveal changes associated with the corresponding
APOE allele.
Our study provides RNA-seq and mass-spectrometry lipidomics data derived from the inferior parietal lobule of
APOEε2/c,
APOEε3/3, and
APOEε4/c postmortem brains at known age of death and sex, at an advanced stage of AD, and allows interpretations in the context of gene expression and differences in brain lipidomes. We analyzed the changes in the gene expression using two different statistical approaches with their corresponding computational tools: WGCNA [
51] with an initial normalization step executed by DESeq2 [
54], and edgeR [
45]. WGCNA builds gene co-expression networks and reveals the relationship between biologically meaningful modules based on all transcripts excluding those indistinguishable from the sequencing noise, in all samples; edgeR performs RNA-seq profiling and identifies differentially expressed (DE) genes and molecular pathways between two or more biological conditions. In our study, lists of genes that belonged to individual modules within the network—WGCNA—or identified as DE genes based on the comparisons between genotypes in edgeR were further processed to reveal GO terms and categories and to demonstrate differences between
APOE genotypes.
We found that four of the significantly correlated modules of the network contained hub genes that are involved in GO terms with highly significant enrichment. The modules enclosed pathways with biological functions that are considered or suspected as impaired and associated with AD molecular pathology. In MEmagenta, MEgreenyellow, and MEturquoise modules, with highly positive correlations to the network, a number of GO terms remarkably overlapped with GO terms generated by genes found differentially upregulated by edgeR in
APOEε2/c samples when compared to
APOEε3/3 and
APOEε4/c (Figs.
1 and
2). These highly enriched GO terms were represented by pathways associated with proteostasis in ER, response to unfolded protein, intracellular protein, and organelle degradation—selective and basal autophagy, macroautophagy and its regulation, ubiquitination and ubiquitin-mediated proteasomal degradation, and SRP-dependent protein targeting.
Intracellular catabolic processes deliver cytoplasmic components to lysosomes through autophagic vacuoles. During the course of AD, autophagy and macroautophagy have a range of effects—deleterious as well as protective, depending on the stage of the pathologic process [
55,
56]. In recent years, the results of research aiming at a better understanding of proteostasis in neurons have identified interrelated regulatory mechanisms and posttranslational modifications that are part of the ubiquitin proteasomal system and autophagy-lysosomal pathway, operating in concert to achieve intracellular protein balance [
57]. Importantly, as discussed above, in a number of modules of the co-expression network, numerous highly significant GO terms are associated with macroautophagy, regulation of macroautophagy, protein ubiquitination, and proteasome-mediated ubiquitin-dependent catabolic process (Fig.
2b).
We found particularly interesting module MEcyan and the set of its genes—all snoRNAs. Functionally, box C/D and H/ACA snoRNAs play an important role in posttranscriptional modifications of mRNAs, impacting translational machinery and ultimately protein synthesis. C/D guide ribonucleoproteins to conduct the methylation of the 2′-OH group of ribose, while H/ACA rotate and convert C-5 ribosyl isomer of uridine into pseudouridine through a rotational break of C–C glycosidic bond and formation of an N–C one [
53]. The most well-studied box C/D snoRNAs—SNORDs—are located in two large, imprinted gene clusters at human chromosome region 15q11q13 (the SNURF-SNRPN domain) and at 14q32 (the DLK1-DIO3 domain) [
58]. They are expressed respectively only from the paternally and maternally inherited alleles. While there is evidence to consider the altered expression of SNORD115 and SNORD116, a primary cause of Prader-Willi syndrome, most recently those two and some other snoRNAs, has been implicated in the pathogenesis of schizophrenia [
59‐
63]. If and how exactly SNORDs are involved in altered mRNA splicing in the pathogenesis of schizophrenia is not clear yet, but none of those has been so far associated with AD. The biology and function of box H/ACA snoRNAs—SNORAs—have been extensively studied [
64], and their role in cancer is well established [
65]. Studies addressing the role of SNORAs in AD and results of research so as to compare our findings are not available. The role of snoRNAs in the pathogenesis of AD, however, will evolve as an important research topic, and we believe further research will definitely reveal important aspects of their biogenesis, structure, and mechanisms implicated in the pathogenesis of the disease.
There were significant and consistent changes in the total amount of lipids and numerous individual molecular species in 10 of the 14 lipid classes analyzed in this study (Fig.
3). In all of those instances, there was a significant decrease of phospholipids in
APOEε4/c vs either
APOEε2/c or
APOEε3/3 or vs both genotypes, like in PA, PC, SM, and ST. While in agreement with previously published alterations/decrease of phospholipids during the course of pathogenic processes in AD [
66], the differences between the lipidomes revealed in our study become particularly important since they can be correlated with the changes in the transcriptomic profiles of the exact same brain area. These correlations help to better understand the contribution of different
APOE allelic combinations towards differences in the disease progression and possibly AD pathogenesis. In this regard, particularly relevant are the metabolic and regulatory pathways that are involved in the maintaining of a healthy cellular proteome, a process collectively called proteostasis, through highly coordinated intracellular protein and organelle degradation. A fundamental challenge in proteostasis is the protection against misfolded or damaged proteins and protein aggregates that severely disturb cellular functions. If we consider the most significant differences in the transcriptomic profiles associated with
APOE2/c genotype vs
APOEε4/c and
APOEε3/3, we can link the enriched metabolic and regulatory pathways to the differences in proteostasis. Thus, we are suggesting a model explaining the protective effect of
APOEε2 allele in AD by the differences in some well-defined steps of the unfolded protein response, ER stress and ER-associated degradation (ERAD), and proteasomal and lysosomal intracellular degradation. We are assuming that transcriptional upregulation of genes, an important part of the pathways discussed below, facilitates sustained ER homeostasis that provides better protection against misfolded or damaged proteins and organelles. Such a model is supported by the following correlated
APOE genotype-associated lipidomic profiles:
First, key genes in the pathways that target proteins to the ER are differentially upregulated in
APOEε2/c: (a) in the co-translational translocation pathway
SRP68 and
SRP72, encoding the subunits of the SRP; (b)
SEC61 (all three subunits of the heterotrimeric complex),
SEC62, and
SEC63 at key regulatory steps of both co-translational translocation and SRP-independent posttranslational translocation pathways. Importantly,
SEC62 functions as a LC3-II receptor, and the interaction with LC3-II promotes the maintenance and recovery of ER homeostasis through clearance of select ER constituents by autolysosomes [
67]; (c) within the pathway of tail-anchored proteins, gene orthologs
WRB and
ASNA1 that target proteins to ER are significantly upregulated in
APOEε2/c samples, too. Similarly, in ER stress pathways and unfolded protein response activation, genes coding for proteins in all three key axes—transcription factor
XBP1,
HSPA5 (GRP ortholog), and
EIF2K3 (PERK ortholog), and transcription factor
ATF6—are differentially upregulated in
APOEε2/c AD samples; (d) a cellular pathway that recognizes unfolded/misfolded proteins in the ER and targets them for ubiquitination and subsequent degradation by the proteasome in the cytosol is called ERAD. Three of the key genes,
EDEM2,
EDEM3, and
OS9, are upregulated in
APOEε2/c samples. The genes are coding for proteins responsible for recognition of N-glycan structures, targeting and routing misfolded proteins for ubiquitination and subsequent degradation by the proteasome in the cytosol [
68,
69].
Second, LC3-PE conjugation is an indispensable step for autophagy-related genes (ATG) to exert their function in autophagy, and for that reason, the availability of sufficient PE is critical, too. The first step of phagophore formation is the conjugation of PE to the mammalian orthologs of yeast ATG8/LC3. Five of those mammalian orthologs MAP 1LC3A, MAP 1LC3B, GABARAP, GABARAPL1, and GABARAPL2 are upregulated in APOEε2/c brain samples. The subsequent generation of a covalent bond between ATG8 and PE requires a complex composed of ATG5-ATG12/ATG16L1; the genes of this complex are also upregulated in APOEε2/c.
Third, autophagy receptors (similarly to LC3-II/SEC62 complex) bind to cytosolic LC3 conjugated to PE and have a major role in selective autophagy, which is a process that regulates the abundance of specific cellular components [
70]. Autophagy receptors target protein complexes, aggregates, and whole organelles into lysosomes. Selective autophagy pathways, named after the cargo—aggrephagy, mitophagy, xenophagy, and pexophagy—can be ubiquitin (Ub)-dependent and Ub-independent. Four autophagy receptors—
p62,
NBR1,
OPTN, and
BNIP3—that can act on one or several pathways were upregulated in
APOEε2/c brain samples as common genes for both comparisons, against
APOEε3/3 and
APOEε4/c (
p62 only in
APOEε2/c vs
APOEε4/c). Numerous upregulated genes involved in the proteasome-mediated Ub-dependent protein catabolic process were significantly upregulated in
APOEε2/c brain samples, as well.
Fourth,
Beclin1 (
BECN1), acting as a molecular platform assembling an interactome which regulates the initiation of the autophagosome, is upregulated in
APOEε2/c brain samples. Although results from a previous study [
71] that demonstrated decreases in Beclin1 levels in AD midfrontal cortex gray matter still remain to be confirmed [
72,
73], numerous reports show the inhibition of Beclin1 interactome impairs autophagy and promotes AD-like pathology in in vitro and in vivo model systems [
71,
74].
Fifth, but not least, autophagy is highly dependent on the proper lipidation through PE conjugation of several proteins critical for phagophore formation, elongation, and autophagosome generation [
75‐
77]. Significantly lower amounts of PE in
APOEε4/c brains likely provide conditions for less efficient initiation of autophagy [
78,
79].
In the “
Results” section, we indicated that the comparison of
APOEε3/3 vs
APOEε4/c did not reveal differentially expressed genes at FDR < 0.05. While results of a study with a design and selection of groups as in our own have not been published so far, the differences in the expression profiles of
APOEε3/3 vs
APOEε4/4 and
APOEε3/4 (the latter two groups similar to our
APOEε4/c) were a goal of a study published in 2007 by Xu et al. [
27]. The study concluded that the expression pattern of APOE3/4 and APOE4/4 in the hippocampus of AD patients differed substantially from that of APOE3/3 AD patients. Since we have found no difference between the transcriptomic profiles of
APOEε4/c and
APOEε3/3 brain samples, there is an obvious discrepancy. The technologies used for transcriptomic profiling in both studies—SAGE, Xu et al. [
27], and NGS on Illumina platform, together with the methodology to analyze the differential gene expression—edgeR—in our study could be a reason for the differences; other explanations are possible as well: (1) stage of the disease—all our
APOEε4/c samples are at advanced Braak stage 6 vs stages 3–4 for the samples in Xu et al.; (2) brain area used for transcriptomic profiling—the inferior parietal lobule in our case vs MTL in Xu et al. While WGCNA analysis after clustering within
APOEε4/c group in our study was precluded by the insufficient number of samples, the questions raised by the discrepancy of the two studies should be addressed in the future, and hopefully, the answers would elucidate important aspects of the protective effect of
APOEε2 allele in AD.
The most recent study [
80], addressing
APOE genotype-associated differences in transcriptional profiles of postmortem AD samples, was published just a week before the submission of this article. While the most important difference with our study is the relative heterogeneity of their samples (combining traumatic brain injury and AD samples), the authors made very important conclusions that, to some extent, strongly support the results we are presenting here: regardless of the sex, the observed difference in transcription patterns for all brain regions analyzed including parietal cortex significantly correlated to the presence or absence of
APOE4 allele. Moreover, it should be noted that in the group of APOE4/4 brain samples, only a marginal, but statistically non-significant, difference between males and females was revealed.
Altogether, the differences in brain lipidomes and transcriptomic profiles associated with
APOE genotypes demonstrated in our study strongly support the idea that the efficiency of unfolded protein response, response to ER stress, intracellular proteasomal and lysosomal degradation, and better preserved mitochondrial function provides a molecular background for
APOE-associated differences in AD pathology, interpreted as driven by the
APOEε2/c group. In studies like ours, however, significant differences in “omics” profiles could raise a concern that the differences might be either due to age or AD brain pathology, including the integrity of RNA as a PMI-dependent variable. We present results based on the methodology for processing AD brain samples and statistical analyses of high-throughput datasets according to the widely accepted and rigorous standards [
81]. Since the age of patients at the time of death between the groups is statistically indistinguishable (one-way ANOVA), the age as a factor, most probably, does not play a significant role. To discern whether the differences can be clearly attributed to
APOEε2 or there is a significant contribution of AD pathology is a more difficult task. The difficulties are primarily associated with the availability and thus an insufficient number of samples of
APOEε2/2 and
APOEε2/c genotypes. The nearest consequence is that
APOEε2/c cases are overwhelmingly of lower Braak stages, and thus, within a relatively small pool of only several hundred of AD samples, a randomized, yet homogenous group of
APOEε2/c samples, age-matched to the other two groups—
APOEε3/3 and
APOEε4/c—and at advanced level of AD pathology is difficult, or impossible, to construct. An alternative explanation of the demographic structure of
APOEε2/c cases with samples predominantly in lower Braak stages would be that unlike
APOEε3/3 and
APOEε4/c,
APOEε2/c genotype confers genomic and likely epigenomic environment or promotes metabolic pathways that altogether have a protective effect and slow down the progression of AD and neurodegenerative pathology. The initial analysis of the
APOEε2/c group of samples included in this study did not identify differential gene expression between the subgroups based solely on Braak stage—2, 3, and 4 vs 5 and 6 (data not shown). Since
APOEε2/c genotype (excluding
APOEε2/4) is consistently associated with lower Braak stages and less prominent AD brain pathology, early activation and properly functioning autophagic-lysosomal degradation, improved myelination and slower myelin breakdown might explain the better clinical outcomes observed overwhelmingly in patients of
APOEε2/c genotype. With the relatively small sample size of the
APOEε2/c group, intrinsic difficulties in obtaining samples at the early stages of the disease regardless of the genotype and lack of experimental designs allowing functional studies using postmortem AD brain prevent immediate testing of this hypothesis. In a study aiming at gene expression profiles differentially associated with
APOE genotype at the time of death, there are additional limitations: for postmortem samples, age matched at the time of death and segregated by APOE genotype, the age when the cognitive decline was first recorded, and thus the duration of the disease remains unknown. It is known, however, that age is an important variable in the earlier stages of the disease, and significantly affects the progression, depending on the
APOE genotype [
7,
82] particularly if
APOEε2/c is included in the comparisons. Finally, while we are far from understanding the role of remote mechanisms above local interactions in the evolution of AD [
83], the pattern of metabolic brain alteration is likely a result of changes in the gene expression including brain areas far from MTL. Availability and transcriptomic analysis of samples of other brain areas would certainly strengthen the conclusions of a study like ours.
Despite the limitations, the results presented here support the future investigation to reveal the significance of improved myelination, more efficient autophagic-lysosomal degradation, response to ER stress, and reduced levels of intracellular toxic Tau oligomers in APOEε2/c individuals, ultimately slowing down the development and progression of the disease. While we still do not know if an impaired autophagic-lysosomal pathway and ER stress response, per se, is critical in prodromal AD, and how important relevant changes of the genome-wide regulatory networks are for AD progression, a systematic multi-omics approach, using postmortem AD brain samples provided by multiple AD Research Centers, will greatly facilitate the next steps towards identifying meaningful therapeutic targets.