Introduction

Cerebrovascular pathology is highly prevalent in different degrees across general population1, acting as a comorbidity in other types of dementia2,3. It is still debated whether cerebrovascular lesions add incrementally to the cognitive burden, or if vascular dysfunction acts synergistically with other neurodegenerative disorders4. Besides the shared risk factors5 that may concur for the development of more than one disorder, some evidence supporting a role for blood-brain barrier dysfunction6 and other vascular changes preceding Alzheimer’s Disease (AD) changes also exists7.

For example, neuropathological studies have shown that cerebrovascular pathology is a major risk factor for clinically diagnosed AD8. In fact, 80% of patients diagnosed with AD have some degree of cerebrovascular pathology9, even if they do not meet the criteria for mixed (i.e. vascular) dementia. Vascular dysfunction also appears early in AD10 and is associated with several hallmarks of AD, such as amyloid-beta deposition11 and tau pathology12.

This overlap of cerebrovascular and traditional AD pathologies is not exclusive to the late-onset form of AD but is also present in autosomal dominant AD13. Recently, NOTCH3 (Neurogenic locus notch homolog protein 3, the gene associated with the most common cause of inherited vascular dementia - CADASIL) has been associated with AD14,15, indicating that genes associated with cerebral Small Vessel Disease (cSVD) may also play a role in neurodegenerative dementias. Other studies have also implicated COL4A1 (Collagen, type IV, Alpha 1) and HTRA1 (High Temperature Requirement serine protease A1) in AD pathogenesis16, further supporting this view. Proteins coded by these genes can impact on other pathways that may be involved in AD pathogenesis, such as NOTCH3, that is cleaved by the presenilins that cleave APP17, HTRA1 on APOE (Apolipoprotein E) metabolism18, or GLA (Galactosidase Alpha) on autophagy, lysosomal function, lipid metabolism and inflammation19.

In this study, we aimed to find rare variants in genes associated with monogenic small-vessel disease in a cohort of Portuguese early-onset AD patients.

Methods

Subjects

In this study, we included 104 index early-onset AD patients recruited from a memory clinic in Centro Hospitalar e Universitário de Coimbra, Portugal, a tertiary reference hospital from the central region of Portugal. The diagnosis was performed according to the most widely accepted criteria for AD20 and supported by extensive characterization. Our clinical protocol includes (1) a complete and systematic clinical and neuropsychological evaluation; (2) a structural neuroimaging (computed tomography and/or magnetic resonance imaging (MRI)); (3) lumbar puncture to determine AD cerebrospinal fluid biomarkers, and/or amyloid PET (Positron Emission Tomography) in selected cases. Since we are studying genes leading to cSVD, patients who displayed signs of significant vascular burden (large cortico-subcortical infarct, extensive subcortical white matter lesions superior to 25%, uni- or bilateral thalamic lacunes, lacunes in the head of the caudate nucleus and more than 2 lacunes)21 demonstrated by brain computed tomography or MRI were not considered in this study.

Ethics

The study was approved by the ethics committee of Coimbra University Hospital and biological samples were obtained following written informed consent from the patients’ legal representatives. Research was performed in accordance with the Declaration of Helsinki.

Exome sequencing

Exome sequencing was performed on all patients using a HiSeq2500 with 75–100 bp pair-end reads after library preparation with the SureSelect Exome Capture Kit v4 (Agilent). Exome data processing was done following GenomeAnalysisTK best practices. Alignment was done to the hg19 genome assembly using the Burrows-Wheeler Aligner (bwa-mem) v0.7.12. Duplicates were identified with samblaster v0.1.21 and bases were recalibrated with GenomeAnalysisTK v3.8-122. Variant Quality Score Recalibration23, and annotation with snpEff v4.2 and dbNSFP v2.9 were applied to all variants24,25. Variants were filtered according to their quality metrics as described by Patel and collaborators26. Identical by descent (IBD) statistics were inferred with KING27 and only unrelated patients were included.

Variant analysis

We explored variants in the nuclear genes associated with inherited cSVD28: NOTCH3 (associated with CADASIL), HTRA1 (associated with CARASIL), CTSA (Cathepsin A, associated with CARASAL), TREX1 (Truncated three prime Exonuclease-1, associated with Retinal Vasculopathy with Cerebral Leukodystrophy and Systemic manifestations), COL4A1 (associated with Pontine Autosomal Dominant Microangiopathy and Leukoencephalopathy), COL4A2 (encoding Alpha 2 chain of type IV Collagen) and GLA (encoding α-galactosidase A, associated with Fabry disease). All patients were previously screened for pathogenic mutations in APP (Amyloid Precursor Protein), PSEN1 (Presenilin 1), PSEN2 (Presenilin 2), GRN (Progranulin) and MAPT (Microtubule Associated Protein Tau), and C9ORF72 (chromosome 9 open reading frame 72) expansions, which were not found. Exome data was also surveyed for definite pathogenic changes in the other genes associated with frontotemporal dementia and other forms of dementia (listed in29), and amyotrophic lateral sclerosis (as reported in30).

Only rare variants were included in the study. Variants with a minor allele frequency (MAF) > 0.001 in GnomAD v2.1.1 overall sample, or in any of the detailed subpopulations were excluded, except when the variants have been reported to be pathogenic before. In silico pathogenicity prediction was assessed using SIFT (Sorting Intolerant From Tolerant)31, Polyphen-232, PROVEAN (Protein Variation Effect Analyzer)33, Mutation Taster202134 and CADD (Combined Annotation Dependent Depletion)35. Cutt-offs were as defined by each tool. For CADD PHRED scores we did not classify as pathogenic or not, as there is not an established cut-off, the 10% most pathogenic variants have CADD of 10, the 1% most pathogenic variants have CADD 20 and so on. The relevant findings were comprehensively assessed by considering MAF (Minor allele frequency), predicted pathogenicity, disease association, and family history. Literature and available databases were also surveyed to check for previous reports on the found variants.

Neuropathological analysis

One patient donated his brain for research to the Portuguese Brain Bank of Centro Hospitalar Universitário de Santo António. It was fixed in a 4% aqueous solution of formaldehyde for three weeks prior to diagnostic cutting and selection of brain regions of interest for paraffin-embedded tissue blocks.

Histological studies with haematoxylin and eosin, Klüver- Barrera, and immunohistochemistry [\(\:{\upalpha\:}\)-synuclein, ubiquitin, SQSTM1/p62, phosphorylated tau (AT8), TDP-43, GFAP, amyloid-β, CD77] were performed in 6 μm paraffin-embedded formalin-fixed sections from selected brain regions. Appropriate consent procedures for the collection and use of human brain tissue by the PBB were approved by the National Ethics Committee.

Results

Sixty-four out of 104 (61.5%) early-onset AD patients were female. The average age of onset was 54.8 ± 5.5 years. We found 12 rare variants in 18 patients (17.3% of the cohort), as described in Table 1.

Table 1 Listing of the rare variants (MAF < 0.001) in the selected genes found in our cohort.
Full size table

In NOTCH3, we found 6 rare variants in 9 patients (p.Met2280Val, p.Arg2234His, p.Arg1895Leu, p.Val1141Met, p.Leu691Ile, p.Glu74Lys). Of these, p.Arg1895Leu and p.Val1141Met, both present in two patients each, were predicted to be damaging by in silico tools. Further, two additional patients carried the p.Met2280Val variant.

We further identified 6 rare variants in 9 patients (1 patient carried 2 variants; and two variants were present in more than one patient) in GLA, COL4A2 and HTRA1, accounting for 8.7% of the cohort (Table 1). We did not find rare variants in COL4A1, CTSA or TREX1.

GLA variants

We identified a previously reported pathogenic GLA variant (p.Arg118Cys) in 3 male AD patients (Table 2). Two had CSF (cerebrospinal fluid) collected, with CSF biomarkers suggestive of AD. None of the patients had an ischemic lesion burden considered to be the cause of the dementia. None had other reported Fabry symptoms, but this was not systematically assessed, since Fabry disease was not suspected in any of the patients. Alpha-galactosidase levels were not assessed for any of the patients, as they had died prior to the present study. One had AD since the age of 61 and died of subarachnoid hemorrhage at the age of 71. Other had AD since the age of 62 years and the third one since the age of 64, none with clinical vascular events. All were APOE ε3ε3.

One of the patients (patient A) underwent postmortem neuropathological examination, which revealed a 1034 g brain with global atrophy, particularly involving the occipital-parietal and hippocampal regions (Fig. 1A). The locus coeruleus was severely depigmented. Histological examination revealed severe AD neuropathology (NIA – AD neuropathological change high likelihood: A3, B3, C3; NFT (Neurofibrillary tangle) stage V36; Aβ-amyloid phase 537; CERAD (Consortium to Establish a Registry for Alzheimer’s Disease): frequent neuritic plaques) (Fig. 1B and C). Additionally, there was Aβ CAA (Cerebral Amyloid Angiopathy) (Vonsattel grade 2)38, moderate atherosclerotic disease, small vessel change without vascular lesions, and Lewy body pathology (α-synuclein) in the medulla at the level of the dorsal motor nucleus of the vagus nerve and medullary reticular formation compatible with stage 1 Parkinson’s Disease (PD)39.

Fig. 1
figure 1

Neuropathological findings in patient A. (A) Brain coronal section; (B) Aβ, temporal neocortex; (C) Tau (AT8), hippocampus (CA1); (D) CD77 (arrows) – CA1 region (D) and subiculum (E).

Full size image

After WES results, we performed immunohistochemistry with a monoclonal mouse antibody anti-CD77 (1:200; Clone 5B5; BioLegend, San Diego, CA) for recognition of Fabry disease-associated Gb3 deposits. We found CD77 immunoreactivity in hippocampal sectors CA1-CA4 and, to a lesser extent, in the subiculum (Fig. 1D and E).

Table 2 Characteristics of the 3 AD patients harboring GLA pathogenic variants.
Full size table

Discussion

Here, we aimed to assess the possible role of genes associated with monogenic small vessel disease and AD. We identified three AD patients carrying a known pathogenic GLA variant (p.Arg118Cys), which causes Fabry disease, a monogenic cSVD. One of these patients had neuropathological data showing both AD and Fabry pathology, linking the two diseases for the first time. Fabry pathology was identified in regions typically associated with AD, further supporting the linking of both phenotypes.

We also found several rare variants in other genes associated with cSVD (NOTCH3, COL4A2 and HTRA1) in AD patients, corroborating previous studies and providing further support for the possibility that cSVD genes may play a role in AD pathogenesis.

GLA

GLA, located at the Xq22.1, encodes α-galactosidase A, a lysosomal enzyme classically associated with Fabry disease (FD). α-Galactosidase A main function is to catalyze the removal of terminal α-galactose groups from substrates such as glycoproteins and glycolipids. In FD, its deficiency leads to accumulation of neutral glycosphingolipids. The mechanism of tissue damage is considered to be at least partly due to poor perfusion caused by accumulation in the vascular endothelium, particularly in the kidneys, heart, nervous system, and skin. However, defects in autophagy, lysosomal function, lipid metabolism, and inflammation have also been implicated28. FD can present as a severe, classical phenotype, most often seen in males without residual enzyme activity, and a generally milder nonclassical phenotype40. This phenotype, also referred to as late-onset or atypical FD, is characterized by a more variable disease course, in which patients are generally less severely affected and disease manifestations may be limited to a single organ40.

During the past decade, GLA has been increasingly associated with PD41,42,43, suggesting a role for it in neurodegeneration. In fact, reduced alpha-galactosidase A levels have been reported in PD patients and correlate with α-synuclein deposition, hinting at an association of GLA changes with autophagy-lysosome pathway dysfunction44, which has also been shown to be involved in the pathophysiology of AD. Interestingly, it has been shown that, in the brain, the accumulation of globotriaosylceramid is not restricted to vessels, but also occurs in the substantia nigra, neocortex, hippocampal formation, brainstem, amygdala, hypothalamus, and entorhinal cortex45,46, possibly suggesting that these areas may be more susceptible to alpha-galactosidase A dysfunction. In fact, Fabry disease has been linked with hippocampal atrophy and global atrophy47. Noteworthy, although sexual chromosomes have been largely understudied in AD48, X-chromosome appears to carry some of the risk for AD49.

All 3 of our patients carrying a rare variant in GLA had the same variant (Arg118Cys). Although this variant’s pathogenicity has been disputed50,51,52, it has been shown to lead to a mild/moderate deficiency of α-Galactosidase A activity50 and to some changes characteristic of the disorder50,52, with a later age of onset50,52. It is now postulated that this and other variants may be associated with milder forms with later onset of dysfunction53. Our neuropathological data supports this variant’s pathogenicity. Moreover, the localization of FD pathology in CA1, an area associated with early AD pathology neurodegeneration process36 raises the possibility of the association of this variant with AD and should prompt further studies to determine the frequency of AD-related pathology in FD patients. The presence of α-synuclein-positive lesions restricted to the medulla oblongata and corresponding to PD stage 1 indicates that FD patients also can display Lewy pathology and, thus, ultimately may be capable of developing parkinsonism46. Unfortunately, we were unable to assess the presence of Lewy body pathology at the level of the spinal cord and peripheral autonomic or enteric nervous systems. None had a positive first-degree family history.

We can draw a parallel with GBA, also a lysosomal enzyme. GBA homozygous pathogenic variants lead to Gaucher disease, whereas heterozygous carriers have an increased risk of PD54. Different mechanisms have been postulated for the association of GBA with PD, different from the one that causes Gaucher Disease. The most probable are (1) retainment of the misfolded protein in the endoplasmic reticulum, activating the unfolded protein pathway; (2) the lysosomal dysfunction and subsequent impairment of the autophagic-lysosomal pathway; (3) accumulation of aberrant lipid forms that affect the membrane composition; (4) neuroinflammation; and (5) accumulation of defective mithocondria55. In the case of GLA, we hypothesize that some (more disruptive) variants lead to the accumulation of neutral glycosphingolipids and FD. Some other less disruptive variants, which do not cause early-onset Fabry disease, can cause cellular dysfunction that increases the accumulation of misfolded proteins and, ultimately, AD (as well as PD). The same pattern of different phenotypes at different ages caused by different degrees of pathogenicity in the same gene is seen in other autosomal genes, such as GRN and TREM2.

NOTCH3

Mutations in the EGFr domain on the external surface of the NOTCH3 protein lead to misfolding and gradual accumulation in the vascular smooth muscle56. This mechanism fits the Cerebral Autosomal Recessive Arteriopathy with Subcortical infarcts and Leukoencephalopathy (CADASIL) pathobiology, but does not fit well with the known AD mechanisms. NOTCH receptors undergo proteolytic processing via γ-secretase in a manner comparable to that of APP. This suggests that interactions between these signaling pathways may be related with AD mechanisms57. We found six rare NOTCH3 variants in nine patients (8.7% of the cohort). None of these variants involved cysteine residues in the EGFr domain of the protein, which are typically associated with CADASIL58.

The p.Arg1895Leu variant leads to a change in a residue in a highly conserved ANK domain59. Interestingly, this region is vital for GSK3β binding59. GSK3β is considered to be central in AD pathology, namely in tau phosphorylation60. The impact of NOTCH3 phosphorylation by GSK3β is yet to be determined, but this relationship may be relevant for AD pathogenesis. The Met2280Val has been found in one AD patient and none of the 95 controls in a previous study61. We found this variant in two of our patients. It changes a residue just before the PEST domain. Mutations in the PEST domain have been associated with activation of NOTCH signalling62, which has been implicated in adult hippocampal neurogenesis63. Interestingly, one of the other variants (p.Arg2234His) we found is located in the same region. A similar trend is seen in other similar studies64 adding interest to our findings.

Although cysteine-changing variants were absent from our cohort, we demonstrate that rare variants in NOTCH3 may be frequent in EOAD patients. Of the ones we report, those with the higher probability of association with disease lie outside the EGFr region, typically associated with CADASIL, and are located in regions associated with the intracellular NOTCH signaling.

COL4A2 and HTRA1

COL4A1 and COL4A2 encode the α1 and α2 chains of type IV collagen, a major component of the basement membrane65. Mutations in COL4A1 and COL4A2 are associated with a variety of diseases, including cSVD. COL4A2 is known to harbor less variants reported as pathogenic65, most of them changing a Glycine residue. We found 4 rare variants in our cohort (reported in Table 1). Most of these variants are predicted to be pathogenic by in silico tools, but further studies would be needed to assess their possible role in disease.

HTRA1 encodes a serine protease that is involved in a variety of cellular processes, including apoptosis, angiogenesis, and inflammation. Mutations in HTRA1 cause Cerebral Autosomal Recessive Arteriopathy with Subcortical infarcts and Leukoencephalopathy (CARASIL), a rare autosomal recessive cSVD. Heterozygous HTRA1 mutations have also been shown to cause autosomal dominant small vessel disease66, sometimes leading to dementia. HTRA1 is an allele-selective ApoE-degrading enzyme, that degrades ApoE ε4 faster than APOE ε3, having a potential role in the mechanisms underlying AD. We found one AD patient carrying the p.Val433Ile variant in HTRA1. This variant has not been previously reported in the literature67.

The findings of our study suggest that COL4A2 and HTRA1 variants may play a role in AD pathogenesis. Further studies are needed to confirm these findings and to elucidate the underlying mechanisms.

Overall, our study provides new insights into the genetic overlap between cSVD and AD. Our findings suggest that cSVD associated genes may be a more important risk factor for AD than previously thought, and that targeting cSVD pathways may be a relevant therapeutic approach for AD. However, our study has some limitations. First, our sample size was relatively small. Second, we did not perform functional validation of the genetic variants that we identified, particularly of the association of the reported variants with the corresponding phenotype. These are thoroughly studied patients, most of them with very early onset dementias without a genetic cause identified by exome sequencing, increasing the potential for an association of these genes with the observed AD phenotype. Still, we can neither confirm pathogenicity nor establish a definite link between the changes and the phenotype. Third, we do not have pathological confirmation of the diagnosis of AD in the majority of patients. Yet, we combine CSF and/or PET biomarkers and a long follow-up from an experienced team. Moreover, the main finding of our study was corroborated by neuropathological data.

Conclusion

We found several rare variants on the genes of interest in our cohort. We show that three unrelated carriers of the GLA p.Arg118Cys variant, without other probable genetic cause responsible for the disease, have developed early-onset AD. One of these cases had post-mortem confirmation of the diagnosis and we further show the colocalization of Fabry pathology in areas relevant for AD pathogenesis, suggesting a role for GLA in AD. The association of GLA with AD suggested in this study relies on two main points. First, the finding of Fabry pathology in a patient’s brain carrying the GLA p.Arg118Cys variant supports the potential pathogenic role of this variant in the disease. Second, the fact that we identified this same GLA rare variant in 3 out of 104 early-onset AD patients (2.9% vs. 0.0008% of Non-Finnish European in gnomAD) where exome sequencing did not show any other probable genetic cause for the disease, and considering the rarity of the variants and the phenotype, indicates that this overlap is not likely to be an incidental finding.

In the future, functional data and animal models can give further support to this association. Fabry disease has some forms of treatment, and can indicate an additional genetic therapeutic target for EOAD patients. If confirmed, this association would also add stronger evidence to the role of lysosomal involvement in AD pathogenesis.