Introduction

Alzheimer’s disease (AD) is the most common type of dementia in our clinical setting. It mostly affects the elderly, and its prevalence reaches up to 45% in the population older than 85. Characterized by a progressive and serious decline in cognitive, memory, and language functions, it leads to severe disability and dependence with an enormous personal, familial, and social impact [1].

Although the ultimate cause of AD remains unknown for most cases, the pathogenesis of the disorder has been widely studied: an abnormal excision of amyloid precursor protein (APP) leads to the parenchymal and vascular deposition of insoluble fibrils of beta-peptide and to the formation of neuritic plaques and amyloid angiopathy. This phenomenon is accompanied by an abnormal activity of several neuronal kinases, which in turn lead to oxidative stress, mitochondrial dysfunction, activation and recruitment of microglia and astrocytes, and in some cases, to hyperphosphorylation, abnormal configuration, and accumulation of Tau protein. This last event destabilizes neuronal microtubules and generates neurofibrillary tangles, the second main histological hallmark of AD [2]. About 15–20% of patients with AD dementia also present cortical and subcortical Lewy bodies (LB) [3]. It has been suggested that subjects with AD and LB comprise a distinct subgroup referred to as LBV/AD (Lewy bodies variant/Alzheimer’s disease) [4].

Misfolded and aggregated proteins interact with microglia and astrocytes, triggering the innate immune response. This is part of the neuroinflammatory process in AD, which contributes to disease progression and severity and is associated also with neurotoxicity and synaptic dysfunction [5–7]. In fact, microglia has been pointed to have a major role in the clearance of oligomeric beta-amyloid, and different microglia phenotypes, such as microglial priming, may be associated to different disease outcomes [8–11].

For many years, only a few genes were associated with susceptibility to AD: APP, PSEN1, and PSEN2 with genetically determined early-onset familial AD; and APOE for the most common late-onset AD (LOAD). Since the introduction of genome wide association studies (GWAS) in 2009, the knowledge of a much more complex genetic background has emerged. More than 20 loci have been identified which increase the susceptibility/risk to LOAD with small effect, including BIN1, CR1, CLU, CD33, EPHA1, MS4A4/MS4A6, ABCA7, CD2AP, HLA-DRB5/DRB1, PTK2B, SLC24A4-RIN3, INPP5D, MEF2C, NME8, ZCWPW1, CELF1, FERMT2, CASS4, TREM2, PLD3, TBK1, CTNNA3 PICALM, SORL1, GAB2, and MTHFR [12–16]. Interestingly, some patterns emerge when analyzing the functional relations of these genes, and three main pathways in the pathogenesis of AD have been established: the lipid-processing pathway (APOE, CLU, and ABCA7), the immune system (CLU, CR1, ABCA7, CD33, TREM2, IL1A, IL1B, 2, and EPHA1), and the synaptic-cell-functioning pathway (PICALM, BIN1, CD33, and CD2AP) [17–19]. Also, single nucleotide polymorphisms (SNPs) in the SNCA genes have been reported to contribute to LB pathology in LBV/AD patients, possibly via interaction with SNPs in the LRRK2 gene [20].

Recently, two rare mutations (in TREM2 and CD33) support the view that microglia has a fundamental role in AD. The SNP rs75932628 of TREM 2 is associated with increased risk of AD [21, 22]. TREM2 encodes for the triggering receptor highly expressed by microglia which mediates phagocytic activity and debris clearance [14]. Similarly, CD33-rs3865444 carriers have been shown to have increased β amyloid deposition [23], caused by a defective microglial β-amyloid uptake [24].

The CX3CR1 gene (chemokine (C-X3-C motif) receptor 1, also known as fractalkine receptor, OMIM: 601470) in the CNS is a microglia-specific gene that mediates its neurotrophic functions, and it is involved in the microglia-neuron crosstalk [25]. Its ligand, fractalkine (CX3CL1), appears to be neuroprotective in some settings, whereas it contributes to neuronal damage in others. AD, and many progressive neuroinflammatory disorders associated with increased microglial activation, shows disruption of the CX3CL1/CX3CR1 communication system [26]. The impairment of CX3CL1/CX3CR1 signaling has a direct influence on the processes of neuroinflammation, neurotoxicity, and synaptic plasticity in the pathogenesis of AD and other neurodegenerative diseases [7, 8, 26]. Fractalkine has been suggested as an endogenous neuronal modulator that may limit microglial activity in AD [27, 28] by reducing the inflammatory reaction through fractalkine/NRF2/heme oxygenase 1 pathway [29]. Two of its genetic variants (V249I and T280M) have been described to affect CX3CR1 protein activity [30, 31]. Recently, our group has found an association of the CX3CR1 V249I and T280M variants with the survival and disease progression in amyotrophic lateral sclerosis (ALS) [32]. These variants also have been associated with several inflammatory and degenerative human conditions, including multiple sclerosis [33], Crohn’s disease [34], AIDS [35], and age-related macular degeneration [36].

Despite numerous evidence showing involvement of CX3CR1 in AD, and in several central nervous system conditions, no human genetic studies have been performed to investigate its influence on the pathogenesis of Alzheimer’s disease. We hypothesized that CX3CR1 genetic variants may influence AD outcome. Here, we analyzed the association between these two CX3CR1 variants, and the disease risk, age at onset, and pathologic progression rate in a Spanish late onset AD (LOAD) cohort, including a subgroup of LOAD patients with LB (LBV/AD).

Subjects and Methods

Study Population

A total of 955 subjects were recruited for the study. 205 brain donors with pathologically proven late onset AD (LOAD) were recruited from the Neurological Tissue Bank of the Biobank-Hospital Clinic-IDIBAPS, Barcelona, Spain. Their DNA was obtained from fresh frozen cerebellar cortex. 750 healthy controls were obtained from the Spanish National DNA Bank (BNADN), and their DNA sample obtained from peripheral blood samples (270 controls were randomly selected by age and gender to match the cases).

The diagnosis of LOAD was defined as an age of onset or diagnosis ≥65 years according to clinical charts and required a high or intermediate score of AD neuropathological change according to the current National Institute on Aging, Alzheimer’s Association guidelines, characterized by the presence of moderate or severe neuritic plaque density and neocortical or limbic neurofibrillary pathology (tangles and threads) in the context of dementia [37]. The presence of concomitant Lewy body (LB) pathology was assessed in brainstem nuclei, limbic system, and cortical areas by hematoxylin-eosin stain and anti-alpha-synuclein immunohistochemistry in selected regions. LB pathology was subclassified as brainstem, limbic, neocortical, or amygdala only [38]. Information regarding age, gender, and clinical diagnosis was collected for all samples with available information. Age at onset (AAO), survival time, and rate of progression were recorded for AD cases (see demographic data at Table 1). Braak’s staging system of neurofibrillary pathology (transentorhinal stages I and II, limbic stages III and IV, and neocortical stages V and VI) was used for the assessment of progression [38].

Table 1 Demographic data of the cohorts: late-onset AD (LOAD) or LOAD without LB (LB−/AD) or LOAD with LB (LBV/AD) patients
Full size table

DNA Purification and Genotyping

DNA was isolated from postmortem fresh frozen cerebellar tissue using a variant of the silica affinity in guanidine tiocianate method developed in-house by our group. Briefly, 25–50 mg of brain tissue were thawed in 100 μl of lysis buffer (EDTA 20 mM, guanidine tiocianate 6M, K3PO4 20 mM, 1% triton X-100, and DTT 6.5 mM, pH = 6.8) and were broken down using an automatic pestle. The sample was further processed by squeezing the tissue homogenate five times through a 20 G syringe after adding 600 μl lysis buffer. After incubation for at least 3 min at room temperature, centrifugation (1 min at >13,000×g) and removal of the pellet, the supernatant was incubated in a silica-based affinity chromatographic column (Econospin, Epoch Life Sciences) for 1 min at room temperature. The DNA retained in the silica column was washed twice with 300 μl wash solution (25%ethanol, 25% isopropanol, 10 mM Tris-Cl, and 100 mM NAcl, pH = 8.0) and was centrifuged (1 min at 13000×g). DNA was finally eluted in 50 μl TE-reduced buffer (10 mM Tris-Cl and 0.1 mM EDTA, pH = 8.0), was prewarmed at 65 °C, and was centrifuged (1 min at 13000×g). This elution step was repeated once.

CX3CR1 V249I (rs3732379) and T280M (rs3732378) alleles were genotyped using the KASPar® SNP Genotyping system (LGC genomics, UK) according to the standard provider procedures. CX3CR1 gene T280M and V249I variants were identified after amplification with V249I primers set (249_V:CTT CTG GAC ACC CTA CAA CG; 249_I:CCT CTT CTG GAC ACC CTA CAA CA; 249rev:GAG CTT AAG YGT CTC CAG GAA AAT CAT) and T280M primer set (280_T:GGC CCT CAG TGT GAC TGA GAC; 280_M:GGC CCT CAG TGT GAC TGA GAT; 280_rev:GAG AGG ATT CAG GCA ACA ATG GCT A). Fluorescence was measured at 25 °C (final point) in a 7300 real time PCR System (Applied Bioscience). Genotype calling was carried out using 7300 system SDS software v1.4 (Applied Biosytems, USA) and Klustercaller software (LGC genomics, UK).

Statistical Design, Analyses, and Statistical Power

We designed an open, pragmatic, case-control retrospective study with a total of 475 subjects divided into two cohorts: the control cohort of 270 individuals without neurodegenerative diseases (randomly selected by age and gender according to cases), and the case cohort of 205 patients with pathologically proven LOAD. The case cohort was further divided in 68 LOAD patients with LB (LBV/AD subgroup) and 137 LOAD patients without LB pathology (LB−/AD subgroup). Statistical data analyses were performed using the linear generalized models implemented in the SNPassoc R software package [39]. For progression analysis, ordinal logistic regression was carried out using SPSS software.

Sample population was tested for Hardy-Weinberg equilibrium using Fisher’s exact test. The analyses of allelic frequencies proved that cases and controls’ genotypes were in Hardy-Weinberg equilibrium (V249I p = 0.912, T280M p = 0.243).

In the LOAD cohort, the effect of the two CX3CR1 genetic variants was analyzed for association with AD risk, survival time, progression, and AAO. Analyses were performed for five different inheritance models: dominant, co-dominant, additive, recessive, and over-dominant. Sex, age, and/or AAO were introduced as adjusting variables. P values were modified by a factor of 4.3 to correct for multiple test comparison with Bonferroni method [39, 40]. Using this criterion, the uncorrected level for statistical significance was established at p < 0.0116. The uncorrected and corrected p values (adjusted or adj-p) are shown.

This study has a prior statistical power of 80% (alpha = 0.05) for detecting a difference in age of onset between groups of patients (mean population age of onset = 75.25 ± 5.87 years; 280M minor allele frequency = 0.145) greater than 2.3 years. Concerning the risk to suffer AD, this study has a power of 80% (alpha = 0.05) for detecting an OR > 1.49 between cases and controls for the 249I (minor allele frequency of 0.28).

Results

CX3CR1 Variants and AD Susceptibility

In the analysis of CX3CR1 variants V249I and T280M with the disease risk in the whole LOAD group, or LBV/AD or the LB−/AD subgroups, we did not find any statistical significant association of neither variant with AD risk under different inheritance models (Tables 2 and 3). But nevertheless in the LBV/AD subgroup, we observed a statistical association tendency of V249I variant with LOAD risk (p = 0.074, Table 2).

Table 2 Summary of P values for genetic analyses of the CX3CR1 V249I variant with the different phenotypes analyzed in LOAD or LB−/AD or LBV/AD groups using different genetic models
Full size table
Table 3 Summary of p values of the genetic analyses of the CX3CR1 T280M variant with the different phenotypes analyzed in LOAD or LB−/AD or LBV/AD groups using different inheritance models
Full size table

CX3CR1 Variants and Neurofibrillary Pathology Progression

We found a statistically significant association of the V249I variant with neurofibrillary pathology stages, when assuming a co-dominant, over-dominant, and recessive inheritance pattern (Tables 2 and 4). Patients with CX3CR1 249V/I genotypes were over-represented in Braak stage IV subgroup (limbic pathology stage) and were under-represented in the higher neocortical neurofibrillary Braak stage VI subgroup (Table 4, Fig. 1). The protective effect of the 249V/I genotype for neurofibrillary pathology progression was higher (OR = 0.416, 95%CI [0.232, 0.737], p = 0.003, adj-p = 0.013) under an over-dominant genetic model. Instead, under a recessive model, patients with the CX3CR1 249I/I genotypes were more frequent (80%) in the highest neocortical pathology subgroups (Braak stage VI that includes primary cortical fields) than in the lower neocortical pathology group (20%; Braak stage V, that includes cortical association areas but preserves primary cortical fields) while no 249I homozygous individuals were observed in the lower limbic stages (Table 4, Fig. 1). The risk effect of the CX3CR1 249I/I genotype on the progression of neurofibrillary pathology was OR = 3.967 (95%CI [1.271, 12.391], p = 0.018) under a recessive model.

Table 4 CX3CR1 V249I marker analysis for Braak’s neurofibrillary pathology progression in LOAD patients assuming different genetic models
Full size table
Fig. 1
figure 1

Scatterplot for age at onset and Braak’s neurofibrillary stage severity of the different CX3CR1-V249I (a) and T280M (b) genotypes, according to a recessive genetic model

Full size image

Analysis of the pathology progression in the LOAD LB- and LBV/AD subgroups showed the same effect of the V249I variant only in those patients that did not develop LB during AD pathology (Table 2 and Online Resource 1). Despite the smaller sample size of the LOAD LB- subgroup compared with the whole LOAD group, the heterozygous genotype 249V/I showed a larger protective effect for neurofibrillary pathology progression (OR = 0.352, 95%CI [0.173, 0.715], p = 0.004, adj-p = 0.017) under an over-dominant model (Online Resource 1). The homozygous genotype 249I/I also showed a larger risk effect for disease progression under a recessive model (OR = 10.57 (95%CI [1.315, 84.94], p = 0.027; Online Resource 1).

No association has been detected between the variant T280M and neurofibrillary pathology stage for LOAD group or the LBV/AD and LB−/AD subgroups (Table 3).

CX3CR1 Variants and Age at Onset

Our analyses indicated that LB−/AD patients homozygous for the 280M allele showed a later AAO (84.0 years ±3.61) than individuals carrying one (74.85 ± 0.82) or both wild-type alleles (75.12 ± 0.53). Assuming a recessive inheritance model, the mean difference of 8.71 years (95%CI [2.06 to 15.37] years was statistically significant (p = 0.0099, adj-p = 0.43) (Tables 3 and 5, Fig. 1). No statistically significant association was found for T280M variant with the AAO for LOAD (all) group or LBV/AD subgroup (Table 3).

Table 5 CX3CR1-T280M variant for age at onset (AAO) for LOAD, LBV/AD, and LB−/AD, under a recessive model of inheritance
Full size table

In the LBV/AD subgroup, patients with 249I/I genotype showed an earlier AAO (69.17 ± 1.76, n = 6) compared to genotypes 249V/I (74.76 ± 1.11, n = 21) and 249V/V (74.74 ± 1.15, n = 27). The AAO difference (69.17 ± 1.76 vs 74.75 ± 0.80, n = 48) was −5.04 years (95%CI [−0.28, −9.8], p = 0.038) for the homozygous patients, being statistically nominal significant under a recessive genetic model. No statistical association was found for V249I variant with the AAO for LOAD group or LB−/AD subgroup (Table 2).

CX3CR1 Variants and Survival

Analyses of the two CX3CR1 variants using different genetic models did not show any statistically significant difference regarding survival among LOAD patients or the LB-subgroup (Tables 2 and 3). In the LBV/AD subgroup, under a recessive genetic model, the patients with genotype 249I/I showed a longer survival (163.2 months ± 20.3, n = 5) compared to genotypes 249V/I (110.9 months ± 14.1, n = 21) and 249V/V (113.3 ± 10.7, n = 27). Patients homozygous for the 249I allele had a survival of 55.48 months longer (95%CI [097, 110], p = 0.046). Although there was a nominal statistically significance, the observed difference did not pass the multiple correction test (adj-p = 0.207).

Discussion

In our pathologically confirmed LOAD series, we have detected a novel association between the V249I variant of the microglial-specific gene CX3CR1 and the progression of neurofibrillary pathology in LOAD; while the variants T280M and V249I seem to be associated with the age at onset. These results suggest that the CX3CR1 gene could act as a disease-modifying gene in the pathogenesis of AD.

The most significant observation of our study concerns the neurofibrillary pathology progression. We have observed that the CX3CR1 V249I allele was related to the pathology stage in our Spanish cohort of LOAD patients. Most homozygous patients for 249I allele showed a worse AD neurodegenerative progression at death with high Braak’s neurofibrillary stages (stages VI and V). This result is in accordance with the fact that the same CX3CR1 249I allele was associated with a worse disease progression in a Spanish ALS cohort [32].

Heterozygous patients carrying the two alleles of the 249I variant (249V/I genotype) showed a lower Braak neurofibrillary stage than homozygous patients (249V/V genotype, Fig. 1). This indicates that heterozygous patients had less involvement of neocortical anatomical areas by neurofibrillary pathology; hence, a less severe neurodegenerative process, which could possibly indicate a heterozygous advantage: patients harboring two different alleles of the CX3CR1 V249I variant could have better means and resources in different/specific brain areas, reducing or delaying topographical disease progression. This may be related to the functions of CX3CR1 as a specific microglial gene in the CNS, with a fundamental role in the neuron-microglial crosstalk [26] and on microglial activation [36, 41]. Having two alleles (V and I) at position 249 could influence the versatility of microglia to change from a predominantly cytotoxic phenotype to one mostly neuroprotective, or vice versa; either at the initial neuronal damage or during the progressive chronic lesions.

The CX3CR1 249I variant has been associated with reduced number of fractalkine (CX3CL1) binding sites and reduced binding affinity on peripheral blood mononuclear cells, resulting in a loss of function [42]. It has been suggested that CX3CL1/CX3CR1 may act in a neuroprotective pathway and as first-line of defense response to neuronal injury and neuroinflammation [26]. An altered modulation of CX3CL1/CX3CR1 signaling would promote neuroinflammation, neurotoxicity, and changes in synaptic plasticity in AD pathogenesis [7, 26, 29]. Thus, the decreased CX3CR1 activity of the 249I allele would reflect an impaired microglial function that may contribute to AD pathogenesis, in part, by enhancing inflammatory activity of microglia and also neuronal degeneration leading to disease progression.

Concerning to the association of the CX3CR1 variants with the age at onset (AAO), the development of Lewy bodies (LB) during AD pathology (LBV/AD subgroup) modifies the effect of the variants under a recessive genetic model. In our Spanish LBV/AD subgroup, the V249I/I genotype could be associated with an earlier AAO (−5.04 years). In contrast, in the LOAD LB-subgroup, homozygosis for 280M allele seems to be associated with a later AAO. It has been described that LBV/AD patients show both overlaps and differences with LB−/AD patients, and it has been suggested that LBV/AD subjects may comprise a distinct subset. At the present moment, however, the nosological position of LBV/AD within the broad spectrum of AD and LBD remains disputable [20, 43–46]. Our results indicate that genetic differences exist between late onset LBV/AD and LB−/AD subgroups, but further research is needed to address this issue.

Besides the influence of the presence of LB in AD pathology, another factor that could explain the opposed effects of CX3CR1 variants (the earlier AAO for 249I in LBV/AD and the later AAO for 280M in LB−/AD) is the counter-acting effects of these two SNP’s. Naissner et al. postulated that increased fractalkine concentration would outweigh the moderately reduced binding site density in subjects carrying the 249I variant [47]. Similarly to those findings, McDermott et al. suggested that any harmful effects of 249I allele could be reversed by the protective effect of 280M allele [30]. Therefore, 249I and 280T variants might have counter-acting effects. Something which makes evolutionary sense as the allele 280M occurs mostly in the presence of 249I allele [48].

In our Spanish cohort, 280M/M and 249I/I genotypes seem to modify the AAO, with an unexpectedly high mean difference of 8.7 and −5.4 years, respectively. If this is confirmed by further studies in different cohorts, CX3CR1 variants could be the second strongest genetic factor described up to date which modify disease onset in LOAD. The strongest factor modulating AAO in AD is the APOE ε4 allele, first described in 1993 (0 copies: 84.3 years; 1 copy: 75.5; 2 copies: 68.4) [49] and widely reproduced [50, 51]. Other genes have been proposed as strong modifiers of disease onset in LOAD, such as SORL1(rs1784933AA genotype causes an AAO of 2.5 years earlier in homozygosis) [52] and HMGCR (in women delayed about 3.6 years the AAO) [53].

None of the GWAS studies that analyzed AAO in LOAD have found any association with the CX3CR1 gene or loci [12, 54–57]. However, the magnitude of the effect we observed is unexpectedly high. This might be related to the sample heterogeneity because GWAS studies are using only clinically diagnosed cases and not neuropathologically confirmed cases. Another factor to take into account is the fact that the frequency of the variants are low (14.2% for 280M and 28% for 249I); and even lower for the genotype in homozygosis (2.5% of the cases for 280M and 10.1% for 249I); as such, in our Spanish series, only three patients were homozygous for 280M allele and six for 249I allele. This low frequency could also explain why the effect of CX3CR1 280M variant on AAO remained undetected in previous GWAS. Therefore, while rare variants may remain undetected in GWAS, targeted hypothesis studies have more statistical power to succeed [58].

Achieving more accurate diagnoses is essential to make progress in this field of research, and Brain Banks constitute a highly valuable source of neuropathologically confirmed neurodegenerative diseases. They have the potential to contribute in clarifying the existent unknowns concerning AD’s causes and mechanisms [59]. The homogeneity and appropriate diagnostic classification of patients is always an important point to take into account when studying genetics of neurodegenerative diseases. In this respect, postmortem neuropathological diagnostic confirmation is an added value to our study. It has been previously reported that the origin of the samples regarding clinical vs. pathologically confirmed AD has effect on the results [60]. Strengths of our design are the homogeneity of clinical-pathological diagnoses, and the fact that all samples were analyzed at the same Brain Bank, which minimizes observer variation and sources of bias. Even though our study has enough power to detect the differences observed in disease progression and age on onset, the possibility of a “statistically” spurious association cannot be disregarded completely, and we encourage the replication of our findings in larger cohorts and different populations.

Our finding supports the well-known intervention of neuroinflammation in the pathogenesis of LOAD. These CX3CR1 associated variants, added to the previously described in TREM2 and CD33 [24, 61], provide further evidence of the fundamental role of microglia/macrophages in the outcome of LOAD. Interestingly, TREM2 and CX3CL1 (CX3CR1 ligand) are over-expressed in AD hippocampal samples, thus, suggesting that dysfunction is those pathways may be an important hallmark of pathological neuroinflammatory processes in AD [62]. In this regard, it would be interesting to further investigate the possible interactions of these three microglia-associated genes (CX3CR1, TREM2, and CD33) in LOAD cases.

The reported results could be relevant in the clinical setting, considering CX3CR1variants as genetic markers and prognostic factors for AD progression, and point to CX3CR1 as an interesting target for AD therapy development. Also, CX3CR1 could be proposed as a potential pharmacogenetic marker useful for the development of new targeted AD treatments and personalized therapies, in a similar way that have been applied in the development of therapies for cancer (imatinib metasilate, transtuzumab, and others) or cystic fibrosis (Ivakaftor).

As conclusions, the variants V249I (rs3732379) and T280M (rs3732378) of the CX3CR1 microglial gene could represent new modifying factors of age at onset and pathology progression in a neuropathologically confirmed Spanish LOAD population. Homozygous patients for the 249I allele have a higher neurofibrillary Braak stage at death, i.e., a more advanced AD-related neurodegenerative process than heterozygous patients. Homozygous patients for the 280M allele could have a delayed onset of the illness. Replication of this study with a larger sample size will be required to confirm the observed associations in other populations.