SORL1 Variants in Familial Alzheimer’s Disease

INTRODUCTION

Several genes and proteins involved in amyloid-β protein precursor (AβPP) processing are under investigation as potential genetic causes of Alzheimer’s disease (AD). One such candidate gene is SORL1, which encodes the SORLA protein. This protein prevents trafficking of AβPP from the Golgi to endosomal compartments, where Aβ secretases reside [1–3]. Downregulation of SORLA in cultured cells or in transgenic mice has been shown to enhance the amyloidogenic process [3–6]. Therefore, complete SORLA activity seems crucial for preventing Aβ peptide deposition.

Some clinical data support the relevance of SORLA activity in AD development. Sporadic AD cases show reduced SORLA expression in cortical and hippocampal tissue [1, 7], as well as in cerebrospinal fluid samples [8]. Some single nucleotide polymorphisms (SNP) variants of SORL1 have been reported to increase the risk of developing sporadic AD [9–12], and select variants—which alter Aβ levels in vitro—have been found in early- and late-onset familial cases [12, 13]. Furthermore, potentially pathogenic SORL1 mutations have been described in a few cases with autosomal dominant early-onset AD [14], and variants introducing a premature termination codon have been described in AD cases but not in controls [15]. However, segregation data of missense SORL1 variations in the few carrier kindreds reported are not consistent [12, 13]. Therefore, it is not clear whether particular SORL1 variants could be pathogenic and a direct cause of AD.

In this study, we used an extensive cohort of dementia of Alzheimer type (DAT) cases, including kindreds with several siblings available for study, and screened for SORL1 variations. When variants where suggestive of pathogenicity, segregation patterns were analyzed to assess whether some of these mutations could be causal for AD, with effects comparable to AβPP or presenilin genes (autosomal dominant and with full penetrance).

METHODS

We screened for SORL1 variants in 124 familial (44 early- and 80 late-onset) DAT cases recruited from the memory clinic of Fundación Jiménez Díaz (Madrid, Spain). Cases were considered to be familial when there was at least one first-degree relative also affected with DAT, and early-onset when symptoms were detected at≤65 years of age. All cases had been diagnosed according to standardized criteria after clinical and cognitive assessment, blood analysis, and neuroimaging studies [16]. Table 1 summarizes the demographic data of the families. Family history was graded according to Goldman score [17]. All cases had the APOE genotyped. PSEN1 mutations were excluded in cases with autosomal dominant inheritance and age at onset ≤ 60 years. This study was approved by the Research Ethical Committee at Fundación Jiménez Díaz and cases or surrogates gave informed consent for genetic studies.

DNA was extracted from peripheral blood leukocytes using the QIAamp DNA blood Mini-kit (Qiagen). Sixteen of the cases with the strongest family history (Goldman score 1 or 2) [17] had the SORL1 gene analyzed as included in a Next Generation Sequencing (NGS) panel made up of 18 genes related to dementia (PS1, PS2, APP, TAU, PGRN, VCP, CHMP2B, TARDBP, FUS, ADAM10, SORL1, SNCA, TREM2, UBIQLN2, ITM2B, CSF1R, TYPOBP, and SQSTM1). The exons and intronic boundaries of these genes were amplified by Ampliseq technology and sequenced using Ion Torrent equipment, yielding an average coverage of over 200x (Thermo Fisher LifeTecnologies). The remaining cases were analyzed by a pooling technique. The rationale for this was that, as we were looking for rare changes, we could analyze samples in pools whenever the coverage was deep enough and provided each sample—in exactly the same concentration as measured by spectrofluorometry—was represented in at least three pools. This way, 120 samples (the remaining 108 familial, plus twelve sporadic cases) and one mutation-positive control (Ala2173Thr) were gathered in 11 pools of 11 samples each, and each sample was present in three independent pools. Eleven libraries were constructed with Ampliseq and sequenced by Ion Torrent. All the pools achieved a minimum coverage of 1500x. The technique was validated with the Ala2173Thr sample. Only those changes found in three pools were deconvoluted to identify the sample that contained the change, to be later confirmed by Sanger sequencing.

The SORL1 variants found were first reviewed in genomic databases (dbSNP and ExAc) and analyzed for potential pathogenicity with prediction programs (PolyPhen2, SIFT, M-CAP, and CADD). Potentially pathogenic variants in probands and relatives were confirmed by Sanger sequencing (ABI 3730, Applied Biosystems). We applied the recently described criteria for pathogenicity of SORL1 variants, based on the combination of CADD prediction score and the MAF frequency in the ExAc database [18]. These criteria classify variants into five subtypes, as follows:

Pathogenic: truncating SORL1 variants;

In a further step, the families with some available siblings (affected or healthy at≥70 years of age) were inspected for segregation profiles, and a control population of 200 Spanish elderly subjects (69 to 95 years) was screened for the unreported or rare SORL1 changes. Half of the controls were cognitively preserved nonagenarians recruited at Fundación Jiménez Díaz. Another 100 controls were provided by the Banco Nacional de ADN Carlos III (Salamanca University, Spain).

Finally, we gathered all data (predictions, segregation data, and screening of our control population) and established our conclusions of variant pathogenicity according to the American College of Medical Genetics and Genomics (ACMG) guidelines. These consider a variant as likely benign, likely pathogenic, or uncertain significance [19]. In short, variants considered as “likely benign” fulfilled two or more of the following criteria: variant is not segregating with disease, silent variant with no impact on splice-site, allele frequency greater than incidence of disease, not predicted by multiple in silico programs to affect protein. Variants “likely pathogenic” fulfilled three or more of the following criteria: variant only found in cases, variant segregates with disease, variant is predicted by two in silico programs to affect protein, variant changes the protein-length or is a loss-of-function mutation. Variants that could not be classified were scored as “uncertain significance”.

RESULTS

Nine different potentially pathogenic changes were found in nine unrelated probands (7% of the studied population) (Table 2). Three variants had not been previously reported: Trp848Ter, Gly1871Val, and a splice-site variant (chromosome position 121466486 G>A). Another six were classified as rare variants according to information displayed in the dbSNP and ExAc databases: Glu270Lys, Gly852Ala, Arg1702Met, Asn1809Ser, Asp2065Val, and Ala2173Thr. Two of the changes were detected in the first set of 16 patients studied using the NGS dementia panel (Ala2173Thr and Asp2065Val), and the others were found in the NGS of pooled samples.

Prediction of pathogenicity according to the different software programs is shown in Table 2. The splice-site variant was predicted to alter wild-type donor site (c.4519 + 5 G>A), most probably affecting splicing (Human Splicing Finder). When the Holstage et al. criteria [18] were applied (Table 2), two variants were classified as pathogenic (Trp848Ter and a splice-site, which predict a truncated protein), one variant as likely pathogenic (Ala2173Thr), three variants as likely benign (Arg1702Met, Asn1809Ser, and Asp2065Val), and another three variants as uncertain significance (Glu270Lys, Gly852Ala, and Gly1871Val).

A summary of relevant clinical features in the proband cases (exceeding a DAT phenotype with predominant memory deficits) is shown in Table 3. Detailed information of the pedigrees is provided as Supplementary Material.

DISCUSSION

The screening for SORL1 variants in our cohort of familial DAT cases identified potentially pathogenic variants in 7% of the cases, though many of these variants could not be directly correlated with the disease process, either because they did not segregate in the carrier family or because the change was found in control subjects.

The SORL1 gene is a strong genetic candidate for involvement in AD pathogenesis, particularly as a risk factor. Several studies have related some SNP variants with the risk of developing sporadic AD [9–12, 21]. Moreover, some SORL1 mutations have been related to the disease in familial cases [13, 14], supported by functional studies in vitro showing that these select mutations (Gly511Arg, Thr588Ile, and Thr2134Met) impair the SORLA receptor’s ability to direct newly produced Aβ to lysosomes for degradation [6, 13]. However, to consider a variant as likely pathogenic according to current consensus guidelines [19] and the direct cause of the disease, requires consistent segregation of the genetic change in the carrier families. While Vardarajan et al. [12] have reported segregation in a family with the Glu270Lys mutation, the early-onset families reported by Cuccaro et al. [13], with the Thr588Ile and Thr2134Met changes, include healthy mutation carriers and affected non-carrier siblings.

The main strength of our study is that it allows the implication of some SORL1 missense variations (the non-described Gly1871Val, and the rare variants Glu270Lys, Asn1809Ser, and Asp2065Val) to be ruled out as the direct cause of the AD in these families because of a lack of segregation. The Gly1871Val change, found in a late-onset family, was absent in one affected sibling. The Asp2065Val change was also absent in one affected sibling. We had considered this latter variant (very rare in European population, with unknown clinical correlation) a potential pathogenic candidate because it was present in the proband of an autosomal dominant early-onset family, in which an extensive panel of dementia-related genes was negative except for this change. Moreover, codon 2065 is located in the extracellular fibronectin-type III repeats domain, where reside two of the probable pathogenic changes found by Pottier et al. [14], and a recently reported missense mutation segregating in two Swedish siblings with early onset AD [22]. However, the lack of segregation makes this variant “likely benign”.

The Ala2173Thr change was another potential candidate for pathogenicity. It has been described in extremely low frequency: only one allele in 5.045 African subjects and one allele in 32.871 European subjects. Furthermore, it is located in the functionally significant endocytosis signal domain of the protein and was predicted as pathogenic by all “in silico” prediction programs. The A2173T change was absent in one elder healthy sister of the proband, though there were no affected siblings to further assess segregation. However, the potentially pathogenic impact of this change is challenged by the fact that we found one carrier in the control population (0,5%). We also considered the implication of the Arg1702Met change to be unlikely, because it was found in one control subject.

The nature of another two unreported changes—a premature stop in codon 848 and a splice-site variant found in two early-onset probands—are likely to result in truncated protein and haploinsufficiency, and could be more likely associated with pathology than missense mutations (changes the protein-length or loss-of-function mutation criteria [18, 19]). Both changes were also absent in our control population, though there were no other affected siblings to confirm segregation, only a non-carrier healthy sister of proband Trp848Ter. A recent report [22] has shown cosegregation for a SORL1 splice-site variant resulting in deletion of 58 amino acids (the complete Epidermal growth factor-like domain of the protein), which supports pathogenicity of splice-site changes. Of interest is that this same study [22] identified several SORL1 variants predicted to be “likely benign” or that did not segregate in the studied families.

Finally, we found the Gly852Ala mutation, reported to be also an extremely rare change in genomic databases, and with a high predicted pathogenic score by M-CAP program, but segregation could not be analyzed because of the unavailability of siblings.

A comparison of the Holstege et al. [18] criteria with our conclusions revealed mostly consistent results regarding the effect of seven of the variants. However, we considered as “likely benign” two variants which were predicted by Holstege et al. to be likely (Ala2173Thr) or possibly (Gly1871) pathogenic, based on one positive control and lack of segregation, respectively.

In summary, the study of our familial DAT cohort shows that some SORL1 missense variations are not a direct cause of the disease. The relationship of these allelic variants with increased risk of developing AD in the carrier members of these kindred, as occurs for APOE ɛ4 genotype, remains a possibility. In any case, SORL1 variants resulting in truncated SORLA protein could be a rare cause of early-onset familial AD.