Amyloid-β1–43 cerebrospinal fluid levels and the interpretation of APP, PSEN1 and PSEN2 mutations

Background Alzheimer’s disease (AD) mutations in amyloid precursor protein (APP) and presenilins (PSENs) could potentially lead to the production of longer amyloidogenic Aβ peptides. Amongst these, Aβ1–43 is more prone to aggregation and has higher toxic properties than the long-known Aβ1–42. However, a direct effect on Aβ1–43 in biomaterials of individuals carrying genetic mutations in the known AD genes is yet to be determined. Methods N = 1431 AD patients (n = 280 early-onset (EO) and n = 1151 late-onset (LO) AD) and 809 control individuals were genetically screened for APP and PSENs. For the first time, Aβ1–43 levels were analysed in cerebrospinal fluid (CSF) of 38 individuals carrying pathogenic or unclear rare mutations or the common PSEN1 p.E318G variant and compared with Aβ1–42 and Aβ1–40 CSF levels. The soluble sAPPα and sAPPβ species were also measured for the first time in mutation carriers. Results A known pathogenic mutation was identified in 5.7% of EOAD patients (4.6% PSEN1, 1.07% APP) and in 0.3% of LOAD patients. Furthermore, 12 known variants with unclear pathogenicity and 11 novel were identified. Pathogenic and unclear mutation carriers showed a significant reduction in CSF Aβ1–43 levels compared to controls (p = 0.037; < 0.001). CSF Aβ1–43 levels positively correlated with CSF Aβ1–42 in both pathogenic and unclear carriers and controls (all p < 0.001). The p.E318G carriers showed reduced Aβ1–43 levels (p < 0.001), though genetic association with AD was not detected. sAPPα and sAPPβ CSF levels were significantly reduced in the group of unclear (p = 0.006; 0.005) and p.E318G carriers (p = 0.004; 0.039), suggesting their possible involvement in AD. Finally, using Aβ1–43 and Aβ1–42 levels, we could re-classify as “likely pathogenic” 3 of the unclear mutations. Conclusion This is the first time that Aβ1–43 levels were analysed in CSF of AD patients with genetic mutations in the AD causal genes. The observed reduction of Aβ1–43 in APP and PSENs carriers highlights the pathogenic role of longer Aβ peptides in AD pathogenesis. Alterations in Aβ1–43 could prove useful in understanding the pathogenicity of unclear APP and PSENs variants, a critical step towards a more efficient genetic counselling.

Background AD is the most common cause of dementia, characterized by progressive cognitive decline and memory loss, accounting for 50 to 75% of all dementia patients [1]. Based on the disease onset, AD is classified into earlyonset AD (EOAD, < 65 years) and late-onset AD (LOAD, > 65 years) [2]. Mutations in amyloid precursor protein (APP), presenilin 1 (PSEN1) and presenilin 2 (PSEN2) have been identified as a cause of both EOAD and LOAD [3,4] explaining 10% of all EOAD and about 2% of LOAD patients [4,5]. The overproduction and aggregation of amyloid-β (Aβ) in the brain are thought to be the major causal events triggering AD that ultimately lead to neuronal loss [6]. Aβ peptides aggregate in the brain forming amyloid plaques, which together with neurofibrillary tangles (NFTs) of hyper-phosphorylated tau protein are the AD pathological hallmarks [6,7]. Aβ is generated from cleavages of APP through at least two distinct and mutually exclusive pathways. In the socalled non-amyloidogenic pathway, APP is cleaved by αsecretase and γ-secretase to produce three fragments: a secreted C-terminal fragment (sAPPα), p3 and the APP intracellular domain (AICD) [4]. In the amyloidogenic (pathogenic) pathway, APP is cleaved by β-secretase, followed by γ-secretase cleavage. The cleavage by βsecretase generates a large soluble extracellular secreted domain (sAPPβ) and C99. The latter undergoes additional cleavages by γ-secretase to generate a series of Aβ peptides 39-43 amino acids long, following two different pathways: Aβ 1-49 > Aβ 1-46 > Aβ 1-43 > Aβ 1-40 and Aβ 1-48 > Aβ 1-45 > Aβ 1-42 > Aβ 1-38 [8]. Shorter peptides can also be produced, including Aβ 1-41 from Aβ 1-43 [9] and Aβ 1-34 from either Aβ  or Aβ 1-40 [10], which is considered a biomarker of Aβ clearance and AD progression [11]. Pathogenic mutations in APP and PSEN1 and 2 (γ-secretase's catalytic subunits) are known to influence APP metabolism leading to the deposition of Aβ peptides. The most abundant Aβ peptides in the cerebrospinal fluid (CSF) resulting from APP processing are Aβ 1-38 , Aβ 1-40 and Aβ  . The latter is considered the most pathological peptide in AD as it is most prone to aggregation into amyloid plaques [12]. Aβ 1-42 levels have been found to be reduced in CSF of patients with AD [13] as a result of Aβ 1-42 increased production and subsequent accumulation into plaques [14]. Furthermore, CSF Aβ 1-42 levels are usually reduced up to decades before the clinical symptoms of dementia appear [15]. Thus, CSF Aβ 1-42 levels are considered a core biomarker for early AD [16], together with total tau protein (T-tau) (non-AD specific) and tau phosphorylated at threonine 181 (P-tau181) (AD specific) [17]. In addition to Aβ peptides, α and β cleaved soluble APP (sAPPα and sAPPβ) are products of the APP metabolism which also have been investigated as possible AD biomarkers, though with contradicting results [18,19]. Aβ 1-42 levels have been used for the pathological classification of AD mutations, both in vivo and in vitro [20]. However, some of the established AD pathogenic mutations do not show altered Aβ 1-42 levels [21]. For example, in the brain of the PSEN1 p.R278I knock-in mice, a decrease of Aβ  was accompanied by an increase of another Aβ, i.e. species, Aβ  , which showed higher aggregative properties than Aβ 1-42 [21]. Interestingly, Aβ  was also detected in the brain of sporadic and familial AD patients [22][23][24] supporting the hypothesis that the generation of relatively long Aβ peptides (>Aβ  ) could explain part of the pathogenic effect of the known deleterious PSENs and APP mutations [25,26]. Furthermore, several variants identified in the causal AD genes remain of uncertain significance (VUS) (www.alzforum.org/mutations; AD/FTD Mutation Database [27]), due to lack of functional studies and co-segregation with disease in relatives. Understanding the role of these variants is important for a correct clinical diagnosis, for genetic counselling and for the selection of well-stratified patient groups for clinical trials. The investigation of longer Aβ peptides, including Aβ 1-43 , in CSF of individuals carrying AD mutations or VUS could disclose possible alterations of the Aβ peptide production and explain their possible role in the AD pathogenesis. Recent studies showed that CSF Aβ 1-43 levels are significantly reduced in individuals with AD and mild cognitive impairment (MCI) [16,28], as well as in EOAD patients compared to LOAD [29], independently from the mutation status. Studies assessing Aβ 1-43 levels in CSF of PSENs and APP mutation carriers are therefore lacking. In the present work, we measured for the first time the Aβ 1-43 levels in CSF of carriers of APP and PSENs pathogenic and VUS mutations and control individuals and we compared these results to Aβ 1-42 and Aβ  . We also measured, for the first time in mutation carriers, sAPPα and sAPPβ in relation to Aβ 1-43 , as they are less investigated, but still part of the APP processing.

Study population
The study population consisted of 1431 AD patients (62.03% [n = 889] women, average age at onset (AAO) 73.51 ± 9.81 years, range 29-96 years), ascertained in Belgium through the neurology centres of the clinical partners of the Belgian neurology (BELNEU) consortium. All patients received a diagnosis of possible, probable or definite AD according to the criteria described by the National Institute of Neurological and Communication Disorders and Stroke Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA) criteria and the National Institute on Aging-Alzheimer's Association (NIA-AA) [30]. A positive family history of dementia (i.e. at least one first-degree relative affected) was documented in 22.22% (n = 318) of the patients. Based on the onset age (< 65 years), 19.57% (n = 280) of the patients were classified as having early-onset AD (EOAD, 54.64% [n = 153] women, average AAO 58.22 ± 6.14, age range 29-65 years). A Belgian control cohort of 809 individuals (73.54% [n = 595] women, average age at inclusion (AAI) 70.12 ± 10.71, 31-96 years) was included in the study. The control individuals were primarily community-dwelling volunteers or spouses of patients. Subjective memory complaints, neurologic or psychiatric antecedents, and a familial history of neurodegeneration were ruled out by means of an interview. Cognitive screening was performed using the Mini-Mental State Examination (cut-off score > 26) [31] and/or the Montreal Cognitive Assessment test (cut-off score > 25) [32]. The spouses of patients were examined at the Memory Clinic of ZNA Middelheim and Hoge Beuken in Antwerp, Belgium. CSF of additional 64 controls individuals (AAI 68.05 ± 5.97), used in this study for comparison purposes, were available at the Reference Centre for Biological Markers of Dementia (BIODEM), Department of Biomedical Sciences, University of Antwerp. These were primarily community-dwelling volunteers enrolled in the BACEi program (ClinicalTrials.gov identifiers: 54861911ALZ1005 (NCT01978548); 54861911ALZ2002 (NCT02260674)) by Janssen Pharmaceutica NV, Beerse, Belgium [33]. Subjective memory complaints, neurologic or psychiatric antecedents, were ruled out by means of an interview. Cognitive screening was performed using the mini-mental state examination (cut-off score > 26). The Clinical Dementia Rating Scale (CDR) score was 0, and AD CSF biomarkers were within the normal range. Genetic screening in these 64 individuals was not performed due to the unavailability of DNA. All research participants or their legal guardian provided written informed consent for participation in genetic and clinical studies. Clinical study protocols and informed consent forms for patient ascertainment were approved by the local medical ethics committees of the collaborating medical centres in Belgium. Genetic study protocols and informed consent forms were approved by the ethics committees of the University Hospital of Antwerp and the University of Antwerp, Belgium.

Mutation screening
Genomic DNA of patients and control individuals was analysed for mutations in APP, PSEN1 and PSEN2 using a targeted re-sequencing gene-panel as we previously described [34]. We selected non-synonymous variants with a minor allele frequency (MAF) < 1%, including exonintron boundaries for possible variants affecting splicing. The allelic frequencies of these variants, observed in the patient group, were assessed in the control cohort and Genome Aggregation database (gnomAD, lastly accessed in January 2020) [35]. In addition, PSEN1 rs17125721 A > G, p.E318G variant (MAF 1.927%, gnomAD), considered a possible risk modifier (https://www.alzforum.org/ mutations/psen1-e318g) and Apolipoprotein E (APOE) genotypes were analysed. Variants with a read depth below 20X, with Genoqual value below 99 and with an imbalanced wild-type/variant read depth (cut-off > 3) were considered false positives. All the selected variants were validated by Sanger sequencing (BigDye Terminator Cycle sequencing kit v3.1) on the ABI 3730 DNA Analyser (both Applied Biosystems). Sequences were analysed using SeqManII or novoSNP software packages [36]. To screen for APP locus duplications, multiplex amplicon quantification (MAQ) was used on 280 EOAD patients, as previously described [37]. Briefly, multiplex PCR amplification of 5 target and 11 reference amplicons was performed using fluorescently labelled primers. The amplification products were size separated on ABI 3730 automatic sequencer using GeneScan-500 LIZ (Applied Biosystems) as internal size standard. Data analysis was performed using the MAQ software (MAQs) package (www.vibgeneticservicefacility.be). One sample carrying the APP duplication and four different control samples were used as positive and negative controls, respectively.
Lumbar puncture, CSF sampling and handling have been performed according to standard protocols [12]. Samples were stored at − 80°C until analysis. CSF concentrations of Aβ 1-42 and Aβ 1-40 were measured using enzymelinked immunosorbent assay (ELISA) with commercially available single parameter ELISA kits at the BIODEM laboratory as previously described [38]. Aβ 1-43 levels were measured using the amyloid beta (1-43) (FL) Aβ kit (IBL) and both sAPPα and sAPPβ using sAPPα/sAPPβ kit (Meso Scale Discovery). Concerning T-tau and P-tau181 levels, these were previously measured using hTAU Ag and PHOSPHO-TAU(181P) (INNOTEST), respectively, and the BIODEM laboratory performed the ELISA for the 38 mutation carriers and the Sahlgrenska University Hospital (Sweden) for the 64 control individuals. For this reason, we did not perform statistical analysis on either T-tau or P-tau181 amongst the groups, but we only reported the values of the measurements, which are part of the phenotypic characterization of the cohort. All calibration standards and CSF samples were analysed in duplicate. Only mean values with a coefficient of variation (CV) of the replicates less than or equal to 20% were included in the analysis.

Statistical analyses
Statistical analyses were performed in SPSS version 24, and graphs were made using GraphPad Prism8. A logistic regression analysis was performed in the AD and control cohorts, to determine whether patients had higher probability than controls to carry both the common PSEN1 p.E318G and APOE ε4 genotype. With a similar model, we further tested the prevalence of APOE ε4 between EOAD and LOAD cohorts and of APOE ε4ε4 between EOAD and LOAD cohorts.
For the CSF biomarker analysis, the mutation carriers were divided in three different groups: carriers of known pathogenic mutations, of VUS and of PSEN1 p.E318G. First, a Kolmogorov-Smirnov test was performed to check for normal distribution. Since most variables did not follow a normal distribution, non-parametric tests were used. Differences amongst groups were tested using Kruskal-Wallis. Post hoc analysis with pairwise comparisons was carried out (adjusted p value < 0.05). To identify whether an association between two markers was present, the Spearman's rho correlation tests were performed. Correlation coefficients were extracted from each group (controls, known pathogenic, VUS, PSEN1 p.E318G) separately. After correction for multiple testing (Bonferroni's correction), p values < 0.005 or below were considered to be statistically significant. Receiver operating characteristic (ROC) curves were additionally performed to assess the diagnostic accuracy of the individual biomarkers, and the area under each ROC curve (AUC) was calculated. Finally, explanatory cut-offs were identified as the concentration of specific biomarker that maximizes sensitivity and specificity of the test (Youden's index).

Transcript analysis
To inhibit non-sense-mediated mRNA decay (NMD), 1 × 10 6 lymphoblast cells of the double mutation carrier (patient 16; PSEN1 p.G183V, PSEN1 p.P49L), of the single carrier (sibling of patient 16; PSEN1 p.G183V) and of 4 non-carriers as negative controls were incubated with 150 mg/mL cycloheximide (CHX) (Sigma-Aldrich) at 37°C for 4 h. To generate cDNA, total RNA was extracted using the RiboPureTM kit (Life Technologies) followed by a DNase treatment (TURBODNase Kit, Invitrogen). Firststrand complementary DNA was synthesized using the SuperScript III First-Strand Synthesis System (Life Technologies) using random hexamer primers. Full-length PSEN1 transcript sequencing was subsequently performed on a MinION sequencing platform (Oxford Nanopore Technologies (ONT)) based on a modernized cDNA sequencing protocol [39]. Briefly, cDNA amplification of PSEN1 full transcript was carried out with Platinum® Taq DNA Polymerase (Clontech Laboratories) using exonic primers, designed with Primer3Plus [40], with a 5′ ONT adapter. In a second PCR round, using PCR Barcoding Kit 96 (EXP-PBC096; ONT), sample-specific barcodes were added. Samples were pooled equimolar and processed according to ONT SQK-LSK109 library preparation. Sequencing was performed on a Mk1 MinION (MIN-101B), using FLO-MIN106 flow cells. Base calling and barcode de-multiplexing were performed with Albacore (v2.2.5). Sequencing reads were subsequently aligned using minimap2 [41], with splice-aware parameters. Only full-length PSEN1 spanning sequencing reads were retained for further analysis. Relative quantifications of splice junctions were calculated by dividing the number of junction-supporting reads by the total number of reads spanning the PSEN1 transcript in R (R Core Team, 2017).

CSF biomarker analysis
The CSF biomarkers were analysed comparing three groups of mutation carriers: (1) known pathogenic, (2) VUS and (3) PSEN1 p.E318G to control individuals. A significant reduction of CSF Aβ 1-43 levels was detected in all three carrier groups (all p < 0.05; Table 1; Fig. 1). For each carrier of a known pathogenic mutation or a VUS, Aβ 1-43 levels are shown in the bar plot (Fig. 2). Similarly, Aβ 1-42 CSF levels were significantly reduced in all three carrier groups (all p < 0.05; Table 1; Fig. 3). Comparison of CSF Aβ 1-40 levels did not show a significant difference amongst all groups (p = 0.66).

PSEN1 full transcript analysis
The effect of the PSEN1 p.G183V and PSEN1 p.P49L double mutation on PSEN1 alternative transcript generation in patient-derived RNA was examined with ONT MinION sequencing, because of a previously reported effect of this mutation on exon skipping in both HEK293 cells [42] and mice brain [43]. Both the double mutation carrier (patient 16; PSEN1 p.G183V, PSEN1 p.P49L) and the single PSEN1 p.G183V carrier show the skipping of exon 6 in 20% of the sequencing reads from the cDNA obtained from the lymphoblast cells treated with CHX and 5% from the untreated (Fig. 5). Exons 6-7 skipping was also detected, but it was a rarer event ( Figure S4, Additional file). The analysis of the transcript surrounding the PSEN1 p.P49L variants did not show splicing alterations (data not shown).

Mutation screening
The APP, PSEN1 and PSEN2 mutation screening in 1431 AD patients and 809 control individuals identified known pathogenic mutations in 5.7% of EOAD (16/280) and 0.3% of LOAD (4/1151) patients and in 0.12% (1/ 809) of the controls, in line with what it is usually observed [4,5]. The carrier of the APP p.A713T mutation considered with unclear pathogenicity but actually showing segregation with the disease in another study [44] was asymptomatic at the age of inclusion in our cohort (age > 65 years) [34]. This APP mutation is known to have a wide onset age range; therefore, it is not surprising to identify asymptomatic carriers [34,44]. Amongst the known pathogenic mutations, two were also found in AD patients with more than 65 years of onset: the known pathogenic PSEN1 p.C263F, identified in five patients with onset age range of 53-70, and the PSEN1 p.G183V, previously identified in a patient with frontotemporal dementia and Pick-type tauopathy [42] (further discussed later in the text). These findings highlight the importance to screen the causal AD genes also in LOAD patients [5]. The genetic screening identified also two double mutation carriers: patient 11 carrying the APP p.G625_S628del and the PSEN1 p.P355S, both VUS, and patient 16 having the PSEN1 p.G183V and the novel p.P49L.

PSEN1 p.G183V and p.P49L mutation carrier
Patient 16 received a diagnosis of probable AD (age at onset > 65). Clinically, patient 16 showed amnestic presentation without other remarkable signs or symptoms. Single-photon emission computed tomography (SPECT) displayed a moderate hypoperfusion of the bilateral parietal, temporal and frontal lobe, compatible with AD. Magnetic resonance imaging (MRI) showed age-related atrophy and multiple supratentorial lacunary infarcts. Neuropathological examination showed a mild atrophy of the frontotemporal gyri ( Figure S1, Additional file), but no FTD symptoms were reported in patient 16. The patient was pathologically diagnosed with definite AD (Montine stage A3B3C3), with an accent on the neurofibrillary pathology ( Figure S1). The sibling of patient 16 also carried the PSEN1 p.G183V, but not the PSEN1 p.P49L, and was diagnosed with probable AD. Both PSEN1 mutations therefore segregated independently in the family [45]. The PSEN1 p.G183V was previously identified in a patient diagnosed with frontotemporal dementia (FTD) and Pick-type tauopathy [42]. The brain lesions in patient 16 were however different from those of the published FTD patient, where severe frontotemporal atrophy and Pick-like pathology were described [42]. There were no intranuclear neuronal inclusions in patient 16 and no signs of Pick's disease. Thus, the Pick's pathology detected in the FTD patient [42] is probably independent of this PSEN1 p.G183V mutation. The PSEN1 p.G183V is located in the last nucleotide of the exonic splice donor site of exon 6, and experiments in HEK293 cells and mice showed the formation of alternative transcripts with skipping of exons 6 and exon 6-7, which are likely degraded by the non-sense-mediated mRNA decay control mechanism (NMD) [42,43]. Our PSEN1 transcript analysis with ONT minION sequencing, in patient-derived biomaterials, confirmed that PSEN1 p.G183V led to the formation of transcripts lacking exons 6 and 6-7, which are indeed degraded by NMD. Only a small amount of these alternative transcripts remains in the cells (≤ 5%) and they unlikely interfere with the wild-type PSEN1. The presence of the second PSEN1 mutation (p.P49L) could drive the disease in patient 16.    While longer Aβ peptides (e.g. Aβ 1-42 and Aβ 1-43 ) promote aggregation and neurotoxicity, Aβ 1-40 appears to act protectively [46]. Moreover, CSF Aβ 1-40 levels did not differ significantly between AD patients and controls in a meta-analysis [19]. Therefore, we speculated that the positive correlation of Aβ 1-43 with Aβ 1-42 , but not with Aβ 1-40 (except in PSEN1 p.E318G carriers), further supports the involvement of Aβ 1-43 in AD. It is however still unclear why Aβ 1-40 would be unaltered and lacking of correlation with Aβ 1-43 in the mutation carrier groups, as it is related to the process of Aβ 1-43 . It is plausible that a possible decrease of Aβ 1-40 in CSF would only be visible or measurable during the earlier stages of the disease when mostly all neurons are still intact, rather than in later stages (when it is normally measured), when Aβ production is higher but less neurons are preserved. Alternatively, it is possible that the cleavage of the C99 fragment by γ-secretase may be heavily impaired at the third cleavage step when Aβ 1-40 is generated from Aβ 1-43 , as suggested by Kakuda et al. [47] for the pathogenic PSEN1 p.R278I, which would result in negligible levels of Aβ 1-40 and unusually high levels of Aβ 1-43 . Furthermore, a recent study found a correlation between Aβ peptide length and plaque load (Aβ 1-43 > Aβ 1-42 > Aβ 1-40 ), indicating that longer Aβ peptides have an increased tendency towards accumulation in the brain [23], thus explaining their lower CSF levels. Further analysis of the shorter Aβ species Aβ 1-41 , Aβ 1-37 (from Aβ 1-43 ) Aβ 1-38 (from Aβ 1-42 ) and Aβ 1-34 (from either Aβ 1-43 or Aβ 1-42 ) are however needed to better clarify the role of longer Aβ peptides in AD and the pathogenic events linked to Aβ process, degradation and clearance in the presence of APP and PSENs mutations [9][10][11].
The production of longer Aβ peptides (> Aβ 1-42 ), including Aβ 1-43 , in the pathogenesis of AD is receiving increased attention [23,25,26,29,48]. Studies showed that the PSEN1 p.L435F produces primarily Aβ 1-43 , which was detected in the brain plaques [49] and induced pluripotent stem cell (iPSC) neurons [50] of patients carrying this mutation. Another work suggested that the APP mutations affect the endopeptidase activity of γ-secretase, leading to Aβ 1-42 formation, while PSEN1 mutations inhibit its carboxypeptidase activity, releasing multiple longer peptides including Aβ 1-42 and Aβ 1-43 [25]. Unfortunately, the number of mutations we investigated was not sufficient to make similar observations. Based on the CSF biomarker analyses performed, some of the PSENs VUS mutations analysed showed biomarker levels comparable to the known pathogenic mutations: for instance, the PSEN1 p.40del, which was described for the first time in an EOAD patient with prominent frontal features and no family history of dementia [51]. This variant was considered "most likely benign" or causing a small increase in risk based on its frequency in gnomAD [52].
However, an in vitro assay, testing the mutant PSEN1 ability to cleave the APP-C99 fragment, revealed a robust decrease in Aβ 1-42 production and undetectable levels of Aβ 1-40 [8]. Our data is in favour of an involvement of PSEN1 p.40del in Aβ alteration, showing decreased CSF levels of both Aβ 1-43 and Aβ 1-42 and unaltered Aβ 1-40 levels in the patient carrier, and also in line with a recent study showing increased Aβ 1-42 and unaltered Aβ 1-40 in the medium collected from mice neuro-2A cells transfected with the plasmid containing the PSEN1 p.40del mutation [53]. The PSEN2 p.S130L variant was first reported in an Italian family, but segregation with disease could not be determined [54]. Another study described a carrier of this variant who had an autopsy-confirmed AD diagnosis [55]. The pathogenicity of PSEN2 p.S130L is currently unclear, since it did not affect Aβ 1-42 levels or Aβ 1-42 /Aβ 1-40 ratio in in vitro experiments [56]. According to an algorithm proposed by Guerreiro et al. [57], this variant has been considered "probably pathogenic." In line with this classification, we found a significant reduction of both Aβ 1-43 /Aβ 1-40 and Aβ 1-42 /Aβ 1-40 ratios in the patient carrier, pointing out to a possible pathogenicity of PSEN2 p.S130L. Finally, the carrier of the novel PSEN2 p.T153S variant also had reduced CSF levels of Aβ 1-43 and Aβ 1-42 , comparable to the known pathogenic mutations. In light of these data, we could consider those mutations as "likely pathogenic", since they seem to affect or interfere with Aβ production. Furthermore, it is important to note that the three mutation carriers were all APOE ε4 negative, meaning that the detected Aβ alteration is likely due to the mutations themselves without an effect of APOE. For the APP variants, instead, we could not drive conclusions on the variants pathogenicity. In fact, all the APP variants analysed were located outside the pathogenic exons 16-17, encoding for the Aβ domain. Furthermore, two carriers of APP variants were considered controls at the moment of inclusion: control 3 carried the APP p.S198P variant and CSF analysis did not reveal Aβ alterations; control 4, carrying the APP p.R328W and having altered CSF marker levels, developed vascular dementia at follow-up. Of note, control 3 but not control 4 was APOE ε4 negative. We would like to stress the importance to further investigate a possible interference of the APP variants outside the Aβ domain in AD. There was variability of Aβ 1-43 CSF levels between the carriers of the same mutations (PSEN1 p.A79V, PSEN1 C263F, APP R328W, APP E599K, PSEN2 R62H, Fig. 2; Table 2). We initially hypothesized an effect of APOE ε4 on the decreased Aβ 1-43 CSF levels as already demonstrated for Aβ [58]. This seemed a valid hypothesis for all mutations but one (PSEN1 p.C263F, Fig. 2; Table 2). However, a recent study showed a reduced influence of APOE on Aβ 1-43 aggregation in cerebrovascular cells [48].

Aβ 1-43 CSF levels in PSEN1 p.E318G carriers
To better understand the role of PSEN1 p.E318G on AD pathology, we investigated the CSF Aβ 1-43 levels in carriers, detecting a reduction. This PSEN1 variant is suggested to be an AD risk modifier as it was associated with high levels of P-tau181P and T-tau [59,60]. Moreover, carriers of both APOE ε4 and PSEN1 p.E318G variant were reported to have a higher risk of developing LOAD than APOE ε4 carriers without the PSEN1 variant [59,60]. In our study, however, the PSEN1 p.E318G variant did not show an association with AD regardless of APOE ε4 genotype, which is in line with the study of Hippen et al. [61]. A recent analysis of CSF Aβ 1-42 , Ttau and P-tau181 in asymptomatic p.E318G carriers of a large LOAD Italian family did not reveal any changes in CSF markers [62]. The levels of Aβ 1-43 were not measured and therefore not allowing a direct comparison.

Soluble species sAPPα and sAPPβ
We also detected reduced levels of sAPPα and sAPPβ in the three carrier groups analysed compared to controls, but statistical significance was not detected in the group of the known pathogenic mutations. Our results are different from other studies where no difference in sAPPα and sAPPβ levels was observed between patients and controls [19]. Of note, in these studies, the mutation status of the patients was not described [19].
We report that both sAPPα and sAPPβ positively correlated with Aβ 1-43 and Aβ 1-42 only in the control group but not in the mutation carriers. This result is in line with another study where the correlation between Aβ 1-42 and sAPPα or sAPPβ in AD patients was not significant [63]. We cannot explain why both sAPPα and sAPPβ levels would be reduced in the mutation carriers and correlate with the control group. In fact, sAPPα and sAPPβ should not be affected by the presence of AD mutations, which affect γ-secretase activity, while sAPPα and sAPPβ are respectively generated by αand βsecretases in the earlier steps, before Aβ production. However, the data of CSF sAPPα and sAPPβ levels in AD are unclear [19] and there are no studies so far that have measured CSF sAPPα and sAPPβ in AD mutation carriers. The possible involvement of sAPP fragments in AD is still not fully understood [18], and the interpretation of the results remains therefore challenging. It is suggested that sAPPα enhances neurogenesis and has a neuro-protective role, and it is involved in neurotransmission and synaptic plasticity [64]. A recent study, however, showed that all fragments of sAPP acted as agonists of a specific GABA receptor (GABAB-R1a) and would therefore explain their role in neurotransmission and plasticity [65]. In fact, a fragment derived from sAPPα inhibited synaptic transmission in mouse hippocampus. Additional studies in carriers of AD mutations are essential to fully understand the involvement of sAPP in AD.

Limitations
A potential limitation of this study is the small cohort size used to analyse Aβ 1-43 and the other markers in CSF of APP and PSENs mutation carriers, despite the genetic screening was performed in a relatively large AD population. As we know, APP and PSENs mutation carriers are extremely rare [4], as well as the availability of their biomaterials. Therefore, a replication study was not performed. The availability of a larger control cohort allowed us to reach satistical significance when analysing Aβ 1-43 CSF levels. Unfortunately, age at onset, gender and APOE genotype could not be used as covariates in the analysis, because the number of mutation carriers was not sufficient for the statistical test. Our pilot study is nevertheless relevant because it highlights the involvement of Aβ 1-43 in AD and adds missing information regarding specific mutations to very recent studies published during these last 2 years [23,25,48]. As CSF Aβ 1-43 has never been studied before in CSF of APP and PSENs mutation carriers, the results need to be interpreted with caution and more studies are needed to replicate these findings in larger cohorts. Analysis of the shorter Aβ peptides (e.g. Aβ 1-34 , Aβ 1-37 , Aβ 1-38 , Aβ 1-41 ) could provide a more comprehensive interpretation of our results. Lastly, in our study, it was not possible to assess the difference between PSEN1 and APP mutations in affecting γ-secretase and Aβ generation [25]; therefore, this aspect warrants further follow-up, also in terms of interpretation of novel and VUS mutations in the AD genes.

Conclusions
In this study, CSF levels of the long Aβ peptide Aβ 1-43 were investigated for the first time in carriers of known pathogenic and unclear (VUS) APP and PSENs mutations. We observed a significant reduction of CSF Aβ 1-43 levels and a positive correlation with Aβ 1-42 in all mutation carrier groups. We suggested the re-classification of three VUS into "likely pathogenic", as their biomarker levels were comparable to the known pathogenic mutations. We added important information on the debatable genetic modifier PSEN1 p.E318G and we were able to clarify the role of PSEN1 p.G183V, using ONT long-read sequencing, considered so far pathogenic, but probably not involved in AD.
From a clinical perspective, our data could prove useful. A recent study showed that Aβ43 was cleared more than Aβ42 in plaques of patients treated with Aβ immunotherapy [23]. These are important data that open new possibilities for personalized medicine in patients with AD, who have a high Aβ 1-43 load in the brain. For these reasons, Aβ 1-43 could be considered an added AD biomarker together with the others already in use. Despite the small study cohort, the data presented here corroborate previous findings on Aβ   [21,25,26,29,49], suggesting a possible involvement of longer Aβ peptides in the AD pathophysiology.
CSF levels of sAPPα and sAPPβ were also analysed for the first time in AD mutation carriers, where their levels were reduced compared to controls. Based on the recent analysis on sAPP fragments as regulators of (GABA) neurotransmission, further investigation in AD is important to study more in detail this pathway to open new venues for therapy strategies.
Finally, functional work using patient biomaterials can prove valuable to better understand the pathogenicity of unclear AD mutations. This will be useful for patient stratification for clinical trials, genetic counselling and therapy development.
Additional file 1: Table S1. List of the identified APP, PSEN1 and PSEN2 mutations. Table S2. Diagnostic accuracy of the different markers and ratios to discriminate between controls and mutation carriers, measured by ROC curve analysis. Table S3. Descriptive features of PSEN1 p.E318G mutation carriers with CSF. Figure S1. Neuropathology of Patient 16. Right lateral (A) and right medial (B) hemispheres showing brain atrophy. Atrophy can be observed also at the ventricles and thalamus (C). The 4G8 staining shows amyloid plaques in the hippocampus CA4 (D). Classic neurofibrillary tangles are present in the hippocampal CA4 (AT8 stain) (E). Figure S2. Area under the curve (AUC) calculated for the three mutation carrier groups compared to controls of Aβ 1-43 , Aβ 1-43 /Aβ 1-40 , Aβ 1-42 , Aβ 1-42 /Aβ 1-40 , Aβ 1-40 and Aβ 1-43 /Aβ  . AUC are calculated for the know pathogenic (red), VUS (orange) and PSEN1 p.E318G (green) mutation carrier groups compared to the control group. The AUC values and the ones for sensitivity and specificity are listed in Table S2. Figure S3. Area under the curve (AUC) calculated for the three mutation carrier groups compared to controls of sAPPα, sAPPβ, Aβ1 -43 /sAPPα and Aβ 1-43 /sAPPβ. AUC are calculated the know pathogenic (red), VUS (orange) and PSEN1 p.E318G (green) mutation carrier groups compared to the control group.
The AUC values and the ones for sensitivity and specificity are listed in Table S2. Figure S4. Transcript analysis of PSEN1 in Patient 16. The bar graph shows the relative quantifications of exon 6-7 in the double carrier (Patient 16; PSEN1 p.G183V, PSEN1 p.P49L), the single carrier (sibling of Patient 16; PSEN1 p.G183V), and 4 non-carriers lymphoblast cells CHX treated (CHX) and untreated (UNT). Relative quantifications of splice junctions were calculated by dividing the number of junction-supporting reads by the total number of reads spanning the PSEN1 transcript. The quantifications for both CHX and UNT of the non-carriers are reported as averages (values of SD for CHX ± 0,001052869 and for UNT ± 0,000671837).