A family (n = 14) originally from Peru, with five members diagnosed with neurological and neuropsychiatry disorders, was enrolled in this study. Non-familial patients with Alzheimer’s disease (n = 8) and healthy individuals (n = 50) were recruited for the variant validation study. The selection criteria for the healthy individuals were as follows: age > 60 years, without signs of dementia, and no familial history of Alzheimer’s disease. Probable Alzheimer’s disease was diagnosed according to the guidelines of the National Institute of Neurological and Communicative Disorders and the Stroke and Alzheimer Disease and Related Disorders Association (McKhann et al., 2011). Whenever possible, the cognitive status of each family member was diagnosed based on a neuropsychological test (MoCA blind test and the clock drawing test). The cutoff for a normal MoCA score was 18, and for a normal clock drawing test was 6.

Genomic DNA was extracted from saliva samples using the prepIT.L2P reagent (Genotek, Cat. No PT-L2P-5) according to the manufacturer’s instructions. The qualifying genomic DNA samples were randomly fragmented using Covaris Technology, obtaining a fragment of 350 bp. The DNA nanoballs (DNBs) were produced using rolling circle amplification (RCA), and the qualified DNBs were loaded into the patterned nanoarrays. The WGS was conducted on the BGISEQ-500 platform (BGI Genomics, Shenzhen, China). Raw sequencing reads were aligned to the human reference genome (GRCh38/HG38) with the Burrows–Wheeler Aligner (BWA) software and variant calling was performed with the Genome Analysis Toolkit (GATK v3.5) according to best practice. On average, 88.10% mapped successfully and 93.23% mapped uniquely. The duplicate reads were removed from the total mapped reads, resulting in a duplicate rate of 2.48% and a 30.72-fold mean sequencing depth on the whole genome, excluding gap regions. On average per sequencing individual, 99.34% of the whole genome, excluding gap regions, were covered by at least 1 × coverage, 98.78% had at least 4 × coverage, and 97.42% had at least 10 × coverage. The whole-genome sequencing analysis pipeline is presented in Supplementary Figure 1.

Based on the results of the WGS, the variants that were present in affected members of the family but absent in healthy individuals were validated using Sanger sequencing. All PCR products were sequenced using an ABI 3130 Genetic Analyzer. Sequence analysis was performed with the Chromas program in the DNASTAR analysis package. The PCR and sequencing primers are shown in Table 1.

Primary human vascular smooth muscle cells (VSMCs) from carotid of healthy donors were purchased from Cell Applications Inc. (Cat. No 3514k-05a, neural crest origin). Human brain vascular pericytes were purchased from ScienCell (Cat. No 1200). Human neurons (SH-SY5Y) were purchased from ATCC (Cat. No CRL-2266). Human astrocytes were purchased from Cell Applications Inc. (Cat. No 882A05f).

Total RNA was extracted using a miRNeasy kit (Qiagen, Cat. No 217084) following the manufacturer’s protocol instructions. The BGISEQ platform was used for RNA-seq, generating some 4.28G Gb bases per sample, on average. The average mapping ratio with the reference genome was 97.01% and the average mapping ratio with the gene was 74.05%; 17,029 genes were identified. We used HISAT to align the clean reads to the reference genome and Bowtie2 to align the clean reads to the reference genes. A total of 100 ng of total RNA was used for qPCR as the starting template for cDNA synthesis. The cDNA was prepared by reverse transcription (RT), and gene expression was analyzed by quantitative PCR (qPCR) on an SYBR green system (Applied Biosystems). Expression results were analyzed using the DDCT method, and GAPDH (encoding glyceraldehyde-3-phosphate dehydrogenase) was used as a housekeeping gene. Fold changes were calculated as the average relative to the control carotid as the baseline.

Genealogic investigations allowed us to identify five members of a Peruvian family with a strong clinical history of ADRD, including dementia, Alzheimer’s disease, and schizophrenia across two generations (Figure 1A; Supplementary material Table 1). To detect genetic risk factors associated with the development of the neurologic disorders observed in this family, we performed a WGS analysis on affected and healthy members (n = 14) of the family. By using the BGISEQ-500 platform, we obtained an average of 113,895.38 Mb of raw bases. After removing low-quality reads, we obtained an average of 106,676.25 Mb clean reads, identifying a total of 3,933,470 SNPs. We then selected coding variants that met the following two criteria: first, candidate variants that harbored at least one “disruptive” or missense variant, and second, variants that were present in affected probands, but not in unaffected members of the family. As a result, we identified three coding variants that segregated across two generations of affected individuals (Supplementary material Table 2). These variants were found to be located at chr6:167728791 (UNC93A; rs7739897), at chr6:170047902 (WDR27; rs61740334), and at chr6:170058374 (WDR27; rs3800544; Figures 1B,C). Two different Sanger PCR sequencing platforms were used to validate the presence of these SNPs (Figure 1D). Several studies have also found SNPs in genes located on chromosome 6q with a significant connection to neurological diseases (Kohn and Lerer, 2005; Naj et al., 2010); however, the UNC93A and WDR27 variants have not yet been associated with ADRD. It is worthy of note that none of the currently known Alzheimer’s disease-associated variants were present in this Peruvian family.

To further confirm an association between the UNC93A and WDR27 variants and familial genetics risk for neurologic disorders, we analyzed their presence in unrelated healthy Peruvians (n = 50) and unrelated individuals with neurological disorders (probable Alzheimer’s disease, n = 8). As shown in Supplementary material Table 3, the UNC93A variant (V409I) was present in 1/50 of the healthy group, and the WDR27 variants (Thr542S and Arg467His) were present in 2/50 of the healthy group. Interestingly, the three variants did not co-exist in any healthy individuals and were absent in volunteers diagnosed with probable Alzheimer’s disease with no familial history of ADRD. These findings suggest that the co-occurrence of these three variants may be related to neurological disorders in a Peruvian family.

UNC93A genes encode a transmembrane protein (457 amino acids) that has 11 alpha-helices and is mainly expressed in the brain, kidney, and liver (Ceder et al., 2017). The WDR27 gene encodes a scaffold protein with multiple WD repeat domains and is ubiquitously expressed in the human body. We used in silico approaches to provide insights into the molecular and structural effect of the UNC93A (V409I) and WDR27 (Arg467His and T542S) variants associated with familial ADRD. We, therefore, built the human structure of the UNC93A and WDR27 proteins by homology modeling and performed site-direct mutagenesis to generate the mutated proteins using the UCSF Chimera software (Figures 2A,B). Molecular dynamics simulations (MDS) for 500 ns were performed to stabilize the physical motions of atoms in both proteins to mimic physiological conditions. Importantly, we observed that at the beginning of the MDS (0–50 ns), the residue V409 of the wild-type UNC93 protein interacts with the lipid bilayer of the cell membrane. After 500 ns of MDS, the V409 is internalized toward the protein-active transmembrane conduct region in which the subsequent interactions with ligands or ionic exchanges occur. These movements were characterized by high structural vibration and epitope exposure of the UNC93A’s ectodomain (aa 200–300) of the protein toward the surface of the cell membrane (Figure 2C). In contrast, the mutant I409 blocks the internalization of this residue and loses its capacity to move toward the protein-active transmembrane conduct region due to the loss of about 50% of the global residual vibration (Figure 2D). As a result of this change in the amino acid, the UNC93A loses its capacity for ion exchange and interaction with potential ligands or partners. Regarding the WDR27 protein, both His467 and S542 variants induce the internalization of these residues in the part of the hinge domain of the protein predicting the loss of function and capacity to interact with other partners (Figure 2E). We also observed that the wild-type and mutant proteins are very stable due to their close residual vibration and epitope exposure patterns (Figure 2F). The solvent accessible surface area (SASA) analysis demonstrated that the UNC93A (I409) protein increased the surface area of the mutated amino acid and its environment (Figure 2G), while the WDR27 variants reduced its SASA (Figure 2H). The effect on the amino acid substitutions for both proteins was determined using the Molecular Mechanics Poisson–Boltzmann Surface Area (MM/PBSA) calculation of free energies and energy contribution using the last 150 ns of MD trajectories. Remarkably, the I409 increased the affinity of the protein to the membrane, indicating a reduced capacity for internalization, while the WDR27 variants reduced the protein stabilization of both His467 and S542 mutated amino acids. Together, these findings indicate that the UNC93A and WDR27 variants have a strong effect on the functionality and ability to interact with their environment, and thus potentially affect brain homeostasis.

We next investigated how the co-occurring inhibition of UNC93A and WDR27 gene expression affects the cellular homeostasis of brain cell lines, including neurons, pericytes, astrocytes, and VSMCs. To achieve this goal, we simultaneously silenced both UNC93A and WDR27 genes using a siRNA approach and performed a high throughput RNA sequencing (Supplementary material Figure 2). Differential expression analysis demonstrates that VSMCs had the highest numbers of differentially expressed genes (DEG = 2,231), followed by pericytes (DEG = 2091), while astrocytes (DEG = 226) and neurons (DEG = 191) showed a modest change in gene expression compared with control groups (Figure 3A), suggesting that mutations promoting a loss of function of the UNC93A and WDR27 genes affect the neurovascular unit of the brain. Pathway analysis of VSMC and pericyte gene signatures identified enrichment for multiple known molecular pathways associated with neurological disorders, such as impaired autophagy pathways signaling (Li et al., 2016), ubiquitin-mediated proteolysis (Upadhya and Hegde, 2007), unproductive metabolism signaling (Van Der Velpen et al., 2019) inflammation, and necroptosis (Zhang et al., 2021; Figure 3B), while neurons showed a gene signature enriched for cellular senescence, cytokine–cytokine interactions, and apoptosis. Astrocytes showed only modest enrichment for necroptosis and the proteasomal degradation pathway (Figure 3B). Similar to our results, previous in vivo studies have reported a direct connection between autophagy activation and UNC93A levels in the healthy brains of mice under starvation, indicating the potential role of UNC93A in metabolic stability, energy uptake, and nutrient transport in the brain (Ceder et al., 2017). Interestingly, several neurological diseases have been associated with defects of autophagy and metabolism homeostasis, including Alzheimer’s, Parkinson’s, and Huntington’s diseases (Wang et al., 2016; Croce and Yamamoto, 2019; Xu et al., 2021). We also observed that multiple genes previously related to Alzheimer’s disease, such as CLU (Jun et al., 2010) SQSTM1 (Cuyvers et al., 2015) GPC6 (Kunkle et al., 2021), and ABCA7 (De Roeck et al., 2019), were dysregulated in brain cells deficient in UNC93A and WDR27 (Figure 3C). Interestingly, the strongest genetic risk factor for Alzheimer’s disease, APOE, was also modulated in pericytes and neurons (Figure 3C).