Integrative genomic and functional analyses reveal NINL as a modulator of tau aggregation

Abstract

INTRODUCTION

Proteostasis dysfunction is a hallmark of frontotemporal dementia (FTD) and Alzheimer's disease (AD), yet the genetic and molecular pathways that disrupt protein homeostasis remain poorly understood.

METHODS

We integrated human genetics, transcriptomics, and functional studies to identify proteostasis network components involved in tauopathy.

RESULTS

We identified 18 proteostasis network genes harboring 75 rare, damaging variants enriched in FTD and/or AD. These genes, spanning multiple proteostasis pathways, were differentially expressed in microtubule associated protein tau (MAPT) mutant neurons and dysregulated in FTD and AD brains. NINL, which encodes Nlp, emerged as the only gene consistently upregulated across all datasets. NINL overexpression reduced tau seeding and enhanced lysosomal proteolytic activity, whereas two FTD‐enriched NINL frame shift variants impaired Nlp expression and abolished these protective effects.

DISCUSSION

We identified a set of proteostasis genes with genetic and transcriptional links to neurodegeneration and revealed NINL as a novel regulator of tau aggregation.

Highlights

• Rare proteostasis variants are enriched in frontotemporal dementia (FTD) and Alzheimer's disease (AD) cases.

• Proteostasis genes are dysregulated in microtubule associated protein tau (MAPT) neurons and tauopathy brains.

• NINL is consistently upregulated across stem cell and human brain datasets.

• NINL overexpression reduces tau seeding and boosts lysosomal proteolysis.

• FTD‐linked NINL variants abolish NINL's protective effects on tau seeding.

1. BACKGROUND

Alzheimer's disease (AD) and a subset of frontotemporal degeneration (FTD) cases, known as frontotemporal lobar degeneration with tau inclusions (FTLD‐tau), share the hallmark feature of selective neuronal vulnerability and tau aggregation in the brain and are collectively referred to as tauopathies. Although common genetic risk loci and causal mutations have been identified in tauopathies, ¹ , ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ a substantial proportion of disease heritability remains unexplained, suggesting that additional genetic factors, particularly rare variants, contribute to disease susceptibility. Defining these genetic contributors provides critical insight into the molecular mechanisms that drive neurodegeneration, revealing core pathogenic pathways and potential therapeutic targets.

Disruption of proteostasis is a feature of both FTD and AD. Impaired autophagy‐lysosome function, defective chaperone activity, and the accumulation of aggregation‐prone proteins, such as tau, converge to drive neuronal dysfunction and disease pathophysiology in FTD and AD. ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² While prior studies implicate proteostasis pathways in disease risk, ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ the extent to which rare genetic variation affects components of the proteostasis network and how such variation translates into molecular dysfunction in human neurons and brain tissue remains poorly understood. Moreover, it is unclear whether specific proteostasis genes play direct, mechanistically actionable roles in modulating tau pathology.

By combining human genetics, disease model systems, and functional biology, our study aimed to define proteostasis components that contribute to tau pathophysiology. Here, we integrated rare variant analyses from FTD and sporadic, early‐onset AD cohorts with transcriptomic data from MAPT mutant induced pluripotent stem cell (iPSC)‐derived neurons and human tauopathy brain tissue to identify proteostasis network genes with coordinated genetic and transcriptional alterations. Across these datasets, we identified NINL, a motor‐associated and trafficking‐related protein, as consistently upregulated in tau‐associated disease. Using functional assays, we validated NINL as a novel regulator of tau aggregation.

2. METHODS

2.1. Whole genome sequencing and rare variant analysis

We performed secondary analyses of previously reported whole genome sequencing (WGS) data of early onset AD (EOAD), FTD, and control patients (Table S1). ⁷ Cases and controls were clinically evaluated as previously described. ²¹ , ²² All participants or their surrogates provided written informed consent to participate in the study, and institutional review boards approved all aspects of the study.

Genome sequencing methods are provided in more detail in Cochran et al., 2020. ⁷ Briefly, short‐read, paired‐end WGS was performed using Illumina HiSeq X. On average, 92% of bases were covered at a sequencing depth of 20x, along with a mean sequencing depth of 34x. ⁷ Sequencing reads were aligned to the reference genome hg19, and variants were called using GATK 3.8. ⁷ Population frequency was determined using 1000 Genomes Phase 3, TOPMed Bravo, dbSNP (release 151), WGSA 0.7, which includes ExAC, gnomAD, ESP, and UK10K. ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ Variants were filtered using SnpSift 4.3s, along with local and population frequency, predicted deleteriousness (annotated using CADD v1.3), ²⁹  and segmentation for function. ⁷ Variants analyzed had a maximum minor allele count of three (0.1% frequency in the cohort) and a minor allele frequency of 0.01% in population databases.

2.2. RNA sequencing analysis

RNA sequencing of iPSC‐derived neurons was previously performed and described in Minaya et al 2023. ³⁰ Briefly, samples were sequenced by an Illumina HiSeq 4000 Systems Technology with a read length of 1 × 150 bp and an average library size of 36.5 ± 12.2 million reads per sample. Salmon (v. 0.11.3) ³¹ was used to quantify gene expression (GRCh38.p13). Differential expression analyses were performed using DESeq2 (v.1.22.2) ³² .

Proteostasis gene expression in tauopathy patient brains was analyzed using summary statistics from a publicly available dataset: the temporal cortex of 80 control, 84 progressive supranuclear palsy (PSP), and 82 AD brains (syn6090811 and syn6090813). ³³ Summary statistics from a previously published transcriptomic dataset from the middle temporal gyrus of FTLD‐tau patients with MAPT IVS10+16 mutation (n = 2) and neuropathology free controls (n = 3) were also analyzed using DESeq2 (v.1.22.2). ³⁰ , ³² To define the neuronal specific proteostasis gene expression changes, summary statistics generated using the R package nebula (v1.1.5) ³⁴ from all neuronal nuclei were analyzed in control (n = 9) and AD (n = 28) brains as described in Brase et al 2023 ³⁵ .

RESEARCH IN CONTEXT

1. Systematic review: Prior literature suggests that proteostasis dysfunction is involved in frontotemporal lobar degeneration (FTD) and Alzheimer's disease (AD); however, few studies have integrated rare variant analyses, stem cell models, and human brain data to identify convergent pathways or mechanistic regulators.

2. Interpretation: Our findings demonstrate genetic and transcriptional perturbations in proteostasis network genes in FTD and AD. We identify NINL as a novel modulator of tau aggregation and lysosomal proteolytic activity. These results advance understanding of tau clearance mechanisms.

3. Future directions: Future work will be necessary to define the impact of NINL variants in neuronal and in vivo models and to evaluate whether enhancing NINL function improves tau degradation in vivo. Additionally, systematic functional testing of additional proteostasis genes containing rare variants may uncover broader therapeutic targets for FTD and AD.

2.3. Cell culture

HEK293T or HEK293T‐tau biosensor cells ³⁶ were cultured in Dulbecco's modified Eagle medium (DMEM; Gibco, Cat #: 61965‐059), supplemented with 10% fetal bovine serum (FBS; Gibco, Cat #: 10270106) and 1% penicillin‐streptomycin (Gibco, Cat #: 15140122). Cells were passaged by incubating cells in 0.05% Trypsin‐ethylenediaminetetraacetic acid (EDTA) (ThermoFisher, Cat #: 25300054) for 5 minutes at 37°C. Trypsin was quenched using DMEM with 10% FBS, and cells were centrifuged at 300 x g for 5 minutes and resuspended in fresh media before replating. Cells were maintained in a humidified atmosphere with 5% CO₂ at 37°C and were used within 5–15 passages. Cells were confirmed to be negative for mycoplasma contamination.

2.4. Plasmids and transient transfection

To evaluate the impact of NINL on cellular phenotypes, we used a pCMV6 plasmid containing NINL wild‐type (WT) with a C‐terminal myc‐DDK‐tag (OriGene, Cat #: RC206094; Figure S1). To model rare variants in NINL, p.T275fs (c.823_824delAC) and p.R1202fs (c.3604dupA) variants were introduced into the pCMV6 NINL WT‐Myc‐DDK plasmid using site‐directed mutagenesis (mutagenesis performed by Azenta; Figure S1). Control (vector and NINL WT) and mutagenized plasmids were grown using an endotoxin free maxi prep kit (Qiagen, Cat #: 12362) following the manufacturer's instructions, and whole plasmid sequencing was performed.

HEK293T or HEK293T‐tau biosensor cells ³⁶ were transiently transfected with pCMV6 NINL WT, p.T275fs, p.R1202fs, or an empty vector control using Lipofectamine 2000 following the manufacturer's instructions (ThermoFisher, Cat #: 11668019). Briefly, cells were seeded in a poly‐D‐lysine (PDL; ThermoFisher, Cat #: A3890401) coated plate at 50% confluence. Plasmid (25 ng) was mixed with 1ul of Lipofectamine 2000 in 1 mL of serum‐free Opti‐MEM (Gibco, Cat #: 31985062) and incubated at room temperature for 15 minutes. Eighteen hours after adding the DNA/Lipofectamine complexes to the cells, the medium was replaced with fresh DMEM containing 10% FBS. Cells were utilized for downstream analysis at 24‐ or 48‐hours post‐transfection as indicated below. Plasmids are available upon request.

2.5. Quantitative polymerase chain reaction

Gene expression was evaluated using quantitative polymerase chain reaction (qPCR) via SYBR Green iTaq Universal SYBR Green Supermix (Bio‐Rad, Cat #: 1725121). Primers specific to NINL included 5′‐ACCTGGGATTCTGAGGACTTTG‐3′ (forward) and 5′‐ACTTTGCCGTCTCCGTCTTGAT‐3′ (reverse). Technical replicates were performed for each sample. NINL assays were run in separate wells from the housekeeping gene, GAPDH, to prevent interference. Primers for GAPDH included 5′‐TGCACCACCAACTGCTTAGC‐3′ (forward) and 5′‐ GGCATGGACTGTGGTCATGAG‐3′ (reverse). Data were analyzed via the ∆∆C_T (comparative) method. Relative expression was quantified using the comparative Ct method. NINL Ct values were normalized to glyceraldehyde‐3‐phosphate dehydrogenase (GAPDH) for each sample and then compared to the mean normalized expression of the control cohort. Only samples for which the standard error was less than 20% between technical replicates were analyzed.

2.6. LysoTracker and LysoSensor

Acidic vesicles were measured by flow cytometry using LysoTracker and LysoSensor. HEK293T cells were transiently transfected with control or NINL plasmids for 48 hours using Lipofectamine 2000 (ThermoFisher, Cat #: 11668019) as described above. Cells were then incubated with LysoTracker Red (ThermoFisher, Cat #: L7528) at a dilution of 1:10,000 in DMEM supplemented with 10% FBS for 25 minutes at 37°C. LysoSensor Green (1:1,000; ThermoFisher, Cat #: L7535) was added to the cells for 60 s at the end of the LysoTracker incubation. Cells were then rinsed with phosphate buffered saline (PBS) and collected via trypsinization (0.05%). The live cell population was labeled with Far Red Live/Dead (ThermoFisher, Cat #: L34974) stain at 1:1,000 in PBS for 25 minutes. Flow cytometry was performed using a BD FACSymphony™ A1. LysoTracker Red was measured using a 405 nm laser with a BV480 filter, LysoSensor Green was measured using a 561 nm laser and a PE filter, and Far‐Red Live/Dead was measured using a 637 nm laser and an APC filter. We analyzed 10,000 cells per biological replicate. Data were processed using FlowJo software (v10).

2.7. Immunoblotting

Immunoblotting was performed to measure expression of proteins in the autophagy‐lysosome pathway. HEK293T cells were transiently transfected with control or NINL plasmids for 48 hours using Lipofectamine 2000 (ThermoFisher, Cat #: 11668019) as described above. Cells were harvested using trypsinization (0.05%), washed with ice‐cold 1x PBS containing a protease inhibitor cocktail (ThermoFisher, Cat #: P2714), and centrifuged at 500 x g for 5 min at 4°C. Cells were lysed by sonication in RIPA buffer containing 50 mM Tris (pH 7.4), 150 mM NaCl, 1% Triton X‐100, 1% sodium deoxycholate, 0.1% sodium dodecyl sulfate (SDS), 50 mM NaF, 1 mM Na2VO4, 5 mM EDTA, 1x phosphatase inhibitor (Sigma‐Aldrich, Cat #: 4906845001), and protease inhibitor cocktail (1:500; Sigma‐Aldrich, Cat #: P2714). Lysates were then centrifuged at 16000xg for 10 minutes at 4°C and transferred to a fresh tube. Protein concentration was determined using BCA (ThermoFisher, Cat #: 23225). Equal amounts of total protein (15 ug) were incubated at 70°C for 10 minutes in SDS sample buffer (ThermoFisher, Cat #: B0007) containing 10% β‐mercaptoethanol. Proteins were resolved by SDS‐PAGE (ThermoFisher, Cat #: NW04120BOX) and transferred onto a 0.2 µm PVDF membrane (Millipore Sigma, Cat #: ISEQ00010). Membranes were blocked with 3% bovine serum albumin (BSA) in PBS‐T (phosphate buffered saline with 0.1% TWEEN20) and incubated with primary antibodies overnight at 4°C. Primary antibodies used in this study include Nlp (Sigma‐Aldrich, Cat #: HPA000686; 1:3000); Lamp1 (Cell Signaling, Cat #: 9091; 1:3000); LC3B (Cell Signaling, Cat #: 2775; 1:3000); 9E10 C‐Myc (Sigma‐Aldrich, Cat #: M4439; 1:3000); and β‐actin (Cell Signaling, Cat #: 4970; 1:5000). After rinsing in PBS‐T, membranes were incubated with horseradish peroxidase‐conjugated secondary antibodies for one hour at room temperature. Secondary antibodies used in this study include anti‐mouse immunoglobulin (Ig) G, horseradish peroxidase (HRP)‐linked Antibody (Cell Signaling, Cat #: 7076S; 1:3000), and anti‐rabbit IgG, HRP‐linked Antibody (Cell Signaling, Cat #: 7074S; 1:3000). Blots were visualized using Lumigen ECL Ultra (TMa‐6) chemiluminescent substrate (Fisher Scientific, Cat #: NC0240697) on a Bio‐Rad ChemiDoc Imaging System. Intensity measurements of each band were generated in ImageJ by drawing a rectangle around each band and measuring the area under the curve for each peak. LAMP1 was normalized to b‐actin and LC3B ratios were generated by dividing the intensity of LC3B II by the intensity of LC3B I. Each value was then normalized to the vector control within its own experiment.

2.8. Measuring autophagic vesicles

Autophagic vesicle size and number were measured by confocal microscopy using CytoID Autophagy detection kit (Enzo, Cat #: ENZ‐51031). ³⁷ HEK293T cells were transiently transfected with control or NINL plasmids. Twenty‐four hours after transfection, cells were replated at a density of 200,000 cells per well into 35 mm live imaging dishes with a 20 mm glass bottom well (Cellvis, Cat #: D35‐20‐1.5‐N) coated with Poly‐D‐Lysine. Cells were grown in imaging dishes for an additional 24 hours. Two hours prior to imaging, cells were treated with ViaFluor Live Cell Microtubule Stain (Biotium, Cat #: 70063), and Verapamil HCl (Enzo, Cat #: ENZ‐51031) was added at a final concentration of 10 µm according to the manufacturer's instructions. CytoID Green Reagent (Enzo, Cat #: ENZ‐51031) and Hoechst 33342 Nuclear Stain (Enzo, Cat #: ENZ‐51031) was added 30 minutes prior to imaging per the manufacturer's instructions. In a subset of wells, cells were treated with 500 nM Rapamycin and 10 µM Chloroquine for 6 hours prior to imaging to stimulate the formation of autophagosomes and impair autolysosome fusion. Wells were imaged on a Nikon AX‐R with NSPARC confocal microscope using a 60x oil immersion objective. Approximately 15 cells were imaged per well per condition. Images were analyzed using Imaris version 10.2. Cell borders were defined using ViaFluor staining. CytoID puncta were defined using the “spots” function in Imaris. Spots statistics (number and area) were exported for each individual cell.

2.9. Monitoring protease activity

Substrate uptake and degradation by proteases were measured via live imaging using DQ‐BSA. ³⁸ , ³⁹ Twenty‐four hours after transient transfection with control or NINL plasmids, HEK293T cells were replated at a density of 10,000 cells per well into a 96 well plate coated with PDL. Cells were treated with DQ‐BSA Red (1:1,000; ThermoFisher, Cat #: D12051) 24 h after replating and imaged every 2 h for 48 h in a Sartorius Incucyte S3. Five images per well were acquired using a 10x objective. Images were analyzed using Incucyte analysis software (version 2021A and 2025B).

To measure Cathepsin B activity, HEK293T cells were transiently transfected with control or NINL plasmids and replated at a density of 10,000 cells per well into a 96 well plate 24 h later. Cells were treated with 1x Magic Red following the manufacturer's instructions (Antibodies Inc., Cat #: 938) 24 h after replating. Cells were imaged hourly for 12 h in a Sartorius IncuCyte S3. Five images per well were taken using a 10x objective. Images were analyzed using IncuCyte analysis software (version 2021A and 2025B).

2.10. Tau seeding activity

To investigate whether changes in NINL expression modulate tau seeding activity, we utilized a HEK293T‐tau biosensor cell line stably expressing tau repeat domain (RD) with the FTD‐linked P301S variant of tau fused to either cyan fluorescent protein (CFP) or yellow fluorescent protein (YFP). ³⁶ , ⁴⁰ Cells were cultured in a 12 well plate. Twenty‐four hours after transient transfection with control or NINL plasmids, cells were treated with 10 µM of human recombinant Tau‐441 (2N4R) wild‐type protein pre‐formed fibrils (StressMarq, Cat #: SPR‐471E) incubated with Lipofectamine 2000. After an additional 24 h, cells were harvested using 0.05% trypsin, fixed in 4% paraformaldehyde for 10 minutes, and resuspended in flow cytometry buffer (2% FBS, 0.5% BSA in PBS). Flow cytometry was conducted using a BD LSR Fortessa™ Cell Analyzer (BD Biosciences). CFP and FRET were measured using a 405 nm laser with emissions collected at 405/50 nm and 525/50 nm, respectively. YFP fluorescence was measured using a 488 nm laser with emission at 525/50 nm. FRET signals were quantified as previously described, ⁴⁰ using cells expressing only RD‐CFP or RD‐YFP for spectral compensation. A bivariate plot of FRET versus CFP was generated, and FRET‐positive cells were identified using a triangular gating strategy. This gate was calibrated using biosensor cells treated with Lipofectamine alone as FRET‐negative controls. For each experimental condition, 20,000 cells per biological replicate were analyzed. Data were processed using FlowJo software (v10).

3. RESULTS

3.1. Rare variants occurring in proteostasis genes are enriched in FTD and AD

FTD and AD exhibit a high degree of heritability that remains unexplained, suggesting additional genetic factors likely contribute to the disease that have not yet been identified. ³ , ⁴¹ Defining how rare variants modify FTD and AD risk can reveal core pathogenic pathways and therapeutic targets, while reciprocal analyses (e.g. testing for enrichment of rare variants within these pathways) enable a deeper understanding of genetic convergence in disease. Given the genetic and molecular evidence for the role of proteostasis genes in FTD and AD risk, ¹⁰ , ¹¹ , ¹⁴ , ¹⁶ we sought to determine whether rare variants enriched in FTD and AD occur in genes involved in the proteostasis network. Here, we leveraged summary statistics from our prior study of rare variant enrichment performed in unrelated FTD cases (n = 208), early onset AD cases (n = 227), and healthy adult controls (n = 671; Table S1). ⁷ We specifically asked whether proteostasis network genes, defined using the Proteostasis Consortium ⁴² , ⁴³ and prior literature, ⁴⁴ harbored rare variants enriched in FTD/AD cases. Rare variants were defined as those with a minor allele frequency of 0.0001 in all populations. Priority was given to variants present in three or more cases; present in 0 or 1 controls; and with a CADD score > 15 (Figure 1A). CADD scores above 15 are predicted to be damaging to the protein. ²⁹ We identified 75 variants in 18 proteostasis network genes with rare variants that were enriched in FTD and/or AD cases and predicted to be damaging (Figure 1B). Across the 18 proteostasis genes, 75 unique variants were identified and include missense, frame shift, splicing, and stop gain variants (Table S2). The 18 proteostasis genes are involved in regulation of chaperone functions, autophagy‐lysosome pathway, autophagosome and lysosome positioning, ubiquitin proteosome system, and stress response, among others (Figure 1C). These genes are known to encode proteins that localize throughout the cell: the lysosome (IDUA); the mitochondria (TRAP1); the cellular membrane (SLC2A7, NPHS1, SNTA1, MPDZ, and PATJ); the extracellular matrix (COL5A2); the cytoplasm (SHKBP1, DLC1, PDZD2, NINL); the nucleus (KMT2D, ORC3, PER2, SYNE2); or shuttle between the cytoplasm and nucleus, playing a role in transcription, translation, and processing (SND1 and UBR2). Together, we show that rare variants involved in proteostasis network genes occur in FTD and AD.

FIGURE 1.

Rare variants in proteostasis associated genes are identified in AD and FTD cases. (A) Schematic depicting workflow of variant identification from whole genome sequencing from Cochran et al ⁷ . (B) Table of genes that fall within proteostasis related pathways and that contain a rare variant that are enriched in FTD and AD. (C) Diagram of pathways involving proteostasis related genes. AD, Alzheimer's disease; FTD, frontotemporal dementia.

3.2. Proteostasis gene expression is altered in tauopathy

We next sought to determine whether proteostasis network genes harboring rare variants associated with FTD and AD also exhibit altered expression in disease models. This approach provides functional context for genetic risk, which could reveal how sequence variation translates into dysregulated molecular pathways. A subset of FTD and AD share a common pathological feature, tau aggregation. Thus, we analyzed transcriptomic data from human iPSC‐derived neurons harboring MAPT mutations, which have been established as a powerful cellular system to model early stage tauopathy. ³⁰ , ⁴⁵ , ⁴⁶ , ⁴⁷ Three MAPT mutations were analyzed in comparison to their isogenic controls: p.P301L (exon 10 point mutation), IVS10+16 (splicing), and p.R406W (C‐terminal point mutation; Figure 2A). ³⁰ One gene, SLC2A7, was not expressed in the iPSC‐derived neurons. Of the remaining 17 genes, 14 genes were differentially expressed in MAPT p.P301L neurons, and 10 genes were differentially expressed in MAPT IVS10+16 neurons, which is more than we would expect by chance (p = 0.0127 and 0.0003, respectively; Figure 2B). Only two of the 17 genes were differentially expressed in MAPT p.R406W neurons: NINL and IDUA (Figure 2B). Nine of the 18 genes were significantly differentially expressed in at least two of the three MAPT mutations (p < 0.05; Figure 2B; Table S3–S4). Strikingly, one gene, NINL, exhibited increased expression in all three MAPT mutations (Figure 2B). These results demonstrate that proteostasis genes containing rare variants enriched in FTD and AD are altered in a stem cell model of tauopathy, suggesting these genes are more broadly related to pathways associated with tau pathophysiology.

FIGURE 2.

Proteostasis genes are differentially expressed in models of tauopathy. (A) Schematic. FTD‐associated MAPT mutant neurons and isogenic controls were differentiated from iPSC and subjected to RNAseq. (B) Heatmap of logfold change. Differential gene expression data from MAPT mutant neurons. The 18 genes identified via whole genome sequencing (Figure 1B) are shown. (C) Schematic depicting tauopathy (FTLD‐tau, AD, and PSP) and control brains. (D) Heatmap. Differential gene expression data from human brains. The 18 genes identified via whole genome sequencing (Figure 1B) are shown. AD, Alzheimer's disease; FTD, frontotemporal dementia; FTLD‐tau, frontotemporal lobar degeneration with tau inclusions; iPSC, induced pluripotent stem cell; MAPT, microtubule associated protein tau; PSP, progressive supranuclear palsy.

Due to their immaturity, human iPSC neurons likely capture early events associated with disease pathophysiology prior to protein aggregation. ³⁰ , ⁴⁵ , ⁴⁸ , ⁴⁹ , ⁵⁰ To determine whether the 18 proteostasis network genes were associated with pathological events late in disease and in the presence of tau aggregates, we used transcriptomic data from FTLD‐tau patients carrying a MAPT IVS10+16 mutation (n = 2) and controls (n = 3; Figure 2C). In the brains of FTLD‐tau patients, SLC2A7 was not detected, and seven of the 17 remaining genes were differentially expressed (Figure 2D; Table S5). To determine whether the 18 proteostasis network genes were associated with AD, we analyzed transcriptomic data from AD (n = 82) and control (n = 84) brains (Figure 2C). We identified 13 differentially expressed genes in AD brains (Figure 2D; Table S5). NINL, SYNE2, KMT2D, TRAP1, COL5A2, and DLC1 exhibit consistently significant differential expression across FTLD‐tau and AD brains (Figure 2D; Table S5). To determine whether the 18 proteostasis network genes were altered in other forms of tauopathy, we examined transcriptomic data in brains from the primary, sporadic tauopathy, PSP (n = 84; Figure 2C). In PSP brains, we identified 4 differentially expressed proteostasis network genes (of the 17 genes present in the dataset; Figure 2D; Table S5). UBR2 and PDZD2 were significantly upregulated in PSP and AD brains (Figure 2D). NINL and KMT2D were significantly upregulated in FTLD‐tau, AD, and PSP brains (Figure 2D). Together, we show that proteostasis genes with rare variants enriched in FTD and AD are transcriptionally dysregulated at disease end stage.

3.3. NINL reduces tau seeding

We identified a single gene that contains rare variants enriched in FTD and that was significantly elevated across MAPT mutant neurons as well as in FTLD‐Tau, PSP, and AD brains: NINL. Bulk RNAseq analyses capture composite, population‐level gene expression changes across brain cell types. To determine how NINL expression changed in neurons, we evaluated neuronal subclusters from publicly available snRNA sequencing data from control (n = 9) and late onset AD (n = 28) brains (Figure 3A). ³⁵ NINL was significantly upregulated in AD neurons compared to controls (Figure 3B, Table S6). Interestingly, only one other gene among the 17 remaining proteostasis genes, ORC3, was significantly differentially expressed in AD neurons (Table S6).

FIGURE 3.

NINL reduces tau seeding. (A–B) SnRNAseq of control and late onset AD cases was previously reported ⁸¹ . Neuron clusters were analyzed for NINL expression. (A) Schematic of snRNAseq workflow. (B) Volcano plot of differential expression in neuronal nuclei. NINL is significantly elevated in AD compared with controls (p = 4.9 × 10⁻⁴; FDR = 0.0476). (C–E) FRET‐based tau seeding assay in HEK293T tau biosensor cells. (C) Schematic. (D) qPCR for NINL RNA. ***, p = 0.0006. Student's t‐test. n = 4 biological replicates per condition. (E) Integrated FRET density (FRET MFI x percent FRET positive cells) normalized to vector controls, ***, p = 0.0009. Student's t‐test. n = 8 biological replicates per condition. Graphs represent mean ± SEM. AD, Alzheimer's disease; MFI, median fluorescence intensity; qPCR, quantitative polymerase chain reaction.

Given the observation that NINL has increased expression in the neurons of tauopathy patients, we asked whether NINL plays a role in a central feature of tau pathophysiology: tau aggregation. To determine whether NINL expression impacts the propensity of tau to seed new aggregates, we used an engineered HEK293T line that stably expresses two constructs encoding a mutagenized form of the tau repeat domain (RD‐P301S) conjugated to either CFP or YFP ³⁶ (i.e. tau biosensor cells). Adding tau aggregates to these cultures has been shown to cause nucleation of the endogenous tau reporter proteins and aggregation that can be measured by FRET using flow cytometry ⁴⁰ (Figure 3C). Surprisingly, overexpression of NINL WT in tau biosensor cells (Figure 3D) resulted in a significant reduction in tau seeding compared with vector controls (Figure 3E). Thus, NINL dampens tau seeding capacity, reducing tau aggregation. Considering these findings along with the observed increase in NINL in MAPT mutant neurons and tauopathy brains, we hypothesize that up regulation of NINL reflects a proteostatic response to accumulating tau that ultimately becomes overwhelmed.

3.4. NINL impacts lysosomal load

NINL encodes Nlp, which is a motor‐associated protein commonly known for its role in the centrosome and cell division. ⁵¹ , ⁵² , ⁵³ Beyond mitosis, accumulating evidence indicates that Nlp also contributes to intracellular trafficking in post‐mitotic neurons through interactions with motor complexes, including dynein and dynactin. ⁵⁴ , ⁵⁵ , ⁵⁶ This is notable because efficient autophagic degradation depends on the coordinated, motor‐driven positioning and transport of autophagosomes and lysosomes along microtubules. ⁵⁷ Consistent with a role for Nlp in this process, loss of NINL has been reported to disrupt autophagic flux, ⁵⁸ suggesting that Nlp may help couple motor function to lysosome‐autophagosome dynamics. Multiple studies in MAPT mutant neurons have described abnormalities in molecular motors, stalled lysosomal motility, altered autophagic flux, and delayed cargo degradation, ⁴⁶ , ⁵⁹ , ⁶⁰ , ⁶¹ , ⁶² raising the possibility that Nlp could be a mechanistic link between cytoskeletal transport and autophagy‐lysosome dysfunction in tauopathy.

We propose that NINL overexpression reduces tau seeding by promoting protein clearance via the autophagy‐lysosome pathway. HEK293T cells are a highly tractable immortalized cell line for mechanistic interrogation that enables precise manipulation of gene expression and robust quantitative readouts. HEK293T cells were transiently transfected with NINL WT to evaluate how NINL overexpression impacts components of the autophagy‐lysosome pathway (Figure 4A). We used an acidotropic probe, LysoTracker Red, which accumulates in acidic vesicles, such as the lysosome, to evaluate the impact of NINL expression on lysosomes. Flow cytometry analyses revealed that LysoTracker intensity was significantly increased in cells overexpressing NINL compared to vector controls (Figure 4B–C). An increase in LysoTracker intensity could reflect a change in lysosomal load, acidity, or abundance. To determine whether increased NINL expression alters the acidity of lysosomal vesicles, we used a pH sensitive dye, LysoSensor Green (Figure S2A). LysoSensor levels were similar in cells overexpressing NINL and vector controls (Figure S2B). Thus, NINL overexpression does not cause hyperacidification of lysosomes. We next evaluated lysosomal abundance using immunoblotting for a lysosomal membrane marker, LAMP1 (Figure 4A). NINL overexpression resulted in similar levels of LAMP1 protein levels compared to vector controls (Figure 4D–E), suggesting neither lysosomal biogenesis nor mass is altered. Together, these findings suggest that NINL overexpression increases lysosomal functional properties, without shifting basal lysosomal acidity or lysosomal abundance.

FIGURE 4.

NINL expression enhances the autophagy‐lysosome pathway. HEK293T cells were transiently transfected with pCMV6 NINL‐myc plasmid or vector control plasmid for 48 hours. (A) Schematic of the components of the autophagy‐lysosome pathway that were evaluated. (B) qPCR for NINL RNA. **, p = 0.0022. Student's t‐test. n = 3 biological replicates per condition. (C) LysoTracker MFI quantified by flow cytometry plotted relative to vector control. ***, p = 0.0001. Student's t‐test. n = 9 biological replicates per condition. (D) Representative immunoblot for Nlp (NINL), Lamp1, LC3B, and B‐actin. (E) Immunoblot quantification of Lamp1 protein levels normalized to vector controls. ns, not significant. Student's t‐test. n = 7 biological replicates per condition. (F) Immunoblot quantification of LC3B II/I ratio normalized to vector controls. *, p = 0.0224. Student's t‐test. n = 6–7 biological replicates per condition. (G) Representative images from live imaging with CytoID. Both vector and NINL expressing cells were treated with rapamycin and chloroquine for 6 hours prior to imaging as a positive control. CytoID (autophagic vesicles) in green, Hoescht (nuclei) in blue, and ViaFluor (cytoskeleton) in red. Scale bar, 10 µm. (H) Quantification of the number of CytoID puncta per cell, normalized to the cell size (56–102 cells quantified per condition; four to six biological replicates). ****, p < 0.0001. Two‐way ANOVA and post hoc testing using Tukey's HSD. (I) Quantification of the average CytoID puncta size per cell (56–102 cells quantified per condition; 4–6 biological replicates). ****, p < 0.0001. Two‐way ANOVA and post hoc testing using Tukey's HSD. Graphs represent mean ± SEM. ANOVA, analysis of variance; HSD, honestly significant difference; MFI, median fluorescence intensity.

3.5. NINL alters autophagy

The observed increase in lysosomal functional properties, reflected by higher LysoTracker signal, without changes in lysosomal pH (LysoSensor) or overall lysosome abundance (LAMP1) could reflect altered cargo trafficking through the lysosomal system. Given that lysosomes represent the terminal compartment of the autophagy‐lysosome pathway, an increase in lysosomal load could point to changes in autophagic input (Figure 4A). Thus, we next examined markers of autophagy to determine whether NINL expression alters autophagosomes. Initiation of autophagy involves the conversion of LC3BI to LC3BII, which integrates into the membrane of the forming phagophore and is ultimately degraded after autophagolysosomes undergo cargo degradation. ⁶³ , ⁶⁴ Thus, ratios of LC3BII/I can provide insights into the initiation and completion of autophagy. The ratio of LC3B II/I was significantly elevated in HEK293T cells overexpressing NINL compared to vector control (Figure 4D and 4F; Figure S3). Thus, at baseline, NINL likely modulates autophagy initiation or LC3 processing. We next sought to determine whether NINL expression alters autophagosome number or size using the cationic amphiphilic tracer, CytoID, that selectively labels autophagic vesicles. HEK293T cells overexpressing NINL WT or a control vector produced a similar number of autophagic vesicles (Figure 4G–4H; Figure S4A–B) that were similar in size (Figure 4G and 4I; Figure S4A–C) under basal conditions. This finding, along with the change in LC3B II/I, suggests that NINL expression alters autophagy initiation or LC3 processing without expanding the autophagic vesicle pool under basal conditions. ⁶³ To directly examine the impact of NINL on autophagosome dynamics, we used rapamycin to induce autophagy and chloroquine to block autophagosome‐lysosome fusion in cells expressing control and NINL WT plasmids (Figure 4G–I; Figure S4). CytoID‐positive vesicle number and size were similar in cells overexpressing NINL and vector controls, suggesting NINL expression does not shift autophagic vesicles at baseline (Figure 4G–I; Figure S4). Upon stimulation with rapamycin and chloroquine, the number of CytoID‐positive vesicles was significantly increased in vector control conditions, as expected (Figure 4G–H; Figure S4A–B). Yet stimulation with rapamycin and chloroquine failed to alter the number of CytoID‐positive vesicles when NINL was overexpressed (Figure 4G–H; Figure S4A–B). CytoID vesicle size was similarly increased in vector and NINL expressing cells upon treatment with rapamycin and chloroquine compared to vehicle controls (Figure 4G and 4I; Figure S4A and C). Taken together, the increase in CytoID vesicle size in response to rapamycin and chloroquine in both vector and NINL overexpressing cells indicates that the pharmacologic perturbations were effective; however, the absence of a corresponding change in CytoID puncta number with NINL overexpression suggests that NINL constrains vesicle abundance, consistent with a role in trafficking, organization, or maturation of autophagy‐related compartments.

3.6. NINL promotes proteolytic activity

Our findings point to a role for NINL in the activation of structural components of the degradative pathway without revealing the functional impact on degradative capacity. We next assessed lysosomal proteolytic activity to determine whether NINL promotes effective lysosomal degradation or leads to accumulation of undegraded material. DQ‐BSA is a fluorogenic protease substrate that is taken up by endocytosis and degraded by proteases, at which point it fluoresces. ⁶³ Live imaging was performed in HEK293T cells overexpressing NINL or vector controls treated with DQ‐BSA Red. Cells were imaged every 2 hours for 48 hours. We observed a similar increase of DQ‐BSA signal in control and NINL expressing cells for the first 10 hours after treatment (Figure 5A; Figure S5A). Beginning at 12 hours, NINL expressing cells produced significantly more DQ‐BSA signal compared with cells expressing a vector control (Figure 5A–B; Figure S5A). DQ‐BSA signal is mediated by substrate uptake and proteolytic degradation. To further evaluate the impact of NINL expression on proteolytic activity, a membrane permeable fluorogenic substrate, Magic Red, was used. ⁶⁵ Hourly live imaging revealed that HEK293T cells overexpressing NINL produced significantly more Magic Red signal than vector controls (Figure 5C–D; Figure S5B). Thus, we show that NINL promotes proteolytic activity.

FIGURE 5.

Increased NINL expression alters protease activity. HEK293T cells were transiently transfected with pCMV6 NINL‐myc plasmid or vector control for 48 hours. (A–B) Immediately after treatment with DQ‐BSA Red, wells (96 well plate) were imaged every 2 hours at 10x (five images per well). (A) Representative graph of DQ‐BSA integrated intensity over cell area across a 48 hour period. (B) DQ‐BSA integrated intensity over cell area normalized to vector controls at 48 hours. Each datapoint is representative of the average of five pictures per well. ***, p = 0.0002. Student's t‐test. n = 16 biological replicates per condition. (C–D) Immediately after treatment with Magic Red, wells (96 well plate) were imaged every hour at 10x (five images per well). (C) Representative graph of Magic Red integrated intensity over cell area across a 12 hour period. (D) Magic Red integrated intensity over cell area normalized to vector controls at 12 hours. Each datapoint is representative of the average of five pictures per well. ****, p < 0.0001. Student's t‐test. n = 14–16 biological replicates per condition. Graphs represent mean ± SEM. BSA, bovine serum albumin.

3.7. NINL variants block protective effects on tau seeding

Our findings suggest that NINL overexpression reduces tau seeding via the autophagy‐lysosome pathway by promoting degradative activity. Three rare NINL variants were identified in FTD cases and were absent from AD and control subjects (Figure 6A; Table S2). Interestingly, the clinical research team established clinical diagnoses and predicted underlying neuropathology for all three cases as tau, either due to FTLD‐tau or corticobasal degeneration (Table S7). Of the three variants, NINL p.T275fs and p.R1202fs were predicted to be the most deleterious with high CADD scores (22.5 and 26.6, respectively), suggesting the variants are damaging. Both variants are predicted to cause a frameshift that results in an early stop codon and truncation of the Nlp protein. To determine whether NINL p.T275fs and p.R1202fs alter the Nlp protein, we mutagenized the NINL WT plasmid to introduce each variant (Figure 6A; Figure S1). Transient overexpression of plasmids containing NINL WT, p.T275fs, and p.R1202fs or vector control in HEK293T cells illustrated that the variants indeed reduced Nlp expression compared to NINL WT (Figure 6B, Figure S6). We next sought to determine whether NINL variants alter tau seeding. Consistent with our prior experiments, expression of NINL WT in HEK293T tau biosensor cells resulted in a significant reduction of tau seeding (Figure 6C). Expression of NINL p.T275fs and p.R1202fs in tau biosensor cells failed to reduce tau seeding (Figure 6C). Thus, FTD variants in NINL eliminate the protective effect of NINL on tau seeding.

FIGURE 6.

The protective effect of increased NINL expression on tau seeding is lost in NINL variants. Cells were transiently transfected with vector control, NINL WT, NINL p.T275fs, or NINL p.R1202fs. (A) Schematic of the NINL gene and the variants identified in this study. (B) Representative immunoblot for C‐Myc tagged Nlp (NINL). (C) Integrated FRET density (FRET MFI x percent FRET positive cells) normalized to vector controls. **, p = 0.0014. One‐way ANOVA with Dunnett correction. n = 10 biological replicates per condition. (D) LysoTracker MFI quantified by flow cytometry plotted relative to vector control. ****, p < 0.0001. One‐way ANOVA with Dunnett correction. n = 6 biological replicates per condition. (E) Representative graph of DQ‐BSA integrated intensity over cell area across a 48 hour period. (F) DQ‐BSA integrated intensity over cell area normalized to vector controls at 48 hours. Each datapoint is representative of the average of five pictures per well. ****, p < 0.0001. Student's t‐test. n = 16 biological replicates per condition. Graphs represent mean ± SEM. ANOVA, analysis of variance; MFI, median fluorescence intensity; WT, wild‐type.

NINL variants may fail to reduce tau seeding by disrupting the effects of NINL in the autophagy‐lysosome system. Consistent with our prior experiments, expression of NINL WT in HEK293T cells resulted in a significant increase in LysoTracker intensity, pointing to increased lysosomal functional properties (Figure 6D). However, expression of NINL p.T275fs and p.R1202fs failed to alter LysoTracker intensity compared to vector controls (Figure 6D). To understand the impact of NINL p.T275fs and p.R1202fs on protease activity, HEK293T cells were transiently transfected with NINL WT, p.T275fs, p.R1202fs, or vector control, treated with DQ‐Red BSA, and imaged every two hours for 48 h (Figure 6E; Figure S7). Cells expressing NINL p.T275fs and p.R1202fs behaved similarly to vector controls, exhibiting less protease activity than NINL WT (Figure 6E–F; Figure S7). Together, we discovered that NINL p.T275fs and p.R1202fs block the protective effect of NINL WT on tau seeding by failing to promote lysosomal protease activity.

4. DISCUSSION

Proteostasis dysfunction is a hallmark of FTD and AD, yet the genetic factors and molecular pathways that disrupt protein homeostasis remain poorly understood. Here, we integrated human genetics, transcriptomics, and mechanistic studies to define proteostasis network components contributing to tauopathy. Using genome sequencing from unrelated FTD, early‐onset AD, and control subjects, we identified 18 proteostasis network genes harboring 75 rare, predicted‐damaging variants enriched in FTD and/or AD. These genes span diverse proteostasis functions, including autophagy‐lysosome regulation, ubiquitin‐proteasome activity, and chaperone regulation. Many of these proteostasis network genes were differentially expressed in MAPT mutant neurons and dysregulated in FTLD‐tau, AD, and PSP brains, demonstrating convergence of genetic and transcriptional perturbations. A single gene, NINL, emerged as consistently upregulated across MAPT mutant neurons, FTLD‐tau, AD, and PSP brains. Functional studies demonstrated that NINL overexpression reduces tau seeding, increases LC3B lipidation, elevates LysoTracker signal without altering lysosomal pH or mass, and enhances proteolytic activity, supporting a role for NINL in promoting lysosomal degradation. FTD‐enriched NINL variants (p.T275fs, p.R1202fs) disrupted Nlp protein expression and abolished NINL‐mediated reductions in tau seeding and lysosomal protease activity. Together, these findings identify a set of proteostasis genes genetically and transcriptionally linked to neurodegeneration and reveal NINL as a novel regulator of tau aggregation, potentially upregulated as part of an adaptive proteostatic response to proteotoxic stress.

Proteostasis, the coordinated regulation of protein synthesis, folding, trafficking, and degradation, is essential for neuronal health, and its breakdown is a defining feature of neurodegenerative diseases, such as FTLD‐tau and AD. Since these disorders typically arise in mid‐to‐late adulthood, proteostasis decline has often been viewed as an age‐related vulnerability that acts as a secondary hit in cells already burdened by misfolded proteins. However, emerging evidence indicates that defects in core proteostasis machinery can occur much earlier, prior to clinical onset, and may actively contribute to the initiation or acceleration of tau pathology ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁹ , ²⁰ .

Our sequencing analyses identified rare variants enriched in FTD and AD across multiple functional classes of proteostasis genes, including those encoding lysosomal enzymes, trafficking regulators, and components of the autophagic machinery. This suggests that genetic variation in protein clearance pathway genes is relevant to neurodegeneration and, specifically, that rare variation in these genes may directly affect degradation capacity by altering distinct but converging proteostasis nodes. Transcriptomic analyses further demonstrated that these proteostasis genes are consistently dysregulated in iPSC‐derived neurons carrying MAPT mutations, which model early disease processes, and in FTLD‐tau, AD, and PSP brains, which represent end‐stage pathology. This convergence across early, pre‐aggregation neuronal states and advanced human brain pathology suggests that proteostasis genes harboring rare variants represent components of degradation pathways systematically stressed across disease stages. Directionality of the proteostasis gene expression changes (upregulation vs. downregulation) may reflect compensatory responses, pathway overload, or direct mutation effects. Additional functional analyses are required to elucidate these mechanisms. Together, this convergence across genetics, patient‐derived neurons, and diseased brains supports a model in which both inherited variation and disease‐driven cellular stress perturb common proteostasis pathways, underscoring the biological relevance of these genes to neurodegenerative processes.

Among the candidate genes, NINL emerged as a compelling regulator of proteostasis given its consistent upregulation across MAPT mutant neurons and tauopathy brains. NINL encodes the motor‐associated protein Nlp, originally characterized for its role at the centrosome. ⁵¹ , ⁵² , ⁵³ Nlp can interact with dynein–dynactin complexes in neurons ⁵⁴ , ⁵⁵ , ⁵⁶ ; however, dynein and dynactin subunit expression was unchanged in the presence of MAPT mutations (data not shown). ⁶⁶ Instead, we have observed major transcriptional and protein changes in the autophagy‐lysosome system in these mutant neurons. ⁵⁹ , ⁶⁰ , ⁶⁶ Recent studies more directly implicate Nlp in the autophagy pathway itself, showing that it promotes autophagy by facilitating interactions among autophagosome sorting proteins. ⁵⁸ These findings position Nlp as a regulator of autophagy‐related compartment dynamics and motivated our focus on autophagy‐related phenotypes rather than primarily motor‐based readouts. Our functional studies revealed that NINL overexpression remodels autophagy‐related compartments: it increases LC3BII/I ratio at baseline and elevates LysoTracker signal without altering lysosomal abundance or pH. NINL also enhanced lysosomal functional properties and proteolytic activity, reflected by increased LysoTracker retention and accelerated degradation of DQ‐BSA and Magic Red substrates. These findings build on a developing literature to support a role for NINL in strengthening the degradative arm of the autophagy‐lysosome pathway. ⁵⁴ , ⁵⁵ , ⁵⁸ Future studies will be necessary to dissect the role of Nlp in trafficking of autophagic vesicles and its other canonical functions. Together, these data support a model in which NINL promotes effective clearance of aggregation‐prone proteins by improving autophagy‐lysosome coupling and degradative efficiency.

Our genetic, transcriptomic, and functional evidence supports a model in which NINL is a component of the neuronal proteostasis response to tau stress which is necessary, but not sufficient, to fully restore cellular homeostasis in the setting of chronic tau burden. We show that NINL reduces tau aggregation. Yet, two FTD‐enriched NINL frameshift variants abolished the protective effects of NINL on tau seeding and failed to enhance lysosomal proteolysis. These variants reduced Nlp protein expression, indicating loss‐of‐function effects. Their inability to reduce tau seeding or promote proteolytic activity suggests that the genetic disruptions in NINL observed in FTD cases may impair the cell's capacity to counteract pathological tau accumulation. It is well established that genes central to neurodegenerative disease biology are often upregulated as part of a protective or compensatory response, ⁶⁷ , ⁶⁸ , ⁶⁹ , ⁷⁰ yet this response is frequently insufficient to halt disease progression. Moreover, these same genes commonly harbor rare deleterious variants that increase disease risk, which is often how their biological importance is first identified. ⁶⁷ , ⁶⁸ , ⁶⁹ , ⁷⁰ , ⁷¹ , ⁷² , ⁷³ , ⁷⁴ Canonical examples include TREM2, SORL1, BIN1, PICALM, and PLD3 in Alzheimer's disease, where increased expression or activity can be protective in some contexts, yet loss‐of‐function variants confer risk and do not phenocopy a simple absence of disease. This paradigm is also well documented in lysosomal storage diseases and related proteostasis disorders, where transcriptional upregulation of lysosomal and autophagy genes frequently occurs in affected neurons and glia, despite progressive pathology. For example, in disorders such as Gaucher disease (GBA1), GM1/GM2 gangliosidoses, and neuronal ceroid lipofuscinoses, lysosomal genes are robustly induced through stress‐responsive pathways (including TFEB/TFE3 activation), yet this compensatory response is unable to overcome the underlying enzymatic or trafficking defect. ⁷⁵ , ⁷⁶ Importantly, these diseases illustrate that upregulation of lysosomal machinery does not always imply functional rescue, but instead reflects an engaged, and often overwhelmed, attempt to restore proteostasis. These principles extend to late‐onset neurodegenerative diseases, where partial lysosomal insufficiency intersects with age‐related decline and disease‐specific protein stressors. We propose that NINL follows a similar paradigm. Our data support a model in which NINL is a component of the neuronal proteostasis response to tau stress. In this framework, upregulation of NINL reflects an attempted compensatory mechanism engaged in response to accumulating tau pathology. However, as with AD risk genes and lysosomal storage disease associated pathways, this response is necessary, but not sufficient, to fully restore cellular homeostasis, particularly in the setting of progressive and chronic tau burden. In contrast, rare NINL variants that reduce function likely compromise this adaptive capacity, lowering the threshold at which neurons fail to maintain proteostasis, and thereby accelerating or exacerbating tau pathology. Thus, loss‐of‐function variants and disease‐associated upregulation are not contradictory but instead reflect two sides of the same biology: one revealing the importance of NINL through genetic insufficiency, and the other reflecting compensatory activation in response to proteostatic stress. Autopsy tissue is not yet available for the participants carrying rare NINL variants. Systematic analyses of genotype‐pathology relationships, including regional burden and tau strain characteristics, represent an important direction for future work

Together, this work broadens the spectrum of proteostasis components implicated in disease, demonstrating that rare variants and disease‐driven dysregulation extend beyond canonical autophagy and lysosome genes. Our findings reveal a coordinated pattern of genetic burden, dysregulation of gene expression, and impaired proteostasis involving a subset of proteostasis genes in FTD and AD. By identifying NINL as a regulator of tau aggregation and lysosomal degradation, we highlight a previously uncharacterized molecular contributor to disease pathophysiology. These results suggest that enhancing NINL function or strengthening downstream trafficking and degradative pathways may represent promising therapeutic strategies for tauopathies. Future studies using stem cell and mouse models will be important for extending our understanding of NINL’s mechanism in cell specific and disease contexts.

Several limitations warrant consideration. While rare variant enrichment highlights genes of potential importance, further functional validation of these variants will be required to establish their disease risk effect. Whether rare variants in NINL or other proteostasis genes are specific to tauopathies or extend to other neurodegenerative contexts remains to be resolved. Robust assessment of disease specificity of rare variants will require systematic analyses across multiple, well‐powered cohorts. We have not comprehensively examined cell type‐specific regulation of NINL across diverse human tauopathy datasets. Our analysis of snRNAseq data supported that changes in bulk RNA expression could not simply be explained by shifts in cellular composition. Additional disease‐specific datasets may provide further insight into how NINL expression relates to tau pathology across brain regions and cell populations. Recent snRNAseq studies of PSP and AD offer opportunities to examine the relationship between NINL expression, cell identity, and tau burden at higher resolution. ⁷⁷ , ⁷⁸ , ⁷⁹ , ⁸⁰ Preliminary exploration of publicly available datasets suggests that NINL regulation may vary across neuronal subtypes and disease contexts, consistent with a complex and potentially cell type‐specific role in tauopathy. Systematic integration and analysis of these datasets will be required to fully define the relationship between NINL expression and tau pathology in human brain tissue. Future studies incorporating disease‐specific single nuclei and spatial transcriptomic approaches will be important to further establish the pathological context and cell type‐specific functions of NINL across tauopathies. Finally, while iPSC‐derived neurons and overexpression assays capture many critical aspects of human proteostasis networks, in vivo studies will be needed to determine whether NINL modulation alters tau propagation, neuronal survival, and cognitive outcomes.

In summary, our integrative approach demonstrates that rare genetic variation, transcriptional dysregulation, and functional perturbation of proteostasis genes converge on pathways critical for tau aggregation. We identify NINL as a mechanistically relevant regulator of autophagic degradation and tau aggregation and propose that its upregulation in disease reflects an adaptive attempt to bolster proteostasis capacity. These insights advance our understanding of tau clearance mechanisms and nominate proteostasis pathways as actionable targets for therapeutic intervention in FTD and AD.

AUTHOR CONTRIBUTIONS

Designed experiments: Samantha K. Swift, Jennifer S. Yokoyama, Celeste M. Karch. Performed and analyzed experiments: Samantha K. Swift, Guangming Huang, J. Nicholas Cochran, Ricardo D'Oliveira Albanus, Miguel A. Minaya, Patricia A. Castruita, Rui Zhang, Katherine J. Miller, Emma Starr, Grant Galasso, Jacob A. Marsh, Aimee W. Kao, Oscar Harari, Jennifer S. Yokoyama, Celeste M. Karch. Provided funding: Aimee W. Kao, Jennifer S. Yokoyama, Celeste M. Karch. Wrote the manuscript: Samantha K. Swift, Celeste M. Karch. Revised and approved manuscript: Samantha K. Swift, Guangming Huang, J. Nicholas Cochran, Ricardo D'Oliveira Albanus, Miguel A. Minaya, Patricia A. Castruita, Rui Zhang, Katherine J. Miller, Emma Starr, Grant Galasso, Jacob A. Marsh, Aimee W. Kao, Oscar Harari, Jennifer S. Yokoyama, Celeste M. Karch.

CONFLICT OF INTEREST STATEMENT

C.M.K. serves as an advisor for Eisai Co. Ltd and Synapticure Inc. J.S.Y. serves on the scientific advisory board for the Epstein Family Alzheimer's Research Collaboration and the Charleston Conference on Alzheimer's Disease and is the editor‐in‐chief of npj Dementia. The following authors have no competing interests: S.K.S., G.H., J.N.C., R.A., M.A.M., P.A.C., R.Z., K.J.M., E.S., G.G., J.A.M., A.W.K., O.H., Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

All participants provided informed consent for their data to be used in this study. The study was approved by the institutional review board of Washington University School of Medicine in St. Louis.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information