Cellular redox status is linked to olfactory alterations in cognitive decline Authors: Riikka Lampinen, Alexey M Afonin, Laura Mussalo, Simone Avesani, Tana Zavodna, Emily Feneberg, Zdenek Krejcik, Jan Topinka, Sweelin Chew, Heikki Löppönen, Tarja Malm, Rosalba Giugno, Anthony R White, Anne M Koivisto, Elina Penttilä, Katja M Kanninen Published: 2026 ABSTRACT AbstractGrowing evidence demonstrates that olfactory dysfunction and disease-related pathology in the olfactory areas occur early in Alzheimer's disease (AD), which is the most common neurodegenerative disorder. Oxidative stress is strongly linked to AD-associated changes in the brain, yet very little is known about redox homeostasis in the peripheral olfactory tissues, and its link to olfactory dysfunction associated with early AD. The olfactory mucosa (OM), located at the rooftop of the nasal cavity is the primary organ for olfaction through odorant detection and transmission of these signals to the brain via the olfactory nerves. This study collected patient-derived biopsies of the OM from cognitively healthy donors, individuals with mild cognitive impairment (MCI) and AD. Shotgun proteomics of the biopsies provided new insights into the molecular mechanisms associated with early AD and olfactory dysfunction, which was assessed by odor identification testing. Furthermore, microRNA sequencing and quantitative analyses of electron carriers, cofactors and glutathione from the biopsy-derived primary cells of the OM were used to link these findings to cellular redox status. The results revealed mitochondrial changes and alterations to cellular defense against oxidative stress, specifically the potentially dampened NRF2 pathway, in MCI and AD OM, which were linked to impairments in odor identification. Understanding the early cellular and molecular changes occurring in the olfactory system and olfaction could aid in the early identification of individuals at risk of developing AD and pathophysiological changes of the OM in AD.Keywords: Alzheimer's disease, Mild cognitive impairment, microRNA sequencing, Nasal biopsy, NRF2, Olfactory mucosa, Olfaction, Olfactory, Proteomics, Redox homeostasis Background Pathological brain accumulation of beta-amyloid (Aβ) as extracellular plaques and intracellular neurofibrillary tangles of tau initiate decades before the occurrence of clinical symptoms and diagnosis of dementia due to Alzheimer's disease (AD) [1]. Therefore, deciphering early pathological mechanisms to facilitate the development of effective strategies for intervention is needed. There is growing evidence that olfactory processing areas, such as the olfactory bulb (OB) and olfactory mucosa (OM), are early sites of AD-associated pathology [2,3]. Olfactory dysfunction is implicated with aging and is often a prodromal symptom for several neurodegenerative diseases [4,5]. Cells of the OM are located at the roof of the nasal cavity and connected with both the external environment, the OB, and brain via the axons of the olfactory receptor neurons located in the OM [6]. The relatively easy accessibility of OM tissue from living individuals holds potential for studies of early molecular alterations of neurodegenerative diseases [7]. Previous work has confirmed the high prevalence of hyperphosphorylated tau and Aβ in autopsy-derived samples of OM obtained from individuals with AD and significant correlation with the AD brain pathology [8,9]. Moreover, a low baseline score in odor identification tests has been reported to predict the diagnosis of AD at follow-up studies of individuals with mild cognitive impairment (MCI) [[10], [11], [12]], highlighting the need to further monitor individuals with olfactory deficits and MCI. One of the earliest alterations in AD brain tissue is oxidative stress [13]. Oxidative stress is harmful for cell proteins, membranes, lipids, lipoproteins and DNA, and emerges as an outcome of imbalance between the number of free radicals and the cell's capacity to clear them. Evidence supporting elevated oxidative stress in AD has been reported in autopsy-derived samples of OM [[14], [15], [16], [17]]. Oxidative stress is also connected to the pathological hallmark proteins and mitochondrial dysfunction observed in AD [18,19]. Notably, mitochondria are the main source of reactive oxygen species (ROS) in the cell and dysfunctional mitochondria can further contribute to accumulation of oxidative damage by mitochondrial ROS [20]. To resolve the potential cellular harm caused by the excess of ROS, transcription factor nuclear factor erythroid 2–related factor 2 (NRF2) can translocate to the cell nucleus and transcriptionally regulate several cellular antioxidant and anti-inflammatory responses [21]. In addition to the classical antioxidant defense, NRF2 also regulates cellular energy metabolism by taking part in modulating mitochondrial function and homeostasis [22]. The above-mentioned mitochondrial alterations can possibly be involved in the observed olfactory dysfunction in individuals with MCI and AD dementia. However, little is known about the early redox homeostasis related changes in the OM in AD. In this study we analyzed biopsies of the OM derived from living individuals with MCI or AD, and cognitively healthy study participants, to provide new insight into the early molecular alterations of AD [7]. By using shotgun proteomics, we aimed to investigate protein alterations occurring in the OM in MCI and AD. Next, we explored the putative redox-homeostasis-related alterations in cells of the OM by assessing a set of coenzymes and reducing agents of redox reactions and tested for the associations of the observed protein alterations to the biopsy donor's odor identification test score. Lastly, we observed the individuals with MCI and AD to have deficits in identification of certain odors when compared to cognitively healthy control subjects and identified a microRNA potentially contributing to the observed mitochondria-related proteomic and metabolic alterations in the AD OM mucosa. Methods Study participants, odor identification test, and sample collection The study was performed with ethical approval from the Research Ethics Committee of the Northern Savo Hospital District (permit number 536/2017). The study participants were divided into cognitively healthy individuals (N = 53), MCI (N = 47), and AD (N = 44) groups based on their diagnostic examinations as previously described [23,24]. Briefly, diagnostic examinations, recruitments, interviews, and classification of the study participants to study groups (cognitively healthy, MCI, AD) were conducted at the Brain Research Unit or at the Department of Neurology, Kuopio University Hospital by the study clinicians and psychologists specialized to memory disorders. The cognition of all the study participations was evaluated utilizing the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) neuropsychological battery and Clinical Dementia Rating (CDR) scale interview. Cognitively healthy control subjects were recruited during a dacryocystorhinostomy (DCR) surgery, or from the already existing registries of the Brain Research Unit of the University of Eastern Finland and demonstrated no indications of cognitive problems in the above-mentioned cognitive tests. Individuals with AD type mild (CDR 1) dementia fulfilled the NIA-AA clinical criteria of progressive AD and MRI imaging, or a FDG-PET study showed degenerative process, or biomarker (beta-amyloid, tau, and phosphorylated-tau) changes were found in CSF biomarker examinations to be typical to AD. The MCI study group was formed from study participants that demonstrated reduced performance in at least one memory domain in the CERAD-neuropsychological battery but had no indication in changes in daily activities and did not meet the National Institute on Aging and Alzheimer's Association (NIA/AA) criteria for dementia. These differential diagnostic laboratory tests had been performed for the study participants with AD and MCI before this study. If significant incidental findings were made in initial studies, for example in cognitive tests and demographic interview of a control subject, a person with indications of cognitive problems or decline in daily functions was excluded from the control group. People with neurodegenerative diseases other than AD were not included in the study. All participants underwent a 12-odor identification test for the sense of smell (Sniffin’ Sticks, Heinrich Burghart GmbH, Wedel, Germany) [25]. For each of the assessed odors we recorded the number of individuals with a correct answer as well as the number of study participants with incorrect/no answer for the multiple-choice task identification of the odor. Furthermore, all study participants had a venous blood sample taken for APOE-genotyping [26]. A subset of the study participants with MCI (N = 16), AD (N = 16), and cognitively healthy individuals (N = 16) underwent an OM biopsy from the nasal septum in olfactory cleft, close to the roof of the nasal cavity either during the dacryocystorhinostomy surgery or in the clinic-based setting as described previously [7,27,28]. The OM biopsies were either frozen at - 70 0C prior further analysis or processed directly to primary cell cultures. Biopsies were not taken from subjects taking anticoagulation medication. Study participants were provided with oral and written information about the study and written informed consent was collected from the study volunteers. In addition, proxy consent was collected from the family members/legally acceptable representatives of the study volunteers with MCI or AD. Analysis of odors separating the study groups The 12-odor Sniffin’ Sticks odor identification test results (N = 144 individuals) were analyzed with SPSS Statistics software (IBM, v. 29.0.2.0) using Chi-square test and analysis of frequencies in the dataset. To identify the minimal subset of highly predictive odors (variables) separating the three study groups, namely cognitively healthy, MCI and AD study participants, Recursive Feature Elimination (RFE) coupled with a Random Forest algorithm using the caret (v. 7.0.1) and randomForest (v 4.7-1.2) packages in R were performed. This is a machine learning technique to find the best combination of the odors in our dataset. The RFE algorithm evaluated subsets ranging from 1 to 12 variables as we have assessed 12 odors in total by Sniffin’ Sticks odor identification test. To prevent overfitting and selection bias, feature evaluation was performed using k-fold cross-validation (k = 5) implemented via the rfeControl function. To visualize the hierarchical interactions and conditional combinations of a representative subset of these features, a Classification and Regression Tree (CART) model was subsequently generated using the rpart R package. The complexity parameter (cp) was set to 0.02 to restrict tree depth and prevent overfitting. Immunohistochemistry A representative biopsy specimen was cryosectioned and stained with hematoxylin-eosin and antibodies against olfactory marker protein (OMP) and Ulex Europaeus Agglutinin I (UEA I) to confirm the presence of OM in the obtained OC mucosa biopsies [29]. The biopsy was placed directly in 4 % paraformaldehyde solution for approximately 20 h. After fixation it was submerged in 30 % sucrose solution (in 0.1 M phosphate buffer) for 24-h period. The biopsy was then mounted in OCT cryoembedding matrix (Epredia) and the sample was frozen on dry ice before storing the OCT block at −70 °C until prepared for 8 μm cryosections. Cryosections were stained with hematoxylin-eosin solution following a standard protocol and imaged with 40Χ objective on the Zeiss Axio Imager M2 using a digital camera and ZEN software (Zeiss, Germany). Next, the cryosections within ∼16 μm distance from the sections stained with hematoxylin-eosin were processed. First, glass slides containing the frozen sections were thawed and rinsed in 1Χ phosphate-buffered saline (PBS) before placing the slides to 10 mM citrate buffer (pH 6.0) for antigen retrieval for 45 min. Sections were rinsed with 1Χ PBS containing 0.2 % Tween20 (PBST) and blocked with 5 % normal goat serum diluted in 1Χ PBST for 1 h at room temperature. The sections were incubated with primary antibodies diluted in 3 % normal goat serum in 1Χ PBST overnight at room temperature (anti-OMP: sc-365818, Santa Cruz Biotechnology, 1:50. and anti-UEA I conjugated with DyLight595: DL-1067-1, Vector Laboratories, 20 μg/ml) and washed with 1Χ PBST. Incubation with goat anti-mouse Alexa Fluor 488 - conjugated secondary antibody (A-11001, Invitrogen, 1:500) was performed for 2 h at room temperature, and after washing with 1Χ PBST the sections were embedded with Vectashield mounting medium (Vector Laboratories) for fluorescence with 4′,6-diamidino-2-phenylindole (DAPI). Lastly, the fluorescent sections were imaged with 40Χ objective on the Zeiss Axio Imager M2 using a digital camera and ZEN software (Zeiss). Analysis of plasma phosphorylated tau 217 Plasma phosphorylated protein tau 217 (pTau217) was measured from a subset of OM biopsy donors (N = 6 individuals/study group) using a single molecular array (Simoa) immunoassay with a fully automated HD-X Analyser (Quanterix, Lexington, MA). Measurements were performed using the ALZpath Simoa pTau-217 V2 Kit from Quanterix according to the manufacturer's instructions (Lot: 999024). Samples were analyzed in singlets on a single plate in random order and by staff blind to the clinical diagnosis of the sample donors. Proteomics of the OM biopsies Tissue preparation and protein digestion Frozen OM biopsy pieces were crushed with percussive force with pre-cooling the required tools and tissue pieces with liquid nitrogen. 10 mg of tissue powder/sample was used for quantitative shotgun proteomics with liquid chromatography with electrospray ionization and tandem mass spectrometry (LC-ESI-MS/MS). First, the grinded tissue powder samples were lysed in 8 M urea in 50 mM Tris-HCl, pH 8.0, sonicated 3 × 5 min, and finally digested overnight with sequencing grade modified trypsin (Promega). After digestion the peptides were desalted with a Sep-Pak tC18 96-well plate (Waters) and evaporated to dryness prior they were dissolved in in 0.1 % formic acid and the peptide concentration was determined with a NanoDrop device. LC-ESI-MS/MS analysis Equal masses of peptides per sample were injected and analyzed in random order for data independent analysis (DIA) on a nanoflow HPLC system (Easy-nLC1200, Thermo Fisher Scientific) coupled to the QExactive HF mass spectrometer (Thermo Fisher Scientific, Bremen, Germany) and equipped with a nano-electrospray ionization source. Firstly, the peptides were loaded on a trapping column and subsequently separated inline on a 15 cm C18 column (75 μm × 15 cm, ReproSil-Pur 3 μm 120 Å C18-AQ, Dr. Maisch HPLC GmbH, Ammerbuch-Entringen, Germany). The mobile phase consisted of water with 0.1 % formic acid (solvent A) or acetonitrile/water (80:20 (v/v)) with 0.1 % formic acid (solvent B). A 120 min gradient was used to elute peptides (60 min from 5 % to 21 % solvent B followed by 48 min from 21 % to 36 min solvent B and in 6 min from 36 % to 100 % of solvent B, followed by 6 min wash stage with solvent B). The mass spectrometry data was acquired by Thermo Xcalibur 4.1 software (Thermo Fisher Scientific). In the used DIA method, a duty cycle contained one full scan (400 -1000 m/z) and 40 DIA MS/MS scans covering the mass range 400 -1000 with variable width isolation windows. Protein identification and quantification DirectDIA approach of Spectronaut software (v. 17.0.2, Biognosys) was utilized to identify proteins and label-free quantifications of protein abundances were performed with MaxLFQ of Spectronaut software. The parameters in Spectonaut for the data analysis identified the Trypsin/P as the enzyme used for the protein digestion allowing up to two missed cleavages. Fixed modification included Carbamidomethyl and variable modifications of Acetyl (protein N-term) and oxidation (M). Both the precursor and protein false discovery rate (FDR) cutoffs were set to 0.01 and the quantification MS level MS2 was used, quantifying the area under the curve within integration boundaries for each targeted ion. Human proteome (Swiss-Prot 2022_02 Homo Sapiens) and Universal Protein Contaminant database [30] were used for the protein mapping. All found contaminant proteins were removed. Median normalization was performed with proBatch package [31]. Proteins missing from more than a total of three samples were excluded from the following analyses. Missing values remaining in the dataset were imputed using the sample minimum method. Protein differential abundance and overrepresentation analysis between the study groups Differentially abundant proteins were identified with Limma v. 3.54.2, including age, sex and information of the study group in the linear model in order to mitigate the possible effects of these confounding factors. Proteins with P value < 0.05 and absolute fold change (FC) > 1.2 were considered as significantly differently abundant and visualized with Volcano plots generated using the ggplot2 package in R v. 4.4.1. The effect size was calculated using topconfects R package (v. 1.18.0). All identified proteins found to be significantly altered in AD when compared to controls, in MCI compared to controls, or between the AD and the MCI, were assessed for overrepresentation analysis with Gene Ontology and Reactome pathways with PANTHER Classification system platform (v. 18.0) and Fisher's exact test with FDR correction [32]. Enrichment FDR values less than 0.05 were considered as significant. The protein–protein interaction information for the significantly differentially abundant proteins was analyzed using STRING database (v. 12.0) [33]. Mitochondria related protein sub-networks were visualized in Cytoscape v. 3.10.1. Odor identification test results and differential protein expression correlation analysis was performed using Limma. OM primary cell cultures Primary cell cultures of the OM-derived cells were established from biopsies of the OM according to a previously published protocol [23]. The cells derived from biopsies of the OC mucosa were cultured at 37 °C, 5 % CO2 in cell culture medium (DMEM/F12, #11320033, Gibco, Waltham, MA, USA) supplemented with 10 % heat-inactivated fetal bovine serum (FBS) (#10270106, Gibco) and 1× Penicillin–Streptomycin (#15140122, Gibco) and seeded as 5000 cells/cm2 in the cell culture flasks. The cells were used for experiments at primary passages four to five. The cells were gently detached from the cell culture flasks with TrypLE Express enzyme (12604-021, Gibco) when passaged or plated for experiments. Cell redox profiling Quantitative analysis of NAD+, NADH, NADP+, NADPH, and reduced and oxidized forms of glutathione in cultured cells was performed as a service in NADMED laboratory (Helsinki, Finland) using a proprietary technology (for further information see www.nadmed.com and [34]). Briefly, the cells were collected from N = 6 individuals/study group with one technical replicate from each. The cultured cells were washed with 1Χ PBS to remove proteins derived from the culture media, followed by the addition of pre-warmed lysis solution to quench cellular metabolism. Lysis solution was a water-based complex mixture of organic solvents, which force protein unfolding at temperatures above +45 °C and release of all non-covalently bound metabolites into the solution. Next, the lysed cells were carefully collected by scraping. The homogenates were frozen at −70 °C and shipped on dry ice to the NADMED laboratory. Before analysis, the homogenates were equilibrated to room temperature and centrifuged at 20000 x g for 10 min at +4 °C to remove the protein pellet. Next, NAD+, NADH, NADP+, NADPH, GSH pool, and GSSG were measured individually from cell extracts using modified cyclic enzymatic reactions with colorimetric detection. For normalization of the results, protein content in the sample was measured using the pellet obtained after centrifugation of the homogenate. RNA extraction and RT-qPCR Primary OM derived cells were seeded on cell culture plates five days prior to RNA isolation (cognitively healthy controls (N = 10), MCI (N = 11), AD (N = 11)). Total RNA was extracted with RNeasy Plus Mini kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions with β-mercaptoethanol added to the RLT Plus reagent. The RNA integrity was analyzed with RNA 6000 Pico kit together with the Agilent 2100 Bioanalyzer (Agilent, Santa Clara, CA, USA). Before cDNA synthesis 500 ng of the extracted RNA was treated with DNAse I (Thermo Scientific) and converted to cDNA with High-Capacity cDNA Reverse Transcription Kit according to the manufacturer's instructions (Applied Biosystems). The gene expression changes of genes encoding for selected NRF2 target genes were quantified as duplicates with StepOne Real-Time PCR System (Applied Biosystems) with 40 amplification cycles and TaqMan real-time PCR assay mixes. The following TaqMan primer mixes were used: NQO1 (Hs00168547_m1), HMOX1 (Hs01110250_m1), GCLC (Hs00155249_m1), GCLM (Hs00157694_m1), and GAPDH (Hs02758991_g1). microRNA sequencing miRNA libraries were prepared directly from the total RNA with QIAseq miRNA library kit (Qiagen) as described in the kit manual. The miRNA libraries were sequenced (single-end, 85 cycles) with NovaSeq 6000 platform and SP Reagent Kit (100 cycles) (Illumina, San Diego, CA, USA). UMI read deduplication, and miRNA identification and quantification were done using the QIAseq miRNA Primary Quantification pipeline (QIAGEN). DeSeq2 package was used for identification of differentially expressed miRNAs. Only miRNAs with P value < 0.05 and absolute FC of at least 1.2 were considered significant and used for subsequent analysis. Human target genes of the differentially expressed miRNAs were detected using multiMiR R package (v. 1.26.0) [35]. The interactions between differentially expressed proteins and miRNAs were analyzed using STRING database (v. 12.0) [33]. A sub-network of hsa-miR-142-5p and differentially expressed proteins in AD and MCI OM biopsies was visualized in Cytoscape v. 3.10.1. Statistics and graphical illustrations The odor identification test results and study cohort characteristics (N = 144 individuals) were analyzed with SPSS Statistics software (IBM, v. 29.0.2.0) using Chi-square test and analysis of frequencies in the dataset. Both the analysis for shotgun proteomics and miRNA sequencing were performed with Limma (v. 3.54.2) and were considered significant with cut-off value of unadjusted P values < 0.05 based on previous evidence [36,37] and absolute FC difference > 1.2. The cell redox profiling and RT-qPCR experiments with primary OM cells were performed with a single technical replicate or as duplicates of samples collected from each primary cell donor, respectively. The statistical analysis for these experiments were performed with GraphPad Prism (v. 10.0.2) using one-way analysis of variance (ANOVA) for overall groupwise difference followed with Dunnett's or Tukey's multiple comparisons test. The data is presented as mean ± SD or as box plot representing median and individual datapoints. PANTHER overrepresentation test for the DEPs was performed with Fisher's exact test with FDR correction. Graphical illustrations were created with open-source software Inkscape (v. 1.2.2) and Cytoscape platform (v. 3.10.1). Results Study cohort Characteristics of all the study participants (N = 144) involved in this study are summarized in Table 1. A subpopulation of all the study participants underwent a biopsy of the OM (cognitively healthy individuals (N = 16), MCI (N = 16), AD (N = 16)), and the biopsy derived tissue samples were subjected either to shotgun proteomics or processed to primary cultures of the OM cells. Table 1Study participant characteristics (N = 144).Table 1CharacteristicsCognitively healthy control (CTRL)Mild cognitive impairment (MCI)Alzheimer's disease (AD)Number of participants534744Age, Mean (SD)70.6 ± 5.569.7 ± 6.369.8 ± 7.5Sex of participantsFemales, N (%)37 (69.8)17 (36.2)19 (43.2)Males, N (%)16 (30.2)30 (63.8)25 (56.8)Number of abnormalities in CERAD (mean)0.92.76.0APOE ε4 carriers, N (%)heterozygous,N (%)19 (35.8)14 (29.8)10 (22.7)homozygous,N (%)1 (1.9)5 (10.6)18 (40.9)No ε4 allele,N (%)24 (45.3)24 (51.1)13 (29.5)No genotyping result available9 (17.0)4 (8.5)3 (6.8)Odor identification test resultNormal, N (%)45 (84.9)30 (63.8)11 (25.0)Hyposmia, N (%)8 (15.1)15 (31.9)28 (63.6)Anosmia, N (%)0 (0)2 (4.3)5 (11.4)Common comorbiditiesDiabetes, N (%)5 (9.4)8 (17.0)5 (11.4)Hypertension, N (%)24 (45.3)25 (53.2)17 (38.6)Hypercholesterolemia, N (%)22 (41.5)20 (42.6)24 (54.5)Nasopharyngeal diseasesAllergic rhinitis, N (%)6 (11.3)6 (12.8)4 (9.1)Chronic rhinitis, N (%)0 (0)0 (0)3 (6.8)Nasal polyps2 (3.8)0 (0)0 (0) Study participant characteristics (N = 144). Individuals with MCI and AD fail to identify certain odors correctly The sense of smell has a role in several key areas of life, from physical and mental well-being aspects to preserving life itself via everyday safety. Moreover, the chemical cues of different smells also have strong power in memory retrieval, and olfactory impairments are often reported in individuals with several types of neurodegenerative diseases [4,5]. Thus, we first performed a test for nasal chemosensory function for odor identification with the Sniffin’ Sticks test, where 12 different odors were presented in felt-tip pens to all the participants of this study (total N = 144). The results demonstrated that individuals with MCI or AD have more difficulties in the odor identification test compared to the cognitively healthy individuals (Table 2). Odor identification results for the odors of orange, banana, coffee, and fish resulted in the most significant differences between the individuals with AD and cognitively healthy study participants (P value < 0.001, Chi-square test), followed by identification of the odors of leather, peppermint, lemon, liquorice, pineapple, and rose (P value < 0.01). Odor identification results for the odor of cinnamon and clove did not differentiate between any of the study groups in this study. Notably, like the individuals with AD, a test for the smell of coffee stratified individuals with MCI from the cognitively healthy study participants (P value < 0.001). Individuals with MCI also encountered difficulties in identifying the odors of fish (P value < 0.01), banana, and lemon (P value < 0.05). These results show that both individuals with MCI or AD gave significantly more incorrect answers or gave no answer specifically to the odors of coffee, fish, banana, and lemon of the 12 item Sniffin’ Sticks test, when compared to the cognitively healthy study participants. Our additional analysis with a machine learning technique, namely Recursive Feature Elimination (RFE) coupled with a Random Forest algorithm, suggested the odors of coffee, orange, banana, fish and lemon as the minimal subset of highly predictive odors (variables) separating the three study groups. The hierarchical interactions and conditional combinations of a representative subset of these features are visualized with a Classification and Regression Tree (CART) model (Supplementary Fig. 1). In our cohort 34 individuals could not identify the odor of coffee (coffee = 0) and the majority, 19 in total, were individuals with AD. From the rest 110 individuals, which identified the coffee odor (coffee = 1), 17 could not identify the odor of orange. The majority of those were also AD cases. Then from the remaining 93 study participants, 50 could not identify the odor of lemon and a nearly half of these individuals belong to the MCI study group. This emphasizes the added value of certain odor combinations in stratification of the individuals with AD or MCI from the cognitively healthy individuals. Table 2Sniffin’ Sticks odor identification test results for 12 odors for individuals with MCI and AD, and the cognitively healthy individuals (Chi-square test, where * P value < 0.05, **P value < 0.01, ***P value < 0.001).Table 2GroupTotalΧ2 test: MCI vs ControlΧ2 test: AD vs ControlControlMCIADN%N%N%N%ValuedfAsymptotic Significance (2-sided)Significance (P value < 0.05)ValuedfAsymptotic Significance (2-sided)Significance (P value < 0.05)Smell 1 (Orange)Correct4788,7 %3778,7 %2454,5 %10875,0 %183711,75E-01ns14,27711,58E-04***Incorrect/no answer611,3 %1021,3 %2045,5 %3625,0 %Smell 2 (Leather)Correct4381,1 %3268,1 %2352,3 %9868,1 %226111,33E-01ns920812,41E-03**Incorrect/no answer1018,9 %1531,9 %2147,7 %4631,9 %Smell 3 (Cinnamon)Correct3973,6 %3472,3 %2761,4 %10069,4 %0,02018,89E-01ns165111,99E-01nsIncorrect/no answer1426,4 %1327,7 %1738,6 %4430,6 %Smell 4 (Peppermint)Correct4992,5 %4187,2 %3170,5 %12184,0 %0,75413,85E-01ns804914,55E-03**Incorrect/no answer47,5 %612,8 %1329,5 %2316,0 %Smell 5 (Banana)Correct4890,6 %3472,3 %2352,3 %10572,9 %560611,79E-02*17,96912,25E-05***Incorrect/no answer59,4 %1327,7 %2147,7 %3927,1 %Smell 6 (Lemon)Correct3260,4 %1736,2 %1227,3 %6142,4 %584111,57E-02*10,63011,11E-03**Incorrect/no answer2139,6 %3063,8 %3272,7 %8357,6 %Smell 7 (Liquorice)Correct4177,4 %2859,6 %2352,3 %9263,9 %368315,50E-02ns674019,43E-03**Incorrect/no answer1222,6 %1940,4 %2147,7 %5236,1 %Smell 8 (Coffee)Correct5196,2 %3472,3 %2556,8 %11076,4 %11,14718,42E-04***22,01112,71E-06***Incorrect/no answer23,8 %1327,7 %1943,2 %3423,6 %Smell 9 (Clove)Correct3362,3 %2553,2 %1943,2 %7753,5 %0,84213,59E-01ns352016,06E-02nsIncorrect/no answer2037,7 %2246,8 %2556,8 %6746,5 %Smell 10 (Pineapple)Correct3667,9 %2553,2 %1840,9 %7954,9 %227311,32E-01ns711017,67E-03**Incorrect/no answer1732,1 %2246,8 %2659,1 %6545,1 %Smell 11 (Rose)Correct3973,6 %3063,8 %1943,2 %8861,1 %110812,92E-01ns924412,36E-03**Incorrect/no answer1426,4 %1736,2 %2556,8 %5638,9 %Smell 12 (Fish)Correct5094,3 %3370,2 %2659,1 %10975,7 %10,27711,35E-03**17,61012,71E-05***Incorrect/no answer35,7 %1429,8 %1840,9 %3524,3 % Sniffin’ Sticks odor identification test results for 12 odors for individuals with MCI and AD, and the cognitively healthy individuals (Chi-square test, where * P value < 0.05, **P value < 0.01, ***P value < 0.001). Shotgun proteomic analysis of OM biopsies supports a role for mitochondrial changes in MCI and AD olfactory function Since the nasal chemosensory function for odor identification was affected in individuals with MCI and AD, we were curious to investigate the cellular and molecular changes occurring in the upper nasal mucosal tissue, important for the sense of smell. We applied shotgun proteomics in an un-targeted fashion to assess potential alterations in the proteome of the olfactory mucosal tissue of living individuals. In participants donating OM biopsies for the proteomic analyses, plasma pTau217 levels were significantly increased in individuals with AD compared to cognitively healthy control subjects (P value ≤ 0.001). (Fig. 1A). Plasma pTau217 is elevated in preclinical-AD, associated with beta-amyloid pathology, and correlates with the cerebral neurofibrillary tau-tangle density score (27). Plasma pTau217 levels in individuals with MCI were on average between the levels measured in plasma of cognitively healthy study participants and individuals diagnosed with AD, further assuring the correct categorization of our study groups. The biopsies were obtained from the septal wall of olfactory cleft mucosa, a location which is illustrated in Fig. 1B, taken by the same otorhinolaryngologist. Presence of the OM in the biopsy samples was confirmed by morphology with the aid of hematoxylin-eosin staining and cells positive for olfactory marker protein (OMP) and Ulex Europaeus Agglutin I (UEA I) [29] on a representative tissue specimen derived from one cognitively healthy individual (Fig. 1C–F). LC-MS/MS analysis of OM biopsies derived from a subset of study participants with MCI and AD and cognitively healthy individuals (N = 6 individuals per group) demonstrated that a total of 6031 different proteins were identified in OM biopsies after filtering the list of identified proteins found in ≥ 15 of the total 18 biopsy samples. A differential abundance (Limma, P value < 0.05, absolute FC difference > 1.2) was observed for a total of 184 protein candidates between the OM biopsies derived from individuals with MCI and cognitively healthy control subjects, and for 901 proteins between the biopsies derived from individuals with AD and cognitively healthy controls, respectively. Furthermore, the OM biopsies derived from individuals with AD displayed differential abundance for a total of 663 protein candidates when compared to individuals with MCI (Fig. 1G–Supplementary Tables 1–3). We identified a total of 40 protein candidates, the expression of which were altered (Limma, P value < 0.05, absolute FC difference > 1.2) in both MCI and AD OM biopsies when compared to the tissue derived from the cognitively healthy control subjects (Fig. 1H–Supplementary Table 4). For example, protein products of C18orf63 (Uncharacterized protein C18orf63), BICD2 (Protein bicaudal D homolog 2), PID1 (PTB containing, cubilin and LRP1-interacting protein), VPS41 (Vacuolar protein sorting-associated protein 41 homolog), CEP41 (Centrosomal protein of 41 kDa), TGFBRAP1 (Transforming growth factor-beta receptor-associated protein 1), SPAG1 (Sperm-associated antigen 1), and HBG2 (Hemoglobin subunit gamma-2) showed a distinct and coordinated downregulation both in MCI and AD OM biopsies with Log2 FC differences < −3 in both comparisons to biopsies derived from cognitively healthy control subjects. Furthermore, from these 40 protein candidates significantly altered both in MCI and AD, the abundance of six proteins; vesicle-associated membrane protein 5 (VAMP5), 5-demethoxyubiquinone hydroxylase, mitochondrial (COQ7), cathepsin B (CTSB), small ribosomal subunit protein eS21 (RPS21), clathrin light chain A (CLTA), and 3-hydroxyacyl-CoA dehydrogenase type-2 (HSD17B10) were associated with the odor identification test score of the OM mucosa biopsy donor (Fig. 1I–Supplementary Table 5). Shotgun proteomic analysis of the olfactory mucosa (OM) biopsies derived from individuals with MCI and AD or from cognitively healthy controls (CTRL). (A) Plasma SIMOA pTau217 levels of those individuals, from which OM biopsies were analyzed with LC-ESI-MS/MS (N = 6 individuals/group, one-way ANOVA with Tukey test). (B) Illustration of the location of the OM biopsy sampling from the nasal septum of the olfactory cleft. (C-F) Example images of the organization of the biopsied tissue and showing the presence of olfactory neurons visualized by hematoxylin-eosin staining and immunostaining with anti-OMP and anti-UEA I antibodies using 8 μm cryosections of the OM biopsy derived from a cognitively healthy individual. Distance between the sections stained with hematoxylin-eosin and anti-UEA I or anti-OMP antibodies is ∼16 μm. Scale bar 20 μm. (G) Volcano plots for differentially expressed proteins (DEP) identified in the shotgun proteomics analyses of OM biopsies. (H) Scatter plot visualizing unique and shared DEPs between the study groups. Significant DEPs (P value < 0.05, absolute FC for differential expression > 1.2) shown in orange (AD), magenta (MCI) or blue (shared between the AD and MCI). (I) Six DEPs, which were shared between the MCI and AD when compared to controls, were significantly (P value < 0.05, Limma, absolute FC for differential expression > 1.2) associated with the odor identification test score of the biopsy donor. OM, olfactory mucosa; LC-ESI-MS/MS, liquid chromatography with electrospray Ionization and tandem mass spectrometry; DEP, differentially expressed protein; FC, fold change. **P value ≤ 0.01, ***P value ≤ 0.001. Mitochondrial dysfunction has been identified as a pathological feature of AD as summarized in Ref. [38]. A total of 116 of the 901 differentially expressed proteins (DEPs) in AD OM biopsies, 48 of the 184 DEPs in MCI OM, and 12 of the 40 DEPs altered in both MCI and AD, were associated with the mitochondria (GO:0005739) (Fig. 2A, Supplementary Tables 1, 2, and 4). Overrepresentation analysis of the significant DEPs identified several biological processes, and Reactome pathways suggested alterations related to cellular respiration, namely to the citric acid cycle and mitochondrial electron transport chain (ETC), especially in OM biopsies derived from individuals with MCI. In individuals with AD, the most significantly associated pathways were found to be related to metabolism and processing of various forms of RNA (Fig. 2B–Supplementary Tables 6–8). Mitochondrial changes and cellular defense against oxidative stress are evident in MCI and AD olfactory mucosa (OM). (A) 12 of the 40 significant DEPs (P value < 0.05, Limma, absolute FC for differential expression > 1.2) shared with AD-CTRL and MCI-CTRL comparisons are associated with mitochondria (GO:0005739). Shape border color indicates the Log2 FC differences in protein expression between the AD and control derived biopsies, and, respectively, shape fill color is the indicator of the Log2 FC difference between the MCI and cognitively healthy individuals. (B) PANTHER overrepresentation analysis of the DEPs (P value < 0.05, absolute FC for differential expression > 1.2) between the OM biopsies derived from individuals with AD or MCI and cognitively healthy controls. The graphs visualize the top 10 biological processes and Reactome pathways based on the FDR. Protein expression alterations of reported candidate proteins of NRF2 signalling in (C) AD and (D) MCI OM biopsies. The shape color indicates the Log2FC difference in protein expression between AD or MCI and control OM biopsies as the shape size indicates the P value of the observed difference. OM, olfactory mucosa; DEP, differentially expressed protein; FC, fold change; FDR, false discovery rate. Mitochondrial metabolism is a significant source of ROS in the cell; the ROS are mainly produced by cellular respiration at the mitochondrial ETC, and they have the potential to initiate redox dependent signaling and gene expression networks [20]. Accumulation of dysfunctional mitochondria, as has been reported for aging brains affected with AD [38], can cause cellular damage due to the imbalanced levels of ROS and the oxidative potential of ROS for all major cellular biomolecules. To resolve the oxidative stress, nuclear factor erythroid 2–related factor 2 (NRF2) can translocate to the cell nucleus and function as a transcriptional regulator controlling several cellular antioxidant and anti-inflammatory responses [21]. In this study, 56 proteins identified as significantly differentially expressed between the AD and control OM biopsies were candidate gene/protein markers of NRF2 signaling, as recently compiled by Morgenstern et al. [21]; acting as either direct NRF2 targets, downregulated by inhibition of basal NRF2 activity, or upregulated by NRF2 activation (Fig. 2C). Additionally, MCI OM biopsies displayed alterations in protein expression of 16 candidate markers of altered NRF2 signaling (Fig. 2D) [20]. We found significantly reduced expression of protein products of CD99 antigen (CD99), aryl hydrocarbon receptor (AHR), ferritin light chain (FTL), claudin-4 (CLDN4), phosphatidylinositol 4-phosphate 5-kinase type-1 gamma (PIP5K1C), and 5′-AMP-activated protein kinase catalytic subunit alpha-2 (PRKAA2) in OM biopsies derived from individuals with AD (Supplementary Table 1), all shown to be directly regulated by NRF2 by previous chromatin immunoprecipitation experiments [21]. The protein abundance of CD99 was also reduced in the MCI OM biopsies. Together, these findings suggest that cells of the OM of individuals with MCI, at a time when significant odor identification problems are also present, undergo a variety of proteomic alterations related especially to the functions of cell mitochondria, and the cellular defense against oxidative stress. Primary cultures of olfactory mucosa cells reaffirm altered redox homeostasis in AD Primary cultures of the OM biopsy derived cells recapitulate several key features of neurodegenerative diseases associated with disturbances in the sense of smell, for example in Parkinson's disease [[39], [40], [41]]. We have earlier reported AD OM – derived cells to have impairments in ATP production, homeostasis of biometals, and several transcriptional alterations in mitochondrial genes [23,24]. To further support the observed energy metabolism and NRF2 signaling-related proteomic alterations in the OM in MCI and AD, we first focused attention to the levels of selected NRF2 target genes responsible for maintenance of redox homeostasis; NAD(P)H Quinone Dehydrogenase 1 (NQO1), Heme Oxygenase 1 (HMOX1), Glutamate-Cysteine Ligase Catalytic Subunit (GCLC), and Glutamate-Cysteine Ligase Modifier Subunit (GCLM), in primary OM cell cultures derived from individuals with AD, MCI, and cognitively healthy individuals. Similarly to the frontal cortex of 6-month-old APP/PS1 mice [42], the primary cells of the OM exhibited significant reductions in gene expression of NQO1, GCLM and GCLC in both MCI and AD when compared to the cells derived from cognitively healthy individuals (Fig. 3A), suggesting alterations in NRF2 signaling and gene expression of its target genes. Primary cultures of olfactory mucosa cells reaffirm altered redox homeostasis in AD and its potential regulation by microRNAs. (A) RT-qPCR results for selected NRF2 target genes in primary cell cultures of OM derived cells from individuals with MCI (N = 11) and AD (N = 11), and cognitively healthy study participants (N = 10). Each dot represents cells derived from one biopsy donor after samples tested to be outliers (Grubb's test, alpha 0.05) were removed. One-way ANOVA (#) with Dunnett's test (*). (B) Quantification of the levels of reduced and oxidized forms of glutathione, (C) NADP+, and (D) NAD+ and NADH in primary cultures of OM-biopsy derived cells. Each dot represents cells derived from one biopsy donor (N = 6 individuals/study group). One-way ANOVA (#) with Dunnett's test (*); * or #P value ≤ 0.05, ** or ##P value ≤ 0.01, *** or ###P value ≤ 0.001. (E) Significant DEPs (P value < 0.05, Limma, absolute FC for differential expression > 1.2) in MCI and AD OM biopsies identified to interact with hsa-miR-142-5p based on miRTarBase, TarBase and miRecords databases. Shape fill color indicates whether the miRNA target is identified as DEP in AD (orange), MCI (turquoise) or in both MCI and AD OM biopsies (yellow) when compared to biopsies derived from cognitively healthy biopsy donors. (F) Visualization of the DEPs in MCI and AD OM interacting with hsa-miR-142-5p based on their differential protein expression to biopsies derived from cognitively healthy study participants. Shape border color indicates the Log2 FC differences in protein expression between the AD and control derived biopsies, and, respectively, shape fill color indicates the Log2 FC difference between the MCI and control. DEP, differentially expressed protein. NRF2 also takes part in the synthesis of glutathione and NADPH [21], which are key electron carrier molecules facilitating redox defense against ROS. Most of the ROS originates as by-products of the life-sustaining cellular respiration in mitochondria. Since our proteomics results of the OM-derived tissue samples suggested the importance of mitochondria and altered NRF2 signaling in MCI and AD, we were interested to quantify the levels of NAD+, NADH, NADP+, NADPH, and reduced and oxidized forms of glutathione in primary cultures of the cells of the OM. Furthermore, the study participants of the odor identification test with AD or MCI made more errors especially in identification of odor of coffee, which is known to contain thiols as volatile compound [43]. Notably, glutathione is also a small molecule thiol itself. The total levels of glutathione (GSH pool) were not altered between the groups. However, the cells of individuals with AD showed a slight but not statistically significant reduction (P value 0.0617) in the ratio of reduced glutathione over its oxidized form (GSH/GSSG) (Fig. 3B). In previous studies with postmortem tissue samples, levels of GSH were shown to be reduced with increasing age in hippocampal tissue samples but left unaltered in frontal cortex and cerebellum with aging [44], suggesting brain region specific differences. Alterations in GSH/GSSG ratio in OM might be an early event in the AD disease course to be investigated further, as the MCI OM biopsy-derived primary cells showed a trend towards a reduced GSH/GSSG ratio as well. NADPH was found to be very labile in OM biopsy-derived primary cells. It was detected in a very limited number of all the analyzed samples; these samples had lower NADP + levels, but the pool of NADP (NADP + combined with NADPH) was close to the level of NADP + measured in samples with non-detectable NADPH. This may suggest that NADPH is easily converted into NADP+ in these cells. In most samples, the measured NADP + represented the pool of both forms of NADP. Interestingly, the levels of NADP+ were significantly increased both in the cells derived from individuals with MCI and AD when compared to the cells derived from cognitively healthy individuals (Fig. 3C). In physiological conditions NADP+ is required as an electron donor for various anti-oxidative defense systems of the cells, including the generation of GSH from GSSG, and as a precursor of agents involved in intracellular calcium (Ca2+) release and mobilization, usually kept at low levels [45]. We have previously reported increased levels of intracellular Ca2+ in AD OM-derived primary cell cultures when compared to the cells derived from cognitively healthy controls [24]. In comparison to the NADP+, the levels of NAD+ and NADH were not significantly altered in primary cells derived from individuals with MCI or AD (Fig. 3D). Our shotgun proteomic analysis of the OM, as tissue samples from separate individuals to the primary cell cultures and with confirmed levels of pTau217 in plasma, indicated increased protein expression of mitochondrial NADP + -dependent isocitrate dehydrogenase, IDHP encoded by IDH2 gene, and NADP+ and NADH producing nicotinamide nucleotide transhydrogenase (NNT) in MCI when compared to the controls (P value < 0.05, Limma, Log2 FC to control 0.718 and 0.553, respectively), and increased expression of NADPH consuming antioxidant enzyme thioredoxin (TXN) (P value < 0.01, Limma, Log2 FC 0.425). These proteomic changes in the OM tissue support the observation of increased levels of oxidized NADP in OM-derived primary cell cultures, possibly indicating either the increased need of NADPH or enhanced oxidation of NADPH in these cells as the levels of NADP + seemed to be highest especially in cells derived from individuals with MCI. Aberrant NRF2 signaling is linked to AD [42,46]. Expression of NRF2 can be also negatively regulated post transcriptionally by various microRNA (miRNA) molecules directly targeting the NRF2 mRNA, and the expression of several NRF2 targeting miRNAs have been implicated also in the brain of individuals affected by AD [47]. Furthermore, interestingly we detected DEPs related to pathways and biological processes of RNA metabolism and RNA processing to be overrepresented in AD OM biopsies. To study the expression of potentially NRF2 targeting miRNAs in the context of the upper nasal mucosa, we performed miRNA sequencing to our OM biopsy-derived primary cell cultures derived from cognitively healthy study participants and individuals with MCI or AD. The alterations in miRNAs were analyzed from the same samples that were used previously for RT-qPCR for selected NRF2 target genes. Altogether 1225 different miRNAs were detected in the primary cultures of OM cells out of 2656 known mature miRNAs in the microRNA database miRBase [48] (mirbase.org, release 22.1). We found 72 miRNAs to be differentially expressed (P value < 0.05, DeSeq2, absolute FC difference > 1.2) between the primary cultures of OM cells of individuals with AD and cognitively healthy individuals (Supplementary Table 9), and similarly, 31 miRNAs differentially expressed between individuals with MCI and controls (Supplementary Table 10). MicroRNA hsa-miR-142-5p was found upregulated among the top 10 of the most differentially expressed miRNAs (assessed both by the Log2FC and P value) when comparing AD and control cells (hsa-miR-142-5p Log2FC 1.78; P value < 0.01). In donors with MCI the expression of hsa-miR-142-5p was also upregulated (Log2FC 0.62), but the observed upregulation did not reach the set level of significance for P value. Notably, hsa-miR-142-5p has been previously reported to be upregulated in prefrontal cortex of individuals with late-onset AD [49] and to target human NFE2L2, encoding for NRF2, in position 83-90 of NFE2L2 3′ UTR based on miRDB [50,51] and TargetScan databases [52]. From all the identified interactions between miRNAs and DEPs (Supplementary Figs. 2 and 3), a total of 67 DEPs in the OM biopsies derived from either individuals with AD or MCI were observed to interact with hsa-miR-142-5p. Most of these interacting proteins were downregulated mainly in AD OM biopsies when compared cognitively healthy controls (Fig. 3E and F), suggesting that the reduced protein expression might be due to post transcriptional regulation of their mRNA by hsa-miR-142-5p. Seven miRNAs (hsa-miR-133a-3p, hsa-miR-146b-5p, hsa-miR-7976, hsa-miR-126-3p, hsa-miR-29b-3p, hsa-miR-199b-3p, and hsa-miR-361-5p) were found to be upregulated and shared between MCI and AD when both were compared to the primary cells derived from cognitively healthy individuals, and two miRNAs (hsa-miR-668-3p and hsa-miR-1180-3p) were downregulated both in MCI and AD OM-derived cells (Table 3). Interestingly, the microRNAs hsa-miR-29b-3p and hsa-miR-199b-5p, upregulated in MCI and AD OM primary cells, can both target glycogen synthase kinase-3 beta (GSK-3β) with aggregate PCT scores 0.80 and 0.71, respectively (targetscan.org, [52]). In AD, GSK-3β inhibition could be a posttranscriptional response attempting to induce NRF2, increase cellular antioxidant capacity, and reduce of tau hyperphosphorylation [53,54]. Table 3MicroRNAs, the expression of which were significantly altered (P value < 0.05, absolute FC for difference between groups > 1.2) both in MCI and AD OM biopsy derived primary cultures when compared to the cognitively healthy control subjects.Table 3miRNAMCI-CTRL Log2FCMCI-CTRL P valueSignificance MCI-CTRL (P value < 0.05)AD-CTRL Log2FCAD-CTRL P valueSignificance AD-CTRL (P value < 0.05)Previous findings in AD (human biosamples)hsa-miR-133a-3p1.9110.00158**1.7910.00308**hsa-miR-146b -5p1.0220.0444*1.6170.00147**Downregulation in cortex [55, 56] and blood [57]hsa-miR-79761.1070.0122*1.2360.00494**hsa-miR-126-3p0.8120.0195*1.0050.00382**Downregulation in cortex [56] and plasma extracellular vesicles [58]hsa-miR-29b-3p0.8930.0398*0.9620.0268*Downregulation in blood (meta-analysis) [37]hsa-miR-199b-3p0.7220.041*0.9390.0079**hsa-miR-361-5p0.5540.0263*0.5030.0439*Upregulation in blood (meta-analysis) [37]hsa-miR-668-3p−0.5390.0224*−0.6810.00421**hsa-miR-1180-3p−0.340.0387*−0.3670.0257*Upregulation in plasma [59]FC; fold change MicroRNAs, the expression of which were significantly altered (P value < 0.05, absolute FC for difference between groups > 1.2) both in MCI and AD OM biopsy derived primary cultures when compared to the cognitively healthy control subjects. Discussion Although olfactory dysfunction is common and occurs early in AD, and the olfactory processing areas are early sites for AD-associated pathology, the protein alterations of the human OM biopsies in AD and MCI have not previously been studied in an untargeted manner. This study provides evidence of proteomic and molecular alterations occurring in the human OM in MCI and AD. First, we tested the sense of smell of 144 individuals by an odor identification test. The odors of coffee and fish stratify both the individuals with MCI and AD from controls with the highest confidence. Notably, these odors are present in the Sniffin’ Sticks 12-odor identification test used in the current study, but not in the 40-odor University of Pennsylvania Smell Identification Test (UPSIT) test [60]. Individuals with AD are impaired in naming the odor of freshly ground coffee [61], but this has not previously been reported in individuals with MCI. Evidence is increasing for the intrinsic vulnerability of the olfactory system to tau accumulation and thus developing pathology associated with AD [62], highlighting the importance of studies of individuals with MCI. The combination of the 12-item Brief Smell Identification Test (BSIT) with demographics and the short global cognitive test showed predictive value for cognitive decline, similar to 11C-Pittsburgh compound B < pet imaging combined with the demographic information [63]. Furthermore, the identification of the odor of orange to strongly predict longitudinal tau increases in olfactory and medial temporal regions, and impairment specifically in Braak stage II [62]. Recently, the identification of the odor of orange was reported to be impaired already in preclinical stages of AD [64], similarly to our study with significantly impaired identification of the odor of orange in AD. These results provide insight into odor-specific identification difficulties in cognitively impaired individuals, highlighting the potential of odor-specific identification impairments in combination with other biological fluid biomarkers to identify and stratify individuals with preclinical AD from aged, cognitively healthy individuals. Protein expression changes have the potential to alter the disease course. By using explorative shotgun proteomics we identified notable numbers of DEPs in MCI and AD OM biopsies. The DEPs were suggestive of alterations to mitochondria and the transcription factor NRF2. Nearly a third of the 40 protein candidates, the differential expression of which were altered in both MCI and AD, were associated with mitochondria. From the shared DEPs between MCI and AD OM biopsies, when compared to biopsies from controls, the abundances of six proteins, such as 17β-HSD10, were associated with the biopsy donor's odor identification test score. Similar to our results with OM biopsies, increased expression of Aβ and NAD-binding mitochondrial enzyme 17β-HSD10 [65,66] has been previously reported in astrocytes in AD-affected brains [67] and in the hippocampi of the AD-modelling Tg2576 mice [68]. Taken together, the proteome alterations in the MCI OM are limited in comparison to AD, in which close to 1000 proteins are altered in comparison to cognitively healthy controls, yet the MCI and AD OM also harbor some specific protein changes that are common to both. Our data suggested changes to NRF2 signaling in the MCI and AD OM. For example, several DEPs especially in AD OM biopsies are reported as candidate markers of NRF2 signaling and the MCI and AD primary OM cell cultures displayed reduced gene expression of selected NRF2 target genes and were suggestive to changes in amounts of oxidized glutathione and NADPH. Levels of nuclear NRF2 are reduced in hippocampal neurons of AD brains [46], and both Nrf2 protein and gene expression of the Nrf2 target genes are reduced in APP/PS1 mice, occurring along with increased brain Aβ burden [42]. Furthermore, Yoshikawa et al. reported reduced levels of Parkinson disease protein 7 (PARK7) in elderly subjects when compared to young individuals in shotgun proteomic analysis of the mucus layer of the olfactory cleft [69]. Interestingly, PARK7 is required for the stability of NRF2 [70]. We detected reduced abundances of several proteins directly regulated by NRF2 in OM biopsies, such as the CD99 antigen in MCI and AD and protein product of PRKAA2 in AD OM biopsies. Previously, increased levels of the CD99 antigen are reported in CSF in asymptomatic familial AD mutation carriers when compared to the related noncarriers [71]. In addition, the brain-specific suppression of 5′-AMP-activated protein kinase catalytic subunit α-2 (AMPKα2), encoded by PRKAA2, has been linked to cognitive impairment with conditional knockout mice for AMPKα2 [72]. Altogether, these results suggest that the NRF2 pathway might be dampened in the OM of cognitively impaired individuals, similar to the AD brain, possibly due to the close vicinity of the OM to the brain and the intrinsic vulnerability of the olfactory system to pathological hallmarks of AD. Interestingly, several studies support specifically NRF2-based treatment options for AD [[73], [74], [75], [76], [77]] - in the case of OM, this treatment could potentially be administered intranasally. It has been demonstrated that hsa-miR-142-5p, when overexpressed in SH-SY5Y cells, downregulates the NRF2 protein, and reduces endogenous expression of the 3′ UTR mRNA of NRF2 and NRF2 transactivation. Accordingly, NQO1 expression was reduced in cells overexpressing hsa-miR-142-5p, and furthermore, exogenous hsa-miR-142-5p expression led to a reduction of GCLC, and a reduced ratio of GSH over GSSG [78]. We also identified significantly increased levels of hsa-miR-142-5p in primary cultures of AD OM cells. Our results combined with previous literature indicate that upregulation of hsa-miR-142-5p could be related to an impairment of the redox homeostasis in AD OM, as we report also other similar findings as previously observed by Narasimhan et al. [78]. Additionally, the overrepresentation analysis of the DEPs demonstrated altered RNA metabolism and processing in AD OM mucosa biopsies. Several of other miRNAs altered both in MCI and AD OM cells are also altered in other AD-derived biosamples such as blood or CSF. However, the direction of the change tends to be the opposite in the OM compared to other tissues. The associations between the miRNA expression in periphery and brain remain poorly understood in AD, partly hindered by the availability of brain tissue samples from living individuals and the likelihood of regional dependence of the miRNA periphery-brain relationships. Previous findings with cells derived from individuals with PD and the results of this study suggest that altered NRF2 signaling in the OM might be another common feature of these neurodegenerative diseases. Human olfactory neurosphere-derived (ONS) cells from individuals with sporadic PD display metabolic deficiencies, such as decreases in the content of reduced GSH and altered mitochondrial metabolism, when compared to cells derived from control subjects. Notably, these are reported to be unaltered in ONS cells derived from individuals with schizophrenia [39,40] and silencing of NRF2 recapitulated those metabolic alterations also in the ONS cells derived from control subjects [40]. As olfactory impairment is also characteristic of other neurodegenerative disorders, a combination of several biological measures might be needed to find more disease-specific differences. Although the data presented here gives insights into alterations in the OM in living individuals with MCI and AD, it should be noted that this study is associated with some limitations. Firstly, we assessed alterations in odor identification, but there exist also odor threshold tests and methods to evaluate differences in odor discrimination. A reduced identification rate for the banana and lemon in UPSIT have been reported also previously for cognitively healthy participants of The Harvard Aging Brain Study [62], indicating some limitations in these odors for distinguishing disease-related phenotypes. Furthermore, it needs to be further studied whether different combinations of the observed odor identification impairments are unique to preclinical AD, or more commonly neurodegeneration related. Odor identification tasks with spontaneous odor naming are considered more cognitively demanding and could further aid future risk stratification of non-demented individuals when combined with cued olfactory identification tests [79]. Secondly, in this study the number of biopsies and down-stream experiments with them were limited to a smaller cohort of individuals compared to the initial odor identification testing. Even though the biopsies were all taken by one ENT surgeon, sample-to-sample and individual variation in the proportions of OM and respiratory mucosa, both consisting of several types of cells, cannot be totally avoided. Furthermore, it should be noted that the area of OM is subject to an age-related decrease and increase in proportions of mixed epithelium upon aging [80]. In our study, all the biopsied study participants were elderly individuals between 61 and 81 years of age, aiming to control the possible variation in the tissue material due to age. Thirdly, the sample size used for molecular profiling (N = 6 per group), which was constrained by biopsy availability and quality requirements, potentially leads to limited statistical power of the proteomics analysis and increasing the potential for variability. Consequently, the DEPs and miRNA in AD and MCI are reported based on their nominal P values < 0.05 and the set cut-off for FC difference. Differences based on nominal P value < 0.05 has been also used in for example previous meta-analysis of miRNAs and compilation of proteomic studies of brain tissue in AD [36,37]. Despite this low observed power, functional enrichment analysis revealed significantly overrepresented GO categories related to mitochondria and respiration, supporting the existence of real signal within the data. Therefore, these omics findings should be interpreted as exploratory and hypothesis-generating. Future validation in larger, independent, multicenter cohorts is required to confirm the robustness of the specific alterations observed here. Studies with nasal and OM exfoliates or brushings are emerging [64,81,82]. However, to our knowledge the proteomic studies of the AD OM biopsies, also accessing the lamina propria and supporting structures of the neuroepithelial layer, have not been performed before. The lamina propria included in the OM biopsies contain for example the axons of the olfactory sensory neurons and has been demonstrated to contain tau pathology in AD [9,83,84]. As the tissue is heterogenous by nature, it would be optimal to assess the differences in single-cell resolution across different cell types with newly developing techniques such as single-cell proteomics. The main findings with AD OM samples were similar regardless of whether the samples were studied as tissue samples or primary cell cultures. We acknowledge that replication of the studies with a larger cohort of individuals will be important along with future studies of cell-type specific untargeted and targeted proteomic differences in MCI and AD OM. Future studies should also assess the levels of oxidative damage to the genomic and mitochondrial DNA and lipids. It is also important to consider that given the location of the OM in the human body, it is in direct contact with the environment. Therefore, it is plausible that inhaled environmental pollutants coming in contact with the tissue trigger increases in free radicals at the OM. Previous studies showing immunohistochemical stainings on the biopsied OM suggest increased levels of protein adduct of lipid peroxidation, namely HNE—pyrrole, and heme oxygenase-1 in the various cell types in AD cases [16]. Similar alterations referring to oxidative damage have been reported also in the primary cultures of olfactory neuroblasts derived from biopsied tissue, highlighting that these changes also persist the culturing process of the cells [85]. In summary, the results of this study highlight common odor-specific identification impairments in MCI and AD, and energy metabolism and oxidative stress -related pathways in the human OM. Gathering knowledge of the proteomic alterations in the OM in AD could aid in understanding the early pathological processes of AD. Furthermore, odor-specific identification difficulties in cognitively impaired individuals in combination with other more traditional biological fluid biomarkers could be utilized identify and stratify individuals with preclinical AD from aged, cognitively healthy individuals. CRediT authorship contribution statement Riikka Lampinen: Writing – review & editing, Writing – original draft, Visualization, Methodology, Investigation, Conceptualization. Alexey M. Afonin: Writing – review & editing, Visualization, Formal analysis. Laura Mussalo: Writing – review & editing, Visualization, Investigation. Simone Avesani: Writing – review & editing, Formal analysis. Tana Zavodna: Writing – review & editing, Investigation. Emily Feneberg: Writing – review & editing, Resources, Investigation. Zdenek Krejcik: Writing – review & editing, Investigation. Jan Topinka: Resources, Funding acquisition. Sweelin Chew: Writing – review & editing, Investigation. Heikki Löppönen: Resources. Tarja Malm: Resources. Rosalba Giugno: Supervision, Resources. Anthony R. White: Writing – review & editing, Conceptualization. Anne M. Koivisto: Supervision, Resources, Data curation. Elina Penttilä: Writing – review & editing, Resources, Methodology, Investigation. Katja M. Kanninen: Writing – review & editing, Supervision, Project administration, Funding acquisition, Conceptualization. Ethics declaration Written informed consent to take part in the study and to publish the article has been obtained from all participants or their legal representatives. The privacy rights of participants have been observed. This study included organ or tissue donors. This study includes human biological material and consent was obtained by donors, or their next of kin or legal representatives, for use in this study and for publication of the article. The samples used in this research were not sourced from executed prisoners or prisoners of conscience. This study was performed in compliance with relevant laws, regulatory frameworks and guidelines where the research took place. This study was approved by the Research Ethics Committee of the Northern Savo Hospital District. (Approval No. 536/2017) Consent statement Informed consent was obtained from all subjects involved in the study. Funding This publication is connected to work from COST Action CA20121, BenBedPhar, supported by COST (European Cooperation in Science and Technology). Furthermore, this study was financially supported by the Sigrid Jusélius Foundation, Emil Aaltonen Foundation, and Finnish Cultural Foundation, North Savo Regional fund. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements The authors would like to thank the team in Turku Proteomics Facility supported by Biocenter Finland for the mass spectrometry analyses, Ms. Mirka Tikkanen, Ms Donya Behzadpour, Ms Sonja Mykkänen and Ms. Veronika Górová for laboratory assistance, Ms. Pirkko Kanerva for assistance with the clinical work, Ms. Ylli Törn and Ms. Hanna Härkönen for the APOE genotyping, and NADMED laboratory (Helsinki, Finland) for the redox-profiling of the cells. Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.redox.2026.104393. Supplementary data The following are the Supplementary data to this article. Multimedia component 1 Multimedia component 2 Data availability The data is not publicly available due to the restrictions of informed consent. The data can be shared for non-commercial purposes upon a reasonable request to the corresponding author. References Associated Data Supplementary Materials Multimedia component 1 Multimedia component 2 Data Availability Statement The data is not publicly available due to the restrictions of informed consent. The data can be shared for non-commercial purposes upon a reasonable request to the corresponding author.