Reduced olfactory performance is associated with changed microbial diversity, oralization, and accumulation of dead biomaterial in the nasal olfactory area

ABSTRACT The partial or complete loss of the sense of smell, which affects about 20% of the population, impairs the quality of life in many ways. Dysosmia and anosmia are mainly caused by aging, trauma, infections, or even neurodegenerative disease. Recently, the olfactory area—a site containing the olfactory receptor cells responsible for odor perception—was shown to harbor a complex microbiome that reflects the state of olfactory function. This initially observed correlation between microbiome composition and olfactory performance needed to be confirmed using a larger study cohort and additional analyses. A total of 120 participants (middle-aged, no neurodegenerative disease) were enrolled in the study to further analyze the microbial role in human olfactory function. Olfactory performance was assessed using the Sniffin’ Stick battery, and participants were grouped accordingly (normosmia: n = 93, dysosmia: n = 27). The olfactory microbiome was analyzed by 16S rRNA gene amplicon sequencing and supplemented by metatranscriptomics in a subset (Nose 2.0). Propidium monoazide (PMA) treatment was performed to distinguish between intact and non-intact microbiome components. The gastrointestinal microbiome of these participants was also characterized by amplicon sequencing and metabolomics and then correlated with food intake. Our results confirm that normosmics and dysosmics indeed possess a distinguishable olfactory microbiome. Alpha diversity (i.e., richness) was significantly increased in dysosmics, reflected by an increase in the number of specific taxa (e.g., Rickettsia, Spiroplasma, and Brachybacterium). Lower olfactory performance was associated with microbial signatures from the oral cavity and periodontitis (Fusobacterium, Porphyromonas, and Selenomonas). However, PMA treatment revealed a higher accumulation of dead microbial material in dysosmic subjects. The gastrointestinal microbiome partially overlapped with the nasal microbiome but did not show substantial variation with respect to olfactory performance, although the diet of dysosmic individuals was shifted toward a higher meat intake. Dysosmia is associated with a higher burden of dead microbial material in the olfactory area, indicating an impaired clearance mechanism. As the microbial community of dysosmics (hyposmics and anosmics) appears to be influenced by the oral microbiome, further studies should investigate the microbial oral-nasal interplay in individuals with partial or complete olfactory loss. IMPORTANCE The loss of the sense of smell is an incisive event that is becoming increasingly common in today’s world due to infections such as COVID-19. Although this loss usually recovers a few weeks after infection, in some cases, it becomes permanent—why is yet to be answered. Since this condition often represents a psychological burden in the long term, there is a need for therapeutic approaches. However, treatment options are limited or even not existing. Understanding the role of the microbiome in the impairment of olfaction may enable the prediction of olfactory disorders and/or could serve as a possible target for therapeutic interventions.

wanted to determine the potential influence of dead microbial material on olfactory function and discover whether this dead material originates from the gut.Furthermore, FIG 1 Olfactory mucosa-sampling area of interest located at the ceiling of the nasal cavity.(A) The olfactory microbiome is embedded in the mucus of the olfactory epithelium, including the olfactory receptor cells.The axons of these receptor cells reach through the cribriform plate into the olfactory bulb in the brain [image sources: nasal cavity and olfactory bulb-Servier Medical Art (19,20), microbes: freepik (21); images modified with Inkscape (22)].(B) Scanning electron micrographs (SEM) of a normosmic nasal sample: 1, a nasal epithelial cell; 2 a ciliated nasal epithelial cell; 3, cilia; 4, nasal phlegm/mucus; 5, bacteria-like structures; 6, nasal phlegm; 7, squamous epithelial lining; 8, squamous epithelial cell covered with mucus; 9, squamous nasal epithelial cells with bacteria-like structures; yellow arrow, cilia; green arrow, bacteria-like structures; blue arrow, nasal phlegm.
we were also interested in measuring the effect of a putative changed diet on the gastrointestinal microbiome, due to the reported changes in the diet of dysosmics.
Indeed, dysosmics tend to have a higher alpha diversity with a higher fraction of microbial material as compared with normosmics.However, the gut has not been identified as the source of these dead microbial signatures.Only Methanobrevibacter signatures belonging to a similar clade as fecal references have been found to be more prominent in dysosmics.In general, the diet of normosmics seems to be healthier than that of dysosmics.

Design and subjects of the study
To replicate our pilot study (13), we recruited an additional 53 healthy volunteers aged 18-45 years.Great efforts were made to allow for comparability between the pilot study and Nose 2.0.Participants of both studies had to meet the following criteria: non-smoker, absence of psychiatric, neurological, or internal medicine conditions, use of probiotics and antibiotics within the month prior to sampling, no presence of nasal polyps, no use of nasal spray on the day of the study, and an absence of allergies (including acute hay fever or pollen) or acute illness (e.g., rhinitis).The only addition was the exclusion of participants with gastrointestinal disorders in Nose 2.0.Furthermore, no participant of the pilot study was included in the Nose 2.0 study.Because of the susceptibility of swab samples from the nasal olfactory region to contamination from other sites, we used sinus secretion collectors in Nose 2.0.While only 16S rRNA gene sequencing was performed in the pilot study for the microbiome analyses, the nose samples collected in Nose 2.0 underwent additional metagenomics and metatranscrip tomics analyses.Furthermore, stool samples were collected for 16S rRNA gene-based amplicon sequencing and metabolomics.General metadata information (e.g., sex, age, and BMI) was collected in both studies (Table S1).In addition, the participants completed questionnaires in order to collect dietary as well as cognitive and emotional information (LEIDEN-R, SPM, and FAIR; no significant differences observed; data not included herein) and underwent olfactory testing (Sniffin' Sticks Battery).For study schemata, see also Fig. S1.

Olfactory testing and olfactory group definition
The olfactory function of the study participants was investigated using the Sniffin' Sticks battery (Burghart Instruments, Wedel, Germany).Detailed information about this method can be found in reference (27).Briefly, a combined TDI score was defined by adding three different odor scores: the detection threshold (lowest odor concentration that can be detected), odor discrimination (ability to discriminate between odors), and odor identification (ability to name odors) (27).

Dietary questionnaires
All participants had to complete the "German Food Frequency Questionnaires"' (Robert Koch Institute (29), to evaluate nutritional intake information.The responses were analyzed using the Austrian specific nutrition software (30).Values of nutritional intake were normalized based on the kcal intake (nutritional value per 100 kcal food intake) before being correlated with the olfactory performance measures (TDI scores and olfactory groups; Table S14) and the microbiome (nose and stool).

Nasal samples
Samples were collected by ENT physicians at the University Department of Otolaryng ology, Medical University of Graz, Austria, using a sinus secretion collector (Medtronic Xomed Inc.).Samples were immediately placed on ice until further processing.Nasal samples were dissolved in 500 µL of an aqueous 0.9% (wt/vol) NaCl solution and divided between two tubes.One aliquot (250 µL) was treated with propidium monoacid (PMA).PMA is used to mask freely accessible DNA for subsequent PCR reactions and thus allows one to gain a reliable insight into the intact/viable fraction of microbiomes (31).The PMA treatment was performed in the dark.Samples were mixed with the PMA solution (final concentration: 50 µM), shaken briefly, and then incubated on a shaker for 10 minutes in a PMA-Lite LED photolysis instrument (Biotum).Subsequently, both aliquots (PMA treated and untreated) were stored at −20°C.

Stool samples
The stool samples were collected by the participants themselves on the same day that the nasal samples were collected.Samples were stored on ice until further processing.Within a maximum of 2 h, the samples were aliquoted: 0.1 g of stool was dissolved in 1 mL of an aqueous 0.9% NaCl solution and an aliquot of 500 µL was treated with PMA solution as previously described.Again, both aliquots (PMA treated and untreated) were stored at −20°C.

DNA and RNA extraction
DNA from the nasal (PMA-treated and untreated) and stool (PMA-treated) samples was extracted using the DNeasy PowerSoil Kit (QIAGEN, USA).Untreated nasal samples which were frozen in liquid nitrogen immediately after sample collection were used for RNA extraction (RNeasy RNA Extraction Kit, QIAGEN, USA).DNA and RNA extractions were performed according to the manufacturer's protocol with minor modifications: instead of vortexing the samples, a MagNaLyser was used at 6,500 rpm for two cycles of 30 seconds.The nucleic acid concentration obtained was quantified using the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific, USA) before samples were stored at −20 (DNA) and −80°C (RNA).

16S rRNA gene and ITS region-based amplicon sequencing
Three different approaches were used for amplicon sequencing: the universal (primer pair: 515F-806R) and the archaeal approaches (nested PCR: 344F-1041R, 519F-806R) were used to amplify the V4 region of the 16S rRNA gene and the fungal approach (primer pair: ITS86F-TS4R) was used to amplify the ITS region.PCR was performed on DNA extracted from stool samples using all approaches, whereas only the universal approach was used to amplify DNA extracted from nasal samples [see (32,33) for detailed information on protocol and primers].PCR products were sequenced using the Illumina MiSeq sequencing platform (Illumina, Eindhoven, The Netherlands) at the Core Facility for Molecular Biology of the Center for Medical Research in Graz, Austria (34).

Sequence data procession and control
Raw sequences (paired-end reads) were quality filtered and processed using QIIME 2 (Quantitative Insights Into Microbial Ecology) v2020.8.DADA2 (Divisive Amplicon Denoising Algorithm) in QIIME 2 was used to denoise the data before sequences were annotated using the reference databases SILVA v138 (universal approach), SILVA v138 (archaeal approach), and Unite v8.3 (fungal approach) as a Naïve Bayes classifier (detailed information can also be found here: Supplemental text) (35)(36)(37)(38)(39).After data processing, the R package decontam with default settings, a threshold of 0.5, and the prevalence method were used to treat extraction blanks and PCR negative controls (40) (https:// github.com/benjjneb/decontam).In addition to the controls, mitochondrial signatures, unassigned sequences (and also human mitochondrial assigned ones), and features with zero or only one read were removed.
SRS normalization (SRS = scaling with ranked subsampling) (41) was then applied to the nasal samples with sample depths of 500 and 1006 (PMA-treated and untreated, respectively) and to the stool samples (bacterial approach only) with a cut-off at 2,222 read counts.The archaeal and fungal approaches provided read counts that were too low for some samples; therefore, these were not normalized.The remaining ASVs (Data set S1-S5) were visualized using various tools [including Microbiome Explorer v1.6.0 (42), Rstudio v4.1.2(43), and RawGraphs (44)].The colors and legends of charts and graphs were adjusted using Inkscape (22).

Metatranscriptome sequencing
RNA (without PMA treatment) extracted from 12 samples (6 normosmics and 6 anosmics) was sent to a company (Macrogen, South Korea) for metatranscriptomics sequencing.The TruSeq Nano DNA Construction Kit (Illumina, Eindhoven, The Netherlands) was used to construct the library before sequencing with the Illumina NovaSeq 6000 technique (Illumina, Eindhoven, The Netherlands).
Results from applying the gene-centered approach were filtered by removing all traits with fewer than three reads.To examine changes in all life domains, the following additional filters were set: filtering all (i) non-bacterial, (ii) non-archaeal, (iii) non-fungal, or (iv) non-viral signatures (Data set S6 to S9).

Metabolomic analysis
Metabolomic measurements of PMA-untreated stool samples were performed at the Gottfried Schatz Research Center for Cell Signaling, Metabolism and Aging, Molecular Biology and Biochemistry, Medical University of Graz.A subset of 49 stool samples of the Nose 2.0 study was analyzed using untargeted NMR (nuclear magnetic resonance) spectroscopy for several metabolites in-house.Metabolites were extracted using a methanol/water solution and prepared for NMR measurements as described previously (55).NMR was performed on an AVANCE Neo Bruker Ultrashield 600-MHz spectrometer equipped with a TXI probe head at 310 K. Briefly, the 1D CPMG (Carr-Purcell-Meiboom-Gill) pulse sequence (cpmgpr1d, 512 scans, 73,728 points in F1, 11,904.76Hz spectral width, 512 transients, and recycle delays of 4 s) with water suppression by pre-saturation was used for 1 H 1D NMR experiments.Bruker Topspin version 4.0.2 was used for NMR data acquisition.Spectra for all samples were automatically processed (exponential line broadening of 0.3 Hz), phased, and referenced using TSP at 0.0 ppm, using the Bruker Topspin 4.0.2software (Bruker GmbH, Rheinstetten, Germany).Spectra were imported to Matlab2014b, and the regions around the water, TSP, and remaining methanol signals were excluded.To correct for the sample metabolite dilution, a probabilistic quotient normalization step was performed.Metabolite quantification was based on the signal integration of normalized spectra as described previously (56).

Statistical analysis
IBM SPSS Statistics v26 (58) was used to perform statistical analyses (P-values).Param eters were tested for normal distribution before selecting the appropriate statistical test.In general, uncorrected significant values are reported as P-values (P < 0.05) and Bonferroni-corrected values as q-values (q < 0.05) in the manuscript.

Source tracking
Source tracking was performed using sourceTracker2 (https://github.com/biota/sourcetracker2).The original version of SourceTracker was described in Knights et al. (65).For source tracking, only the data sets from Nose 2.0 were used (reads from PMA-treated stool samples and non-PMA-treated nasal samples from both olfactory groups).Reads from respective data sets were co-processed with QIIME 2 as described above.Stool samples were assigned as the "source, " whereas nasal samples served as the "sink." The selected rarefaction depth for source tracking was 480 (for both sink and source), to allow all samples to be kept in the analysis.

Methanobrevibacter tree
For the tree, all Methanobrevibacter sequences from the nasal samples (Nose 2.0) were used.Additional sequences (e.g., representatives of M. oralis and M. smithii), as well as environmental sources (rumen), were retrieved from SILVA (38).Environmental sequen ces were used to expand the tree.All archaeal sequences were subjected to the Silva Alignment, Classification, and Tree Service Tool (ACT (38)).The tree was computed using the "Compute tree" function in SILVA ACT with standard settings (no positional variability filter).The tree was then annotated using iTOL (66), and legends were added using Inkscape (22).

Description of the study and the study cohorts
A new study (Nose 2.0) was conducted to add to the results of our published pilot study (13); Fig. S1) which indicated the existence of a potential correlation between the nasal microbiome composition and olfactory function.In the following, the new study and the pilot study are referred to as "Nose 2.0" and "pilot, " respectively.In the new study, we recruited an additional 53 participants (22 women) and analyzed samples collected from these participants together with the already-collected 67 nasal samples (50 women) from the pilot study.
The observed different olfactory performance in both cohorts could not be explained by any observed metadata, such as age, BMI, or sex (Table S2).The homogeneity of the cohorts was further supported by the results of analyses of the metadata information based on the olfactory groups (major and subgroups) (Table S3).
Participants with dysosmia were selected based on the loss of smell they had experienced due to an infection that had occurred at least 1 year earlier.Based on inflammatory markers in the blood, no active inflammation was detected in any subject (Table S1).Notably, samples were collected before the COVID-19 pandemic-from April 2018 to November 2019.

Diversity of nasal microbial community differs based on olfactory perform ance
To add to the collection of already-published data (13), the raw sequencing data from the pilot study had to be re-analyzed together with the Nose 2.0 data before combin ing/comparing the data sets (see Material and Methods).From both studies (combined data of both studies), an overall of 2,480 unique microbial features (120,720 reads) were obtained by 16S rRNA gene sequencing the "olfactory microbiome." Notably, archaeal information made up approximately 2% of the data set (see data set 1).
The predominant taxonomic signatures and microbial composition were similar in both studies, with Proteobacteria, Actinobacteriota, Firmicutes, Bacteroidota, and Euryarchaeota representing the predominant phyla and Corynebacterium, Ralstonia, Staphylococcus, Lawsonella, and Dolosigranulum representing the most abundant genera (Table S4; Fig. 2A).However, only 243 features were shared between the studies, while 1,529 and 708 ASVs (amplicon sequence variants) were found to be unique for the pilot and Nose 2.0 data sets, respectively (Fig. 2B; Fig. S3; detailed list in Table S5).
Consistent trends in the alpha diversity of both cohorts (Nose 2.0 and pilot) were observed.The increased Shannon index of the dysosmics in both cohorts (compared to normosmics) was driven by the evenness; however, these differences were not significant.The increase in microbial richness (Chao1 index) based on the TDI score was significant.This result was supported by the trend observed in the Shannon index (Fig. 3A; P Nose 2.0_TDI_richness = 0.022; P pilot_TDI_richness = 0.017; Mann-Whitney U).
The differences observed in alpha diversity were even more pronounced if the main groups were further divided into olfactory subgroups (good and weak normosmia, hyposmia, and anosmia), showing a gradual/stepwise increase as the olfactory percep tion decreased; again, this result may have been mainly driven by the data richness (Chao1) rather than evenness (Fig. S4; not significant).
The microbial community information indicated a partial overlap of the clusters of both cohorts in a PCA plot (not significant).The Nose 2.0 cohort revealed a significant difference between normosmics and dysosmics; however, this was not reflected in the pilot cohort (Fig. 3B; P NBA = 0.002; P pilot = 0.505 TDI; Adonis test).
Similar trends in both cohorts that were found included an increased alpha diversity, indicating more microbial signatures in dysosmic than in normosmic participants with additional fluctuating beta diversity seen in dysosmics.These differences are possibly explained by the presence of biomarkers and especially for dysosmics.Dysosmics were characterized by an increase in the relative abundance of gut-and oral-associated microbes (e.g., Methanobrevibacter, Fusobacterium, Porphyromonas, and Selenomonas (33,37)] and mainly intracellular living microbes [e.g., Rickettsia and Spiroplasma (67,68)].
These results raise the question of whether the observed microbial signatures belong to viable and actual inhabitants of the olfactory area or whether these signatures reach this area through respiration-these questions will be addressed later in the manuscript.

The olfactory area of dysosmics contains many signatures of dead cells
In order to approach the mechanistic question of the microbiome in the olfactory area in normosmics and dysosmics, we concentrated our analyses on the viable [propidium monoazide (PMA)-treated] and active (metatranscriptomics) fractions of the microbial community.
PMA is used to mask freely accessible DNA for subsequent PCR reactions and thus allows one to gain a reliable insight into the intact/viable fraction of microbiomes (31).Many of the abovementioned key findings were still detected after PMA treatment, such as the proportion of archaea (2%) and the most abundant phyla and genera (including the phyla Proteobacteria, Firmicutes, and Actinobacteria and the genera Ralstonia, Corynebacterium, Staphylococcus, and Corynebacterium), where most showed similar trends for olfactory groups (Table S7; Data set S2).However, the trends seen for Bacteroi dota, Euryarchaeota, Ralstonia, and Dolosigranulum were exactly the opposite of those observed in the PMA-untreated samples (Fig. S6A).
The analysis of the PMA data set revealed opposite trends for microbial alpha diversity measures as compared with the untreated data set.In untreated samples, as explained above, we observed an increased Shannon index, richness (Chao1), and evenness that correlated with dysosmia; the PMA-treated samples showed decreases in terms of all measured alpha diversity analyses from good to bad olfactory performance.However, the results [Shannon index and richness (Chao1)] were not significant except for microbial evenness (P = 0.05; Fig. S6A).In contrast to the previous results on the overall microbial community, the signatures of the viable community did not show significant differences in terms of their beta diversity (Fig. S6A).An analysis of microbial biomarkers for olfactory performance did not reveal significant differences between the olfactory groups using ALDEx2.
Comparing the PMA and non-PMA data set at the genus level, we revealed that 121 unique genera are shared between the data sets (PMA-174 unique genera; non-PMA-316 unique genera).The read counts belonging to normosmia that are shared between both data sets made up 50% and 53%, respectively, whereas the proportion of normos mia and dysosmia in the non-PMA data set was lower in the non-PMA (0.32 or 24% normosmia) than in the PMA (0.85 or 46% normosmia) data set (Fig. S7).Two of the five genera found in the dysosmics, namely, Rickettsia and Spiroplasma, were found only in non-PMA samples (Fig. S7; Table S8).
All of these results lead us to conclude that dysosmics carry an increased proportion of dead microorganisms as compared with normosmics.To further evaluate the quantity and quality of the dead microbial material, 12 nasal samples (6 normosmia and 6 dysosmia) were selected for further metatranscriptomics analysis.
The information derived from PMA-treated analyses was further confirmed by our metatranscriptomic analysis (Table S9).Again, the alpha diversity showed the same opposite trend as seen in the PMA-treated as compared with untreated samples, and FIG 4 Significantly different taxa found based on olfactory main groups and TDI scores.(A) The data were CLR transformed before ALDEx2 was performed.No significant differences were observed in the pilot cohort.For detailed information, see also Table S6.(B) Several taxa were positively and negatively correlated with TDI scores (Spearman's rho correlation; all P-values were not adjusted).
the core microbiome at the phylum level was similar to that detected using the other methods (Fig. S6B).
No feature that was identified as a biomarker for dysosmics in the amplicon data set from the Nose 2.0 study could be found in the metatranscriptomics data set (Fig. 4; Fig. S5; Amplicon, Data set S1-S6).This is likely due to the difficult nature of metatran scriptomics analysis of nasal samples, small sample size, and many unannotated taxa.However, the phylum Actinobacteriota (P = 0.045) and the genus Corynebacterium (P > 0.05) were found to be indicative for normosmics in the metatranscriptomics output (DeSeq; ALDEx2, data not shown).An analysis of information about the non-bacterial taxonomy (fungi, archaea, and viruses; Data set 8 and 9) and functional capacity did not reveal statistically significant differences between the olfactory groups (Fig. S8; Table S10; Data set 7).
Both the metatranscriptomics and PMA-based 16S rRNA gene analyses revealed a decrease in alpha-diversity measures that correlated with dysosmia.As this result contrasts with the observations made in the classical microbiome analysis (non-PMAtreated samples), we conclude that the olfactory area of dysosmics contains a higher proportion of signatures of dead/non-intact cells as compared with that of normosmics.

Signatures of dead microbes in the nasal cavity do not come from the gut
An increased relative abundance of butyrate producers in the nasal samples (usually found in the gastrointestinal tract) of dysosmics as compared with normosmics was already indicated by the results of our pilot study (13).These results suggest that dead microbial material is transferred from the gut to the nasal cavity.To determine whether microbes stray from the intestine into the nose, we also investigated the viable fecal microbiome (PMA) of all participants.
Furthermore, no significant correlations between the key taxa of nasal and stool samples were found (Fig. S10), and source tracking (at ASV and genus levels) did not reveal a substantial overlap in the overall microbial signatures found in gut and nasal samples (0.49% for anosmics, 0.45% for normosmics; genus level).
For completeness, we also analyzed the information on the fungi and archaea from the stool data set (Data set S4 and S5).The fungal communities in both groups consisted mainly of food-borne fungi, including Saccharomyces (bread), Penicillium (cheese and meat), and Debaryomyces (production of vitamin B2 found on food) (Fig. S11).Signatures of Debaryomyces (P = 0.05) and Scopulariopsis (P = 0.03; infections including sinusitis) were more common in dysosmics.Almost all archaeal signatures found (32 out of 45 features) belonged to the genus Methanobrevibacter (Fig. S11).Due to the nested archaeal PCR approach taken, no differential abundance analysis of the olfactory groups could be performed.
One notable result is that Methanobrevibacter signatures were found to be the most abundant archaea in the nasal (PMA and non PMA) and stool samples (universal and archaeal approach) in our study.Methanobrevibacter are widely distributed in the human aero-digestive tract.In particular, the archaeal signatures from anosmics clustered with signatures found in fecal samples, indicating a potential overlap (Fig. S12).

Dysosmics tend to have a more heavily meat-based diet than normosmics
Although one might assume that dietary habits would vary based on the different olfactory performance (i.e., the sense of smell is strongly linked to the sense of taste), we only found tendencies in this regard.The evident trends calculated based on TDI scores (P < 0.05; Spearman's rho) included mainly the food categories meat (P = 0.002; ρ = −0.431)and fish (P = 0.018; ρ = −0.339),which were eaten more frequently in dysosmics, whereas the intake of fruits (P = 0.012; ρ = 0.359) and legumes (P = 0.012; ρ = 0.359) was higher in normosmics.When considering the microbiome aspect, only the processed meat intake (dysosmics; P = 0.025; ρ = −0.315)was significantly correla ted with the fungal community of the dysosmics' stool samples ( Fig. 13A, B) and the meat consumption, with the nasal microbiome of dysosmics (P = 0.018; non PMA, Fig. 13C, D).Several trends were also observed for nutrients (e.g., fatty acids, vitamins, and carbohydrates), but only the essential omega-3 fatty acid ⍺-linolenic acid (q = 0.03; ρ = 0.495) had a significantly higher correlation with olfactory function than dysfunction after P-value correction (Tables S14).Interestingly, omega-3 fatty acids, in general, have been associated with an improvement in olfactory function in other studies (69,70).All in all, normosmic participants seem to eat "healthier" (more fruits, less meat) than the dysosmics.
Even though we detected trends, the changes in the diet could not explain the differences observed in the microbiome composition.However, in general, it seems as though normosmics are more likely to follow a vegetarian diet.

DISCUSSION
The full spectrum of the physiological effects of olfactory loss is still largely unknown, and, according to our findings, the olfactory microbiome is also affected or even involved in this devastating condition.By carrying out two independent studies, we could show that the microbiome composition varies with olfactory capacity.This was found to be mainly associated with an increased diversity in the dead/non-functional microbial cells at the olfactory epithelium.The source of these dead/non-functioning microbial cells has not been fully elucidated, but we found evidence that the oral cavity, rather than the gut, is a possible source.
The overall composition of the microbial community in the olfactory area varied as a function of olfactory performance.The same trends were observed in this study (Nose 2.0) and the pilot study, indicating reproducible results.Although the olfactory groups overlapped in terms of their microbial composition, the composition in dysosmics was more scattered with respect to beta diversity, suggesting that the microbial composition was not as stable as in the normosmics.Consistently, the analysis of the alpha diversity also revealed higher microbial richness (Chao1) (more different taxa) in samples with reduced olfactory performance.
In contrast to our results, Biswas et al. (2020) reported a loss of diversity in individ uals with poor olfactory function; however, in this case, the authors investigated the microbiome of the sinuses, and thus, the findings are not fully comparable (23).Other diseases of the upper respiratory tract have also been reported to be associated with lower levels of microbial diversity [e.g., chronic Rhinosinusitis (71) and cystic fibrosis (72)].Only allergic rhinitis has displayed the same trend as we found in our study, with higher levels of microbial diversity observed in diseased as compared with healthy subjects (73).
Interestingly, analyses of the data on viable/intact cells in the nasal samples revealed a nearly opposite trend, suggesting that dysosmics carry a higher load of dead/nonfunctional microorganisms.This result was also obtained when taking two independent molecular approaches, namely, PMA-treatment and metatranscriptomics.This finding is further supported by the fact that several taxa were found in both PMA-treated and untreated nasal samples, respectively; overlaps were observed among the groups at a genus level but also among an enormous number of the signatures that were specific for PMA-untreated samples (62% of total non-PMA signatures).Of the genera found in the non-PMA-specific data set, only 24% belonged to the normosmic samples, while the normosmic proportion of the 121 genera found in both data sets (PMA and non-PMA) accounted for about 50%.
The excess number of ASVs might have been introduced from other body parts or the external environment via inhalation.However, the impact of the gut microbiome was clarified by source tracking, and the results indicate that the dead material originated from other sources, such as other nasal areas, skin, the oral cavity, or the environment.We could even find evidence that in particular, oral (or specifically subgingival) taxa such as Gemella, Capnocytophaga, Fusobacterium, and Porphyromonas were enriched in the olfactory area of dysosmics.Some of these have even been associated with periodontitis (e.g., Fusobacterium or Porphyromonas) in other studies (74).
Chemosensory conditions have been associated with oral conditions in earlier studies (75) and, in particular, with halitosis (e.g., Fusobacterium and Porphyromonas) that was associated with smell and taste disturbance (76).Thus, our findings might indi cate problematic dental conditions that would require additional and more thorough analyses.In any case, the higher load of dead/non-functional material might be due to a potentially reduced mucociliary clearance (i.e., where the dead material can no longer be transported toward the throat and/or microbes can migrate into the nasal cavity from lower body parts) in dysosmics.
Notably, dysosmics displayed a higher number of signatures for (dead/non-func tional) intracellular microbes (e.g., Spiroplasma and Rickettsia (67, 68)); the reason for this observation is unclear, and the higher number of signatures is not correlated with increased inflammation.Although the cause of dysosmia in our participants was an infection, no significant difference in inflammatory markers (glucose, CRP, and IL6-L1) between the olfactory groups was observed.This result suggests that the initial inflammation/disease had been overcome, and the long-term smell loss was not driven by an active inflammation.
Interestingly, we found a significant difference in the stool microbiome based on olfactory performance.This stepwise decrease in alpha diversity from low to high olfactory performance and the distinctive microbial composition (normosmia: Bacter oides, Colidextribacter, and Phascolarctobacterium; dysosmia: Alistipes, Oscillospiraceae UCG 002, and Ruminococcus torques group) might be explained by changes in the diet, favoring different kinds of microbes.Overall, the dysosmics' diet was characterized by a rather higher consumption of (processed) meat and fish as compared with the "healthier" diet of normosmics (i.e., more fruits and legumes).Furthermore, we found a decreased intake of the omega-3 fatty acid, alpha-linolenic acid, in dysosmics.Consistent with our findings, other studies could show an improvement in olfactory function as a result of omega-3 fatty acid supplementation (69,70).Since omega-3 fatty acids have neuroprotective and antioxidant effects, as well as boost the anti-inflammatory amino acid production, supplementation might serve as a putative therapy option for olfactory loss (77,78).

Study limitations
To draw appropriate conclusions about the nasal microbiome and olfactory performance in humans, the olfactory mucosa was the sampling site of choice.This is a tiny area (8-10 mm) on the ceiling of the nasal cavity.Hence, the small volume of the specimens obtained from this area presented a major challenge, which was tackled for the first time in this study.Major issues that needed to be overcome were to find a less-con tamination-prone, swab-less sampling procedure and a highly efficient DNA extraction method, both of which needed to be tested and established.However, the small amount of biomass retrieved still presented problems during the subsequent analyses.Thus, only 12 samples (6 from 33 normosmics plus 6 from 20 dysosmics) were suitable for RNA sequencing and did not offer the possibility to re-run and improve the analyses.Therefore, we cannot make any statements about the link between olfaction and the functional profile of the microbiome; we were restricted to drawing conclusions from the taxonomic information.
We are aware that the overall number of participants (in total n = 53) is still low; the recruiting process was difficult due to a discrepancy in the personal olfactory perception of the study participants and the outcome of the sniffing-stick-based olfactory measure ments.However, we were still able to include more participants with dysosmia than in our pilot study, which was included in order to run comparative analyses and to increase the total number of individuals analyzed.To minimize the study bias, the pilot cohort was completely re-evaluated along with the new cohort.
The study design did not include analyses of the aerobiome of the environment of the study participants.Therefore, no conclusions can be made about the aerobiome as the source of the dead microbial fraction.

Ongoing work
The results of our study show that the nasal microbiome differs according to olfactory performance and that the load of dead cells is significantly higher in dysosmics.Several independent studies have shown that the sense of smell can be improved by performing olfactory training (79)(80)(81), but potential effects on the nasal microbiome have never been investigated.In the future, we plan to investigate whether the smell training can be used to revert the "dysosmic" microbiome to a more favorable "normosmic" one (e.g., less dead material) in a longitudinal study using three time points [before (TP0), during (TP1), and after (TP3) smell training].Furthermore, we want to use the gained knowledge to combine two distinct fields of the human life sciences, namely, microbiome research and neuroimaging, to investigate whether a reorganization of the brain structure takes place and to increase predictive and therapeutic opportunities.In addition, future studies should also consider exposure to the aerobiome via inhalation as a potential source of the dead microbial fraction of the olfactory microbiome.

Conclusion
People suffering from a loss of smell display a higher burden of dead microbial material near their olfactory receptor cells.The upper respiratory tract is equipped with highly efficient transport mechanisms, the so-called mucociliary clearance defense mechanism [reviewed in reference (15)], which is responsible for the efficient removal of various kinds of material.This clearance process appears to be impaired in dysosmic subjects, which could lead to the increased migration and accumulation of microbes from other body sites or the external environment into the nasal passages.
We were able to exclude the gut as the source of these signatures but linked specific periodontitis-associated microbial signatures to dysosmia.Investigating the relationship between olfactory function, the oral microbiome, and the functionality of mucociliary clearance may help researchers understand the reason for the high burden of non-func tioning microbes in the olfactory mucosa of dysosmics.

FIG 2
FIG 2 Relative abundance of the most abundant taxa found in the study cohorts.The plots show the 5 and 15 most abundant phyla and genera, respectively, (A) shared in both studies and (B) in the Nose 2.0 and pilot studies.Information on the olfactory performance is based on the main groups.Circle sizes indicate the number of reads of the specific microbial taxa found in the different groups.

FIG 3
FIG 3 Microbial diversity differs between olfactory groups in both cohorts.(A) Higher alpha diversity (Shannon index) in dysosmic samples is driven by richness (Chao1) rather than evenness (P-values not corrected).(B) Clusters of nasal samples of normosmics and dysosmics overlap in a PCA plot (based on TDI).

FIG 5
FIG 5 Composition of stool microbiome differs based on olfactory performance.(A) Bubble plot showing the most abundant taxa at the phylum and genus levels.(B) Alpha diversity indices [Shannon index and richness (Chao1)] were higher in dysosmics as compared with normosmics, except for evenness.(C) No clustering was observed based on the olfactory main groups and subgroups.(D) The LEfSe analysis revealed that more taxa were associated with dysosmics than with normosmics.