Cyanoacrylate glue as an effective skin-decontamination method for hair follicle microbiome sampling: insights from a nonhuman model (Capreolus capreolus) Authors: Dita Gudra, Megija Lunge, Edmunds Skinderskis, Ance Roga, Silvestrs Rubins, Davids Fridmanis Published: 2026 ABSTRACT IntroductionIn humans, microorganisms within hair follicles differ from those on the skin surface and may contribute to autoimmune skin diseases such as alopecia areata and vitiligo. Owing to the low biomass of hair follicle microorganisms, precise sample acquisition and strategies to avoid contamination with genetic material are critical. To address this, we developed and evaluated a sampling methodology using a nonhuman wildlife proxy model – the wild roe deer (Capreolus capreolus) - to identify effective approaches for minimizing skin contamination prior to biopsy. Animal was obtained during licenced hunts and frozen at -20°C shortly postmortem.Four sets of biopsy and skin surface swab samples were collected from the right front leg: 1) untreated skin; 2) skin cleaned with isopropyl alcohol (IPA) wipes; 3) skin treated with cyanoacrylate glue (CAG); and 4) skin treated with CAG followed by IPA wipes. Swabs were collected via Copan FLOQSwabs, whereas biopsies were obtained with single-use 1 mm biopsy punches. Nucleic acids were subjected to droplet digital PCR (ddPCR) for 16S rRNA and 18S rRNA gene copy number quantification. Additionally, the swab samples were subjected to 16S rRNA V3-4 region and ITS-1 sequencing.ResultsddPCR analysis revealed greater bacterial presence on the skin surface (11,661±5,181 copies/±L) than fungal presence (87±35 copies/μL), and the microbial load was greater in the swab samples (16S: 11,661±5,181; 18S: 87±35 copies/μL) than in the biopsy samples (16S: 76±56; 18S: 8±5 copies/μL). Notably, the use of CAG reduced the microbial load in biopsy samples by 3.7-fold. 16S and ITS sequencing analysis revealed slightly greater alpha diversity in swab samples than in biopsy samples, particularly in areas treated with CAG. Beta diversity revealed distinct clustering of swab and biopsy samples, with shorter distances between CAG-treated samples and untreated samples. Fewer bacterial (n = 8) and fungal (n = 1) genera were detected in CAG-treated biopsies compared with untreated (n = 21, n = 9, resp.) and IPA-treated (n = 45, n = 7, resp.) biopsies.ConclusionsThese results demonstrate a high probability of biopsy contamination with microbial DNA originating from the skin surface. They also confirmed that CAG effectively reduces such contamination and can serve as a practical decontamination step before biopsy. Although based on a nonhuman wildlife model, this feasibility study provides methodological insights that can inform the design of future human hair follicle microbiome investigations.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12866-026-05268-1. Introduction In the past decade, human hair follicles have been recognized as environmentally protected niches for microbial colonization and are distinct from the surrounding skin surface [1, 2]. Studies have shown that hair follicles are colonized by a diverse range of microorganisms, including bacteria, fungi, viruses, and mites, with Staphylococcus, Cutibacterium, Streptococcus, and Malassezia identified as predominant residents [2, 3]. In contrast, the classic human skin microbiome is dominated by Cutibacterium, Corynebacterium, Micrococcus, Streptococcus, Staphylococcus, Acinetobacter, and Malassezia, although their relative abundance varies across different skin sites and types [4–6]. Increasing evidence indicates that the hair follicle microbiome plays a significant role in hair physiology, influencing hair growth, health, and regeneration. The microbial communities inhabiting hair follicles contribute to maintaining a balanced scalp environment, which is crucial for optimal follicular function and hair development [1, 7, 8]. These microbes exist in a symbiotic relationship with the host and help shape local immune responses [1, 8, 9]. However, disruptions in this equilibrium have been associated with pathological conditions. The follicular microbiome has subsequently been investigated in connection with dermatological disorders such as acne vulgaris [3, 10] and various forms of alopecia [11, 12]. Current methods for obtaining hair follicles for microbiome research primarily include laser-capture microdissection [2, 13], skin punch biopsy [14], depilation [15], cyanoacrylate glue [3], and surgical excision of follicles from skin biopsies [16]. However, standardized and contamination-aware sampling strategies for hair follicle microbiome studies remain limited. In most cases, these methods do not sufficiently address the issue of microbial contamination from the skin surface. To the best of our knowledge, only one study has evaluated the impact of different sampling techniques on the hair follicle microbiome by comparing pore strips and cyanoacrylate glue. While the authors reported no significant differences in microbial relative abundance between methods, they did observe marked compositional differences between the skin surface and follicular compartments [3]. Therefore, in this study, we aimed to address the current knowledge gap regarding the impact of different skin pretreatment methods prior to hair follicle acquisition via punch biopsies. We will evaluate the effects of these treatments by quantifying bacterial and fungal loads and performing targeted microbiome sequencing to identify the most effective sample collection strategy. The microbial load and community composition were compared across four sampling conditions: untreated skin, skin cleaned with isopropyl alcohol, skin treated with cyanoacrylate glue, and skin pretreated with cyanoacrylate glue followed by cleaning with isopropyl alcohol. For ethical reasons, the model organism used in this study was the wild roe deer (Capreolus capreolus). Although no information is currently available in the literature regarding the roe deer hair or hair follicle microbiome, this study therefore also provides the first insight into the composition of the wild roe deer hair and hair follicle microbiome. Several previous studies have shown that the human skin microbiome is distinct from that of animals and is significantly less diverse [17, 18]. Moreover, the composition of the skin microbiome differs substantially among animal species [17–19]. In humans, these differences can be largely attributed to personal hygiene practices, environmental exposure, clothing, ethnicity, and other factors associated with modern lifestyles [17, 20]. Despite the substantial differences between wild animal and human skin microbiomes, the roe deer provides a valuable and ethically feasible model for evaluating and improving hair biopsy sampling methodologies. Materials and methods Sampling material and study design To develop an optimal strategy for hair follicle acquisition, we aimed to use a nonhuman sample source, as employing human samples for such a pilot experiment poses various ethical challenges. Owing to the similarity of their skin and hair, our initial choice was the domestic pig (Sus scrofa domestica). However, Latvian local regulations mandate rigorous cleaning procedures for the bodies of slaughtered animals, aimed at hair removal and reducing the microbial load on the skin surface. Consequently, we selected a wild animal, the roe deer (Capreolus capreolus), which was obtained during licenced hunts and frozen at -20 °C shortly after death. Although the hair of the roe deer is thinner, its follicle density is greater, and its lifestyle significantly differs from that of humans – resulting in a microbiome that is not directly comparable to that of humans – it still serves as a suitable proxy for assessing skin surface sterilization procedures. All subsequent sampling was carried out on the forearm skin of the animal, covering the underlying muscles. We developed four strategies for sample acquisition (Fig. 1): (1) a biopsy taken from shaved, unsanitized skin as a reference (hereinafter referred to as the reference group); (2) a biopsy taken from shaved skin cleaned with an isopropyl alcohol (IPA) pad to mimic standard disinfection practices used in medical settings; (3) a biopsy taken from shaved skin cleaned with cyanoacrylate (super) glue (CAG) to remove the uppermost layer of the epidermis; and (4) a biopsy taken from shaved skin cleaned with CAG followed by IPA. To account for measurement variability, sample collection for each strategy was repeated at three different sampling sites. Additionally, to evaluate the microbiome composition of the doe’s hair, swab sampling was performed via FLOQSwabs (Copan, Italy), which were placed in eNAT preservation medium for nucleic acids (Copan, Italy) after sample collection. Biopsy samples were collected via disposable 1 mm biopsy punches (KAI Industries Co., Japan) and transferred to prelabelled 2.0 mL tubes filled with RNALater media (Thermo Fisher, USA). Fig. 1Study design to evaluate the optimal strategy for hair follicle acquisition. Abbreviations: IPA, isopropyl alcohol wipe; CAG, cyanoacrylate (super) glue Study design to evaluate the optimal strategy for hair follicle acquisition. Abbreviations: IPA, isopropyl alcohol wipe; CAG, cyanoacrylate (super) glue Extraction of nucleic acids Prior to extraction, swabs in eNAT medium were thoroughly vortexed to release cells from the swab surface. A preheated syringe needle was then used to puncture a hole at the bottom of the eNAT tube, which was then placed in a prelabelled 15 mL Falcon tube. The samples were subsequently centrifuged for 15 min at 4,000 rpm. After centrifugation, the supernatant was removed without disturbing the pellet, and 600 µL of Buffer RLT Plus (Qiagen, Germany) was added. The pellet was resuspended and transferred to a Lysing Matrix E tube (MP Biomedicals, USA). The samples were homogenized via a FastPrep-24 instrument (MP Biomedicals, USA) at a speed of 6.0 for 40 s. For the biopsy samples, centrifugation was performed at 14,100 rpm for 10 min. After centrifugation, the RNALater was subsequently removed, and 600 µL of Buffer RLT Plus was added. The sample was resuspended and transferred to a Lysing Matrix E tube. Homogenization was carried out as described for the swab samples. Following homogenization, DNA and RNA extraction from both the swab and biopsy samples was conducted simultaneously via the AllPrep DNA/RNA Mini Kit (Qiagen, Germany) following the manufacturer’s instructions. Microorganism genetic material quantification The total microbial genetic content in the samples was quantified by amplifying 16S rRNA and 18S rRNA gene fragments via a previously published primer set [21] on a QX200 Droplet Digital PCR system (Bio-Rad, USA). 18S rRNA primer specificity toward fungi was evaluated using the Primer-BLAST online tool (https://www.ncbi.nlm.nih.gov/tools/primer-blast/index.cgi). Each amplification reaction consisted of 11 µl of 2x ddPCR Supermix for Probes (without dUTP) (Bio-Rad, USA), 600 nM primers, 400 nM probes, 2 µl of DNA, and nuclease-free H2O to achieve a total reaction volume of 22 µl. Droplet generation was performed via a QX200 Droplet Generator (Bio-Rad, USA), followed by amplification in a T100 Thermal Cycler (Bio-Rad, USA) under the following conditions: a ramp rate of 1; 95 °C for 10 min; 40 cycles of 94 °C for 30 s and 60 °C for 1 min; and 98 °C for 10 min. After 4–8 h of equilibration and droplet stabilization at 12 °C, positive and negative droplets were quantified via a QX200 Droplet Reader (Bio-Rad, USA). The samples were analysed in triplicate. The acquired data were analysed with QuantaSoft software (Bio-Rad, USA) and expressed as the number of copies per 1 µl of sample. 16 S rRNA V3-4 and ITS-1 library construction and sequencing analysis A two-stage PCR protocol was applied for MiSeq library preparation. The primers used were designed for PCR amplification of the 16S rRNA gene V3-V4 region specific to bacteria (341 F and 805R, respectively) and amplification of the ITS-1 gene specific to fungi containing Illumina overhang adapters (Supplementary Table 1). DNA (4 ng) was amplified separately by 16S rRNA and ITS-1 primers via Phusion U Multiplex PCR Master Mix (Thermo Fisher Scientific, USA) under the following reaction conditions: denaturation at 98 °C for 30 s, 35 cycles of 98 °C for 10 s, {67 °C for 15 s for 16 S V3-4/65°C for 15 s for ITS}, 72 °C for 15 s, and fragment elongation at 72 °C for 7 min. To ensure the purity and quality of the PCR mixture, negative controls were introduced for each prepared master mixture. The yield of the acquired PCR products was assessed via 1.2% agarose gel electrophoresis, and the products were purified via a NucleoMag NGS Clean-Up and Size Select Kit (Macherey-Nagel, Germany). The concentration of the PCR products was measured via a Qubit dsDNA HS Assay Kit and a Qubit 2.0 fluorometer (both from Thermo Fisher Scientific, USA), and samples were normalized to 4 ng/µl. During the second PCR stage, Illumina MiSeq i7 and i5 indexes were added to the PCR products via custom-ordered Nextera XT Index Kit (Illumina Inc., USA) primers (Metabion International AG, Germany). For this reaction, Phusion U Multiplex PCR Master Mix was used under the same thermal cycler reaction conditions specified for the first PCR stage. The PCR products were then pooled and purified for sequencing via NucleoMag magnetic beads. The quality and acquired amount of the amplicons were assessed via an Agilent high-sensitivity DNA kit, an Agilent 2100 bioanalyzer (Agilent Technologies, USA), a Qubit dsDNA HS Assay Kit and a Qubit 2.0 fluorometer. Prior to sequencing, all samples, including negative controls, were pooled at equal molarities and diluted to 6 pM. They were then paired-end sequenced via a 500-cycle MiSeq Reagent Kit v2 and an Illumina MiSeq (Illumina Inc., USA). Each run was expected to produce at least 10,000 reads per sample. 16 S rRNA and ITS-1 sequence analysis Sequence reads were demultiplexed via Illumina MiSeq Reporter Software (Illumina Inc., USA), quality filtered and trimmed via the fastp tool with options -f 30 -F 30 − 5 20 − 3 20 -q 20 -l 40 --correction. All quality-approved sequences were imported into QIIME2 v.2024.2 [22] for further analysis. The DADA2 [23] plugin was used to pair forward and reverse reads and for quality control; phiX reads and chimeric sequences were filtered via a pooled consensus method. The resulting feature table and sequences were used for de novo clustering, employing the vsearch [24] plugin with a 97% identity threshold. Then, de novo multiple sequence alignment was performed via the MAFFT [25] method, and phylogenetic trees were constructed via FastTree2 [26]. De novo clustered sequences were used for taxonomic assignment with a prefitted sklearn-based taxonomy classifier. Bacterial classification was performed via the Silva v.138.1 database at 99% identity [27], and fungal classification was performed via the UNITE fungal ITS database v.9 at 97% identity [28]. Both reference databases were trained with a naïve Bayes classifier [29]. Both the fungal and bacterial negative controls were processed in the same way as the sample sequences. However, following DADA2 filtering and chimera removal, no sequences remained for downstream analysis; thus, these controls were discarded. Statistics The 16 S and 18 S rRNA gene loads were visualized on a logarithmic scale via the Pandas library [30] within the Python environment. To evaluate differences in microbial load across sample treatments, a post hoc pairwise Conover’s test with Holm p value adjustment was applied via the scikit-posthocs package [31] in Python. The microbiome datasets were filtered to remove low-abundance taxa using the following criteria: a taxon had to be present in more than two samples; have more than five read counts per sample, and exceed a total of five reads across all samples. The datasets were then rarefied to even sequencing depths of 1,500 for 16S rRNA and 200 for ITS-1 via phyloseq v.1.46.0. package [32] in R v.3.11.0. Taxonomic composition at the genus level was visualized using microViz v. 0.12.7. Alpha diversity (observed and Shannon indices) and beta diversity (Bray‒Curtis distances) measures at the genus level were calculated via the phyloseq and vegan v.2.6-8 packages. A Wilcoxon signed rank exact test from R base package stats was used to assess the significance of alpha diversity measurements between sampling types in the 16 S V3-4 dataset. However, because several samples were missing in the ITS-1 dataset, a Wilcoxon rank sum test with continuity correction was applied instead to evaluate differences in alpha diversity between sampling types. Permutational analysis of multivariate dispersions (PERMDISP) with 999 permutations was performed to compare Bray‒Curtis distances between sampling types. Differential abundance analysis of the taxonomic groups was performed via MicrobiotaProcess v.1.14.0 [33]. Briefly, the Kruskal–Wallis test (α = 0.05) was used to identify taxonomic features with significantly different distributions across classes. Pairwise Wilcoxon tests (α = 0.01) were then applied to assess whether subclass differences followed a consistent class-level trend. Significantly different taxa were further evaluated via linear discriminant analysis (LDA). Spearman correlation coefficients and their significance (e.g., two-sided Spearman’s rank correlation test for association) between genus level taxonomic features and sampling types were calculated via Hmisc v.5.2-2 and visualized via ggplot2 v.3.5.1. Results Quantification of microbial genetic material We examined three approaches for sample acquisition: [1] the use of IPA wipes before biopsy (hereinafter referred to as IPA) [2], the use of CAG before biopsy (hereinafter referred to as CAG), and [3] the use of CAG followed by IPA before biopsy (hereinafter referred to as CAG-IPA). As a reference, a biopsy sample was also collected from unsanitized and untreated skin (hereinafter referred to as the reference group). All the samples were collected in triplicate from three distinct sites, resulting in a total of 12 swab samples and 12 biopsy samples. As expected, the highest microbial genetic material load was detected in the swab samples (Fig. 2A), with an average of 11,661.04 ± 5,181.6 copies/µL for bacterial genetic material and 86.80 ± 35.4 copies/µL for fungal genetic material. In contrast, the average microbial genetic material load in biopsy samples was lower, with 76.41 ± 55.9 copies/µL for bacterial genetic material and 8.06 ± 5.4 copies/µL for fungal genetic material. The lowest microbial loads for both 16S rRNA and 18S rRNA were detected in the CAG biopsy samples, with average values of 17.14 ± 29.4 copies/µL and 1.26 ± 0.8 copies/µL, respectively. In comparison, the highest microbial loads were detected in the IPA biopsy samples, with 124.38 ± 102.6 copies/µL for 16S rRNA and 13.77 ± 13.2 copies/µL for 18S rRNA. Interestingly, reference group biopsy samples showed similar microbial loads to the IPA samples, with 124.0 ± 86.6 copies/µL for 16S rRNA and 10.70 ± 13.4 copies/µL for 18S rRNA. Finally, biopsy samples from skin treated with CAG-IPA presented a slightly greater microbial load than did samples from skin treated with CAG alone, with 40.10 ± 31.6 copies/µL for 16S rRNA and 6.50 ± 8.9 copies/µL for 18S rRNA. Fig. 2Digital droplet PCR (ddPCR) quantification results. Panel A – quantification of the 16S rRNA and 18S rRNA genes in swab and hair follicle biopsy samples expressed as copies per microliter on a log scale. Error bars represent 95% confidence intervals. Panel B – pairwise comparisons of nonparametric ANOVA with post hoc tests and Holm p value adjustment for 16S rRNA quantification. Panel C – pairwise comparisons of nonparametric ANOVA with post hoc tests and Holm p value adjustment for 18S rRNA quantification. Abbreviations: IPA, isopropyl alcohol wipe; CAG, cyanoacrylate (super) glue Digital droplet PCR (ddPCR) quantification results. Panel A – quantification of the 16S rRNA and 18S rRNA genes in swab and hair follicle biopsy samples expressed as copies per microliter on a log scale. Error bars represent 95% confidence intervals. Panel B – pairwise comparisons of nonparametric ANOVA with post hoc tests and Holm p value adjustment for 16S rRNA quantification. Panel C – pairwise comparisons of nonparametric ANOVA with post hoc tests and Holm p value adjustment for 18S rRNA quantification. Abbreviations: IPA, isopropyl alcohol wipe; CAG, cyanoacrylate (super) glue Although seemingly high variations were observed among biopsy samples across different acquisition approaches, these differences were not statistically significant for either 16S rRNA (p > 0.05, Fig. 2B) or 18S rRNA (p > 0.05, Fig. 2C) quantification. As expected, significant differences were identified between the biopsy and swab samples. For bacterial quantification, the most notable differences were observed between the CAG-IPA biopsy samples and all four swab sample groups (p < 0.05), as well as between the CAG biopsy samples and all four swab sample groups (p < 0.05). In contrast, no significant differences were detected between the IPA biopsy samples and either the IPA or reference group swabs (p > 0.05). With respect to fungal genetic material, significant differences were identified only between the CAG biopsy samples and the CAG-IPA swab samples (p = 0.023) and between the CAG biopsy samples and the CAG swab samples (p = 0.045). We also estimated the differences in microbial load between biopsy and swab samples across different sampling approaches. For the bacterial load, the difference between the swab and biopsy samples was 119.8-fold for the reference and IPA samples (14,874.9 ± 15,931.9 vs. 124.19 ± 84.9 copies/µL), 236.1-fold for the CAG samples (4,046.9 ± 1,592.3 vs. 17.14 ± 29.4 copies/µL), and 320.4-fold for the CAG-IPA samples (12,847.4 ± 9,212.4 vs. 40.1 ± 31.6 copies/µL). Similarly, for the fungal load, the differences were 5.3-fold (64.8 ± 69.5 vs. 12.2 ± 12.0 copies/µL), 68.8-fold (86.7 ± 37.3 vs. 1.3 ± 0.8 copies/µL) and 20.1-fold (130.9 ± 81.6 vs. 6.5 ± 8.9 copies/µL), respectively. More importantly, comparing biopsy samples, the microbial load in the reference group was 7.2-fold higher than in the CAG biopsies for bacteria (124.0 ± 86.6 vs. 17.14 ± 29.4 copies/µL) and 8.5-fold for the fungi (10.7 ± 13.4 vs. 1.3 ± 0.8 copies/µL). Between the reference and CAG-IPA biopsies, the differences were 3.1-fold for bacteria (124.0 ± 86.6 vs. 40.1 ± 31.6 copies/µL) and 1.7-fold for fungi (10.7 ± 13.4 vs. 6.5 ± 8.9 copies/µL). In contrast, the difference between the reference and IPA biopsies was minimal − 0.99-fold for bacteria (124.0 ± 86.6 vs. 124.38 ± 102.6 copies/µL) and 0.78-fold for fungi (10.7 ± 13.4 vs. 13.8 ± 13.2 copies/µL). Taxonomic diversity All 12 swab samples and 12 biopsy samples were sequenced to target the 16S rRNA V3-4 region and the ITS-1 gene. Sequencing of 16S rRNA V3-4 was successful for all samples, yielding an average raw sequence count of 34,994 (min: 15,939; max: 88,637) per sample. After quality control, an average of 11,558 (min: 2,152; max: 31,752) sequences per sample were retained. The resulting dataset was then filtered for low-abundance taxa and subsequently rarefied to 1,500 sequences per sample, which still maintained Good’s coverage estimates above 98.5% (Supplementary Fig. 1, Supplementary Table 3). In contrast, ITS-1 sequencing presented challenges. The average raw sequence count was 34,688 (min: 185; max: 135,552) per sample, and quality control retained an average of 23,792.4 (min: 26; max: 115,279) sequences per sample. However, nine samples yielded fewer than 1,000 sequences, and three samples yielded fewer than 100 sequences. To proceed with data analysis, we filtered out genera containing fewer than 5 sequences per sample, as these are prone to misclassification errors, and rarified the ITS-1 dataset to 200 sequences per sample (Supplementary Fig. 2). To ensure that rarefaction did not significantly alter the taxonomical composition, we compared the per-sample absolute abundances of the raw ITS-1 table and the filtered table via Student’s t test. The abundances for the majority of sample pairs did not differ significantly (p > 0.05, average p = 0.3997 ± 0.305); however, one sample, a swab from the reference group, was significantly different (p = 0.039). Given the low 18S rRNA gene quantities observed in the biopsy samples (Results Sect. 3.1) and the challenges posed by the low biomass profile in ITS-1 sequencing, we decided to proceed with strictly rarefied datasets. Good’s coverage estimates for the biopsy samples remained high (> 97.0%), whereas those for the swab sampled were lower, ranging from 78.5% to 95.0% (Supplementary Table 3). Microbial diversity The most dominant taxon in the swab samples was the genus Staphylococcus (80.2% in the reference group; 67.0% in the IPA group; 77.9% in the CAG group; and 65.7% in the CAG-IPA group), followed by Bacillus (7.4%; 7.2%; 4.8%; and 9.8%, respectively) and the order of Micrococcales (1.2%; 5.6%; 6.8%; and 12.4%, respectively). With respect to the biopsy samples, the largest number of sequences remained unclassified bacteria (53.3%; 28.3%; 77.5%; and 82.5%, respectively), followed by the genera Staphylococcus (17.3%; 19.9%; 1.6%; and 0.0%, respectively) and Pseudomonas (0.4%; 6.6%; 10.4%; and 12.7%, respectively) (Fig. 3A). Fig. 3Taxonomic composition and diversity analysis of the 16S rRNA V3-4 and ITS-1 sequences. Panel A – Genus-level taxonomic composition of 16S rRNA V3-4 sequencing. Panel B – Alpha diversity measurements from 16S rRNA V3-4 sequencing at the genus level. Panel C – Beta diversity in the form of principal coordinate analysis (PCoA) based on Bray‒Curtis distances for 16S rRNA sequencing analysis results. Panel D – Genus-level taxonomic composition of the ITS-1 sequencing data. Panel E – Alpha diversity measurements from ITS-1 sequencing at the genus level. Panel F – Beta diversity in the form of PCoA based on Bray‒Curtis dissimilarity distances for ITS-1 sequencing analysis results. Panel G – Spearman rank correlation analysis between microbial genus-level taxonomic entities and metadata variables. Panel H – Spearman rank correlation analysis between fungal genus-level taxonomic entities and metadata variables. For Panels G and H, asterisks denote a significant correlation according to these levels: ‘***’ – p = < 0.001; ‘**’ – p = (0.01, 0.001]; ‘*’ – p = (0.05, 0.01]. Abbreviations: IPA, isopropyl alcohol wipe; CAG, cyanoacrylate (super) glue Taxonomic composition and diversity analysis of the 16S rRNA V3-4 and ITS-1 sequences. Panel A – Genus-level taxonomic composition of 16S rRNA V3-4 sequencing. Panel B – Alpha diversity measurements from 16S rRNA V3-4 sequencing at the genus level. Panel C – Beta diversity in the form of principal coordinate analysis (PCoA) based on Bray‒Curtis distances for 16S rRNA sequencing analysis results. Panel D – Genus-level taxonomic composition of the ITS-1 sequencing data. Panel E – Alpha diversity measurements from ITS-1 sequencing at the genus level. Panel F – Beta diversity in the form of PCoA based on Bray‒Curtis dissimilarity distances for ITS-1 sequencing analysis results. Panel G – Spearman rank correlation analysis between microbial genus-level taxonomic entities and metadata variables. Panel H – Spearman rank correlation analysis between fungal genus-level taxonomic entities and metadata variables. For Panels G and H, asterisks denote a significant correlation according to these levels: ‘***’ – p = < 0.001; ‘**’ – p = (0.01, 0.001]; ‘*’ – p = (0.05, 0.01]. Abbreviations: IPA, isopropyl alcohol wipe; CAG, cyanoacrylate (super) glue In general, at the genus level, Shannon diversity did not differ significantly between the swab and biopsy samples (1.34 ± 0.9 vs. 1.02 ± 0.9, respectively; Wilcoxon signed-rank exact test, p = 0.62). However, the number of OTUs differed significantly between these sampling types (22.83 ± 20.6 vs. 9.50 ± 10.5; Wilcoxon signed-rank exact test, p = 0.01). Comparisons between all sample groups did not yield any significant differences (Wilcoxon signed-rank exact test, p > 0.05, Fig. 3B). Among the biopsy samples, the lowest number of OTUs and Shannon index values were detected in the CAG-IPA group (2.33 ± 2.3 and 0.42 ± 0.7, respectively), followed by the CAG group (5.67 ± 5.7 and 0.60 ± 0.9, respectively). In contrast, the highest values were found in the IPA group (21.33 ± 14.6 and 2.00 ± 0.8, respectively), followed by the reference group (8.67 ± 7.0 and 1.05 ± 0.8, respectively). For the swab samples, the greatest number of OTUs and Shannon index values were observed in the CAG-IPA group (36.00 ± 32.9 and 1.76 ± 1.3, respectively) and the CAG group (21.33 ± 17.0 and 1.28 ± 0.9, respectively). Principal coordinate analysis (PCoA) at the genus level revealed distinct clustering between the swab and biopsy samples. However, their difference based on Bray‒Curtis distances was not statistically significant (dispersion test, p = 0.099, n = 999 permutations). Additionally, pairwise comparisons between sample types and treatment groups revealed that the global taxonomical compositions of these sample groups were similar (dispersion test, for all comparisons, p > 0.05, n = 999 permutations; Fig. 3C). To assess differences in taxonomic abundance between sample types (e.g., all swab versus all biopsy samples) and treatment groups (e.g., reference group, IPA, CAG, or CAG-IPA), we performed a differential analysis across all taxonomic levels. However, the results revealed that none of the taxonomic ranks differed significantly between sample types and treatment groups (Kruskal-Wallis test, p > 0.05), likely due to the low number of samples per group. We performed a Spearman correlation analysis between genus-level taxonomic entities and metadata to determine whether any bacterial taxa correlated with the sampling area (n = 12), sample type (swab or biopsy), or treatment group (reference group, IPA, CAG, and CAG-IPA). This analysis revealed one significant positive correlation and three significant negative correlations with sample type (two-sided Spearman’s rank correlation test, p < 0.05), and one significant positive correlation with sampling area (p < 0.05), but no significant correlations with the treatment group (p > 0.05) (Fig. 3G, Supplementary Fig. 3). A significant positive correlation with sample type was detected for unassigned bacteria (r = 0.869, p = 3.5 × 10− 8), whereas significant negative correlations were detected for the genera Staphylococcus (r = -0.849, p = 1.6 × 10− 7), Psychrobacter (r = -0.585, p = 0.003) and the order Micrococcales (r = -0.692, p = 0.0002). A significant negative correlation with sample area was observed for the genus Nocardioides (r = -0.426, p = 0.038). The most prominent differences between the swab and biopsy samples were observed for unassigned bacteria, the genus Staphylococcus and unclassified bacteria from the order Micrococcales (Supplementary Fig. 1). For unassigned bacteria, the average abundance in the swab samples was 0.5% ± 1.1%, whereas in the biopsy samples it was significantly higher − 60.4% ± 30.5%. This suggests that a substantial proportion of bacterial genera associated with the hair follicle microbiome have not been recognized within available databases. Further examination revealed that unassigned bacteria were most abundant in the CAG biopsy samples (77.5% ± 47.1%), followed by the CAG-IPA biopsy samples (82.5% ± 50.0%), possibly reflecting a lower level of contamination by skin surface-related bacteria. The relative abundance of Staphylococcus was higher in the swab samples (72.7% ± 36.1%) than in the biopsy samples (9.71% ± 13.4%). Among the biopsy groups, the highest Staphylococcus abundance was detected in the reference group (17.3 ± 27.9%) and IPA biopsies (19.9% ± 6.6%), whereas the lowest was detected in the CAG (1.6% ± 3.8%) and CAG-IPA biopsies (0%). Unclassified Micrococcales were less abundant in the biopsy samples (0.06% ± 0.18%) than in the swab samples (6.5% ± 11.3%). Among the biopsy samples, unassigned Micrococcales were detected only in the IPA group (0.24% ± 0.4%) and were absent from the other biopsy groups. Overall, theses patterns suggest that CAG treatment might effectively eliminate bacteria associated with the skin surface. Fungal diversity After quality filtering and strict rarefication, several samples were excluded due to an insufficient number of sequences. These included one biopsy sample from the IPA group, one swab sample from the reference group, one biopsy sample and one swab sample from the CAG group, and two biopsy samples and two swab samples from the CAG-IPA group. As a result, only 15 samples remained for the exploration of fungal community diversity under different treatment regimens. The most abundant fungal representatives in the swab samples belonged to the genus Vishniacozyma (38.5% in the reference; 40.7% in the IPA; 9.0% in the CAG; and 46% in the CAG-IPA), followed by Microdochium (11.3%; 12.0%; 3.2%; and 1.0%, respectively), and Seimatosporium (8.5%; 0.67%; 4.0%; and 20.0%, respectively). In the biopsy samples, the dominant genera differed between the reference group and the treatment groups. In the reference group, the most abundant genus was Mrakia (33.3%), followed by Vishniacozyma (31.3%) and members of Incertae sedis (21.7%). In contrast, the IPA biopsy samples were dominated by unclassified fungi (41.8%), Cladosporium (22.0%), and Malassezia (6.8%), whereas all sequences in the CAG biopsy samples belonged exclusively to Malassezia (100.0%) (Fig. 3D). At the genus level, both the number of OTUs and the Shannon diversity index were significantly higher in the swab samples (OTUs: 28 ± 5.9; Shannon: 2.41 ± 0.3) than in the biopsy samples (OTUs: 2.5 ± 1.4; Shannon: 0.46 ± 0.5) (p = 0.0017 and p = 0.0018, respectively). Pairwise comparisons among all sample groups did not reveal any significant differences (p > 0.05, Fig. 3E). Consistent with the taxonomical composition results described above, the lowest OTU count (n = 1) and the lowest Shannon diversity index (Shannon: 0) were observed in the only biopsy sample that was successfully sequenced from the CAG-IPA group. The next lowest alpha diversity values were found in biopsy samples from the reference group (OTUs: 2.67 ± 1.5; Shannon: 0.38 ± 0.42) and the IPA group (OTUs: 3.0 ± 1.4; Shannon: 0.82 ± 0.5). Similar to the microbial beta diversity results, the PCoA at the genus level revealed distinct sample clustering - the swab samples clustered apart from the biopsy samples (Fig. 3F). However, clustering on the basis of the treatment groups was mostly unspecific, and owing to the unequal number of samples per group, pairwise comparisons to test for global dissimilarity could not be performed. In addition, owing to the limited number of biopsy samples in the CAG-IPA (n = 1) and IPA (n = 2) groups, differential abundance analysis was not performed. At the genus level, correlation analysis revealed that the abundances of several fungal entities significantly correlated with the sample source (e.g., biopsy or swab; two-sided Spearman’s rank correlation test, p < 0.05), but not with treatment type (Fig. 3G). Further investigation of these significantly correlated genera revealed that all were enriched in the swab samples and depleted in the biopsy samples (Supplementary Fig. 4). The strongest correlation values were observed for Aspergillus (r = -0.813, p = 0.0002), Ramularia (r = -0.806, p = 0.00029), and Eukaryota incertae sedis (r = -0.798, p = 0.0004). However, considering that only 15 out of 24 samples passed the quality control and rarefaction filtering criteria, the results regarding fungal composition should be interpreted with caution. Discussion The host skin alone is populated by a diverse array of microorganisms, which vary between different anatomical sites. However, the study of microbial communities within hair follicles has only gained attention in recent years, driven by advances in molecular biology techniques. Compared with the skin surface, hair follicles provide a moister and more acidic environment with UV protection, which facilitates the colonization of various bacteria, fungi, and viruses, creating communities that are distinct from those on the skin surface [2, 6, 7]. Given that hair follicles are small and easily contaminable microbial habitats, precise sample collection methods are critically important. In this study, we assessed four different punch biopsy sampling methods: (1) a biopsy from shaved, unsanitized skin; (2) a biopsy from shaved skin cleaned with an IPA pad; (3) a biopsy from shaved skin cleaned with CAG; and (4) a biopsy from shaved skin cleaned with CAG followed by IPA. The selected skin treatment procedures were designed to mimic sample acquisition in clinical settings. Topical disinfection with IPA is a common prophylactic step prior to procedures such as blood specimen collection [34, 35], surgical site disinfection [36], biopsy needle cleaning [37], and inanimate surface disinfection [38, 39], among others. In contrast, skin cleaning using CAG has been less well documented – to our knowledge, only one skin microbiome-oriented study has employed this substance for skin cleaning [3]. To date, CAG has been widely used in surgery primarily for closing lacerations and incisions [40–42] and as a skin substitute [43]. Therefore, our study design addresses a gap in knowledge regarding sample collection approaches that are adaptable to clinical settings. Although this study was conducted using a wild animal model - roe deer - and the results are not fully translatable to human-based studies, the findings may nonetheless inform the selection of hair follicle biopsy methods that minimize contamination risk. As expected, the quantification results indicated that the highest microbial genetic load was present in the swab samples, whereas the lowest was detected in the biopsy samples. In terms of the genetic material source, the bacterial load in the hair follicle was, on average, 9.5 times greater than the fungal load. However, owing to the ultralow biomass of fungal material in both swab and biopsy samples, sequencing of the ITS-1 marker gene was challenging, and the resulting data were of lower quality. Nonetheless, our results are comparable to those of previous studies, which reported that the metagenomic sequencing of various human hair follicle compartments yielded, on average, 76% bacterial reads and 24% fungal reads [2]. However, as expected, the taxonomic composition of the hair follicle microbiome in wild animals differed from that in humans, and there is currently no available information on the hair follicle microbiome of doe deer. Furthermore, both our quantification and targeted sequencing results show that different sample acquisition techniques have a substantial effect on the amount of microbial genetic material recovered as well as on the taxonomic composition. Our findings indicate that the potentially most effective approach for studying the hair follicle microbiome is the prior application of CAG, which reduces the amount of bacterial genetic material by 7.2-fold and the amount of fungal material by 8.5-fold compared with the reference samples (e.g., unsanitized skin). The next most effective strategy was the use of CAG followed by skin cleaning with IPA, which reduced bacterial genetic material by 4.1-fold and fungal material by 1.7-fold compared with the reference group. Overall, the mean reduction in microbial load, calculated as the average fold decrease in bacterial and fungal quantities between CAG-treated (CAG and CAG-IPA) and non-CAG (IPA and reference) samples, was 3.7-fold. These differences were also reflected in the taxonomic composition: biopsy samples from skin treated with CAG were more distinct from those collected from skin cleaned only with IPA or left unsanitized. However, our microbiome analysis also revealed that a large proportion of bacterial sequences, especially in samples treated with CAG, remained unclassified, likely owing to the absence of relevant reference sequences in available databases. Given that human hair follicle microbiomes have been successfully classified even at the species level [2, 3, 44], we suspect that this high proportion of unclassified sequences may be specific to wild animals, as similar studies in wildlife have not yet been conducted. These findings highlight the need for more comprehensive reference databases for wildlife microbiomes to improve taxonomic resolution in future studies. In addition, although the seemingly best approach would be to prepare the skin by applying CAG followed by disinfection with IPA, our taxonomic composition results suggest that this method might increase the likelihood of detecting Staphylococcus and Bacillus, which are common skin inhabitants [45–47], as well as environmental bacteria such as Paenarthrobacter [48] and Exiguobacterium [49]. It is possible that applying IPA may mobilize surface bacteria from areas adjacent to the glue-treated region. Owing to the properties of alcohol, these bacteria can penetrate deeper skin layers and thus become detectable during both quantification and targeted sequencing. Therefore, we speculate that the use of IPA may, to some extent, contribute to the contamination of hair follicle microbiome samples with bacteria from the skin surface. For this reason, the sole application of CAG might be the preferred method to minimize such contamination. In addition, it should be noted that during the application of CAG, the removal of the dried layer from shaved skin causes detachment of the uppermost epidermal layer along with a few hair follicles. Nonetheless, after CAG removal, the remaining intact follicles are clearly visible and can be collected using a biopsy punch for subsequent downstream analysis. Our study had several limitations. First, this study involved a nonhuman source of samples and therefore faced sequence classification challenges for biopsy samples, indicating that the hair follicle microbiome structure may differ significantly from that of humans. Moreover, conducting a similar study in humans would raise several ethical concerns, such as the need to recruit many participants to capture interindividual variation and to estimate the core composition of the hair follicle microbiome. Moreover, approaches for skin preparation involving CAG followed by the collection of hair follicle biopsies are invasive and leave the individual with a substantial wound on the scalp that is prone to infection - which may be difficult to justify given that the goal is to assess the best method for contamination-free hair follicle microbiome sampling. In addition, our study was limited by its relatively small sample size. Although we collected samples in triplicate for all the sampling strategies, we acknowledge that a larger number of biological replicates would increase the statistical power and strengthen our conclusions. Finally, our community characterization was based on targeted sequencing approaches rather than shotgun sequencing, which has been shown to be more cost-effective in cases where high levels of host DNA are present, as in this study. However, while targeted sequencing allows us to estimate community diversity at the bacterial and fungal levels, it does not capture viral communities, which should be addressed in future research. In conclusion, this study revealed that the amounts of both bacterial and fungal genetic material recovered from hair follicles are low and that the microbial community composition is distinct from that of wild animal hair. These results also demonstrate a high risk of contamination of hair follicle biopsy samples with microbial DNA from the skin surface and confirm that the application of CAG to remove the upper layer of skin along with its associated microbiome is an effective method for cleaning the skin before biopsy collection. Supplementary Information Supplementary Material 1. Supplementary Material 1.