Unravelling the hidden side of laundry: malodour, microbiome and pathogenome

Abstract

Background

Recent trends towards lower washing temperatures and a reduction in the use of bleaching agents in laundry undoubtedly benefit our environment. However, these conditions impair microbial removal on clothes, leading to malodour generation and negative impacts on consumer well-being. Clothing undergoes cycles of wearing, washing and drying, with variable exposure to microorganisms and volatilomes originating from the skin, washing machine, water and laundry products. Laundry malodour is therefore a complex problem that reflects its dynamic ecosystem. To date, comprehensive investigations that encompass the evaluation of both microbial community and malodorous volatile organic compounds throughout all stages of the wash-wear-dry cycle are scarce. Furthermore, the microbial and malodour profiles associated with extended humid-drying conditions are poorly defined.

Results

Here we present olfaction-directed chemical and microbiological studies of synthetic T-shirts after wearing, washing and drying. Results show that although washing reduces the occurrence of known malodour volatile organic compounds, membrane-intact bacterial load on clothing is increased. Skin commensals are displaced by washing machine microbiomes, and for the first time, we show that this shift is accompanied by an altered pathogenomic profile, with many genes involved in biofilm build-up. We additionally highlight that humid-drying conditions are associated with characteristic malodours and favour the growth of specific Gram-negative bacteria.

Conclusions

These findings have important implications for the development of next-generation laundry products that enhance consumer well-being, while supporting environmentally friendly laundry practices.

Background

The shift towards more environmentally friendly domestic laundry practices has resulted in the adoption of cooler wash temperatures and shorter wash cycles, with more consumers choosing chlorine bleach-free, enzyme-based detergents [1, 2]. These washing regimes typically use less water and energy, offering clear environmental and economic benefits. However, evidence indicates that cleaning performance and microbial inactivation are impaired, leading to residual malodour with detrimental impacts on consumer well-being [3,4,5] and ultimately, more frequent laundering that counteracts eco-friendly intent [2, 6].

During wear, when clothing is brought into close contact with the skin, the transfer of resident microorganisms, secretions (such as sweat and sebum) and malodorous volatile organic compounds (VOCs) is facilitated [1, 7,8,9,10]. This is highly textile dependent, with synthetic fabrics such as polyester associated with greater malodour impact than natural materials such as cotton and wool [7, 10, 11]. Furthermore, synthetic textiles support stronger adhesion of skin associated bacteria than natural counterparts and enable selective proliferation of specific genera [2, 7, 8, 12]. Some of these bacteria are well-known culprits of malodour generation in the axilla, able to metabolise components of skin secretions into malodorous VOCs [13,14,15,16]. This activity has also been shown in clothing, where adhered bacteria transform adsorbed organic feedstock into VOCs that contribute to malodour perception in laundry [11, 17, 18]. Importantly, some of these bacteria are known opportunistic pathogens, e.g. Staphylococcus aureus and Micrococcus luteus, and are known to produce biofilms, facilitating their enhanced survival and therefore their persistence on textiles [19].

The process of laundering aims to remove soils, skin secretions, microorganisms and malodorous VOCs, leaving clothes hygienically clean and imparting a fragrant scent. However, evidence shows that the effectiveness of this too is textile dependent, with synthetic textiles retaining higher quantities of skin secretions and malodorous VOCs such as fatty acids, aldehydes and aromatic compounds [11, 20,21,22]. In addition, washed fabrics host textile-specific microbial communities that originate from both human and environmental sources. For example, washed polyester is associated with Micrococcus, Moraxella, Staphylococcus, Corynebacterium, Cutibacterium, Enhydrobacter and Acinetobacter spp. [2, 7, 12]. In order to achieve microbial reduction, clothes must be washed at high temperature (≥ 60 °C) and/or in the presence of oxidising compounds (e.g. activated oxygen bleach or chlorine based) [23,24,25]. High-temperature washing not only inactivates microbes, but also activates oxidising compounds, resulting in microbial cell membrane damage and in turn, microbial lysis [23]. However, standard European washing cycles at 30 °C are known to impair antimicrobial performance, leading to bacterial exchange among the laundry load, the washing machine and its influent water [7, 12, 24]. As a result, microbial communities that originate from both human and environmental sources persist, providing opportunity for the establishment of biofilms and posing clear negative implications for textile hygiene as well as continued malodour generation.

Subsequent to washing, the conditions under which clothing is dried are relevant to continued microbial activity and malodour release. High-temperature tumble drying has been shown to be the most effective at microbial inactivation [26, 27], whereas prolonged drying in high humidity promotes bacterial proliferation and is associated with specific malodours (as discussed by others [4, 12, 28]). Clothes that are line dried in a temperate environment host fewer microbes than those dried in humid conditions, particularly in outdoor drying where UV activity contributes towards microbial inactivation [27]. Nonetheless, some bacteria demonstrate high tolerance to desiccation and continue to generate malodour [4]. The retention of malodorous VOCs, skin secretions and microorganisms on laundered clothes, in combination with repeated exposure to the skin and environment during subsequent wear, creates favourable conditions for cumulative malodour development. Over multiple wear-wash-dry cycles this accumulated “permastink”, defined as the permanent malodour in clothing [29, 30], can be detected in even freshly laundered clothes [1].

Taken together, research to date suggests that modern laundry habits, as well as continued market growth of polyester and “athleisure” clothing (i.e. athletic clothing worn as everyday wear) [31, 32] constitute a multifactorial pathway to laundry malodour. However, odour, VOCs and microbial community changes remain to be fully characterised throughout the entire wash-wear-dry cycle. Furthermore, the relative contributions of host and environmental microbial communities to that of worn, laundered and dried textile remain unknown. Many environmental niches are well-known reservoirs for horizontal gene transfer, facilitating the exchange of genes associated with enhanced survival, such as those involved in biofilm formation and in antimicrobial resistance [33,34,35]. These may have implications for textile hygiene as well as malodour development; however to date, the laundered textile pathogenome has not been fully explored. Finally, given its relevance to indoor-drying malodour [4], an investigation of textiles dried in humid conditions warrants further study.

To this end, we assessed axillae, worn and unworn synthetic (93% polyester and 7% elastane) T-shirts and washing machine samples in 31 participants throughout the wear-wash-dry cycle (including humid-drying); an approach seldom used before [36]. An overview of the study design, the sample types collected and the analyses performed is shown in Fig. 1 and participants’ metadata are available in Additional file 1: Table S1. Whole odour character assessment of T-shirts in addition to odour-directed chemical profiling shows that malodours are introduced during wear and removed by washing, but reappear throughout humid-drying. Metagenomic sequencing and flow cytometry results show that new bacteria are introduced during laundering and that bacteria proliferate after washing and throughout humid-drying. For the first time, shotgun sequencing results are presented, enabling greater resolution of microbial identities, their community-level DNA pathways and the associated pathogenome. Earlier studies have set out to explore the prevalence of individual antimicrobial resistance (AMR) genes on laundered textiles in the context of public health. However, our approach enables us to report the prevalence of all known AMR and virulence factor (VF) genes on worn, washed and dried clothing. We discuss the results of our study in the context of malodour persistence, the main focus of this research, since the further dissemination of AMR genes to the community was out of the scope of this study and therefore was not explored.

Fig. 1

Study set-up. Graphical overview of the study set-up, including all the samples collected from the 31 study participants at each time point. Each analysis method is represented by a symbol. The symbols next to the sample names indicate which analyses were performed on each group of samples

Results

Sensory odour profiles of axillae and T-shirts throughout the wear-wash-dry cycle

Malodour in clothing can be broadly assigned into two main subdivisions: (i) a pre-wash malodour of worn clothes that originates from the metabolism of host sweat by microbes in the axilla, and (ii) a post-wash malodour that is at least partially caused by the microbial community introduced into the textiles during washing. Our sensory evaluations aimed to characterise these events by assessing two odour parameters: the first being the hedonic value, referring to the subjective pleasantness of the odour, ranging from − 8 (extremely unpleasant) to + 8 (extremely pleasant), allowing for a nuanced assessment of the odour’s impact. The second parameter was the intensity scale, ranging from 0 (no odour) to 10 (extremely strong odour), which was selected to quantify the strength of the odour.

There was evident variation in the hedonic value and intensity scores between the samples of different participants (Additional file 2: Fig. S1, 2). However, despite the inter-individual variation, significant differences were detected between sample groups (Fig. 2a, b). Axillary odour after wearing the T-shirts for 48 h was significantly worse than the worn T-shirts odour (lower hedonic value: Wilcoxon signed rank test; R = 0.690; FDR < 0.001; and higher intensity: R = 0.662; FDR < 0.001). Based on the hedonic value, the odour of the worn T-shirts improved significantly upon washing (R = 0.869; FDR < 0.001) but remained worse than the unworn-washed T-shirts (R = 0.533; FDR < 0.01). Dry-drying (defined here as drying at 24.5 °C ± 0.2 °C, 50% relative humidity (Fig. 1)) did not significantly impact the hedonic value of the worn-washed T-shirts (R = 0.028; FDR > 0.05) but the intensity of the odour significantly decreased (R = 0.862; FDR < 0.001). On the other hand, humid-drying (defined here as drying at 31.8 °C ± 0.8 °C, 80% relative humidity (Fig. 1)) strongly increased the malodour of the worn-washed T-shirts (lower hedonic value: R = 0.862; FDR < 0.001; and higher intensity: R = 0.811; FDR < 0.001). For the unworn-washed T-shirts, both drying conditions worsened the odour pleasantness (dry-drying: R = 0.535; FDR < 0.01; humid-drying: R = 0.820; FDR < 0.001), but only humid-drying increased odour intensity (R = 0.503; FDR < 0.01). Wilcoxon signed rank test results of all pairwise sample group comparisons can be found in Additional file 1: Table S2a, b.

Fig. 2

Odour and bacterial load variation throughout the wear-wash-dry cycle. a,b Box plots of odour parameters, stratified by sample origin. The hedonic value (HV) indicates the pleasantness of odour (lowest: − 8; highest: 8); the intensity (I) indicates the strength of odour (lowest: 0; highest: 10). c, d Boxplots of membrane-intact and damaged bacterial cell concentrations, stratified by sample origin (axilla samples: cells/mL; T shirt samples: cells/cm²). Pairwise comparisons were calculated using Wilcoxon signed rank test (n = 31 per group). Different letters mean a significant difference between pairwise comparisons (FDR < 0.05). The horizontal box lines represent the first quartile, the median and the third quartile. Whiskers denote the range of points within the first quartile − 1.5 × the interquartile range and the third quartile + 1.5 × the inter-quartile range. Each dot indicates a sample. Legend: A_T0: axilla before wearing the T-shirt; A_T1: arm pit after wearing a T-shirt for 48 h; T: worn T-shirt; T_w: worn-washed T-shirt; Tb_w: unworn-washed T-shirt; T_dd: dry-dried worn T-shirt; Tb_dd: dry-dried unworn T-shirt; T_hd: humid-dried worn T-shirt; Tb_hd: humid-dried unworn T-shirt

Odour-directed chemical profiles of T-shirts throughout the wear-wash-dry-cycles

In order to further characterise the malodour events in the synthetic T-shirts, odour-directed chemical assessments were carried out. The chemical profiles of the T-shirt samples were extremely complex, with some comprising in excess of several hundred separate detectable VOCs. Using the olfaction-directed directed approach outlined in the “Methods” section, a total of ten VOCs were identified: eight short-chain carboxylic acids (n-butanoic, n-pentanoic, n-hexanoic, n-octanoic, 2- & 3-methylbutanoic, 3-methylhex-2-enoic, 4-hydroxyvoxbutanoic and 2-ethylhexanoic) and two aldehydes (n-hexanal and n-octanal) (Table 1). All of these VOCs are well-known malodour-contributing volatiles and many have previously been associated with axillae and/or laundry [1, 13, 28, 37,38,39,40].

Table 1 Variation of olfactively perceivable malodour compounds through the wear-wash-dry cycle, as identified by GC-O-MS (T: worn T-shirt; T_w: washed T-shirt; T_hd; humid dried worn T-shirt)

Overall, there was wide variation among participant samples, ranging from zero malodour active VOCs perceived in one sample up to 11 in another (though it is important to note that the absence of any given VOC in a particular sample may indicate its presence below olfactory detection threshold). However, in general, the number of instances of VOCs identified in each of the test conditions decreased in the order worn (69 instances) > worn humid-dried (39) > worn-washed (20). This demonstrates that the laundry process (washing and drying) is only partially effective at removing some malodour VOCs from the fabric, as several reappear during humid-drying. The distribution of these VOCs across worn, worn-washed and worn humid-dried T-shirts in each participant is shown in Additional file 1: Table S3.

2- & 3-methylbutanoic acids (indistinguishable in our instrument-based analyses) dominated the chemical profile, being the most frequently identified malodour compound. Of additional note in our study were 3-methylhex-2-enoic acid and n-pentanoic acid. These VOCs were identified in many of the worn T-shirts (more than half in the case of 2- & 3-methylbutanoic acids), contributing to the pre-wash malodour (14,15,38). These three acids were efficiently removed during washing from the majority of T-shirts (Table 1); however, it is notable that 2- & 3-methylbutanoic acids were present in the same number of worn humid-dried T-shirt samples as the worn (unwashed) T-shirt samples, suggesting their regeneration during humid-drying and likely contribution to the post-wash malodour in the T-shirts. A similar pattern (albeit at lower overall frequency) was observed for n-butanoic acid.

In addition to 2- & 3-methylbutanoic acids, a number of other substituted carboxylic acids, likely to be products of oxidation routes, were identified, for example 2-ethylhexanoic acid. This material is a known environmental contaminant so it is not possible to definitively assign its generation to the wear-wash-dry process. In the majority of cases, washing appeared to remove these VOCs, with few instances of regeneration during humid-drying.

Finally, we observed that n-octanal was more abundant in the washed T-shirts (nine instances) than in the worn T-shirts (five instances) and humid-dried T-shirts (two instances), indicating a post-wash malodour that only occurs immediately after laundry.

The bacterial load of axillae and T-shirts throughout the wear-wash-dry cycle

Viability staining with flow cytometry analysis was performed on the axilla and T-shirt derived samples to measure the concentrations of membrane-intact and damaged bacterial cells. Similar to the odour assessment, and in keeping with previously published data [2], bacterial cell concentrations of samples from different participants were highly variable (Additional file 2: Fig. S3). However, despite the inter-individual variation, significant differences were detected between sample groups (Fig. 2c, d). Notably, there was a significant increase of membrane-intact bacterial cells in the individuals’ axillae after wearing the T-shirts for 48 h (Wilcoxon signed rank test; R = 0.873; FDR < 0.001), indicative of bacterial growth (Fig. 2c). The bacterial cell concentrations of both membrane-intact and damaged cells on the worn T-shirts were significantly lower than in the axilla samples (T0 and T1). However, membrane-intact cell concentrations increased significantly after washing (R = 0.704; FDR < 0.001), highlighting the impaired antimicrobial performance of standard washing cycles. The membrane-intact cell concentrations on the unworn-washed T-shirts were lower than those on the worn-washed T-shirts (R = 0.873; FDR < 0.001) but at comparable concentrations to those on the worn T-shirts (R = 0.355; FDR > 0.05). During dry-drying, the membrane-intact bacterial cell concentrations significantly decreased in both worn and unworn T-shirts (0.764 ≤ R ≤ 0.873; FDR < 0.001). However, during humid-drying, there was a significant increase in membrane-intact bacterial cell concentrations in the worn and unworn T-shirts (R = 0.873; FDR < 0.001), which is strongly indicative of bacterial growth. Interestingly, the number of membrane-intact bacteria on the unworn T-shirts after dry-drying was significantly less than that on the worn dry-dried T-shirts (R = 0.725; FDR < 0.001); however, after humid-drying, the number of bacteria on the worn and unworn T-shirts were not significantly different. Wilcoxon signed rank test results of all pairwise sample group comparisons can be found in Additional file 1: Table S2c, d.

When studying the changes in bacterial cell concentrations and odour parameters for all samples, there was a moderate but significant correlation between bacterial cell concentrations (membrane-intact and damaged), hedonic value and intensity scores (− 0.46 ≤ R ≤ 0.53; FDR < 0.001) (Additional file 2: Fig. S4). The higher the bacterial cell concentration, the worse was the odour (lower hedonic value and higher intensity). However, no significant correlations were detected between bacterial cell counts and odour parameters within any individual sample group (Additional file 1: Table S4).

The microbial communities of clothing vary more than their functions throughout the wear-wash-dry cycle

In order to characterise the T-shirt microbial community throughout the wear-wash-dry cycle, 155 samples were selected for shotgun metagenomic sequencing, including the axilla after wearing the T-shirt for 48 h, the worn T-shirt, the worn-washed T-shirt, the dry-dried worn T-shirt and the humid-dried worn T-shirt for each of the 31 participants (Fig. 1). Integrated multiomics profiling resulted in metagenome-based taxonomic profiles and metagenomic functional profiles (community-level DNA pathways), enabling us to search for patterns in the presence of pathways that may be relevant to malodour generation.

Based on microbial genus-level composition, sample origin explained 33.2% of the dissimilarity between the sample groups (Fig. 3a; adonis2; R = 0.332; p < 0.001; Additional file 2: Fig. S5a). The microbial genus-level composition of the axilla and worn T-shirt samples were significantly different (Fig. 3a; adonis2; R-adj = 0.054; FDR = 0.002; Additional file 1: Table S5). Moreover, the microbial genus-level composition of the T-shirts changed significantly throughout the wear-wash-dry cycle (Fig. 3a; FDR < 0.05; Additional file 1: Table S5). Notably, humid-drying resulted in a greater shift of the microbial genus-level composition of the washed T-shirts (Fig. 3a; R-adj = 0.102; FDR = 0.002; Additional file 1: Table S5) than from dry-drying (Fig. 3a; R-adj = −0.003; FDR = 0.021; Additional file 1: Table S5).

Fig. 3

Microbiome associations with processes of wearing, washing, dry-drying and humid drying. a Relative abundances of the 20 most common genera and b Significant associations (FDR < 0.05) of metagenomics genera detected by Microbiome Multivariable Associations with Linear Models’ (MaAsLin2’s) default linear mixed effects model. Only associations significant in at least two sample type pair comparisons are shown. Detected associations are adjusted for participant as random effect and sample origin as fixed effect. Values are log-transformed relative abundances with half the minimum relative abundance as pseudo count. Complete linkage clustering of Pearson correlation coefficients was used to hierarchically cluster the features. Dots indicate the significant level of the FDR-adjusted p-values: < 0.001 “●”, < 0.05 “○”. Legend: A_T1: axilla after wearing the T-shirt for 48 h, T: worn T-shirt, T_w: worn washed T-shirt, T_dd: dry-dried worn T-shirt, T_hd: humid-dried worn T-shirt

Based on functional composition, sample origin explained 3.4% of the dissimilarity between the sample groups (adonis2; R = 0.034; p = 0.005; Additional file 2: Fig. S5b). In this case, the overall functional composition of the axilla and worn T-shirt samples were not significantly different (adonis2; R-adj = 0.002; FDR = 0.096; Additional file 1: Table S6). Furthermore, the wear-wash-dry cycle had no significant effect on the overall functional composition of the samples (FDR > 0.05; Additional file 1: Table S6). To test the degree to which microbial genus-level composition was associated with the functional composition, we performed a Procrustes analysis (Additional file 2: Fig. S6). We compared ordinations of the functional composition with the microbial genus-level composition and found them to correlate significantly (protest; corr = 0.357; p < 0.001). However, we observed that the functional profiles were more conserved between samples than the taxonomic profiles for all the sample groups (Wilcoxon signed rank test; FDR < 0.001; Additional file 2: Fig. S5c).

In order to dissect taxonomic changes throughout the wear-wash-dry cycle at a greater resolution, while controlling for inter-individual effects, we applied Microbiome Multivariable Associations with Linear Models (MaAsLin2) [41] (Fig. 3b). All significant differences detected were represented by changes in bacterial genera abundances. This is unsurprising, given that the great majority of reads were assigned to bacterial genomes, with low read counts detected for fungal and viral metagenomes. A comparison of the microbial genus-level composition of the axilla and worn T-shirt samples resulted in 235 significant associations (Additional file 1: Table S7). Almost all of these bacterial genera increased significantly in the T-shirt samples, including Stenotrophomonas maltophilia and Pseudomonas aeruginosa, together with many skin commensals, including Kocuria, Micrococcus (also found in environmental sources) and Moraxella spp. In contrast, only a few skin commensals decreased in abundance compared to the axilla samples (after 48 h wearing the T-shirts), such as Anaerococcus and Cutibacterium spp. (8,40,41)(Fig. 3; Additional file 1: Table S7). This demonstrates that the worn T-shirts promoted selective bacterial enrichment, with many bacteria not only adhering to the fabric but also thriving on it, creating a different microbiome from that of the axillary region. When comparing the microbial genus-level composition of the worn and washed T-shirts, we observed 229 significant associations of bacterial genera (Additional file 1: Table S8). Those that increased significantly in the washed T-shirts are primarily Gram-negative bacterial genera (Fig. 3b; Additional file 1: Table S8) that are typically associated with water, plants and soil [42, 43]. However, those that decreased significantly in the T-shirts after washing were predominantly Gram-positive bacteria that are common inhabitants of human mucosa and skin [44, 45] (Fig. 3b; Additional file 1: Table S8). Therefore, the process of washing resulted in displacement of the relative abundances of typical Gram-positive skin commensals with Gram-negative environment-associated bacteria (Fig. 3, Fig. 4a). Overall, dry-drying had little effect on the microbial genus-level composition of the washed T-shirts, with only three significant associations identified (Fig. 3b; Additional file 1: Table S9). However, humid-drying resulted in 135 associations compared to the washed T-shirts (Additional file 1: Table S10). On the one hand, environment-associated Gram-negative bacterial genera that increased significantly after washing continued to do so during humid-drying (e.g. Brevundimonas, Rhizobium, Stenotrophomonas, Caulobacter) (Fig. 3b; Additional file 1: Table S10). On the other hand, host(human)-associated bacteria that decreased significantly after washing continued to do so during humid-drying (e.g. Prevotella, Actinomyces, Haemophilus, Rothia, Streptococcus, Schaalia, Cutibacterium, Anaerococcus) (Fig. 3b; Additional file 1: Table S10). In addition, other host-associated bacteria that did not change significantly after washing, decreased significantly after humid-drying (e.g. Micrococcus [found in both skin and environmental sources], Dermacoccus, Kocuria, Lactobacillus) (Fig. 3b; Additional file 1: Table S10). For further information on the metagenomic species associated with wearing, washing, dry-drying and humid-drying processes, we refer to Additional file 1: Table S11-14 and Additional file 2: Fig. S7.

Fig. 4

Bacterial exchange in household washing machines. a Graphical overview of the bacterial composition (16S rRNA gene-based genera) of axilla, T-shirt, and washing machine samples throughout a wear-wash-dry cycle. Estimations of source contributions to the b worn and c unworn washed T-shirt samples using fast expectation–maximisation for microbial source tracking (FEAST). Source estimates considering three known sources (A_T1, WM and WM_in) using amplicon OTU-level data. Legend: A_T1: axilla after wearing the T-shirt for 48 h; WM: washing machine rubber door seal; WM_in: washing machine influent water

We additionally applied MaAsLin2 modelling for comparison of the functional composition of the samples throughout the wear-wash-dry cycle. This analysis resulted in few significant associations, with none indicating a direct role in the generation of the malodorous VOCs identified in Table 1 (Additional file 2: Fig. S8; Additional file 1: Table S15-18).

Bacterial exchange occurs during the washing process

In order to assess bacterial exchange between the worn and unworn T-shirts during washing, and between the T-shirts, the washing machine and the influent water, all samples were analysed using 16S rRNA gene amplicon sequencing (Fig. 1). Results highlighted that many microbes in the worn T-shirts were additionally present in the worn-washed and unworn-washed T-shirts (Fig. 4a). This confirms earlier reports that bacterial removal during washing is incomplete and, moreover, that transfer of bacteria between worn and unworn items occurs [7, 12, 24, 46].

To explore the source of bacteria (operational taxonomic unit (OTU)-level) present on washed worn and unworn T-shirts, we used fast expectation–maximisation microbial source tracking (FEAST) [47]. We found that the bacterial profile of the worn-washed T-shirts had the highest known contribution on average from OTUs derived from the axilla (23%) and the washing machine (23%), followed by the influent washing water (3%) (Fig. 4b; Additional file 2: Fig. S9a). The unworn-washed T-shirts had the highest known contribution on average from OTUs derived from the washing machine (25%), followed by the axilla (6%) and the influent washing water (6%) (Fig. 4c; Additional file 2: Fig. S9b). These results highlight that many environment-associated bacteria present on the washed T-shirts (worn and unworn) originate from the intrinsic microbial community of the washing machine and to a minor extent, from the influent washing water (Fig. 4).

Washing and humid-drying alters the pathogenome of clothing

In the metagenomic samples (n = 155), the average number of microbial reads per sample was 9 million reads (range 23 million—288 K). An average of 4% of the reads were assigned to virulence factor (VF) genes. The total VF gene abundance varied throughout the wear-wash-dry cycle, with the highest VF gene levels observed in the washed and humid-dried T-shirt samples (average 0.047). At the lower end of the spectrum were the worn T-shirt samples (average 0.034) (Fig. 5a). Based on VF gene abundance, sample origin explained 52.1% of the dissimilarity between the sample groups (adonis2; R = 0.521; p < 0.001; Additional file 2: Fig. S10a). In total, 574 VF genes were identified; however, only nine VF genes contributed to > 25% of the total VF abundance (Additional file 1: Table S19).

Fig. 5

Virulome and resistome variation throughout a wear-wash-dry cycle. Box plots of relative abundances of a Virulence factor (VF) gene reads and b antimicrobial resistance (AMR) gene reads, stratified by sample group. Pairwise comparisons were calculated using Wilcoxon signed rank test (n = 31 per group). Different letters indicate a significant difference between pairwise comparisons (FDR < 0.05). The horizontal box lines represent the first quartile, the median and the third quartile. Whiskers denote the range of points within the first quartile − 1.5 × the interquartile range and the third quartile + 1.5 × the interquartile range. Each dot indicates a sample. c Relative VF abundance per VF class and d relative abundance of the 10 most common AMR genes, stratified by sample group (n = 31 per group). Legend: A_T1: axilla after wearing the T-shirt for 48 h, T: worn T-shirt, T_w: worn washed T-shirt, T_dd: dry-dried worn T-shirt, T_hd: humid-dried worn T-shirt. AMR gene abbreviations: RND resistance-nodulation-cell division; MDR multidrug resistance; ABC ATP-binding case; MFS major facilitator superfamily; rpoB rifamycin-resistant beta-subunit of RNA polymerase; vanR glycopeptide resistance gene cluster; ileS Antibiotic-resistant isoleucyl-tRNA synthetase; PEA phosphoethanolamine transferase

VF genes belonging to 13 classes were identified and the relative abundance of VF genes were aggregated to the corresponding virulence class level in each sample. Based on VF class abundance, sample origin explained 55.0% of the dissimilarity between the sample groups (R = 0.550; p < 0.001; Additional file 2: Fig. S10b). The process of washing led to an increased abundance of adherence, antimicrobial activity, biofilm, effector delivery system and motility VF classes (Fig. 5c; Additional file 2: Fig. S11a). In contrast, the abundance of exoenzyme, and stress survival VF classes decreased (Fig. 5c; Additional file 2: Fig. S11a). In addition, humid-drying led to a further increase in the abundance of antimicrobial activity, biofilm and motility VF classes (Fig. 5c; Additional file 2: Fig. S11a). To test the degree to which the virulome was associated with the microbial community, Procrustes analysis (Additional file 2: Fig. S12) was performed, comparing ordinations of the VF gene abundance with the metagenomic genera abundance. Results showed that they correlated significantly (protest; corr = 0.901; p < 0.001).

Antimicrobial resistance (AMR) genes contributed to an average of 1.7% of the reads in the metagenomic samples. The total AMR gene abundance also varied across the wear-wash-dry cycle (Fig. 5b), with the highest AMR gene levels observed in the humid-dried T-shirts (average: 0.018), followed by the washed T-shirt and axilla samples (average: 0.017). At the lower end of the spectrum were the dry-dried T-shirts and worn T-shirts samples (average: 0.015). Based on AMR gene abundance, sample origin explained 61.6% of the dissimilarity between the sample groups (adonis2; R = 0.616; p < 0.001; Additional file 2: Fig. S13). In total, 320 AMR genes were identified, but only ten AMR genes contributed to > 80% of the total AMR abundance (Additional file 1: Table S20). The majority of these genes were more abundant in the axilla and worn T-shirt samples (Fig. 5d; Additional file 2: Fig. S11b) and their resistance mechanism included antibiotic efflux pumps and antibiotic alteration conferring protection against a variety of drug classes (Additional file 1: Table S20). One AMR gene in particular increased considerably after washing and continued to do so with humid-drying, a gene encoding for a membrane transporter belonging to the resistance-nodulation-cell division (RND) antibiotic efflux pump protein superfamily (Additional file 1: Table S20). This drug efflux pump confers intrinsic multidrug resistance to Gram-negative bacteria [48, 49]. Moreover, it plays an important role in virulence and biofilm formation since it can also transport toxins, dyes, detergents, lipids and molecules involved in quorum sensing [48,49,50]. In addition, the degree to which the resistome was associated with the microbial community was tested using Procrustes analysis (Additional file 2: Fig. S14). Ordinations of the AMR gene abundance were compared with the metagenomic genera abundance, showing that they correlated significantly (protest; corr = 0.843; p < 0.001).

Discussion

The results presented in our study confirm previous findings that a standard European household washing cycle without oxidising laundry products does not reduce the microbial load in clothing but shifts the textile microbial community [7, 12, 24, 46]. We show that the intact bacterial load was in fact increased in washed T-shirts compared to worn T-shirts, with Gram-negative environment-associated bacteria displacing typical Gram-positive skin commensals. Many of these environment-associated bacteria originate from the intrinsic microbial community of the washing machine and, to a minor extent, from the influent washing water. Therefore, our observations also support that microbial exchange occurs during the washing process between the laundry load, washing machine and influent washing water [7, 12, 46, 51]. This microbial exchange implies that both worn and unworn clothing acquire a ‘washing’ microbial community, leading to microbial growth and specific malodour generation trends during the subsequent drying process, as mirrored in a recent study in towels by Lam et al. [52]. Dry-drying conditions (lower temperature and relative humidity) led to a decrease in bacterial load, a slight improvement in odour parameters and a subtle shift of the microbial profile in clothing. In contrast, humid-drying (high temperature: 31.8 ± 0.8 °C, and relative humidity: 80 ± 5% RH) led to bacterial growth, worse odour parameters and a significant change of the textile microbial community. These observations could ultimately be related to how quickly and well the clothes dry. In dry conditions, the water in the textile evaporates faster and moisture in the air is less likely to settle in the fabric than in humid conditions in which clothing is known to develop a musty smell [12, 26]. Moreover, humid and high-temperature conditions create a warm and moist environment that promotes microbial growth, particularly Gram-negative bacteria [8, 53]. However, it is important to note that these observed effects could be attributed to elevated temperature, elevated humidity or a combination of both factors, as no tests were conducted isolating elevated temperature or elevated humidity to differentiate between these parameters.

Through olfactive screening, we identified ten malodorous VOCs that showed interesting trends throughout the wear-wash-dry cycle for several of the study participants. Other studies that have assessed malodorous VOCs in laundry have observed a broader array of compounds, including for example, dimethyl disulphide (DMDS) and dimethyl trisulfide (DMTS), which were not observed here. However, some of these other studies assessed only the chemical presence or absence of VOCs, irrespective of their odour detection threshold [37, 52, 54]. Since these VOCs were not present above their odour detection threshold in our study, they were not highlighted here. The remaining previously published work in this area has assessed the microbial production of malodorous VOCs in laundry, either in monoculture or in a multi-species mock model [18, 52, 54]. In these studies, model representatives of laundry associated microorganisms, or microorganisms that have been isolated directly from laundered textiles and washing machines, have been sub-cultured to higher concentrations in vitro, prior to the assessment of VOC generation [18, 54]. Such in vitro systems recreate a limited representation of the in vivo environment, where growth conditions and nutrient availability are very different, and where a complex multi-species microbial community structure exists in constant flux. In our study, we investigated only those VOCs that were above the odour detection threshold on textiles that were naturally soiled and contaminated during wearing, washing and drying. As a result, only those VOCs that specifically contribute to perceivable laundry malodour have been reported.

Of particular note in our study are 2- & 3-methylbutanoic acids (indistinguishable in our instrument-based analyses), 3-methylhex-2-enoic acid and n-pentanoic acid, which have previously been associated with axillary and laundry malodour [13, 20, 28, 37,38,39]. The compounds 2- & 3-methylbutanoic acids and 3-methylhex-2-enoic acid were identified in many of the worn T-shirts and likely resulted from the metabolism of sweat components by axilla commensals [14, 15, 45]. For example, 3-methylhex-2-enoic acid is carried to the skin surface in the apocrine secretions bound to proteins, where axilla microbes release it [16], while 2- & 3-methylbutanoic acids result from amino acid biodegradation [14, 15].

Washing the T-shirts resulted in the efficient removal of these three acids, however during humid-drying, 2- & 3-methylbutanoic acids re-appear, contributing to the post-wash malodour as well. It is possible that 2- & 3-methylbutanoic acids are regenerated via metabolic activity of the specific microbes that remain after washing. Our metagenomic data demonstrate that many axillary commensals, which are known to generate these acids, are removed during washing, however, many new bacteria are added. Furthermore, our flow cytometry results show a significant increase in membrane-intact bacterial load after washing and during humid-drying, suggestive of an increase in viable bacteria during these processes (though further investigation is required in order to confirm viability and metabolic activity of these membrane-intact cells). This implies that humid conditions favour the growth and metabolic activity of specific bacteria that produce 2- & 3-methylbutanoic acids. The same explanation may be applicable in the case of n-butanoic acid, which also appears to be regenerated during humid-drying in our results (albeit with lower overall frequency).

The very low incidence (one instance) of 3-methylhex-2-enoic acid in the humid-dried T-shirts suggests that this VOC in particular contributes to the pre-wash malodour only. The fact that it is not regenerated during humid-drying also suggests that the microbes that produce it are effectively removed during washing. In support of this, earlier reports demonstrate that 3-methylhex-2-enoic acid is produced by specific Gram-positive skin commensals, such as Corynebacterium and Micrococcus [16], which are reduced in relative abundance after washing, according to our results (Fig. 3a). This suggests that any new bacteria that are introduced to the washed T-shirts during laundering and continue to proliferate throughout humid-drying, lack the enzyme Nα-glutamine aminoacylase that is responsible for the release of 3-methylhex-2-enoic acid from its glutamine conjugate. Alternatively, it may be that its precursor is efficiently removed by washing, rendering it unavailable for bioconversion.

Conversely, our results show that n-octanal appeared as a component of a post-wash malodour, occurring with highest frequency immediately after laundry. Others have also described an immediate post-wash malodour; however, no specific VOCs appeared elevated compared to pre-washed and post-wash dried counterparts in their study, and n-octanal was not among the VOCs identified [37]. One possible explanation for our result is that this VOC is generated during the conditions of the washing process via a combination of biotic and abiotically driven lipid oxidation reactions. Octanal is most likely present as a by-product of lipid autoxidation; however, enzyme-mediated (lipoxygenase) catalysis of fatty acids into hydroperoxides also leads to its secondary generation [55]. Evidence in the literature indicates that oleic acid, a component of sebum, may be transformed into octanal and that Pseudomonas aeruginosa is able to express lipoxygenases with oleic acid specificity [56]. Our results demonstrate an increased bacterial load and a significant increase in relative abundance of Pseudomonas on washed T-shirts. Thus, it is possible that octanal is at least in part, generated via microbial activity on residual sebum components present on the washed T-shirts (though we did not assess sebum retention). Others have also identified octanal with an associated fatty odour in the washing machine [54], which is additionally known to host Pseudomonas aeruginosa [57]. Therefore, the washing machine itself may contribute to the presence of octanal on washed T-shirts via two mechanisms: direct transfer of the VOC subsequent to its generation by the washing machine microbial community, and/or transfer of microorganisms to the washed T-shirt, that subsequently generate octanal via biodegradation of sebum components in situ.

Alternatively, although only those VOCs that were perceivable as malodorous were investigated and further identified in our study, at specific concentration, octanal is also a known fragrance constituent. Therefore, while its origin is not clear-cut, it is also important to recognise that its presence on the washed T-shirt could be a result of carryover from personal or home-care fragrance products used prior to the study period [54].

The reduction in the incidence of n-octanal in humid-dried T-shirts is a curious feature, which could be the result of a shift in microbial community members that outcompete those responsible for biological n-octanal generation. Alternatively, a shift in the microbial community may lead to the introduction of microorganisms that further biodegrade n-octanal. Otherwise, its reduction may simply be explainable by its volatility in humid conditions over 72 h, leading to its reduction below the olfactory detection threshold, or a reduction in lipid autoxidation rate associated with high humidity [58].

The results of our functional analysis show that microbial functions were more conserved than the microbial community members throughout the wear-wash-dry cycle and therefore no obvious mechanisms of malodour generation could be deduced at any individual stage. This observation might partly be because we controlled the redundancy effect in pathway abundances explainable by at most a single taxon, by retaining only those features with low correlation with individual microbial genera abundances. Alternatively, it may indicate that the pathways involved in malodour generation are universal to an array of taxa supported by polyester T-shirts and that it is their transcriptional control that is altered under varied environmental conditions throughout the wear-wash-dry cycle. This has been demonstrated to some extent already by Jacksch et al. [51]. Furthermore, it is also possible that malodour generation in laundered T-shirts results from numerous combinations of multi-species metabolic pathways, in addition to abiotic transformations, whereby the product of one transformation acts as the substrate to another. Such complex interactions would necessitate additional data collection (e.g. transcriptomics) and more sophisticated data handling methods (such as unsupervised machine learning) than those employed here, as well as larger datasets to fully untangle. Nonetheless, the connection between the occurrence of malodour and microorganisms was evident in our analyses from the association of higher bacterial cell counts and worse malodour in the T-shirts, in agreement with previous reports [54].

Interestingly, our results show for the first time that washing and humid-drying increase the virulome and resistome in clothing. Others have identified individual AMR genes in household dishwashers, washing machines and laundered textiles; for example, Rehberg et al. and Schages et al. identified and quantified the prevalence of genes associated with class I integrons and β-lacatamase resistance [59, 60]. However, our results report the prevalence of all known AMR and VF genes present on worn, washed and dried clothing. This information was available in our study due to the nature of our metagenomics approach, which aimed to achieve high-resolution identification of microbial communities by employing whole genome sequencing. In doing so, we additionally obtained information detailing the pathogenomic profile of laundry. Our data confirm earlier findings of AMR genes on laundered textiles (albeit different genes) and additionally report the prevalence of VF genes. Throughout the laundry process, the prevalence of both AMR and VF genes altered, likely reflecting the observed changes in the relative abundance of microbial community members present.

One important aspect of our findings is the increase in abundance of known biofilm-forming bacteria and biofilm-associated genes in washed and humid-dried T-shirts. Biofilm formation facilitates microbial survival during unfavourable conditions, e.g. in low nutrient environments and in the presence of detergents; such as that on washed T-shirts. In the context of laundry, biofilms also facilitate bacterial adhesion to fabric. Our data indicate that many bacteria present on the washed T-shirts originated from the washing machine and influent water, and previous work has highlighted that some of those bacteria that are introduced during washing are known to form biofilms in washing machines (for example Pseudomonas putida, Pseudomonas aeruginosa, Brevundimonas diminuta, Brevundimonas vesicularis, Sphingomonas paucimobilis, and Stenotrophomonas maltophilia [51, 54, 61,62,63]). Furthermore, household influent water is a known reservoir for horizontal transfer of virulence genes associated with biofilm formation [33, 34, 64]. It is possible, therefore, that the introduction of biofilm-forming bacteria and biofilm-associated genes during washing could result in the formation of biofilms on washed T-shirts. During successive wear-wash-dry cycles, biofilms containing resilient bacterial species may build up on colonised textiles, directly contributing to the “permastink” (the permanent malodour in clothing) phenomenon [1, 29, 30]. Such build-up has been demonstrated on towels by Kato et al., though was not linked to malodour and remains to be explored in washed T-shirts [65].

Conclusions

In summary, this study represents a multidisciplinary approach combining sensory analysis, bacterial load measurements, metagenomics and odour-directed chemical characterisation of VOCs on synthetic clothing throughout the wear-wash-dry cycle. Our results demonstrate the exchange of specific bacteria during washing and imply that both worn and unworn clothing acquire a “washing” microbiome.

Furthermore, we show for the first time that the introduction of environment-associated bacteria during laundry leads to an altered resistome and virulome on textiles (Fig. 5). This may contribute to biofilm formation on laundered clothes, which are known to enhance microbial survival and may participate in “permastink”. Traditional approaches to eradicate bacteria during laundering include high-temperature washing or the use of antibacterial agents. However, the former is incompatible with modern textiles and while the latter may reduce malodour [17, 52], evidence indicates that horizontal gene transfer is increased, which could facilitate the propagation of AMR and VF genes [66]. Instead, the optimisation of enzyme-based technologies could offer a greener and more attractive route via dissipation of biofilms. Overall, our observations provide critical knowledge, forming the basis for the development of strategies to control microbial survival and malodour generation in standard household laundry processes.

Methods

Study design and participants

Participants were recruited from the Flemish Region of Belgium in 2019. Participants’ inclusion criteria were as follows: (i) clear unpleasant odour evaluation of the axilla and clothing during pre-screening, (ii) presence of malodorous bacteria in the underarms (Anaerococcus, Corynebacterium, Moraxella and Peptoniphilus) during pre-screening via 16S rRNA gene sequencing and (iii) based on a questionnaire including odour self-assessment. From the initial 32 participants, 31 finished the study (participant number 30 dropped out). These included 12 European females (39%) and 19 European males (61%) aged 21–66 years old. A summary of the participants’ metadata is provided in Additional file 1: Table S1. All study participants provided written informed consent and all investigations adhered to Ghent University Ethical Approval B670201940586.

The study was set up as shown in Fig. 1. All participants were provided with deodorant and unscented shower gel to be used for the 5 days before the start of the study. After those 5 days, samples were taken from the participants’ axillae (T0). At that time, participants were given a synthetic white T-shirt (93% polyester, 7% elastane) to wear for 48 h, without taking it off, washing themselves or using deodorant or perfume. We chose such a T-shirt composition because washed polyester exhibits higher odour intensities than washed natural textiles, without differences in microbial load [13, 20, 21]. After 48 h, samples were taken from the participants’ axillae and the axillary region of the worn T-shirts (T1). After sampling, participants returned home to wash the worn T-shirt. They were instructed to use a wool cycle on their washing machines, selecting a cycle that most closely matched 30 °C for approximately 1 h with a spin speed of 600 rpm. Due to variations in washing machine brands and their cycle parameters, the duration ranged from 0.5 to 1.5 h. For laundry detergents and fabric conditioner, participants used non-fragranced, non-bio laundry detergent (5–15%: Potassium Cocoate, Nonionic Surfactant, Below 5%: Anionic Surfactant, Sodium Citrate, Amphoteric Surfactant, Citric Acid) (BioD, Hull, UK) and non-fragranced fabric conditioner (5–15% Cationic Surfactant, Less than 5%: Citric Acid) (BioD, Hull, UK). The use of fabric conditioner was included to reflect typical laundry practices in Europe. To this wash the following garments were added: an unworn synthetic white T-shirt (93% polyester, 7% elastane) provided by the researchers, five pairs of socks, five pairs of underwear, one pair of trousers and five T-shirts, either from the participant or other household members. Moreover, samples were taken from the incoming washing water, the rubber seal in the washing machine’s door opening, and the first washing water leaving the washing machine (T2). The rubber door seal in a washing machine was used due to its moisture retention, accumulation of detergent and dirt residues, and contact with all laundry loads. These factors create a nutrient-rich environment that fosters microbial growth, making the seal a representative and persistent hotspot for microbial activity in the washing machine. Both the worn and unworn T-shirts provided by the researchers were brought back immediately after washing for sampling the axillary region (T3). Afterward, both T-shirts were cut in half: one half was dried in a “dry” environment (50% RH, 24.5 ± 0.2 °C) and the other half was dried in a “humid” environment (80 ± 5% RH, 31.8 ± 0.8 °C) for 72 h; after which again samples were taken from the axillary region of the worn and unworn T-shirts (T4). The T-shirts were dried in a controlled, enclosed chamber tent.

Sampling methods

Axilla samples were taken by scraping both left and right axillary regions of the participants with cotton swabs (Novolab, Belgium) for 20 s on each side. One cotton swab was stored at − 20 °C until DNA extraction and the other at 4 °C and in 1-mL sterile PBS buffer (Sigma-Aldrich, Belgium) until flow cytometry measurements. After wearing the T-shirt for 48 h, participants wore a cotton pad (Novolab, Belgium) in the left axillary region for 5 h for odour assessment.

T-shirt samples were taken by cutting out 9 cm² of both left and right axillary regions of the T-shirts. One square centimetre from each axillary region was used for DNA extraction (stored at − 20 °C), 1 cm² from each axillary region was used for flow cytometry measurements (stored at 4 °C in 1 mL PBS) and 7 cm² was used for gas chromatography–olfactometry-mass spectrometry (GC-O-MS) analysis (stored at − 80 °C). For the dry-dried T-shirts (T_dd and Tb_dd), 2 cm² from the left and right axillary regions were used for DNA extraction and flow cytometry measurements due to lower cell counts.

The rubber seal in the door opening of the washing machine was sampled by scraping with two cotton swabs (Novolab, Belgium) for 30 s. One cotton swab was kept at − 20 °C until DNA extraction and the other at 4 °C and in 1-mL sterile PBS (Sigma-Aldrich, Belgium) until flow cytometry measurements. The washing water was sampled using sterile 1-L buckets (Novolab, Belgium) and kept at 4 °C until downstream analysis. The influent water was concentrated by vacuum filtration, using a 0.2-µm polyethersulfone filter (VWR, Belgium). Subsequently, DNA extraction was performed on the filter. The effluent water was centrifuged (5 min, maximum speed), and DNA extraction was performed on the pellet.

Sensory panel odour assessment

The sensory panel consisted of 10 individuals who were selected based on their sensitivity to dilutions of n-butanol and the triangle test at Ghent University (Ghent, Belgium) [67].

Study samples were assessed for two odour parameters: the hedonic value which refers to the pleasantness of the odour (scale − 8 to +8) and the intensity which refers to the strength of the odour (scale 0 to 10). T-shirts and cotton pads (for axilla samples) were placed in a glass jar for at least 30 min before assessment.

Flow cytometry

Flow cytometry, widely used in microbial research [68,69,70], was utilised to assess bacterial load and viability from axillae and T-shirt samples. It was selected over more traditional methods, such as cultivation, due to its ability to provide additional, valuable information, such as viability staining, which allowed us to distinguish between intact and damaged microbial cells. Furthermore, cultivation for a large amount of samples is challenging. By consistently taking sections of the same size (1 cm²), we ensured reliable comparisons. This technique enables rapid, quantitative, and multiparametric analysis of individual cells, offering a comprehensive overview of cellular characteristics. The simultaneous measurement of multiple parameters significantly enhanced the depth and accuracy of our data.

The samples were diluted in 1 × PBS (0.01 M phosphate buffered saline (NaCl 0.138 M; KCl 0.0027 M), pH 7.4)). For viability staining, we incubated diluted samples with SYBR Green I (100X concentrate, Invitrogen) combined with propidium iodide (50 × 20 mM, Invitrogen, in 0.22 μm-filtered dimethyl sulfoxide) for 20 min at 37 °C. Samples were analysed immediately after incubation. A BD Accuri C6 or C6 + flow cytometer (BD Biosciences, Belgium) was used, which was equipped with four fluorescence detectors (530/30 nm, 585/40 nm, > 670 nm and 675/25 nm), two scatter detectors and a 20-mW 488-nm laser. The flow cytometer was operated with Milli-Q water (MerckMillipore, Belgium) as sheath fluid. Both membrane-intact and damaged cell concentrations were quantified to assess the effect of the wear-wash-dry cycle.

DNA extraction

DNA extraction was performed by a combination of chemical and mechanical lysis as described by Geirnaert et al. [71]. The final DNA pellet was dried and suspended in 100 µL 1X TE buffer (10 mM Tris, 1 mM EDTA). After finishing the extraction protocol, DNA samples were immediately stored at − 20 °C until further analysis. Quality of DNA samples was analysed by 2% (w/v) agarose gel electrophoresis. Accuracy and reliability of the results were ensured by including a blank control sample during the DNA extraction and PCR process. This allowed us to confirm that any amplification observed in the PCR was due to DNA in the sample, and not from contaminating sources.

16S rRNA gene amplicon sequencing

DNA samples were sent to Biofidal (Vaulx-en-Velin, France) for library preparation and sequencing on an Illumina Miseq platform in PE 2 × 300 bp using V3-V4 chemistry. The primers used were 341F (5′-CCTACGGGNGGCWGCAG-3′) and 785Rmod (5′-GACTACHVGGGTATCTAAKCC-3′). The Mothur software package (v.1.42.3) [72] and guidelines were used to process the amplicon data generated by Biofidal, as described in detail by De Paepe et al. [73].

Shotgun metagenomic sequencing

A total of 155 samples were selected for shotgun metagenomic sequencing. These samples included the axilla after wearing the T-shirt for 48 h, worn T-shirt, worn-washed T-shirt, dry-dried worn T-shirt and humid-dried worn T-shirt for each of the 31 study participants. DNA samples were sent to CosmosID Inc. (Rockville, MD, USA) for library preparation and sequencing. DNA libraries were prepared using the Nextera XT DNA Library Preparation Kit (Illumina) and Nextera Index Kit (Illumina) with total DNA input of 1 ng. Genomic DNA was fragmented using a proportional amount of Nextera XT fragmentation enzyme (Illumina). Unique or combinatory dual indexes were added to each sample followed by 12 cycles of PCR to construct libraries. DNA libraries were purified using AMpure magnetic Beads (Beckman Coulter) and eluted in EB buffer (Qiagen). DNA libraries were quantified using Qubit 4 fluorometer and Qubit™ dsDNA HS Assay Kit (Thermo Fisher). Libraries were sequenced on an Illumina HiSeq 4000 2 × 150 bp.

Metagenomic data processing

Pre-processing

Adapter sequences were first trimmed from the reads of the 155 metagenomic samples using Trimmomatic (v.0.39) [74]. Unpaired sequences and those < 30 bp in length were dropped and a rudimentary quality trimming was conducted using a sliding window of size 4 requiring an average Phred score of 15.

Taxonomic annotation

Assembly-free taxonomic annotation was performed using Kraken2 [75] against the entire RefSeq database supplemented with a number of fungal genomes known to be associated with malodour (Additional file 1: Table S21). These genomes, when not already present in RefSeq, were gathered from JGI’s MycoCosm [76]. All human reads were filtered out. To retrieve a relative abundance matrix over all samples, a Bayesian re-estimation of the taxonomic counts as outputted by Kraken2 was conducted using Bracken [77].

Functional annotation

To characterise the functional potential of the metagenomic communities, a marker gene-based functional annotation was performed using the HUMAnN pipeline [78]. The resulting single-sample tables for gene family and pathway abundance were concatenated and sum-normalised to adjust for differences in sequencing depth.

Aside from the general functional annotation using HUMAnN, specialised gene family alignments were conducted using Diamond BLASTX [79] to annotate virulence factors (VF) and antimicrobial resistance (AMR) elements. The BLASTX searches were run against the 2022 VFDB database [80] and the CARD database [81], respectively. The e-value thresholds to call a hit were based on the median e-value of BLASTX searches of 10 samples whose reads were shuffled to retain nucleotide proportions but in randomised order. The lowest e-value per read was kept and sequence variants within gene families were aggregated to result in VF and AMR element abundance tables.

Gas chromatography–olfactometry-mass spectrometry

VOC Sampling

The fabric swatches were placed into individual inert compartments in a microchamber thermal extractor (Markes International, Bridgend, UK) set to 35 °C with a flow rate of 50 mL/min nitrogen. The headspace VOCs from each sample were collected onto individual inert coated ¼” thermal desorption tubes packed with sulficarb sorbent (C2-EAXX-5314, Markes International, Bridgend, UK) over a period of 3 h. Tubes were thermally conditioned prior to collection (held at 280 °C for 45 min in a 50 mL/min nitrogen stream). The microchamber compartments were cleaned between samples using propan-2-ol soaked lint-free tissues followed by heating at 200 °C for 2 h.

Thermal desorption conditions

Samples were introduced by means of a thermal desorption system (TD-100xr, Markes International Ltd, Bridgend, UK). Sample tubes were pre-purged with helium carrier gas (5 min at 50 mL/min), desorbed (120 °C for 2 min followed by 300 °C for 10 min with a flow of 50 mL/min) onto an internal sorbent bed (sulficarb, C2-EAXX-5314, Markes International, Bridgend, UK). The internal sorbent bed was held constant at 25 °C. The volatiles were released from the sorbent bed by purging (3 min at 50 mL/min) and subsequent heating at 16 °C/min to 300 °C (hold for 3 min) with a split flow rate of 5 mL/min. The transfer line to the gas chromatograph was held constant at 120 °C.

Gas chromatograph configuration

Analysis by two-dimensional gas chromatography was conducted using an Agilent 7890A gas chromatograph, fitted with a purged two-way microfluidic splitter plate after the primary column. The effluent was split between an odour assessment port (ODP3, Gerstel GmbH, Mulheim an der Ruhr, Germany) and a flow modulator (Insight, SepSolve Analytical, Peterborough, UK) with a non-purged two-way microfluidic splitter plate coupled to a flame ionisation detector and a time-of-flight mass spectrometer with electron impact ionisation (BenchTOF, SepSolve Analytical, Peterborough, UK).

The column configuration comprised a SolGel-Wax (Trajan, Australia) 30 m × 0.25 mm × 0.25 μm primary column and a 100% poly dimethyl siloxane 5 m × 0.25 mm × 0.1 μm DB-1MS (Agilent J&W, US) as the secondary column. Helium was used as the carrier gas, and the primary and secondary column flow rates were kept constant at 1.25 and 20 mL/min, respectively. The modulation period was set to 3 s, with a fill and flush (load and inject) time of 2.85 s and 0.15 s, respectively. The restrictor from the post-primary split outlet port of the splitter plate to the odour detection port was 3.5 m × 0.1 mm deactivated fused silica, with a constant flow of 1.1 mL/min. The restrictor from the second outlet port of the post primary column splitter plate was 1 m × 0.1 mm deactivated fused silica with a constant flow of 0.25 mL/min. The restrictor from the first outlet port of the second post flow modulator splitter to the flame ionisation detector was 0.8 m × 0.32 mm deactivated fused silica, with a constant flow of 15.6 mL/min and the restrictor from the splitter plate to the mass spectrometer was 1 m × 0.18 mm deactivated fused silica with a constant flow of 4.4 mL/min. The gas chromatograph was held at 40 °C (0.5 min) then programmed to 250 °C at 5 °C/min, held for 20 min.

Chemical detection conditions

The flame ionisation detector was held constant at 300 °C, the make-up gas was purified nitrogen with a flow rate of 5 mL/min, with a purified air flow rate of 400 mL/min and 40 mL/min of hydrogen (BOC, Guildford, UK). Data were collected at 100 Hz. The mass spectrometer transfer line and ion source temperatures were held constant at 280 °C and 200 °C, respectively. The mass scan range was m/z 35–450 at an acquisition rate 100 Hz.

Olfactive assessment

Identification of sensory features was conducted by trained panellists who were part of the trained fragrance sensory panel at Givaudan Ashford (UK). The members of the panel were selected based on their olfactory and gustatory sensory acuity and then trained for a period of 4–6 months to enable them to discriminate between compounds and score consistently. Odours were presented via a heated transfer line held at 175 °C connected to a glass nose port at ambient temperature. Each sample was assessed in duplicate by a combination of two panellists in each run, alternating at 12-min intervals to prevent olfactory fatigue over the 60-min run duration. The chromatographic location of each sensory event was recorded.

Given that the present study is concerned with understanding the mechanistic routes of malodour generation, the primary task was to identify which materials contribute to the overall malodour profile. This requirement directed our focus towards compounds which were as follows: (i) odour active, (ii) present in sufficient quantities above their odour detection threshold to be perceivable and (iii) malodorous in olfactive character. Only those molecules that satisfied all three criteria (and therefore influenced the overall malodour profile) were characterised. Results were further triaged to remove known environmental contaminants and instrument system artefacts. Fragrance materials were also automatically excluded.

Data collection and processing software

Data were acquired in ChromSpace (SepSolve Analytical, Peterborough, UK), Gerstel ODP Recorder (standalone, Gerstel, GmbH, Mulheim an der Ruhr, Germany) and EyeQuestion (EyeQuestion, Netherlands). The data were processed using ChromSpace and ChromCompare + (SepSolve Analytical, Peterborough, UK). Mass spectral identifications were made with in-house (Givaudan UK Ltd) and commercial libraries (NIST20).

Statistical analyses

Sample correlations and dissimilarities

Normality of data was checked with the Shapiro–Wilk test. Since the data were non-normally distributed, we used the Wilcoxon signed rank test to assess differences throughout the wear-wash-dry cycle between paired samples. For correlation analysis, Spearman-Rho correlation coefficient was used. P values were corrected for multiple hypotheses testing using Benjamini-Hochberg (BH) method.

We performed multivariate analysis of variance with the adonis2 function in the vegan R-package (2.6–4) [82] to test for effect size and significance of predictor variables on Bray–Curtis dissimilarity matrices. To account for the potential inter-individual effect, the model included study participant as a predictor variable, followed by the sample origin variable (distance_matrix ~ participant + origin). The tests were run for 999 default permutations, p values were corrected for multiple hypothesis testing using BH method and adjusted R squared were calculated using the varpart function (vegan). Bray–Curtis dissimilarity matrices were calculated with the vegdist function (vegan) on relative abundance of a variety of data including amplicon genera, metagenomic genera, functional DNA pathways, virulome and resistome. Dissimilarity matrices were subjected to classical multidimensional scaling (PCoA) to obtain the principal coordinates and the variance explained by each coordinate.

Fast expectation–maximisation for microbial source tracking

We used fast expectation–maximisation for microbial source tracking (FEAST) [47] to determine the potential origin of the OTUs present in the worn and unworn-washed T-shirt samples. We considered the axilla, the washing machine and the influent washing water as potential known sources.

Procrustes analysis

The protest function in the vegan R-package (2.6–4) [83] was used to test the strength of the association between two ordinations by scaling and rotating one ordination onto a target ordination. The ordinations were principal coordinates (PCoA) of the Bray–Curtis dissimilarity matrix calculated on relative abundance of features. We tested the strength of association between the metagenomic genera (target) and the virulome, resistome and amplicon genera (rotations). For the comparison between metagenomic genera and functional DNA pathways, we set the latter as the target since pathways do not necessarily depend on present taxa.

Differentially abundant taxonomic and functional features

We applied the Microbiome Multivariable Associations with Linear Models (MaAsLin2) [41] method to identify microbial features associated with each step of the wear-wash-dry cycle in textile clothing. For that, we used a variety of data types including metagenome-based taxonomic profiles and metagenome functional profiles. For each of these data modalities, independent filtering was performed before downstream testing to reduce the effect of zero inflation on the subsequent inference. Specifically, features for which the variance across all samples was very low (below half the median of all feature-wise variances), with abundance across all samples < 0.01% or with > 90% zeros, were removed. To further remove the effect of redundancy in pathway abundances (explainable by at most a single taxon), only features with low correlation with individual microbial genera abundances (Spearman correlation coefficient < 0.5) were retained. We applied default method implementation using a log-transformed linear model on TSS-normalised quality-controlled data, accounting for inter-individual variability by specifying between-participant random effects in the model. All associations were corrected for multiple hypotheses testing using BH method, declaring significant associations at a target of FDR 0.05.

Graphics and statistics

Plots were created using ggplot2 R-package (v.3.3.6) [83]. Statistical analyses were carried out in R software (v.4.2.0). For those analyses that included some element of randomness, we used the initial seed ‘123’.