Skin care and hygiene products persist on the skin
Systematic strategies to influence both the skin chemistry and microbiome have not yet been investigated. The outermost layer of the skin turns over every 3 to 4 weeks [8, 9]. How the microbiome and chemistry are influenced by altering personal care and how long the chemicals of personal care products persist on the skin are essentially uncharacterized. In this study, we collected samples from skin of 12 healthy individuals—six males and six females—over 9 weeks. One female volunteer had withdrawn due to skin irritations that developed, and therefore, we describe the remaining 11 volunteers. Samples were collected from each arm, armpit, foot, and face, including both the right and left sides of the body (Fig. 1a). All participants were asked to adhere to the same daily personal care routine during the first 6 weeks of this study (Fig. 1b). The volunteers were asked to refrain from using any personal care product for weeks 1–3 except a mild body wash (Fig. 1b). During weeks 4–6, in addition to the body wash, participants were asked to apply selected commercial skin care products at specific body parts: a moisturizer on the arm, a sunscreen on the face, an antiperspirant on the armpits, and a soothing powder on the foot (Fig. 1b). To monitor adherence of participants to the study protocol, molecular features found in the antiperspirant, facial lotion, moisturizer, and foot powder were directly tracked with mass spectrometry from the skin samples. For all participants, the mass spectrometry data revealed the accumulation of specific beauty product ingredients during weeks 4–6 (Additional file 1: Figure S1A-I, Fig. 2a orange arrows). Examples of compounds that were highly abundant during T4–T6 in skin samples are avobenzone (Additional file 1: Figure S1A), dexpanthenol (Additional file 1: Figure S1B), and benzalkonium chloride (Additional file 1: Figure S1C) from the facial sunscreen; trehalose 6-phosphate (Additional file 1: Figure S1D) and glycerol stearate (Additional file 1: Figure S1E) from the moisturizer applied on arms; indolin (Additional file 1: Figure S1F) and an unannotated compound (m/z 233.9, rt 183.29 s) (Additional file 1: Figure S1G) from the foot powder; and decapropylene glycol (Additional file 1: Figure S1H) and nonapropylene glycol (Additional file 1: Figure S1I) from the antiperspirant. These results suggest that there is likely a compliance of all individuals to study requirements and even if all participants confirmed using each product every day, the amount of product applied by each individual may vary. Finally, for weeks 7–9, the participants were asked to return to their normal routine by using the same personal care products they used prior to the study. In total, excluding all blanks and personal care products themselves, we analyzed 2192 skin samples for both metabolomics and microbiome analyses.
To understand how long beauty products persist on the skin, we monitored compounds found in deodorants used by two volunteers—female 1 and female 3—before the study (T0), over the first 3 weeks (T1–T3) (Fig. 1b). During this phase, all participants used exclusively the same body wash during showering, making it easier to track ingredients of their personal care products. The data in the first 3 weeks (T1–T3) revealed that many ingredients of deodorants used on armpits (Fig. 2a) persist on the skin during this time and were still detected during the first 3 weeks or at least during the first week following the last day of use. Each of the compounds detected in the armpits of individuals exhibited its own unique half-life. For example, the polyethylene glycol (PEG)-derived compounds m/z 344.227, rt 143 s (Fig. 2b, S1J); m/z 432.279, rt 158 s (Fig. 2b, S1K); and m/z 388.253, rt 151 s (Fig. 2b, S1L) detected on armpits of volunteer 1 have a calculated half-life of 0.5 weeks (Additional file 1: Figure S1J-L, all p values < 1.81e−07), while polypropylene glycol (PPG)-derived molecules m/z 481.87, rt 501 s (Fig. 2c, S1M); m/z 560.420, rt 538 s (Fig. 2c, S1N); m/z 788.608, rt 459 s (Fig. 2d, S1O); m/z 846.650, rt 473 s (Fig. 2d, S1P); and m/z 444.338, rt 486 s (Fig. 2d, S1Q) found on armpits of volunteers 3 and 1 (Fig. 2a) have a calculated half-life ranging from 0.7 to 1.9 weeks (Additional file 1: Figure S1M-Q, all p values < 0.02), even though they originate from the same deodorant used by each individual. For some ingredients of deodorant used by volunteer 3 on time 0 (Additional file 1: Figure S1M, N), a decline was observed during the first week, then little to no traces of these ingredients were detected during weeks 4–6 (T4–T6), then finally these ingredients reappear again during the last 3 weeks of personal product use (T7–T9). This suggests that these ingredients are present exclusively in the personal deodorant used by volunteer 3 before the study. Because a similar deodorant (Additional file 1: Figure S1O-Q) and a face lotion (Additional file 1: Figure S1R) was used by volunteer 3 and volunteer 2, respectively, prior to the study, there was no decline or absence of their ingredients during weeks 4–6 (T4–T6).
Polyethylene glycol compounds (Additional file 1: Figure S1J-L) wash out faster from the skin than polypropylene glycol (Additional file 1: Figure S1M-Q)(HL ~ 0.5 weeks vs ~ 1.9 weeks) and faster than fatty acids used in lotions (HL ~ 1.2 weeks) (Additional file 1: Figure S1R), consistent with their hydrophilic (PEG) and hydrophobic properties (PPG and fatty acids) [25, 26]. This difference in hydrophobicity is also reflected in the retention time as detected by mass spectrometry. Following the linear decrease of two PPG compounds from T0 to T1, they accumulated noticeably during weeks 2 and 3 (Additional file 1: Figure S1M, N). This accumulation might be due to other sources of PPG such as the body wash used during this period or the clothes worn by person 3. Although PPG compounds were not listed in the ingredient list of the shampoo, we manually inspected the LC-MS data collected from this product and confirmed the absence of PPG compounds in the shampoo. The data suggest that this trend is characteristic of accumulation of PPG from additional sources. These could be clothes, beds, or sheets, in agreement with the observation of these molecules found in human habitats [27] but also in the public GNPS mass spectrometry dataset MSV000079274 that investigated the chemicals from dust collected from 1053 mattresses of children.
Temporal molecular and bacterial diversity in response to personal care use
To assess the effect of discontinuing and resuming the use of skin care products on molecular and microbiota dynamics, we first evaluated their temporal diversity. Skin sites varied markedly in their initial level (T0) of molecular and bacterial diversity, with higher molecular diversity at all sites for female participants compared to males (Fig. 3a, b, Wilcoxon rank-sum-WR test, p values ranging from 0.01 to 0.0001, from foot to arm) and higher bacterial diversity in face (WR test, p = 0.0009) and armpits (WR test, p = 0.002) for females (Fig. 3c, d). Temporal diversity was similar across the right and left sides of each body site of all individuals (WR test, molecular diversity: all p values > 0.05; bacterial diversity: all p values > 0.20). The data show that refraining from using beauty products (T1–T3) leads to a significant decrease in molecular diversity at all sites (Fig. 3a, b, WR test, face: p = 8.29e−07, arm: p = 7.08e−09, armpit: p = 1.13e−05, foot: p = 0.002) and bacterial diversity mainly in armpits (WR test, p = 0.03) and feet (WR test, p = 0.04) (Fig. 3c, d). While molecular diversity declined (Fig. 3a, b) for arms and face, bacterial diversity (Fig. 3c, d) was less affected in the face and arms when participants did not use skin care products (T1–T3). The molecular diversity remained stable in the arms and face of female participants during common beauty products use (T4–T6) to immediately increase as soon as the volunteers went back to their normal routines (T7–T9) (WR test, p = 0.006 for the arms and face)(Fig. 3a, b). A higher molecular (Additional file 1: Figure S2A) and community (Additional file 1: Figure S2B) diversity was observed for armpits and feet of all individuals during the use of antiperspirant and foot powder (T4–T6) (WR test, molecular diversity: armpit p = 8.9e−33, foot p = 1.03e−11; bacterial diversity: armpit p = 2.14e−28, foot p = 1.26e−11), followed by a molecular and bacterial diversity decrease in the armpits when their regular personal beauty product use was resumed (T7–T9) (bacterial diversity: WR test, p = 4.780e−21, molecular diversity: WR test, p = 2.159e−21). Overall, our data show that refraining from using beauty products leads to lower molecular and bacterial diversity, while resuming the use increases their diversity. Distinct variations between male and female molecular and community richness were perceived at distinct body parts (Fig. 3a–d). Although the chemical diversity of personal beauty products does not explain these variations (Additional file 1: Figure S2C), differences observed between males and females may be attributed to many environmental and lifestyle factors including different original skin care and different frequency of use of beauty products (Additional file 2: Table S1), washing routines, and diet.
Longitudinal variation of skin metabolomics signatures
To gain insights into temporal metabolomics variation associated with beauty product use, chemical inventories collected over 9 weeks were subjected to multivariate analysis using the widely used Bray–Curtis dissimilarity metric (Fig. 4a–c, S3A). Throughout the 9-week period, distinct molecular signatures were associated to each specific body site: arm, armpit, face, and foot (Additional file 1: Figure S3A, Adonis test, p < 0.001, R2 0.12391). Mass spectrometric signatures displayed distinct individual trends at each specific body site (arm, armpit, face, and foot) over time, supported by their distinct locations in PCoA (principal coordinate analysis) space (Fig. 4a, b) and based on the Bray–Curtis distances between molecular profiles (Additional file 1: Figure S3B, WR test, all p values < 0.0001 from T0 through T9). This suggests a high molecular inter-individual variability over time despite similar changes in personal care routines. Significant differences in molecular patterns associated to ceasing (T1–T3) (Fig. 4b, Additional file 1: Figure S3C, WR test, T0 vs T1–T3 p < 0.001) and resuming the use of common beauty products (T4–T6) (Additional file 1: Figure S3C) were observed in the arm, face, and foot (Fig. 4b), although the armpit exhibited the most pronounced changes (Fig. 4b, Additional file 1: Figure S3D, E, random forest highlighting that 100% of samples from each phase were correctly predicted). Therefore, we focused our analysis on this region. Molecular changes were noticeable starting the first week (T1) of discontinuing beauty product use. As shown for armpits in Fig. 4c, these changes at the chemical level are specific to each individual, possibly due to the extremely personalized lifestyles before the study and match their original use of deodorant. Based on the initial use of underarm products (T0) (Additional file 2: Table S1), two groups of participants can be distinguished: a group of five volunteers who used stick deodorant as evidenced by the mass spectrometry data and another group of volunteers where we found few or no traces suggesting they never or infrequently used stick deodorants (Additional file 2: Table S1). Based on this criterion, the chemical trends shown in Fig. 4c highlight that individuals who used stick deodorant before the beginning of the study (volunteers 1, 2, 3, 9, and 12) displayed a more pronounced shift in their armpits’ chemistries as soon as they stopped using deodorant (T1–T3), compared to individuals who had low detectable levels of stick deodorant use (volunteers 4, 6, 7, and 10), or “rarely-to-never” (volunteers 5 and 11) use stick deodorants as confirmed by the volunteers (Additional file 1: Figure S3F, WR test, T0 vs T1–T3 all p values < 0.0001, with greater distance for the group of volunteers 1, 2, 3, 9, and 12, compared to volunteers 4, 5, 6, 7, 10, and 11). The most drastic shift in chemical profiles was observed during the transition period, when all participants applied the common antiperspirant on a daily basis (T4–T6) (Additional file 1: Figure S3D, E). Finally, the molecular profiles became gradually more similar to those collected before the experiment (T0) as soon as the participants resumed using their personal beauty products (T7–T9) (Additional file 1: Figure S3C), although traces of skin care products did last through the entire T7–T9 period in people who do not routinely apply these products (Fig. 4c).
Comparing chemistries detected in armpits at the end timepoints—when no products were used (T3) and during product use (T6)—revealed distinct molecular signatures characteristic of each phase (random forest highlighting that 100% of samples from each group were correctly predicted, see Additional file 1: Figure S3D, E). Because volunteers used the same antiperspirant during T4–T6, molecular profiles converged during that time despite individual patterns at T3 (Fig. 4b, c, Additional file 1: Figure S3D). These distinct chemical patterns reflect the significant impact of beauty products on skin molecular composition. Although these differences may in part be driven by beauty product ingredients detected on the skin (Additional file 1: Figure S1), we anticipated that additional host- and microbe-derived molecules may also be involved in these molecular changes.
To characterize the chemistries that vary over time, we used molecular networking, a MS visualization approach that evaluates the relationship between MS/MS spectra and compares them to reference MS/MS spectral libraries of known compounds [29, 30]. We recently showed that molecular networking can successfully organize large-scale mass spectrometry data collected from the human skin surface [18, 19]. Briefly, molecular networking uses the MScluster algorithm [31] to merge all identical spectra and then compares and aligns all unique pairs of MS/MS spectra based on their similarities where 1.0 indicates a perfect match. Similarities between MS/MS spectra are calculated using a similarity score, and are interpreted as molecular families [19, 24, 32,33,34]. Here, we used this method to compare and characterize chemistries found in armpits, arms, face, and foot of 11 participants. Based on MS/MS spectral similarities, chemistries highlighted through molecular networking (Additional file 1: Figure S4A) were associated with each body region with 8% of spectra found exclusively in the arms, 12% in the face, 14% in the armpits, and 2% in the foot, while 18% of the nodes were shared between all four body parts and the rest of spectra were shared between two body sites or more (Additional file 1: Figure S4B). Greater spectral similarities were highlighted between armpits, face, and arm (12%) followed by the arm and face (9%) (Additional file 1: Figure S4B).
Molecules were annotated with Global Natural Products Social Molecular Networking (GNPS) libraries [29], using accurate parent mass and MS/MS fragmentation patterns, according to level 2 or 3 of annotation defined by the 2007 metabolomics standards initiative [35]. Through annotations, molecular networking revealed that many compounds derived from steroids (Fig. 5a–d), bile acids (Additional file 1: Figure S5A-D), and acylcarnitines (Additional file 1: Figure S5E-F) were exclusively detected in the armpits. Using authentic standards, the identity of some pheromones and bile acids were validated to a level 1 identification with matched retention times (Additional file 1: Figure S6B, S7A, C, D). Other steroids and bile acids were either annotated using standards with identical MS/MS spectra but slightly different retention times (Additional file 1: Figure S6A) or annotated with MS/MS spectra match with reference MS/MS library spectra (Additional file 1: Figure S6C, D, S7B, S6E-G). These compounds were therefore classified as level 3 [35]. Acylcarnitines were annotated to a family of possible acylcarnitines (we therefore classify as level 3), as the positions of double bonds or cis vs trans configurations are unknown (Additional file 1: Figure S8A, B).
Among the steroid compounds, several molecular families were characterized: androsterone (Fig. 5a, b, d), androstadienedione (Fig. 5c), androstanedione (Additional file 1: Figure S6E), androstanolone (Additional file 1: Figure S6F), and androstenedione (Additional file 1: Figure S6G). While some steroids were detected in the armpits of several individuals, such as dehydroisoandrosterone sulfate (m/z 369.19, rt 247 s) (9 individuals) (Fig. 5a, Additional file 1: Figure S6A), androsterone sulfate (m/z 371.189, rt 261 s) (9 individuals) (Fig. 5b, Additional file 1: Figure S6C), and 5-alpha-androstane-3,17-dione (m/z 271.205, rt 249 s) (9 individuals) (Additional file 1: Figure S6E), other steroids including 1-dehydroandrostenedione (m/z 285.185, rt 273 s) (Fig. 5c, Additional file 1: Figure S6B), dehydroandrosterone (m/z 289.216, rt 303 s) (Fig. 5d, Additional file 1: Figure S6D), and 5-alpha-androstan-17.beta-ol-3-one (m/z 291.231, rt 318 s) (Additional file 1: Figure S6F) were only found in the armpits of volunteer 11 and 4-androstene-3,17-dione (m/z 287.200, rt 293 s) in the armpits of volunteer 11 and volunteer 5, both are male that never applied stick deodorants (Additional file 1: Figure S6G). Each molecular species exhibited a unique pattern over the 9-week period. The abundance of dehydroisoandrosterone sulfate (Fig. 5a, WR test, p < 0.01 for 7 individuals) and dehydroandrosterone (Fig. 5a, WR test, p = 0.00025) significantly increased during the use of antiperspirant (T4–T6), while androsterone sulfate (Fig. 5b) and 5-alpha-androstane-3,17-dione (Additional file 1: Figure S6E) display little variation over time. Unlike dehydroisoandrosterone sulfate (Fig. 5a) and dehydroandrosterone (Fig. 5d), steroids including 1-dehydroandrostenedione (Fig. 5c, WR test, p = 0.00024) and 4-androstene-3,17-dione (Additional file 1: Figure S6G, WR test, p = 0.00012) decreased in abundance during the 3 weeks of antiperspirant application (T4–T6) in armpits of male 11, and their abundance increased again when resuming the use of his normal skin care routines (T7–T9). Interestingly, even within the same individual 11, steroids were differently impacted by antiperspirant use as seen for 1-dehydroandrostenedione that decreased in abundance during T4–T6 (Fig. 5c, WR test, p = 0.00024), while dehydroandrosterone increased in abundance (Fig. 5d, WR test, p = 0.00025), and this increase was maintained during the last 3 weeks of the study (T7–T9).
In addition to steroids, many bile acids (Additional file 1: Figure S5A-D) and acylcarnitines (Additional file 1: Figure S5E-F) were detected on the skin of several individuals through the 9-week period. Unlike taurocholic acid found only on the face (Additional file 1: Figures S5A, S7A) and tauroursodeoxycholic acid detected in both armpits and arm samples (Additional file 1: Figures S5B, S7B), other primary bile acids such as glycocholic (Additional file 1: Figures S5C, S7C) and chenodeoxyglycocholic acid (Additional file 1: Figures S5D, S7D) were exclusively detected in the armpits. Similarly, acylcarnitines were also found either exclusively in the armpits (hexadecanoyl carnitines) (Additional file 1: Figures S5E, S8A) or in the armpits and face (tetradecenoyl carnitine) (Additional file 1: Figures S5F, S8B) and, just like the bile acids, they were also stably detected during the whole 9-week period.
Bacterial communities and their variation over time
Having demonstrated the impact of beauty products on the chemical makeup of the skin, we next tested the extent to which skin microbes are affected by personal care products. We assessed temporal variation of bacterial communities detected on the skin of healthy individuals by evaluating dissimilarities of bacterial collections over time using unweighted UniFrac distance [36] and community variation at each body site in association to beauty product use [3, 15, 37]. Unweighted metrics are used for beta diversity calculations because we are primarily concerned with changes in community membership rather than relative abundance. The reason for this is that skin microbiomes can fluctuate dramatically in relative abundance on shorter timescales than that assessed here. Longitudinal variations were revealed for the armpits (Fig. 6a) and feet microbiome by their overall trend in the PCoA plots (Fig. 6b), while the arm (Fig. 6c) and face (Fig. 6d) displayed relatively stable bacterial profiles over time. As shown in Fig. 6a–d, although the microbiome was site-specific, it varied more between individuals and this inter-individual variability was maintained over time despite same changes in personal care routine (WR test, all p values at all timepoints < 0.05, T5 p = 0.07), in agreement with previous findings that individual differences in the microbiome are large and stable over time [3, 4, 10, 37]. However, we show that shifts in the microbiome can be induced by changing hygiene routine and therefore skin chemistry. Changes associated with using beauty products (T4–T6) were more pronounced for the armpits (Fig. 6a, WR test, p = 1.61e−52) and feet (Fig. 6b, WR test, p = 6.15e−09), while little variations were observed for the face (Fig. 6d, WR test, p = 1.402.e−83) and none for the arms (Fig. 6c, WR test, p = 0.296).
A significant increase in abundance of Gram-negative bacteria including the phyla Proteobacteria and Bacteroidetes was noticeable for the armpits and feet of both females (Fig. 6e; Mann–Whitney U, p = 8.458e−07) and males (Fig. 6f; Mann–Whitney U, p = 0.0004) during the use of antiperspirant (T4–T6), while their abundance remained stable for the arms and face during that time (Fig. 6e, f; female arm p = 0.231; female face p value = 0.475; male arm p= 0.523;male face p = 6.848751e−07). These Gram-negative bacteria include Acinetobacter and Paracoccus genera that increased in abundance in both armpits and feet of females (Additional file 1: Figure S9A), while a decrease in abundance of Enhydrobacter was observed in the armpits of males (Additional file 1: Figure S9B). Cyanobacteria, potentially originating from plant material (Additional file 1: Figure S9C) also increased during beauty product use (T4–T6) especially in males, in the armpits and face of females (Fig. 6e) and males (Fig. 6f). Interestingly, although chloroplast sequences (which group phylogenetically within the cyanobacteria [38]) were only found in the facial cream (Additional file 1: Figure S9D), they were detected in other locations as well (Fig. 6e, f. S9E, F), highlighting that the application of a product in one region will likely affect other regions of the body. For example, when showering, a face lotion will drip down along the body and may be detected on the feet. Indeed, not only did the plant material from the cream reveal this but also the shampoo used for the study for which molecular signatures were readily detected on the feet as well (Additional file 1: Figure S10A). Minimal average changes were observed for Gram-positive organisms (Additional file 1: Figure S10B, C), although in some individuals the variation was greater than others (Additional file 1: Figure S10D, E) as discussed for specific Gram-positive taxa below.
At T0, the armpit’s microflora was dominated by Staphylococcus (26.24%, 25.11% of sequencing reads for females and 27.36% for males) and Corynebacterium genera (26.06%, 17.89% for females and 34.22% for males) (Fig. 7a—first plot from left and Additional file 1: Figure S10D, E). They are generally known as the dominant armpit microbiota and make up to 80% of the armpit microbiome [39, 40]. When no deodorants were used (T1–T3), an overall increase in relative abundance of Staphylococcus (37.71%, 46.78% for females and 30.47% for males) and Corynebacterium (31.88%, 16.50% for females and 44.15% for males) genera was noticeable (WR test, p < 3.071e−05) (Fig. 7a—first plot from left), while the genera Anaerococcus and Peptoniphilus decreased in relative abundance (WR test, p < 0.03644) (Fig. 7a—first plot from left and Additional file 1: Figure S10D, E). When volunteers started using antiperspirants (T4–T6), the relative abundance of Staphylococcus (37.71%, 46.78% females and 30.47% males, to 21.71%, 25.02% females and 19.25% males) and Corynebacterium (31.88%, 16.50% females and 44.15% males, to 15.83%, 10.76% females and 19.60% males) decreased (WR test, p < 3.071e−05) (Fig. 7a, Additional file 1: Figure S10D, E) and at the same time, the overall alpha diversity increased significantly (WR test, p = 3.47e−11) (Fig. 3c, d). The microbiota Anaerococcus (WR test, p = 0.0006018), Peptoniphilus (WR test, p = 0.008639), and Micrococcus (WR test, p = 0.0377) increased significantly in relative abundance, together with a lot of additional low-abundant species that lead to an increase in Shannon alpha diversity (Fig. 3c, d). When participants went back to normal personal care products (T7–T9), the underarm microbiome resembled the original underarm community of T0 (WR test, p = 0.7274) (Fig. 7a). Because armpit bacterial communities are person-specific (inter-individual variability: WR test, all p values at all timepoints < 0.05, besides T5 p n.s), variation in bacterial abundance upon antiperspirant use (T4–T6) differ between individuals and during the whole 9-week period (Fig. 7a—taxonomic plots per individual). For example, the underarm microbiome of male 5 exhibited a unique pattern, where Corynebacterium abundance decreased drastically during the use of antiperspirant (82.74 to 11.71%, WR test, p = 3.518e−05) while in the armpits of female 9 a huge decrease in Staphylococcus abundance was observed (Fig. 7a) (65.19 to 14.85%, WR test, p = 0.000113). Unlike other participants, during T0–T3, the armpits of individual 11 were uniquely characterized by the dominance of a sequence that matched most closely to the Enhydrobacter genera. The transition to antiperspirant use (T4–T6) induces the absence of Enhydrobacter (30.77 to 0.48%, WR test, p = 0.01528) along with an increase of Corynebacterium abundance (26.87 to 49.74%, WR test, p = 0.1123) (Fig. 7a—male 11).
In addition to the armpits, a decline in abundance of Staphylococcus and Corynebacterium was perceived during the use of the foot powder (46.93% and 17.36%, respectively) compared to when no beauty product was used (58.35% and 22.99%, respectively) (WR test, p = 9.653e−06 and p = 0.02032, respectively), while the abundance of low-abundant foot bacteria significantly increased such as Micrococcus (WR test, p = 1.552e−08), Anaerococcus (WR test, p = 3.522e−13), Streptococcus (WR test, p = 1.463e−06), Brevibacterium (WR test, p = 6.561e−05), Moraxellaceae (WR test, p = 0.0006719), and Acinetobacter (WR test, p = 0.001487), leading to a greater bacterial diversity compared to other phases of the study (Fig. 7b first plot from left, Additional file 1: Figure S10D, E, Fig. 3c, d).
We further evaluated the relationship between the two omics datasets by superimposing the principal coordinates calculated from metabolome and microbiome data (Procrustes analysis) (Additional file 1: Figure S11) [34, 41, 42]. Metabolomics data were more correlated with patterns observed in microbiome data in individual 3 (Additional file 1: Figure S11C, Mantel test, r = 0.23, p < 0.001), individual 5 (Additional file 1: Figure S11E, r = 0.42, p < 0.001), individual 9 (Additional file 1: Figure S11H, r = 0.24, p < 0.001), individual 10 (Additional file 1: Figure S11I, r = 0.38, p < 0.001), and individual 11 (Additional file 1: Figure S11J, r = 0.35, p < 0.001) when compared to other individuals 1, 2, 4, 6, 7, and 12 (Additional file 1: Figure S11A, B, D, F, G, K, respectively) (Mantel test, all r < 0.2, all p values < 0.002, for volunteer 2 p n.s). Furthermore, these correlations were individually affected by ceasing (T1–T3) or resuming the use of beauty products (T4–T6 and T7–T9) (Additional file 1: Figure S11A-K).
Overall, metabolomics–microbiome correlations were consistent over time for the arms, face, and feet although alterations were observed in the arms of volunteers 7 (Additional file 1: Figure S11G) and 10 (Additional file 1: Figure S11I) and the face of volunteer 7 (Additional file 1: Figure S11G) during product use (T4–T6). Molecular–bacterial correlations were mostly affected in the armpits during antiperspirant use (T4–T6), as seen for volunteers male 7 (Additional file 1: Figure S11G) and 11 (Additional file 1: Figure S11J) and females 2 (Additional file 1: Figure S11B), 9 (Additional file 1: Figure S11H), and 12 (Additional file 1: Figure S11K). This perturbation either persisted during the last 3 weeks (Additional file 1: Figure S11D, E, H, I, K) when individuals went back to their normal routine (T7–T9) or resembled the initial molecular–microbial correlation observed in T0 (Additional file 1: Figure S11C, G, J). These alterations in molecular–bacterial correlation are driven by metabolomics changes during antiperspirant use as revealed by metabolomics shifts on the PCoA space (Additional file 1: Figure S11), partially due to the deodorant’s chemicals (Additional file 1: Figure S1J, K) but also to changes observed in steroid levels in the armpits (Fig. 5A, C, D, Additional file 1: Figure S6G), suggesting metabolome-dependant changes of the skin microbiome. In agreement with previous findings that showed efficient biotransformation of steroids by Corynebacterium [43, 44], our correlation analysis associates specific steroids that were affected by antiperspirant use in the armpits of volunteer 11 (Fig. 5c, d, Additional file 1: Figure S6G) with microbes that may produce or process them: 1-dehydroandrostenedione, androstenedione, and dehydrosterone with Corynebacterium (r = − 0.674, p = 6e−05; r = 0.671, p = 7e−05; r = 0.834, p < 1e−05, respectively) (Additional file 1: Figure S12A, B, C, respectively) and Enhydrobacter (r = 0.683, p = 4e−05; r = 0.581, p = 0.00095; r = 0.755, p < 1e−05 respectively) (Additional file 1: Figure S12D, E, F, respectively).