Combined metagenomic- and culture-based approaches to investigate bacterial strain-level associations with medication-controlled mild-moderate atopic dermatitis

Background: The skin microbiome is disrupted in atopic dermatitis (AD). Existing research focuses on moderate-severe, unmedicated disease. Objective: Investigate metagenomic- and culture-based bacterial strain-level differences in mild, medicated AD, and the effects these have on human keratinocytes (HK). Methods: Skin swabs from anterior forearms were collected from 20 pediatric participants; 11 participants with AD sampled at lesional and nonlesional sites and 9 age- and sex-matched controls). Participants had primarily mild-moderate AD and maintained medication use. Samples were processed for microbial metagenomic sequencing and bacterial isolation. Isolates identified as S. aureus were tested for enterotoxin production. HK cultures were treated with cell free conditioned media from representative Staphylococcus species to measure barrier effects. Results: Metagenomic sequencing identified significant differences in microbiome composition between AD and control groups. Differences were seen at the species- and strain-levels for Staphylococci, with S. aureus only found in AD participants and differences in S. epidermidis strains between control and AD swabs. These strains showed differences in toxin gene presence, which was confirmed in vitro for S. aureus enterotoxins. The strain from the most severe AD participant produced enterotoxin B levels >100-fold higher than the other strains (p<0.001). Strains also displayed differential effects on HK metabolism and barrier function. Conclusions: Strain level differences in toxin genes from Staphylococcus strains may explain varying effects on HK, with S. aureus and non-aureus strains negatively impacting viability and barrier function. These differences are likely important in AD pathogenesis.


University of Wisconsin School of Medicine and Public Health (AMS)
The other authors have no conflicts to declare. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 28, 2023.

111
Atopic dermatitis (AD) is the most common inflammatory skin disease in children, with recent 112 estimates suggesting as many as 1 in 5 children experience AD, with 80% exhibiting symptoms 113 within the first 6 years of life 1-3 . AD is characterized by recurrent eczematous lesional skin sites. 114 During flares, these erythematous patches are associated with pruritis, exudation, blistering, and 115 lichenification, which can cause loss of sleep, poor mental health, and decreases in school/work 116 performance 1,5 . It has been proposed that disruptions to the skin barrier during AD flares 117 promote AD pathogenesis 6,7 . Numerous causes have been proposed for this disruption of the skin 118 barrier, including genetic mutations (for example, in the filaggrin gene) 8 , immune dysregulation 119 (classically characterized by Th2 and IgE predominance) 4,6,7 , environmental triggers (such as 120 detergents and epicutaneous protease exposure) 9-11 , and alterations in the skin microbiome 121 (particularly overgrowth of Staphylococcus aureus) 12, 13 .

123
The microbiota plays a critical role in the maturation and function of the skin barrier 14 , including 124 pH balance 15-17 , water retention 18 , keratinocyte differentiation 19,20 , wound healing 21-24 , innate 25,26 125 and adaptive immune responses 16,27,28 , and pathogen competition [29][30][31] . Commensal skin bacteria, with staphylococcal enterotoxin B (SEB) in particular associated with more severe AD 36,37 . 132 However, existing studies, including those associating increased S. aureus with lesional skin, 133 have largely focused on moderate to severe participants while holding AD medications, and have 134 focused on the microbiome at the genus and species levels. Whether S. aureus plays a similarly 135 important role in less severe or well-controlled disease remains less well understood. 136 Additionally, variability at the strain-level in AD progression is only beginning to be addressed, 137 with early studies suggesting strains from severe AD subjects more negatively impact the skin in 138 a murine model 35,38 .

141
Overall approach 142 Skin swabs were collected from 20 participants aged 0.5-14 years for metagenmic analysis of the 143 microbiome. When present, lesional samples were collected in addition to matched nonlesional 144 samples. In contrast to prior studies, AD participants were primarily mild-moderate and 145 continued use of topical therapies to control disease, mimicking real-world clinical scenarios. 146 Shotgun metagenomic and culture-based approaches were used to evaluate the microbial 147 communities associated with these samples. To determine the effects of strain variation on 148 keratinocyte viability and barrier integrity, keratinocyte cultures were exposed to conditioned 149 media containing secreted metabolites from select Staphylococcus isolates derived from 150 participants with a range of clinical presentations. and tandem repeat removal as described previously 40 . Taxonomic classification and abundance 176 estimation was performed using   41 and Bracken (v2.5) 42 . Reads assigned as 177 Homo sapiens or that were unclassified at the genus level were filtered out. Decontam (v1.16 bitscore with additional filters to retain only alignments exhibiting a maximum E-value of 1e-5, a 199 minimum percent identity of 50%, and minimum query and subject coverages of 70%.

201
Generation of staphylococcal cell free supernatants 202 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 28, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 Staphylococcus isolates were grown from stock on trypticase soy agar overnight at 37 °C. Single 203 colonies were then grown in trypticase soy broth overnight at 37 °C with shaking. The following 204 morning, liquid cultures were used to inoculate 50 mL supplemented keratinocyte growth media 205 (see Supplemental Methods for recipe) to a starting OD 600 of 0.1, followed by incubation at 37 206 °C with shaking. OD 600 was read every 2 hr until an OD 600 of 0.7 was reached (range 0.693-207 1.156, average 0.774; typically 3-7 hrs). The cultures were centrifuged at 3220 x g for 10 min.

208
The supernatants were collected and passed through a Watman cellulose Grade 1 filter paper (GE 209 Healthcare Life Sciences), then through a 0.2 µm pore filter. Supernatants were stored at -20 °C 210 until use. Cultures were serially diluted for CFU plating immediately following inoculation and 211 immediately preceding collection to confirm similar CFUs across strains (average of 2.64 *10 9 212 CFU/mL).

214
In vitro enterotoxin assay 215 Staphylococcus isolates from stock were grown on BHI agar overnight at 37 °C. Single colonies 216 were grown in 3 mL BHI liquid culture overnight at 37 °C with shaking. The next day, the 217 cultures were centrifuged at 3500 x g for 5 min. The supernatants were collected and passed 218 through a pore filter (0.2 µm). Enterotoxin levels were then measured using the BioPharm 219 Ridascreen kit according to manufacturer's instructions.  HKs were also used for the trans-epithelial electrical resistance (TEER) assay. The cells were 230 cultured in HK media on a semipermeable filter insert in a 5% CO 2 atmosphere at 37 °C for 24 231 hrs until they reached confluence. They were then switched to a differentiation medium (DM; 232 DMEM medium supplemented with 1.8 mM calcium ion and 4 mM glutamine (Thermo Fisher 233 Scientific)) alone (Control) or DM containing one of the Staphylococcus supernatants diluted to 234 1*10 6 CFU/mL. The integrity of the HK monolayer was verified by measuring the TEER assay 235 after 2 days of treatment. Three independent replicates were used to calculate the results. The  was used with Bray-Curtis dissimilarity estimates. These calculations were done on relative 246 abundance set to 100% for microbes, bacteria, or Staphylococcus, where appropriate. A t-test 247 was used to compare enterotoxin levels. Following a significant ANOVA, pairwise t-tests were 248 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 28, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 used to compare the MTT and TEER effects of the Staphylococcus isolates to media controls, 249 with a Bonferroni correction used to account for multiple testing.

251
Clinical characteristics and sample collection 252 Twenty pediatric participants, aged 5 months to 14 years, were recruited from the University of 253 Wisconsin Pediatric Allergy-Immunology and Dermatology Clinics (Table 1). One control 254 participant was omitted due to having a parent-reported history of atopic dermatitis (AD) but no 255 physician diagnosis and no symptoms of disease. The remaining participants consisted of 8 256 healthy controls, and 11 with diagnosed AD, primarily mild-moderate in severity. Participants 257 with AD were not asked to halt usage of topical treatments (Supplemental Table 1). Skin swabs 258 collected from participants were classified as control, lesional, or nonlesional and were processed 259 for DNA extraction and metagenomic analysis and untargeted bacterial isolation.

261
Skin metagenomes from well-controlled AD participants resemble controls 262 Metagenomic analysis of skin swabs was completed to provide a comprehensive view of the 263 microbiome across participants. After filtering and quality control, the majority of reads (>97%) 264 were identified as Bacteria ( Fig 1A). The remaining 3% of reads mapped to either Eukaryota or 265 Viral databases. The dominant bacterial genera were consistent with previous reports of pediatric 266 skin-Streptococcus, Cutibacterium, Micrococcus, Staphylococcus ( Figure 1B). Of note, the 267 Eukaryota were dominated by Malassezia restricta, but there were no significant differences 268 across genera (Supplemental Fig 3A). Interestingly, viral relative abundance was elevated in AD 269 (p adj =0.010). This was likely driven by the presence of Escherichia virus T4, which was only 270 identified in AD participants (p=0.004) (Supplemental Fig 3B). Molluscum contagiosum virus, 271 which is associated with skin infection, was also present on the skin of several participants, but 272 was not significantly different among the groups (Supplemental Fig 3B).

274
Alpha (within-sample) diversity of the bacteria, as measured by the Shannon Diversity Index, did Staphylococcus species present in the three groups did not significantly differ, the total number 287 of different Staphylococcus species present in AD participants trended higher (Fig 3B; p=0.15).

288
Notably, we did find a decrease in Shannon Diversity when comparing combined AD vs control 289 (p=0.048; Fig 3C).

291
To determine if specific strains of Staphylococci vary by AD status, we classified the 292 Staphylococcus strains in metagenomes using StrainGST 43 . This analysis showed a striking lack 293 of S. epidermidis FDAARGOS_1361 (strain type (ST) 153) in lesional samples, but was present 294 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 28, 2023. ; https://doi.org/10.1101/2023.05.24.23289041 doi: medRxiv preprint in 5/7 control samples and 4/8 of nonlesional samples ( Fig 3D). Conversely, 3 other strains of S. 295 epidermidis were present only in AD swabs (ST 5,89,387). These data suggest that the 296 association of S. epidermidis with disease activity may be strain dependent. Using StrainGST, we 297 observed S. aureus strains only in AD samples, specifically in the nonlesional swabs. Strains of 298 S. warneri and S. saprophyticus, skin commensals that may act opportunistically 50,51 , were also 299 only observed in AD samples (Fig 3D).

301
Isolation and identification of live bacterial isolates 302 To determine the role different strains may be playing in AD pathogenesis, we performed 303 untargeted bacterial isolation. Using four different media to capture greater microbial diversity, 304 601 isolates from 26 swabs were obtained. Full length 16S rRNA gene sequencing was used to 305 taxonomically classify each isolate, resulting in the classification of 22 genera-13 among the 306 305 isolates from control swabs, 10 from the 84 lesional isolates, and 14 from the 210 307 nonlesional isolates (Supplemental Table 2). Of those, 93% were genera represented in the 308 metagenomes. Actinobacteria were the most abundantly represented phylum, followed by 309 Firmicutes and then Proteobacteria in both the metagenomic-and culture-based approaches. Of 310 note, control swabs had >50% more isolates collected than from nonlesional swabs, which, in 311 turn, had twice the number of isolates as lesional swabs. Micrococcus was the most abundantly 312 cultured genera from control samples, while Staphylococcus spp. (specifically aureus, capitis, 313 epidermidis, hominis, saprophyticus, succinus, and warneri) were the most abundantly isolated 314 in lesional and nonlesional swabs (Supplemental Table 2). Combined, these two genera 315 accounted for >40% of all isolates. Strikingly, the proportion of isolates that are Staphylococcus 316 spp. rose greatly in the AD samples, particularly those from nonlesional sites.

318
Toxins are elevated in AD-associated Staphylococci 319 To better understand the role of Staphylococci strain-level differences on AD, we utilized the 320 above-described isolate library. Staphylococci strains representing species isolated across 321 participants were subject to whole genome sequencing. From this, a phylogenetic tree was 322 generated showing relatedness of the selected strains ( Figure 4A). Of note, all S. aureus isolates 323 were collected from moderate and severe AD participants. Conversely, all S. epidermidis and S. 324 hominis isolates were from mild AD or control participants. The S. capitis strains were split into 325 2 clusters, one comprised of moderate-severe AD participants and the other associated with a 326 single mild participant. Genes encoding toxin production in each genome were predicted using 327 the Virulence Factor Database. We noted three patterns for genes encoding toxins-those 328 universally present (not shown), those present in a specific species, and those present in only 329 select isolates. All isolates, even coagulase negative Staphylococci (CoNS), contained at least 330 five toxin-associated genes. As expected, S. aureus strains contained increased enterotoxin, 331 hemolysin, and leukocidin genes ( Fig 4B). Only 1/3 S. aureus strains isolated encoded for seb. 332 This S. aureus strain, designated LK1493, was isolated from a participant recorded as having 333 severe disease. To confirm the genomic predictions, the production of enterotoxins in each S. 334 aureus isolate was tested in vitro, confirming high levels of SEB production by LK1493 335 (p<0.001) (Fig 4C). 338 To test the in vitro effects of these differences on keratinocytes, cell free conditioned media 339 (CFCM) derived from the Staphylococcus isolates were incubated with keratinocytes to 340 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

342
However, there was wide variability within each group.

344
To investigate the variability within in each group, we first compared among species (Fig 5C,   345 D). S. hominis species isolated from control swabs had no effect on cell viability, while S. 346 hominis species from AD lesional swabs were detrimental. Indeed, many Staphylococcus species 347 isolated from lesional swabs significantly impaired TEER (p adj <0.05). We next examined each 348 isolate (Fig 5E, F). We found intriguing differences driven by the strain of Staphylococcus  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

361
The skin microbiome is critical in maintaining proper functioning of the skin barrier, and 362 previous research has focused on the role of S. aureus in moderate-severe AD while medications 363 are held. In this setting, the relative abundance of S. aureus has been shown to increase during 364 acute disease flares 12, 13 . In addition, S. aureus has also been shown to produce toxins, 365 particularly SEB, which are associated with more severe disease 36,37 . However, there is limited 366 research into medication-controlled AD, which is more reflective of the clinical setting.

368
In our medicated cohort, we did not observe the emergence S. aureus blooms reported in 369 previous studies enrolling unmedicated AD, even in lesional samples. When S. aureus was 370 detected, it was primarily identified from nonlesional AD swabs. This is similar to a previous 371 study that found mitigation of S. aureus overgrowth when taking medication 13  In addition, we observed marked differences in toxin gene presence across the Staphylococcal 387 isolates. All 3 S. aureus strains, isolated from moderate and severe participants, encoded for 388 more toxin genes than other strains. One S. aureus strain (isolated from a severe patient), 389 contained the seb gene and expressed this protein at high levels in vitro. Previous studies have 390 suggested SEB is associated with more severe AD 36,37 , likely due to its cytotoxic effects which 391 disrupt barrier function 56 .

393
Importantly, our data evaluating keratinocyte metabolism and barrier function after exposure to 394 secreted factors from diverse Staphylococci species demonstrates that many Staphylococcal 395 strains are capable of disrupting keratinocyte function, even in the absence of enterotoxin 396 production. Additionally, keratinocyte dysfunction occurred across several species, and was 397 strain dependent. In general, S. aureus and S. capitis were associated with deleterious effects on 398 barrier function, while S. epidermidis and S. warneri had strain-level varied impacts as measured 399 by TEER. Finally, different strains of S. hominis induced opposing effects on keratinocyte 400 integrity as measured by MTT. Ongoing research will attempt to clarify the drivers of these 401 differential effects.

403
Taken together, our data underscore the importance of bacterial strain in AD disease expression 404 and pathogenesis, and suggest that topical steroids can be effective in controlling S. aureus 405 overgrowth. It is worth noting that this was a small cross-sectional study, with participants 406 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 28, 2023. ; https://doi.org/10.1101/2023.05.24.23289041 doi: medRxiv preprint covering a range of developmental stages. The skin microbiome can change significantly from 407 infancy through puberty 60 , and controls were age and sex matched to help address this issue. 408 Additionally, medications used varied somewhat among participants, which may have 409 differential impacts on the skin microbiome. Thus, future research will add additional 410 participants to clarify these effects, and to collect additional strains of Staphylococcus, as well as 411 expanding to genera beyond Staphylococcus.

413
The data presented demonstrate the critical role of strain-level differences in AD pathogenicity.

414
Strains differed with regard to presence and amount of toxins expressed, effect on keratinocyte 415 viability and barrier function. Additionally, deleterious effects were not exclusive to S. aureus. 416 Thus, both inter-and intra-species differences impact AD pathogenesis, suggesting that strain-417 level differences are important considerations in AD pathogenesis and disease expression. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 28, 2023. 1062.e7. Available from: http://www.jacionline.org/article/S0091674920309611/fulltext 508 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted   CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 28, 2023. ; https://doi.org/10.1101/2023.05.24.23289041 doi: medRxiv preprint

693
Patient recruitment and classification 694 Food allergy was diagnosed by a board certified pediatric-allergist immunologist with typical 695 signs and symptoms of immediate Ig-E mediated reaction (ex: hives, angioedema, vomiting, etc.) 696 together with evidence of allergic sensitization or oral food challenge as previously described 61 .

697
Other allergic disease diagnosis was determined by parental report of physician diagnosis and 698 confirmed by medical record review.

700
Sample collection 701 Swabs were obtained from the forearm by rolling the swab with moderate pressure 10 times over 702 the designated area, and then by turning the swab 90 degrees and rolling the swab 10 additional 703 times. Lesional swabs were collected from sites of active lesions and nonlesionsal swabs were 704 obtained from healthy-appearing skin from a similar body site (for example, just adjacent to the 705 lesion, or on the contralateral body site) in a similar fashion, or from the anterior forearm when 706 no lesion was present. Swabs were then submerged in lysis buffer and stored at -80 °C for further 707 analysis.

709
Bacterial isolation 710 Samples were thawed and 100 μ L were pipetted onto 4 media plates-brain heart infusion (BHI) 711 agar, BHI + 0.1% Tween 80 (BHIT) agar, blood agar, and mannitol salt agar. Glass beads were 712 used to spread the sample evenly across plates. After the beads were removed and the plates 713 dried, they were incubated at 28 °C for 2-3 days. At this time, morphologically unique colonies 714 were picked from the plates and streaked for isolation onto fresh plates of the same media type.

715
All plates were then returned to the 28 °C incubator. Every 2-3 days for 1 week, the original 716 culture plates were checked for new colonies. An individual colony was picked from the pure 717 cultures on the isolation plates and grown in BHIT liquid culture overnight at 37 °C. After the 718 cultures were turbid, a 15% glycerol stock was prepared for storage at -80 °C.

720
Bacterial isolates from stock were grown on BHIT agar overnight at 37 °C. From these plates,