Content uploaded by Paola Florez de Sessions
Author content
All content in this area was uploaded by Paola Florez de Sessions on Oct 19, 2017
Content may be subject to copyright.
Whole metagenome profiling reveals skin
microbiome-dependent susceptibility to atopic
dermatitis flare
Kern Rei Chng1, Angeline Su Ling Tay2, Chenhao Li1, Amanda Hui Qi Ng1, Jingjing Wang3,4,5,
Bani Kaur Suri6, Sri Anusha Matta6,NaomiMcGovern
7,BaptisteJanela
7, Xuan Fei Colin C. Wong2,
Yang Yie Sio6, Bijin Veonice Au3,AndreasWilm
1, Paola Florez De Sessions1,ThiamChyeLim
8,
Mark Boon Yang Tang9,FlorentGinhoux
7,JohnE.Connolly
3,5,10, E. Birgitte Lane2, Fook Tim Chew6,
John E. A. Common2*and Niranjan Nagarajan1*
Whole metagenome analysis has the potential to reveal functional triggers of skin diseases, but issues of cost, robustness
and sampling efficacy have limited its application. Here, we have established an alternative, clinically practical and robust
metagenomic analysis protocol and applied it to 80 skin microbiome samples epidemiologically stratified for atopic
dermatitis (AD). We have identified distinct non-flare, baseline skin microbiome signatures enriched for Streptococcus and
Gemella but depleted for Dermacoccus in AD-prone versus normal healthy skin. Bacterial challenge assays using
keratinocytes and monocyte-derived dendritic cells established distinct IL-1-mediated, innate and Th1-mediated adaptive
immune responses with Staphylococcus aureus and Staphylococcus epidermidis. Bacterial differences were complemented by
perturbations in the eukaryotic community and functional shifts in the microbiome-wide gene repertoire, which could
exacerbate a dry and alkaline phenotype primed for pathogen growth and inflammation in AD-susceptible skin. These
findings provide insights into how the skin microbial community, skin surface microenvironment and immune system
cross-modulate each other, escalating the destructive feedback cycle between them that leads to AD flare.
Atopic dermatitis (AD; OMIM 603165) is a dry, itchy, inflam-
matory skin disorder that affects up to 20% of the population
in developed countries1,2. AD is cyclical, with relapsing flare
periods that induce cutaneous phenotypes and non-flare periods
with intact skin and no overt clinical signs of infection3.
Multiple genetic risk factors for AD have been identified, high-
lighting the specific importance of innate and adaptive immune
pathways and skin barrier function4. The strongest risk factor for
AD is loss-of-function mutations in the gene encoding filaggrin
(FLG)5–7, a key component of terminal differentiation and skin
barrier function including pH regulation and epidermal hydration8.
Human studies and in vitro experiments have shown that presence
of Th2 cytokines can also perturb skin barrier function via multiple
mechanisms including downregulation of antimicrobial peptides9,
increased expression of serine proteases10,downregulationofkeratino-
cyte differentiation9,11 and expression of profilaggrin processing
enzymes12. This complex interplay between the immune system
and direct or indirect barrier dysfunction creates a skin micro-
environment where percutaneous sensitization can contribute to
atopic disease13–16.
Skin infections are a frequent problem encountered by AD
patients, with bacterial colonization by Staphylococcus aureus
observed on both lesional and non-lesional skin17, and treatment
with antibiotics or dilute bleach baths often being successful18.
Strong perturbation of the skin microflora during flares, including
massive reduction in microbial diversity has also been demonstrated19.
Antigen transfer through a defective skin barrier can influence host
immunity and, in turn, the host immune system can shape the
microbiome. For instance, skin microbiome variations have been
linked to primary immunodeficiency diseases with AD-like
presentation20, highlighting that alterations in immunity can drive
microbial dysbiosis. The clear differences observed in microbiome
composition between dry, oily and moist body sites in healthy
skin21 suggest that conditions affecting the skin microenvironment
will also have profound effects on the microbiome, even in
asymptomatic states. Thus, while current data strongly indicate a
significant role for the skin microbiome in AD pathogenesis, the
potentially protective or predisposing role of inter-flare microbiota
remains to be explored.
In this study, high-throughput whole-metagenome sequencing
was used to explore between-flare skin flora in a volunteer cohort epi-
demiologically stratified for AD22. This was based on the hypothesis
that, by examining non-flare conditions when the disease is dormant
but the individual is still AD-susceptible, new insights into the role
of the microbiome in AD disease pathogenesis can be determined.
Molecular indicators of skin health are difficult to define, and
predicting early signs of susceptibility to skin disease would
enable early intervention and better outcomes. Skin microflora
1Genome Institute of Singapore, Singapore 138672, Singapore. 2Institute of Medical Biology, Singapore 138648, Singapore. 3Institute of Molecular and Cell
Biology, Singapore 138673, Singapore. 4Department of Neurology, The First Affiliated Hospital of Zhengzhou University, Zhengzhou, Henan 450001, China.
5Institute of Biomedical Studies, Baylor University, Waco, Texas 76798, USA. 6Department of Biological Sciences, National University of Singapore,
Singapore 117543. 7Singapore Immunology Network, Singapore 138648, Singapore. 8Division of Plastic, Reconstructive & Aesthetic Surgery, National
University Health System, Singapore 119074, Singapore. 9National Skin Centre, Singapore 308205, Singapore. 10Department of Microbiology and
Immunology, National University of Singapore, Singapore 117545, Singapore. *e-mail: nagarajann@gis.a-star.edu.sg;john.common@imb.a-star.edu.sg
ARTICLES
PUBLISHED: 11 JULY 2016 | ARTICLE NUMBER: 16106 | DOI: 10.1038/NMICROBIOL.2016.106
NATURE MICROBIOLOGY |www.nature.com/naturemicrobiology 1
© 2016 Macmillan Publishers Limited. All rights reserved
in asymptomatic, high-risk individuals was compared to a control
group to enrich for differences that predispose or protect individ-
uals from flares. Significant bacterial and eukaryotic variations
were observed in skin microbial composition between control
and AD-prone groups. Experimental characterization using
immune challenge assays and functional analysis of microbial
pathways identified a pathogenic niche that is primed for switch
to flares, highlighting the potential functions of the inter-flare
microbiome in AD.
Results
Establishing a robust and convenient skin microbiome analysis
protocol. A selection of widely used skin sampling and
microbiome profiling approaches were evaluated, complementing
earlier work based on marker genes23, but with a specific focus on
whole-metagenome analysis. Three sampling methods were
compared: (1) skin swabbing (http://hmpdacc.org/)(‘swab’), (2) a
modified cup scrub (‘cup scrub’) and (3) a tape-stripping
technique that has not been evaluated for metagenomics (‘tape’).
All sampling approaches were combined with two profiling
approaches: (1) amplification and sequencing of 16S rRNA gene
(‘16S’) and (2) shotgun whole-metagenome sequencing and
analysis of total DNA (‘WM’). Samples were collected from left
and right antecubital fossae (Ac) and retroauricular crease (Ra) of
three healthy volunteers to assess their distinctness and/or
concordance as a measure of the reproducibility of the results
(Supplementary Table 1 and Supplementary Fig. 1).
Overall, high concordance was observed between microbial
abundance profiles derived for left and right Ac from various
sampling approaches (r> 0.86; Supplementary Fig. 2a). Higher
concordance was also observed using whole-metagenome profiles
(r> 0.95) versus 16S (r> 0.86), and slightly higher concordance
using tapes as opposed to swab or cup scrub. Slight biological differ-
ences between left and right Ac were also consistently captured
across sampling approaches. Given the concordance between
sampling approaches, the tape-based protocol stands out for its
ease of use in a clinical setting (see Methods) and robustness for
DNA collection (Supplementary Fig. 1).
Lower concordance was observed between profiling approaches
due to the inherent differences in experimental and analytical
procedures (Supplementary Figs 2b and 3). However, whole-
metagenome profiling showed higher internal consistency than 16S
profiling (Supplementary Figs 2a,b and 3). Whole-metagenome
sequencing also enabled analysis of eukaryota and viruses with
high reproducibility (Supplementary Fig. 2c). Principal coordinates
analysis of profiles from various sampling approaches readily
reflected the distinctness of skin surface niches and demonstrated
their separation from potential environmental contamination
(Supplementary Fig. 2d).
The combination of tape sampling and whole-metagenome
sequencing was validated as an easy-to-use and robust skin micro-
biome analysis protocol by applying it to 80 Asian skin microbiome
samples. The success rate for sequencing libraries was >98.7%,
despite challenges of sample collection in a clinical setting and the
potential for low microbial biomass on skin surfaces, as noted
previously24 (Supplementary Fig. 4). Analysis of control samples
confirmed that our sampling strategy was highly specific
(Supplementary Figs 1 and 4).
Intra- and inter-individual variability in the skin microbiome of
a cohort stratified for AD. Microbiome samples (tape) were
collected from clinically intact, non-inflamed skin in 40 adult
volunteers. Sampling left and right Ac provided a more robust
individual Ac skin microbiome representation. The cohort was
stratified into three groups defined according to the ISAAC study
questionnaire25: (1) AD case group, with volunteers reporting past
AD episodes (n= 19; 1 library failed), (2) Control group, with
volunteers who were skin prick test (SPT) negative and have no
AD history (n= 15) and (3) Skin prick test positive (SPT+) group,
with volunteers (n= 5) having no AD history but SPT+
(Supplementary Tables 2–4). The SPT+ group serves as a
reference for individuals with atopy but not reporting AD. The
three groups were approximately matched according to mean age
(Control: 24.1; Case: 23.1; SPT+: 23.0), gender (Control: 8 males,
7 females; Case: 8 males, 12 females; SPT+: 3 males, 2 females)
and ethnicity (Chinese) (Supplementary Table 2). Trans-
epidermal water loss (TEWL) and pH measurements were
collected to validate the skin health of sampled regions and no
statistically significant differences between cases and controls were
noted (Supplementary Fig. 5a).
Examination of the principal axes of variation in Ac micro-
biomes showed that all three groups clustered together with
similar bacterial composition overall (Fig. 1a). Evaluating intra-
(left versus right) and inter-individual variability across cohorts,
we noted an individual-specific bacterial composition signature
marked by higher intra-individual similarity (P= 4.7 × 10
−10
;
Fig. 1b). These consistent individual-specificprofiles were then
averaged for robustness and compared across groups to characterize
patterns of bacterial variation specific to inter-flare AD skin. Higher
similarity in bacterial composition was observed within the case and
SPT+ groups than compared to the control group (P=2.4×10
−6
and
2.8 × 10
−4
, respectively; Fig. 1c), despite the absence of significant
differences in bacterial diversity between groups (Supplementary
Fig. 5b). A trend for lower bacterial diversity in subjects with
FLG-null mutations (n= 3; two cases, one control) versus FLG
wild-type volunteers (n= 36) was observed (but was not statistically
significant; Supplementary Fig. 5c).
Defining a bacterial signature for susceptibility to AD. To
characterize the distinctness of AD-associated microbiomes and
establish a bacterial signature for AD susceptibility, we
systematically compared non-flare, baseline profiles at the genus
level (due to the limited resolution for some species; see Methods)
across groups. Strong enrichment was observed for potential
pathobionts belonging to the genera Streptococcus and Gemella
in AD and SPT+ groups relative to the controls (FDR-adjusted
P< 0.05, Wilcoxon rank-sum test; Fig. 1d, Supplementary
Table 5), pointing to a role in sensitizing the immune system
through the skin. The genus Dermacoccus was strongly depleted in
the AD group relative to controls and SPT+ individuals (FDR-
adjusted P< 0.05, Wilcoxon rank-sum test; Fig. 1d and
Supplementary Table 5). Dermacocci belong to the order
Actinomycetales, known for producing secondary metabolites with
anti-inflammatory and anti-microbial properties26. In addition, we
detected enrichment in the genera Veillonella and Haemophilus,as
well as depletion in Deinococcus and Methylobacterium,inAD
versus normal controls (Supplementary Fig. 6a and Supplementary
Table 5). Canonical correlation analysis of taxonomic profiles and
clinical information confirmed the predominant association of all
seven identified genera with AD status (ρ= 0.60). Stratification by
gender and cream usage showed significant differences in a subset
and similar enrichment/depletion trends in all identified genera
(Supplementary Fig. 6b,c). Interestingly, different genera were
significantly differentially abundant in males versus females,
potentially due to insufficient sample size or bonafide gender-
specific differences. Patterns observed for SPT+ individuals
reinforced the observations for the AD group, that is, enrichment
in potentially sensitizing pathogens but no depletion of putatively
protective bacteria, highlighting the distinct aetiology of this group.
At the species level, nine bacteria were identified to have
significant AD-associated microbiome differences (Supplementary
Table 5). Among these, three α-haemolytic Streptococci were
ARTICLES NATURE MICROBIOLOGY DOI: 10.1038/NMICROBIOL.2016.106
NATURE MICROBIOLOGY |www.nature.com/naturemicrobiology2
© 2016 Macmillan Publishers Limited. All rights reserved
enriched in AD subjects, but S. pyogenes (β-haemolytic), previously
associated with severe skin infections27, was not detectable in any
subjects studied (Supplementary Fig. 7). The majority of species
enriched in AD-susceptible subjects are known commensals or
opportunistic pathogens of the oral cavity, the significance of
which remains presently unclear (Supplementary Note 1 and
Supplementary Table 5).
Correlation analysis to predict bacterial interactions in skin com-
munities identified >142 interactions (Supplementary Note 2) includ-
ing a known antagonistic relationship (negative correlation) between
P. acne s and S. epidermidis28,andasignificant negative correlation
between Dermacoccus and Methylobacterium species with S. aureus
(see Methods; Supplementary Table 6). Agar-diffusion assays using
bacteria identified here as enriched in AD non-flare skin, in compe-
tition with S. aureus, further support the hypothesis that interactions
between these species have a functional role in AD (Supplementary
Note 3 and Supplementary Fig. 8). Finally, the possibility of using a
purely bacterial signature for stratifying various groups in our
cohort was assessed. By restricting analysis to the most significant
differentially abundant genera (Streptococcus,Gemella and
Dermacoccus), we obtained robust consensus clustering of profiles
into three distinct clusters (Supplementary Fig. 9; see Methods).
Strain-level differences in the AD-associated microbiome. A
systematic approach to catalogue sequence variants was used to
determine whether shifts in strain abundance correlated with AD
susceptibility29 (see Methods). We studied staphylococci and, in
particular, S. aureus, because of its known AD association17,
predominance during flares and correlation with clinical severity19.
An increased percentage of S. aureus carriers was noted in our AD
cohort over controls (Supplementary Fig. 10), showing that these
differences are also present (albeit more subtly) on non-flare skin
of AD-susceptible individuals. Overall, S. epidermidis exhibited
greater strain diversity than S. aureus across cohorts
(Supplementary Fig. 11a,b). An analysis of single nucleotide
polymorphisms (SNPs) also indicated that important strain-level
differences, including in key virulence factors, exist in distinct
groups (Supplementary Table 7). For example, the AD group
preferentially hosted a microbial population carrying a variant of
the staphylococcal lipase gene geh, a well-characterized virulence
factor30,31, with a basic arginine residue (CGC) at amino acid
position 373, rather than the polar threonine residue (ACC)
(Supplementary Fig. 11c). This preferential polymorphism can
modify the enzymatic efficacy of staphylococcal lipase and thus be
positively selected for by the need to adapt to the AD-associated
drier, lipid-reduced skin microenvironment. By analysing data from
a recent study that sampled different human skin microbiomes21,
we reconfirmed this, noting a preference for the arginine residue
(CGC) in dry (volar forearm, Vf) versus moist environments (Ac)
(P= 0.055; Wilcoxon rank-sum test; Supplementary Fig. 11d).
AD-associated bacteria elicit distinct inflammatory and immune
responses. To investigate the impact of bacterial populations
identified from this study on non-flare AD versus normal skin,
the capacity of microbial species to stimulate the immune system
was examined. Six bacterial species (Supplementary Table 8) were
selected, including a known skin pathogen (S. aureus) and a
commensal (S. epidermidis) that have been shown to increase in
abundance during flares19.Deinococcus radiodurans and
Dermacoccus nishinomiyaensis were selected as representative skin
0.0
0.2
0.4
0.6
0.8
1.0
Yue-Clayton theta index
0
2
4
6
8Streptococcus Dermacoccus Gemella
0.0
0.5
1.0
1.5
2.0
3.0
2.5
**
**
0.0
0.2
0.4
0.6
0.8
1.0
**
Relative abundance (%)
Control–case
Control–SPT
SPT–SPT
Case–case
−0.4 −0.2 0.0
−0.25
0.00
0.25
a c
d
PC1 (44.2%)
0.2
PC2 (18.2%)
Control
Case
SPT+
Control
Case
SPT+
P =
2.4 × 10−6
P =
2.8 ×1 0−4
Intra Ac Inter Ac
0.0
0.2
0.4
0.6
0.8
1.0
Yue-Clayton theta index
b
P = 4.7 × 10−10
Figure 1 | Baseline bacterial diversity and signatures associated with AD. a, Principle coordinates analysis based on Bray–Curtis dissimilarity between skin
microbiome profiles (genus level) for all volunteers in an epidemiologically stratified cohort. b, Box plot quantifying the similarity of the skin microbiome within
(left and right Ac) and across individuals (genus-level profiles). c, Box plot quantifying the similarity of the skin microbiome within and across groups in the
cohort (genus-level profiles). d, Box plots showing the relative abundances of the top three differentially abundant genera in AD patients (n= 19) relative to
normal controls (n= 15). SPT+ refers to skin prick test positive non-AD volunteers (n=5). All Pvalues were calculated using the two-sided Wilcoxon
rank-sum test. **FDR-adjusted P< 0.05. Box plot whiskers represent 1.5 × interquartile range or the maximum/minimum data point within the range.
NATURE MICROBIOLOGY DOI: 10.1038/NMICROBIOL.2016.106 ARTICLES
NATURE MICROBIOLOGY |www.nature.com/naturemicrobiology 3
© 2016 Macmillan Publishers Limited. All rights reserved
commensals of their genera (depleted in AD; Supplementary
Table 5), Dermacoccus profundi as control non-skin commensal
from its genus, and Bacillus firmus as non-commensal
immunostimulatory positive control32. Human keratinocytes,
moDCs as well as primary human and mouse Langerhans cells
(LCs) were challenged with live bacteria and supernatants.
In human keratinocytes, a classical IL-1-driven inflammation
signature of cytokines and innate immune response was induced
by live bacteria (Supplementary Fig. 12a) and supernatants
(Fig. 2a). Specifically, IL-1αincreased strongly with challenge
from S. epidermidis and S. aureus, while moderate induction of
IL-1βwas observed with S. epidermidis,S. aureus and B. firmus
(Fig. 2a). In addition, specific chemokines known to play a role in
the recruitment of granulocytes and DCs were stimulated, including
IL-8 in B. firmus and D. profundi, and IP-10 in S. aureus,B. firmus
and D. profundi (Fig. 2a and Supplementary Fig. 12b). The muted
response of D. nishinomiyaensis (skin commensal) in comparison
to D. profundi (non-skin commensal control) is noteworthy and
supports the notion that a protective role for Dermacoccal species
is likely to be specific to skin commensals. Overall, these data are
consistent with the known cytokine profile that induces maturation
of naive DCs33 and supports the hypothesis that keratinocyte acti-
vation by S. epidermidis and S. aureus can lead to recruitment
and stimulation of DCs in the skin.
Challenge assays with human and mouse primary LCs indicated
that all bacteria elicit a muted response (Supplementary Figs 13 and 14),
in agreement with previous results34 and the complex roles
played by LCs in skin35. Challenge of moDCs elicited a more pro-
nounced and graduated immune activation response from different
bacteria, with Staphylococcal species at one end of the spectrum
(most response) and Dermacocci at the other end (Fig. 2b and
Supplementary Fig. 12c). Staphylococcus epidermidis,S. aureus
and B. firmus induced a tumour necrosis factor (TNF) driven
myeloid activating signature of cytokines and chemokines, with che-
mokine profiles predicted to recruit neutrophils and monocytes
(Fig. 2b). We also noted induction of a Th1 polarizing cytokine sig-
nature (IL12p70 and IL12p40) in response to S. epidermidis and, to
a lesser extent, S. aureus,D. radiodurans and D. profundi. Overall,
our results suggest that S. epidermidis and S. aureus invoke a
strong immune inflammatory response in an in vitro setting. In con-
trast, D. radiodurans and D. nishinomiyaensis elicit a muted
response, supporting the hypothesis that the presence of skin com-
mensal Dermacocci and Deinococci plays a protective role on skin.
Changes in the eukaryotic diversity on skin associated with AD.
Whole-metagenome analysis enables the characterization of
eukaryotic and viral constituents of the microbiome, providing an
opportunity to study their biological roles in AD. Overall, we
observed higher relative abundance of eukaryotes in controls
versus cases (P= 0.0045), but no statistically significant differences
in bacterial (P= 0.21) and viral proportions (P= 0.51; Fig. 3a).
Despite their relatively low density (Fig. 3a), a surprising diversity
of viruses/phages was identified (Fig. 3b). As expected21, these
were dominated by phages, although a few viruses could also be
readily detected (for example, human papillomavirus) with no
significant differences observed between cohort groups
(potentially due to low abundances).
Eukaryotic microbiome residents on human skin are dominated
by the Malasseziaceae family (Fig. 4a), and these were clearly depleted
in AD-susceptible individuals (P< 0.01, Wilcoxon rank-sum test;
ab
Log2 conc.
Log2 conc.
0.0 14.0
SA
SE
BF
DR
DP
DN
IL-1α
IL-8
MDC
Eotaxin
TNFα
RANTES
VEGF
GRO
IL-10
IL-12p40
EGF
MIP-1β
FGF-2
PDGF-AB/BB
MCP-3
MIP-1α
G-CSF
sCD40L
PDGF-AA
Fractalkine
TNFβ
Flt-3L
IL-1β
IFNα2
TGF-α
IL-1RA
IP-10
IL-6
IL-12p70
MCP-1
S
A
S
E
B
F
D
R
D
P
D
N
0.0 18.0
DR 0.001%
DP 0.01%
SE 0.1%
SE 0.01%
SE 0.001%
SA 0.1%
SA 0.01%
SA 0.001%
BF 0.1%
BF 0.01%
BF 0.001%
DR 0.1%
DR 0.01%
DP 0.1%
DP 0.001%
DN 0.1%
DN 0.01%
DN 0.001%
DR
0
.
00
1
%
D
P
0
.
01
%
DR
0
.
1
%
DR
0
.
01
%
D
P
0
.
1
%
D
P
0
.
001
%
DN
0
.
1
%
DN
0
.
01
%
DN
0
.
00
1
%
S
E
0
.
1
%
S
E
0
.
01
%
SE
0
.
00
1
%
S
A
0
.
1
%
S
A
0
.
0
1
%
S
A
0
.
00
1
%
B
F
0
.
1
%
B
F
0
.
0
1
%
B
F
0
.
0
01
%
MCP-3
IL-1α
VEGF
IL-1β
IL-6
EGF
Eotaxin
IL-10
TGF-α
MIP-1α
MDC
MCP-1
IL-8
TNFα
PDGF-AA
TNFβ
G-CSF
IP-10
Flt-3L
GRO
FGF-2
PDGF-AB/BB
IFNα2
Fractalkine
IL-12p70
sCD40L
MIP-1β
IL-12p40
RANTES
IL-1RA
SE: Staphylococcus epidermidis
SA: Staphylococcus aureus
BF: Bacillus firmus
DR: Deinococcus radiodurans
DP: Dermacoccus profundi
DN: Dermacoccus nishinomiyaensis
Figure 2 | Hierarchical clustering comparing cytokine secretion profiles of human keratinocytes and dendritic cells with various bacterial supernatants.
a, Heat map representation of log
2
average medium normalized cytokine secretion levels in human keratinocytes (n= 3) that were stimulated for 24 h with
various bacterial supernatants. Cytokines associated with an IL-1αdriven inflammation signature are boxed in red. Hierarchical clustering of cytokine profiles
was performed using the Euclidean distance metric. b, Heat map representation of log
2
average medium normalized cytokine secretion levels in human
moDCs (n= 3) that were stimulated for 24 h with bacterial supernatants as indicated. Cytokines associated with a TNF-driven myeloid activating signature
are boxed in red. Hierarchical clustering of cytokine profiles was performed using the Euclidean distance metric. SE, S. epidermidis;SA,S. aureus;BF,B. firmus;
DR, D. radiodurans;DP,D. profundi; DN, D. nishinomiyaensis.
ARTICLES NATURE MICROBIOLOGY DOI: 10.1038/NMICROBIOL.2016.106
NATURE MICROBIOLOGY |www.nature.com/naturemicrobiology4
© 2016 Macmillan Publishers Limited. All rights reserved
Figs 3a and 4b). The next most common eukaryotic member, the
Aspergillaceae family, was not depleted on AD skin (Fig. 4a).
Several species in the Malassezia genus have been implicated in
skin pathologies including AD36,37, although these associations lack
consensus37. Here, reference genomes for all 14 species in the
genus38 were used to quantitate patterns of Malassezia colonization
in AD and normal individuals. Our data confirmed the predomi-
nance of M. globosa and M. restricta on human skin (Fig. 4c).
Additionally, we observed a higher relative abundance of M. globosa
in controls relative to AD group (P< 0.05, Wilcoxon rank-sum
test; Fig. 4d), in agreement with its preference for lipid-rich environ-
ments. In contrast, M. dermatis (P< 0.1, Wilcoxon rank-sum test)
and M. sympodialis were observed to be at higher relative abundance
in the AD group (Fig. 4d). Our data provide evidence supporting the
previously reported role of M. sympodialis in AD39 and, in addition,
associate M. dermatis with AD skin.
Alterations in the functional potential of AD-associated
microbiome suggest a mechanism for escalation to flare.
Differences in the AD skin microbiome may be due to the
inability of a subset of homeostatic commensals to thrive in the
lipid-depleted habitat of barrier-defective skin, creating ecological
space for colonization by pathogens. Microbial shifts in turn
could modulate the skin microenvironment to affect AD
progression. For instance, elevated skin pH during flares and
surface lipid deficiency in AD have been observed previously40.
We analysed the skin microbiome gene repertoire for functional
capabilities in the control group versus the AD group. We
observed significant differences with the AD samples, showing
lower-than-control levels of potentially beneficial functions
contributing to greater skin moisture (for example, the steroid
biosynthesis pathway) and reduced skin inflammation (for
example, tryptophan metabolism, a modulator of immune
response41), as well as reduced eukaryotic cellular pathways as
expected from the lower eukaryotic relative abundance (Fig. 5a).
In contrast, the AD group was enriched in pathogenesis-
associated pathways such as bacterial secretion systems42 and
DNA recombination and repair43.
Intriguingly, AD samples exhibited enrichment in nitrogen, argi-
nine and proline metabolism pathways, which are associated with
ammonia production (Fig. 5a). Elevated skin pH during AD flares
accelerates desquamation, increases epidermal barrier permeability
and reduces anti-microbial acid stress40. Our data confirm higher
metagenomic abundance of all three enzymes responsible for meta-
bolizing arginine and citrulline into ammonia in the AD-associated
microbiome (Fig. 5b,c), establishing its capacity to promote a less
favourable (high pH) microenvironment for skin health.
Combined with earlier work showing that arginine and citrulline
catabolism is exploited by skin pathogens in combating acid
stress44, this provides a mechanism for enhanced microbial
virulence in AD through increased ammonia production.
Discussion
The involvement of the skin microbiome in various diseases has
been of significant recent interest20,45, although its role in AD has
only been analysed under flare conditions or using marker
genes19,46. This study was based on the hypothesis that deeper
levels of detail can be determined by whole-metagenome analysis
of microbial variations during non-flare periods.
Whole-metagenome sequencing offers the opportunity for
multi-kingdom analysis and functional insights into diseases.
Metagenomic skin studies have been limited due to the complexity
of sampling efficiently21,47, with no known studies in a disease
context. The results presented here demonstrate that whole-meta-
genome profiling of skin with tape-stripping can be achieved in a
clinically practical and robust fashion, while being minimally
invasive47 and avoiding skin-surface disruption with a surgical
blade21. Our benchmarks demonstrate that the results obtained
are consistent with alternative sampling approaches, serving as a
useful guide for future studies.
We observed distinct differences between baseline skin micro-
biomes of AD flare-prone subjects and normal healthy individuals.
Strikingly, we did not observe the marked overall dysbiosis or diver-
sity reduction that has been reported during flares19. As more
microbial diversity is retained in the asymptomatic cohort, their
inter-flare microbiome has provided additional insights. For
example, the potential pathobionts identified here, including
Streptococcus,Gemella and Haemophilus species, have been
reported to be enriched in children compared to adults48, which
concurs with the higher incidence of AD in children.
Relative abundance (%)
a
b
Relative abundance (%)
***
Control Case SPT+
Control Case SPT+
Abelson murine leukaemia virus
100
95
90
85
15
10
5
0
Bacteria
Eukaryota
Viruses
31
19
18
2.0
0.6
0.0
Avian endogenous retrovirus EAV HP
Betapapillomavirus 1
Bombyx mori nucleopolyhedrovirus
C2likevirus unclassified
Gammapapillomavirus 1
Gammapapillomavirus 5
Human adenovirus B
Human papillomavirus 161 like viruses
Lymphocryptovirus unclassified
Merkel cell polyomavirus
Mycobacterium phage Rizal
Porcine parvovirus
Porcine type C oncovirus
Propionibacterium phage P100 1
Propionibacterium phage P100D
Propionibacterium phage P101A
Propionibacterium phage P14 4
Propionibacterium phage PHL112N00
Staphylococcus phage 80alpha
Figure 3 | Eukaryotic and viral diversity in AD-associated skin microbiomes. a, Box plot depicting the relative abundance of bacteria, eukaryotes and
viruses in the skin microbiomes (Ac) of normal controls (n=15),ADcases(n= 19) and SPT+ volunteers (n=5).b, Bar chart depicting the relative abundance
of different viruses and phages present in the skin microbiomes of volunteers in the cohort (stratified by groups). All Pvalues were calculated using the
two-sided Wilcoxon rank-sum test. ***P< 0.01. SPT+ refers to skin prick test positive non-AD volunteers. Box plot whiskers represent 1.5 × interquartile
range or the maximum/minimum data point within the range.
NATURE MICROBIOLOGY DOI: 10.1038/NMICROBIOL.2016.106 ARTICLES
NATURE MICROBIOLOGY |www.nature.com/naturemicrobiology 5
© 2016 Macmillan Publishers Limited. All rights reserved
Our results provide evidence that AD-associated microbiomes can
elevate the risk of flares by influencing the skin surface microenviron-
ment and through interaction with the host immune system, although
additional work is needed to understand strain-specific differences.
Overall, while Staphylococci enriched during flares invoked the stron-
gest in vitro cytokine and chemokine response, bacterial genera
depleted in inter-flare skin elicited a muted response. However, none
of the species invoked a strong anti-inflammatory response, suggesting
any protective role from depleted commensals may be mediated
through alternative mechanisms, such as the ability to crowd outpatho-
gens. These assays have limitations, being in vitro, using cells from
normal individuals and bacterial isolates not from individuals in our
cohort. Additionally, strain variation can significantly impact pheno-
types49. Nevertheless, the challenge and inhibition assays in this
study serve as a preliminary indication that AD-associated bacteria
have different functional roles. Understanding the inter-relationships
between various species/strains and their contribution in priming the
host will further guide the development of rational interventions to
restore microbial balance, improve skin health and avoid AD flares.
The fungal component of the microbiome plays an important
role in defining the skin microenvironment. Malassezia is typically
dominant on human skin, and its presence has often been associ-
ated with AD37. Whole-metagenome data allowed us to characterize,
for the first time, a decreased Malassezia relative abundance in AD-
associated microbiomes. Reduced surface lipids and/or the presence
of higher anti-Malasseziaceae IgE levels in AD could restrict
Malassezia ecological competitiveness50. Additionally, our analysis
revealed an enrichment of Malassezia species (M. dermatis,
M. sympodalis) in AD-susceptible skin, suggesting the biological
relevance of species switching in AD pathology.
The observed depletion of specific bacterial genera in baseline AD-
associated microbiota (and not in SPT+ individuals) indicates that
they may have a protective role against AD. This could be through
immune quiescence (as observed for D. nishinomiyaensis), the
production of metabolites and proteins with anti-inflammatory/anti-
microbial properties (for example, by metabolically versatile
Deinococci with anti-oxidant properties on skin; http://www.deinove.
com), or by nutritionally out-competing pathogenic species in
healthy skin. As specificDermacocci and Deinococci are commonly
present on healthy human skin, they have the potential for safe
application as probiotics for AD, opening up additional therapeutic
strategies for exploration.
In addition to modulating the immune system, we show that the
AD-associated skin microbiome is primed to generate excess
ammonia. This provides a microbial explanation for the higher
skin pH associated with AD flares, which in turn could favour a
pathogenic skin microbiome40. From a therapeutic perspective,
correcting metabolite imbalances is an attractive option; ammonia-
oxidizing bacteria have been proposed to be a component of
healthy human skin, eradicated by the use of modern soaps
(http://www.aobiome.com), and we did not detect any ammonia-
oxidizing bacteria (for example, Nitrosomonas) in this study. The
possibility that the depletion of ammonia-oxidizing bacteria
increases the risk of AD provides an alternative explanation for the
association between increased hygiene and global increase in AD.
This study was initiated to test the hypothesis that microbiome
changes offer a ‘canary in the coal mine’solution to predicting
disease outcomes. The results obtained strongly validate this
hypothesis, providing evidence that the AD-colonizing microbiome
is likely to impact the skin microenvironment and promote a pro-
c
dM. restricta
0
20
40
60
80
100
M. globosa M. dermatis
0
2
4
6
8
M. sympodialis
CaseControl SPT+ CaseControl SPT+ CaseControl SPT+ CaseControl SPT+
** *
0
10
20
30
40
50
60
0
1
2
3
4
Relative abundance (%)
Relative abundance (%)
Relative abundance (%)
Relative abundance (%)
a
b
Control Case SPT+
Control Case SPT+
M. globosa
M. dermatis
M. sympodialis
M. furfur
M. obtusa
M. japonica
M. restricta
M. slooae
M. caprae#
M. cuniculi#
M. equii#
M. nana#
M. yamatoensis#
M. pachydermatis#
0
20
40
60
80
100
10
8
6
4
2
0
Malasseziaceae
*** Control
Aspergillaceae
Malasseziaceae
Case
SPT+
Control
Case
SPT+
7.0
1.0
0.3
0.0
Figure 4 | Changes in Malassezia species composition associated with AD. a, Bar chart depicting the relative abundance of different eukaryotes present
in the skin microbiomes of volunteers in the cohort (stratified by groups). b, Box plot showing the relative abundances of Malasseziaceae in the skin
microbiomes of normal controls (n= 15), AD cases (n= 19) and SPT+ volunteers (n=5). c, Bar chart depicting the relative abundance of 14 different
Malassezia species in the skin microbiomes of normal controls (n= 15), AD cases (n= 19) and SPT+ volunteers (n=5).
#
Absent in all volunteers. d, Box plots
showing the relative abundance of selected Malassezia species (M. restricta,M. globosa,M. dermatis and M. sympodialis) in the skin microbiomes of volunteers
in the cohort (stratified by groups). SPT+ refers to skin prick test positive non-AD volunteers. All Pvalues were calculated using the two-sided Wilcoxon
rank-sum test. ***P<0.01,**P<0.05, *P< 0.1. Box plot whiskers represent 1.5 × interquartile range or the maximum/minimum data point within the range.
Where black bars are not shown the median is exactly 0.
ARTICLES NATURE MICROBIOLOGY DOI: 10.1038/NMICROBIOL.2016.106
NATURE MICROBIOLOGY |www.nature.com/naturemicrobiology6
© 2016 Macmillan Publishers Limited. All rights reserved
inflammatory phenotype (Fig. 5d). The altered microenvironment
and loss of homeostatic microbiota leave the epidermis vulnerable
to transitions from opportunistic to pathogenic microbial coloniza-
tion (Fig. 5d). Although it is hard to calculate the quantal effect of
dysbiosis on population prevalence and severity, it is clear that
microbiome interactions provide additional complexity to AD
pathogenesis. With the ever increasing sensitivity of genomic tech-
niques, we predict that observing microbial changes will increas-
ingly become a feasible, non-invasive and cheap method to
monitor skin health and aid probiotic treatment strategies.
Methods
Cohort recruitment and stratification. Volunteers for the study were from an
established cohort22, recruited using a cross-sectional sampling approach from the
National University of Singapore (NUS) campus. All participants were above the age
of 18 years. Recruitment was approved by the Institutional Review Board (NUS IRB
reference code 07-023 and NUS 10-343) and was in accordance with the Helsinki
Declaration. Informed consent was obtained from all subjects recruited into the
study. Cohort stratification was performed with epidemiological data collected for
clinical history of atopic diseases (according to the ISAAC protocol25) and skin prick
reactions to known allergens as detailed in the following22.
Participants were tested for their atopic status by performing a skin prick
test (SPT). Specifically, participants were tested for four aeroallergens,
Dermatophagoides pteronyssinus,Blomia tropicalis,Elaeis guineensis and
Curvularia lunata. Validation of SPT was performed with positive (histamine) and
negative (saline) controls. SPT was considered positive when the weal diameter was
≥3 mm, whereas no sign of weal and erythema was considered a negative SPT.
Participants with a <3 mm reaction for histamine, or an erythema/weal to saline, or
with a <3 mm reaction to the test allergens were excluded from analysis. The skin-
prick reactivities to dust mites (especially the two species B. tropicalis and
D. pteronyssinus) are highly sensitive and specific markers for allergic sensitization in
Singapore51. Thus, together with the positive and negative controls, a positive
reaction to either of the dust mites indicated positive SPT, a marker of atopy.
Sample collection. Three different collection methods for skin microbiome
sampling were tested: tape stripping, swab sampling and the cup scrub method.
Tape stripping. Adhesive tape discs (D-Squame Standard Sampling Discs;
Cuderm) still attached to the plastic backing (to protect the sticking surface) were
completely soaked in 70% ethanol for 1 min and wiped dry with paper towels before
placement on the antecubital fossa (Ac) or the retroauricular crease (Ra) of subjects.
Pressure was applied on the tape disc with a D-Squame Pressure Instrument
(Cuderm) before peeling off. This cycle of sticking and peeling was repeated
approximately 50 times with the same tape disc over a period of 2 min on a
delimited area (3 cm
2
) of skin regions Ac or Ra to saturate the adhesive tape disc
with skin surface material to maximize DNA collection. We also tested applying
pressure without the D-Squame Pressure Instrument (using the thumb instead), and
no difference was observed in the total amount of DNA collected. Similar
approaches were originally used for proteomic and metabolomic studies52,53.
Swab sampling. In the swab collection method, a sterile swab (Catch-All Sample
Collection Swab; Epicentre) was soaked in phosphate buffer saline (PBS) + 0.1%
vol/vol Triton-X100 and was applied immediately to skin regions Ac or Ra, not
overlapping with the area where tape strip samples were collected. The swab was
rubbed for 1 min to collect biological material from the skin surface. This approach
potentially has the drawback of low DNA yield.
Cup scrub method. A hollow cylinder (internal diameter 25 mm) was soaked in
70% ethanol for 1 min, wiped with tissue and allowed to dry completely before
placement onto a non-overlapping sampling region of the Ac. A volume of 1 ml PBS
+ 0.1% vol/vol Triton-X100 was added into the hollowcylinder. The entire surface of
the skin within the hollow cylinder was rubbed with an inoculating loop for 1 min
and the liquid was then transferred into a 1.5 ml tube. Similar approaches have been
widely used for culture-based studies54,55 but can be cumbersome to use in a clinical
setting. All samples were stored at −20 °C before further processing.
After the initial experiments comparing tape, swab and cup scrub sampling
approaches, all samples from the epidemiologically stratified cohort were collected
with adhesive tape discs adjacent to the Ac region, as described above. Participants
were allowed to maintain a healthy skin care regimen, and only restricted for
antibiotics,based on prior observations that discontinuing medications and emollients
for AD patients frequently leads to dysbiosis similar to what is seen in AD flares19.
DNA extraction. Tape discs and swabs were transferred into Lysing Matrix E tubes
(MP Biomedicals) and 600 µl of ATL buffer (Qiagen) was added. Samples collected
using the cup scrub method were first subjected to centrifugation at 9,300gfor 3 min.
Cell pellets were subsequently re-suspended with 600 µl of ATL buffer (Qiagen).
All samples were subjected to bead-beating with a FastPrep-24 Instrument (MP
Biomedicals) at a speed of 6.0 m s
−1
for 40 s. Samples were centrifuged at 16,000gfor
EC 2.7.2.2
Carbamate kinase
EC 2.1.3.3
Ornithine
carbamoyltransferase
EC 3.5.3.6
Arginine deaminase
H2O
Pi
Ornithine
ADP
ATP + CO2
ab
Case
Control
*
***
*
Combined
P = 0.0029
EC 3.5.3.6
EC 2.1.3.3
EC 2.7.2.2
Relative abundance (×10−4)
0
Lower in AD
than in
controls
Higher in AD
than in
controls
c
Case CaseControl Control
2
4
6
8
10
d
Risk factors
(for example, FLG
mutation, altered
microbiome, allergens)
Colonization by
opportunistic
pathogens
Immune
activation,
inflammation
AD flare skin
log10(LDA score)
−3 −2 −1 0 1 2 3 4
Protein processing in endoplasmic reticulum
D-Glutamine and D-glutamate metabolism
D-Alanine metabolism
Peptidoglycan biosynthesis
Homologous recombination
Lysine biosynthesis
Galactose metabolism
Phenylalanine tyrosine and tryptophan biosynthesis
Mismatch repair
Base excision repair
Protein export
ABC transporters
Nitrogen metabolism
Arginine and proline metabolism
Phosphonate and phosphinate metabolism
Bacterial secretion system
Ubiquitin mediated proteolysis
Meiosis (yeast)
Regulation of autophagy
SNARE interactions in vesicular transport
Ribosome biogenesis in eukaryotes
N-Glycan biosynthesis
Peroxisome
MAPK signalling pathway (yeast)
Tryptophan metabolism
Proteasome
Steroid biosynthesis
Glycosylphosphatidylinositol GPI anchor biosynthesis
Spliceosome
Increase in
pathobionts Loss of water,
altered lipids,
pH change
Arginine
Citrulline
Carbamoyl-P
L
o
s
s
o
f
h
o
m
e
o
s
t
a
s
i
s
L
o
s
s
o
f
i
m
m
u
n
e
d
a
m
p
i
n
g
Ammonia (pH )
Ammonia (pH )
Normal skin
Figure 5 | Functional and metabolic shifts in the AD skin microbiome. a, Pathways that were found to be differentially abundant between the skin
microbiomes of normal and non-flare AD individuals. b, Reaction steps for the metabolic conversion of arginine to ammonia. c, Box plots showing the relative
abundance of core enzymes (EC, Enzyme Commission identifier number) for converting arginine to ammonia in the skin microbiomes of normal controls and
AD cases. All Pvalues were calculated using one-sided student’st-test and combined using Fisher’s combined probability test (***P< 0.01, **P< 0.05,
*P<0.1).d, A model for microbiome-dependent progression from normal skin to AD flare skin involving a feedback loop between skin microbiota, skin
microenvironment and host immune response. Box plot whiskers represent 1.5 × interquartile range or the maximum/minimum data point within the range.
NATURE MICROBIOLOGY DOI: 10.1038/NMICROBIOL.2016.106 ARTICLES
NATURE MICROBIOLOGY |www.nature.com/naturemicrobiology 7
© 2016 Macmillan Publishers Limited. All rights reserved
5 min and the supernatant was treated with Proteinase K (Qiagen) and incubated at
56 °C for 15 min. DNA was extracted using an EZ1 Advanced XL Instrument
(Qiagen) with the EZ1 DNA Tissue Kit (Qiagen) and was quantified using the Qubit
dsDNA HS Assay Kit (Life Technologies) and later stored at −20 °C.
16S PCR. PCR was carried out using primers 338F (5′-ACTCCTACGGGAGGCWG
C-3′) and 1061R (5′-CRRCACGAGCTGACGAC-3′) and the HotStar HiFidelity
Polymerase Kit (Qiagen) according to the manufacturer’s manual with the exception
of primer concentrations (0.5 µM), and the addition of MgSO
4
at a final
concentration of 2 mM. The conditions for PCR were as follows: initial denaturation
at 95 °C for 5 min, 35 cycles of denaturation at 95 °C for 30 s, annealing at 59 °C for
30 s, and extension at 72 °C for 1 min, followed by a final extension at 72 °C for
6 min. PCR products were purified with Agencourt AMPure XP (Beckman Coulter),
and amplicons were visualized using an Agilent Bioanalyzer, prepared with an
Agilent DNA1000 Kit (Agilent Technologies).
Library construction. Adaptive Focused Acoustics (Covaris) was used to shear a
standard volume of 50 µl of the extracted DNA. DNA libraries were prepared using
GeneRead DNA Library I Core Kit (Qiagen) according to the manufacturer’s
protocol with the use of barcode adapters in place of the GeneRead Adapter I Set.
Custom index-primers were used for enrichment of the DNA libraries, which was
performed according to an enrichment protocol adapted from the Multiplexing
Sample Preparation Oligonucleotide kit (Illumina). Libraries were quantified using
an Agilent Bioanalyzer, prepared with the Agilent High Sensitivity DNA Kit (Agilent
Technologies). Of the 80 libraries built for profiling normal controls, the AD group
and SPT+ volunteers, one library was unsuccessful. Thus, its accompanying pair was
omitted from further analysis. Paired-end sequencing (2 × 101 bp reads) was
performed on successful DNA libraries using an Illumina HiSeq 2000 instrument at
the Genome Institute of Singapore to generate around 589 million paired-end reads in
total and 7.55 million paired-end reads on average per library.
Read preprocessing. Sequencing bases with quality scores lower than 3 were
trimmed off the 3′ends of reads. Read pairs with a read shorter than 60 bp were
discarded. For the 16S sequencing data sets, reads from the target region were
identified by mapping (using BWA-MEM with default parameters) to the 16S
database provided with the program EMIRGE56. For the whole metagenome
sequencing data sets, human reads were filtered out by mapping to the human
genome (hg19) using BWA-MEM57 with default parameters (version 0.7.9a).
Overall, around 27.4% of the reads were found to be usable after filtering for
human contamination.
Community profiling, differential abundance and correlation analysis.
16S profiling with filtered 16S reads was performed as described previously58.Inbrief,
EMIRGE56, a probabilistic expectation-maximization-based algorithm, was used to
de novo reconstruct and measure the abundances of 16S rRNA sequences (with default
parameters). The reconstructed sequences were then taxonomically classified using
BLAST against the NR database. For whole metagenome shotgun profiling,
MetaPhlAn59 or MetaPhlAn v2.0 (default parameters) was used to determine
bacterial, viral and eukaryotic community abundances. Pearson correlation matrices
between different sampling and profiling strategies were plotted using the R package
‘corrplot’. Differentiallyabundant taxa between cases and controls were conservatively
identified using the non-parametric Wilcoxon rank-sum test (two-sided) and Pvalues
were corrected using p.adjust in R using the false discovery rate option. Bacterial
genera with mean abundance of less than 0.1% (as determined by MetaPhlAn) across
AD cases and normal controls were excluded for this analysis. Box plots were drawn
with the parameter ‘outpch = NA’(outliers not shown) using R. Principle coordinates
analysis of the genus composition was performed based on Bray–Curtis dissimilarity
using the R package ‘phyloseq’. Correlations between bacterial taxa were used to
identify potential bacterial interactions in skin flora. For this analysis, taxawith a mean
abundance (across all samples) below 0.1% were excluded and taxa pairs with a
Spearman correlation coefficient (ρ) greater than 0.3 (absolute value) and FDR
adjusted P<0.05 were reported (Supplementary Table 6).
Bacteria strain variation analysis. To identify the S. aureus and S. epidermidis
strains present, reference genomes were downloaded from NCBI (ftp://ftp.ncbi.nlm.
nih.gov/genomes/archive/old_refseq/; 39 complete genomes for S. aureus,
2 complete and 61 draft genomes for S. epidermidis). Microbiome reads from all
samples were pooled and mapped to the genomes of S. aureus using Pathoscope2
(default parameters) with the genomes of S. epidermidis as decoys (to identify
S. aureus strains) and vice versa (to identify S. epidermidis strains). Potential false
positives were filtered out based on <10% of the strain genome (in 10 kbp non-
overlapping bins) having read coverage38. Strain genomes with less than 1% of total
mapped reads were also excluded before remapping the microbiome reads from each
individual to selected genomes to quantify strain abundances. The reference strains
selected for S. aureus were CN1 (uid217769; strain-5), JH1 (uid58457; strain-4),
MSHR1132 (uid89393; strain-3), MSSA476 (uid57841; strain-2), ST398 (uid159247;
strain-1), and for S. epidermidis were A487 (uid199704; strain-6), M23864_W1
(uid55897 strain-5), NIHLM023 (uid180512; strain-4), NIHLM039 (uid180510;
strain-3), SK135 (uid42967; strain-2) and VCU129 (uid180073; strain-1).
Differential SNP analysis of staphylococcal virulence genes. Virulence factors of
S. aureus MW2 were downloaded from the virulence factors database60 (http://www.
mgc.ac.cn/VFs/). Microbiome reads from cases and controls were mapped against
virulence genesequences using BWA-MEM57 (version 0.7.9a) with default parameters.
For each sample, we created a per-base pileup profile using the plpsummary module in
LoFreq (version 2.1.1) (with ‘–use-orphan’option to include orphan reads)61. For each
position, we only considered the two most frequent alleles across all the samples, and
we tested the difference in allele frequencies between the two groups (case and control)
using the non-parametric Wilcoxon rank-sum test (two-sided).
Agar diffusion assays. The bacterial species studied here (Staphylococcus aureus
31240, Staphylococcus epidermidis 12228, Streptococcus cristatus 51100,
Streptococcus mitis 49456, Streptococcus pyogenes 49117, Streptococcus salivarius
7073 and Streptococcus sanguinis 49295) were obtained from ATCC. Bacteria were
cultured in BHI broth (Supplementary Table 9) at 37 °C overnight and the optical
density at 600 nm (OD
600nm
) was measured for each culture. Each culture was
diluted to an OD
600nm
of 0.1 with fresh media (with the exception of S. sanguinis,
which was diluted to OD
600nm
= 0.4) in a total of 4 ml BHI broth and incubated
at 37 °C with agitation. Bacterial cultures were allowed to grow to an OD that
corresponds to the log-phase of the bacteria (that is, S. epidermidis, OD = 1.9;
S. mitis, OD = 1.7; S. salivarius, OD = 2.3; S. pyogenes, OD = 1.5; S. cristatus,
OD = 1.8; S. sanguinis, OD =2.1). The S. aureus bacterial lawn was made by diluting
the culture to afinal concentration of 10
6
c.f.u. ml
−1
in 8 ml 0.7% BHI agar and gently
mixing the agar by rotating it in a 50 ml falcon tube. The bacteria/agar mixture was
poured over a BHI agar plate and left to solidify. Filter paper discs (Whatman filter
paper no. 1, 1.5 cm) were then submerged in other bacterial cultures until they were
saturated, and then gently layered over the solidified bacterial lawn. The agar plate was
incubated at 37 °C and checked for clearing zones the following day.
Bacterial cultures for keratinocyte and moDC challenge assays. We cultured
six bacteria (identities confirmed by Sanger sequencing of the 16S rRNA gene):
Dermacoccus profundi BAA-2375, Dermacoccus nishinomiyaensis 29093,
Deinococcus radiodurans 13939, Staphylococcus aureus subsp. aureus 31240,
Staphylococcus epidermidis 12228 and a Bacillus firmus isolate. The bacteria were
grown under culture conditions as summarized in Supplementary Table 8 for either
24 or 48 h. The bacteria were grown overnight or over two nights and we then
measured the OD
600nm
. The ODs were calculated and brought to OD
600nm
= 1 and
were further diluted to 0.01 with fresh media. The OD was measured at 1 or 3 h
intervals, depending on whether the bacterial growth conditions were 24 or 48 h,
respectively. In parallel, bacterial cultures were also plated on agar of their
corresponding growth medium at a dilution of 10
−3
to 10
−7
in replicates of four, for
each time point. Agar plates were incubated overnight or over two nights. The
colony-forming units per millilitre (c.f.u. ml
−1
) were then calculated and the growth
curves were plotted. We next observed the mid-log phase of bacterial growth. This
was the OD we selected at which to collect the bacteria for subsequent bacterial
agonist immunological assays. Three conditions were collected at mid-log phase:
bacterial media alone, bacterial culture and supernatant of bacterial culture. Optical
densities were taken once again for comparison to the original growth curves to
verify that the growth kinetics had not changed between the bacterial growth
experiments. Bacteria collected at the mid-log phase were normalized to the lowest
c.f.u. ml
−1
bacterial count so that equal amounts of c.f.u. were added to the
subsequent bacterial agonist immunological assays.
Keratinocyte cell culture and profiling. Normal human epidermal keratinocytes
(NHEKs) immortalized with the telomerase catalytic subunit, named N/TERT-162 were
cultured in K-SFM (Life Technologies) supplemented with 0.2 ng ml
−1
epidermal
growth factor (EGF), 25 µg ml
−1
bovine pituitary extract (BPE), 0.4 mM CaCl
2
and
penicillin/streptomycin (1×). The cells were tested to be negative for mycoplasma
contamination using the MycoAlert PLUS mycoplasma detection kit (Lonza)
according to the manufacturer’s instructions. For bacterialchallenge assays, N/TERT-1
cells were grown to 80% confluence in 10 cm dishes by switching the culture medium
to an equal ratio mixture of K-SFM medium and DF-K medium, which is Ham’sF-12
and Ca
2+
- and glutamine-free DMEM (1:3) supplemented with 0.2 ng ml
−1
EGF,
25 µg ml
−1
BPE, 1.5 mM glutaminewithout the addition of antibiotics. After 24 or48 h,
supernatants were collected from the wells for cytokine profiling.
Dendritic cell culture and profiling. moDCs were generated as reported
previously63. Briefly, peripheral blood mononuclear cells from healthy donors
(Institutional Review Board approval NUS-IRB 10-250) were isolated by density
centrifugation (Ficoll-Paque; GE Healthcare) and monocytes were isolated by
positive selection of CD14
+
cells with CD14 microbeads (Miltenyi Biotec).
Monocytes were cultured in antibiotic-free medium (RPMI 1640 with 10% FBS,
1 mM pyruvate, 2 mM L-glutamine, 50 mM 2-ME, non-essential amino acids and
15 mM HEPES) and supplemented with GM-CSF and IL-4 (both 1,000 U ml
−1
,
Miltenyi Biotec). At days 4 and 6, fresh cytokines were added into the culture. Cells
were either treated with different concentrations of bacteria, supernatant or bacterial
culture medium at day 6 or left untreated. At day 8, immature (untreated),
bacterial/supernatant/medium-treated moDCs were collected and washed for
further experiments.
ARTICLES NATURE MICROBIOLOGY DOI: 10.1038/NMICROBIOL.2016.106
NATURE MICROBIOLOGY |www.nature.com/naturemicrobiology8
© 2016 Macmillan Publishers Limited. All rights reserved
Cytokine analysis for keratinocytes and moDCs. For cytokine profiling,
supernatants from keratinocyte and moDC culture wells were collected and then
used for multiplex analysis by a human cytokine/chemokine bead panel 1
(MilliplexMap, Millipore) and Flexmap 3D system (Luminex). The levels of 30
protein content (epidermal growth factor (EGF), eotaxin, fibroblast growth factor 2
(FGF-2), Fms-like tyrosine kinase 3 (Flt-3), fractalkine, granulocyte colony-
stimulating factor (G-CSF), growth-regulated oncogene (GRO), interferon-alpha 2
(IFNa2), interleukin-10 (IL-10), IL-12p40, IL-12p70, IL-1Ra, IL-1a, IL-1b, IL-6,
IL-8, interferon gamma-induced protein 10 (IP-10), macrophage inflammatory
protein-1 (MCP-1), MCP-3, macrophage-derived chemokine (MDC), macrophage
inflammatory protein-1 (MIP-1α), MIP-1β, platelet derived growth factor-AB/BB
(PDGF-AA), PDGF-AB/BB, regulated upon activation, normal T cell expressed and
secreted (RANTES), transforming growth factor-α(TGF-α), TNF-α, TNF-β,
vascular endothelial growth factor (VEGF), soluble CD40 ligand (sCD40L)) were
measured from the cell culture supernatants.
Preparation and isolation of epidermal LCs. Human skin samples were obtained
with approval and in accordance with the Singapore National Healthcare Group
Research Ethics Committee. Epidermal cell suspensions were prepared as described
previously64. Briefly, strips of 200 µm whole skin sections were incubated with
2Uml
−1
dispase (ThermoFisher Scientific) in RPMI 1640 at 37 °C for 1 h to
separate the epidermis from dermis. Epidermal sheets were further digested with
0.80 mg ml
−1
collagenase (Type IV, Worthington-Biochemical) in RPMI 1640 with
10% FCS and 0.05 mg ml
−1
DNase (Sigma-Aldrich) at 37 °C overnight. Epidermal
LCs (CD1a
hi
Langerin
hi
) were isolated by fluorescence activated cell sorting (FACS)
with a FACSAriaII (BD Biosciences).
Epidermal LCs from mice were isolated as described previously65. Sex (male) and
age (7–8 weeks) matched C57BL/6 were used. Approval was obtained from the
Institutional Animal Care and Use Committee, Biological Resource Center,
A*STAR, Singapore. Briefly, mouse ears were split into dorsal and ventral halves and
floated in 4 U ml
−1
dispase in RPMI 1640 at 37 °C for 1 h to allow separation of the
epidermal and dermal sheets. Epidermal sheets were cut into small pieces and
incubated in 0.1 mg ml
−1
collagenase type IV (Sigma-Aldrich) at 37 °C for 1 h. Cells
were passed through a 19 G syringe and filtered through a 70 µM nylon mesh to
obtain a homogeneous cell suspension. LCs (CD24
hi
EpCAM
hi
) were isolated
by FACS.
Flow cytometry. The following anti-human antibodies were used: V500-conjugated
CD45 (clone HI30, BD Biosciences), PerCp/Cy5.5-conjugated HLA-DR (clone LN3,
eBioscience), APC/Cy7-conjugated CD14 (clone HCD14, BioLegend), Alexa Fluor
700-conjugated CD1a (clone HI149, BioLegend) and PE-conjugated Langerin (clone
DCGM4, Beckman Coulter). The following anti-mouse antibodies were used: FITC-
conjugated CD45 (clone 30-F11, BioLegend), Alexa Fluor 700-conjugated MHC II
(clone M5/114.15.2, eBioscience), PE-conjugated CD103 (clone 2E7, eBioscience),
PerCp/Cy5.5-conjugated CD11b (clone M1/70, eBioscience), PE/Cy7-conjugated
Ep-CAM/CD326 (clone G8.8, BioLegend), eFluor 450-conjugated CD24 (clone
M1/69, eBioscience).
Bacterial cultures for human and mouse LC challenge assays. Four bacteria,
D. nishinomiyaensis 29093, D. radiodurans 13939, S. aureus 31240 and
S. epidermidis 12228, were grown overnight in BHI media at 37 °C. OD
600nm
was
measured for overnight cultures, and fresh cultures were prepared for each bacteria,
starting from an OD of 0.01. Bacterial cultures were grown to the following ODs:
S. aureus, OD = 1.9; S. epidermidis,OD=2;D. radiodurans,OD=2;
D. nishinomiyaensis, OD = 3. Cultures were centrifuged at 8,000g for 15 min to
sediment the cells. Supernatant was filtered with a 0.22 µm syringe filter, and 100 µl
supernatant was plated onto BHI agar plates to check for the presence of bacterial
cells. Heat inactivation of bacteria was performed by placing cultures in a 70 °C
water bath for 3 h, and heat inactivation was confirmed by the absence of detectable
bacteria on BHI agar plates for up to 3 days. Cells were seeded in 96-well V-bottom
plates (Costar, Corning) for the bacterial challenge assays. Sterile filtered bacterial
supernatants were added to the wells at a final dilution of 100× and incubated for
24 h at 37 °C. For bacterial co-culture, LCs were challenged with bacteria at a
multiplicity of infection (MOI) of 0.1 for 24 h. After incubation, cell supernatants
were collected for cytokine analysis. Negative control cultures contained cells
co-incubated with BHI medium.
Cytokine analysis for human and mouse LCs. Cytokine production was measured
by immunoassay with the human or mouse cytokine/chemokine bead panel
(MilliplexMap, Millipore) and Flexmap 3D system (Luminex). For human samples,
EGF, eotaxin, FGF-2, Flt-3L, fractalkine, G-CSF, GM-CSF, GRO, IFNα2, IFNγ,
IL-10, IL-12p40, IL-12p70, IL-13, IL-15, IL-17A, IL-1RA, IL-1α, IL-1β, IL-2, IL-3,
IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IP-10, MCP-1, MCP-3, MDC, MIP-1α, MIP-1β,
PDGF-AA, PDGF-AB, RANTES, TGF-α, TNF-α, TNF-β, VEGF and sCD40L were
analysed. For mouse samples, eotaxin, G-CSF, GM-CSF, IFNγ, IL-10, IL-12p40,
IL-12p70, IL-13, IL-15, IL-17, IL-1α, IL-1β, IL-2, IL-3, IL-4, IL-5, IL-6, IL-7, IL-9,
IP-10, KC, LIF, LIX, M-CSF, MCP-1, MIG, MIP-1α, MIP-1β, MIP-2, RANTES and
TNF-αwere analysed.
Abundance profiling of Malassezia species. PathoScope 2.0 (ref. 66) was used to
quantify the relative abundance of all 14 known Malassezia species by mapping
microbiome reads to the Malassezia reference genomes (target) and a set of other
fungal genomes that served as decoys (control; Supplementary Table 10). The
approach used in PathoScope first determines reads that have better or unique
alignments to the target genomes (compared to the control references) and then uses
an expectation-maximization algorithm to reassign reads to the most plausible
reference sources and estimate their relative abundances. To further eliminate false
positives, we binned the genomes into 1 kb bins and only kept species assignments
where more than 10% of the bins were covered by at least one read. Species
abundances were then renormalized after this filter.
Pathway analysis. Microbiome reads from cases and controls were mapped to
KEGG orthologues using Bowtie2 (ref. 67; version 2.2.4; parameter setting: -k 10).
Pathway abundances for cases and controls were estimated using HUMAnN68
(version 0.99) with default parameters. The abundance table computed by
HUMAnN was then analysed with the linear discriminant analysis approach in
LEfSe69 (version 1.1.0) to identify differentially abundant pathways.
Accession codes. All sequencing reads have been deposited in the NCBI short read
archive (SRA) under accession no. SRP056821.
Received 12 October 2015; accepted 1 June 2016;
published 11 July 2016
References
1. Tay, Y. K., Kong, K. H., Khoo, L., Goh, C. L. & Giam, Y. C. The prevalence and
descriptive epidemiology of atopic dermatitis in Singapore school children.
Br. J. Dermatol. 146, 101–106 (2002).
2. Asher, M. I. et al. Worldwide time trends in the prevalence of symptoms of
asthma, allergic rhinoconjunctivitis, and eczema in childhood: ISAAC Phases
One and Three repeat multicountry cross-sectional surveys. Lancet 368,
733–743 (2006).
3. Garmhausen, D. et al. Characterization of different courses of atopic dermatitis
in adolescent and adult patients. Allergy 68, 498–506 (2013).
4. Paternoster, L. et al. Multi-ancestry genome-wide association study of 21,000
cases and 95,000 controls identifies new risk loci for atopic dermatitis. Nature
Genet. 47, 1449–1456 (2015).
5. Palmer, C. N. et al. Common loss-of-function variants of the epidermal barrier
protein filaggrin are a major predisposing factor for atopic dermatitis. Nature
Genet. 38, 441–446 (2006).
6. van den Oord, R. A. & Sheikh, A. Filaggrin gene defects and risk of developing
allergic sensitisation and allergic disorders: systematic review and meta-analysis.
Br. Med. J. 339, b2433 (2009).
7. Chen, H. et al. Wide spectrum of filaggrin-null mutations in atopic dermatitis
highlights differences between Singaporean Chinese and European populations.
Br. J. Dermatol. 165, 106–114 (2011).
8. Sandilands, A., Sutherland, C., Irvine, A. D. & McLean, W. H. Filaggrin in
the frontline: role in skin barrier function and disease. J. Cell Sci. 122,
1285–1294 (2009).
9. Howell, M. D. et al. Th2 cytokines act on S100/A11 to downregulate keratinocyte
differentiation. J. Invest. Dermatol. 128, 2248–2258 (2008).
10. Morizane, S. et al. TH2 cytokines increase kallikrein 7 expression and
function in patients with atopic dermatitis. J. Allergy Clin. Immunol. 130, 259–61
e1 (2012).
11. Howell, M. D. et al. Cytokine modulation of atopic dermatitis filaggrin skin
expression. J. Allergy Clin. Immunol. 124, R7–R12 (2009).
12. Hvid, M. et al. Regulation of caspase 14 expression in keratinocytes by
inflammatory cytokines—a possible link between reduced skin barrier function
and inflammation? Exp. Dermatol. 20, 633–636 (2011).
13. Brown, S. J. et al. Filaggrin haploinsufficiency is highly penetrant and is
associated with increased severity of eczema: further delineation of the skin
phenotype in a prospective epidemiological study of 792 school children.
Br. J. Dermatol. 161, 884–889 (2009).
14. Fallon, P. G. et al. A homozygous frameshift mutation in the mouse Flg gene
facilitates enhanced percutaneous allergen priming. Nature Genet. 41,
602–608 (2009).
15. Elias, P. M., Hatano, Y. & Williams, M. L. Basis for the barrier abnormality in
atopic dermatitis: outside–inside–outside pathogenic mechanisms. J. Allergy
Clin. Immunol. 121, 1337–1343 (2008).
16. Kabashima, K. New concept of the pathogenesis of atopic dermatitis: interplay
among the barrier, allergy, and pruritus as a trinity. J. Dermatol. Sci. 70,
3–11 (2013).
17. Leyden, J. J., Marples, R. R. & Kligman, A. M. Staphylococcus aureus in the
lesions of atopic dermatitis. Br. J. Dermatol. 90, 525–530 (1974).
18. Huang, J. T., Abrams, M., Tlougan, B., Rademaker, A. & Paller, A. S. Treatment
of Staphylococcus aureus colonization in atopic dermatitis decreases disease
severity. Pediatrics 123, e808–e814 (2009).
NATURE MICROBIOLOGY DOI: 10.1038/NMICROBIOL.2016.106 ARTICLES
NATURE MICROBIOLOGY |www.nature.com/naturemicrobiology 9
© 2016 Macmillan Publishers Limited. All rights reserved
19. Kong, H. H. et al. Temporal shifts in the skin microbiome associated with disease
flares and treatment in children with atopic dermatitis. Genome Res. 22,
850–859 (2012).
20. Oh, J. et al. The altered landscape of the human skin microbiome in patients
with primary immunodeficiencies. Genome Res. 23, 2103–2114 (2013).
21. Oh, J. et al. Biogeography and individuality shape function in the human skin
metagenome. Nature 514, 59–64 (2014).
22. Andiappan, A. K. et al. Validation of GWAS loci for atopic dermatitis in a
Singapore Chinese population. J. Invest. Dermatol. 132, 1505–1507 (2012).
23. Grice, E. A. et al. A diversity profile of the human skin microbiota. Genome
Res. 18, 1043–1050 (2008).
24. Mathieu, A., Vogel, T. M. & Simonet, P. The future of skin metagenomics.
Res. Microbiol. 165, 69–76 (2014).
25. Asher, M. I. et al. International study of asthma and allergies in childhood
(ISAAC): rationale and methods. Eur. Respir. J. 8, 483–491 (1995).
26. Manivasagan, P., Venkatesan, J., Sivakumar, K. & Kim, S. K. Pharmaceutically
active secondary metabolites of marine actinobacteria. Microbiol. Res. 169,
262–278 (2014).
27. Chiller, K., Selkin, B. A. & Murakawa, G. J. Skin microflora and bacterial
infections of the skin. J. Invest. Dermatol. Symp. Proc. 6, 170–174 (2001).
28. Wang, Y. et al. Staphylococcus epidermidis in the human skin microbiome
mediates fermentation to inhibit the growth of Propionibacterium acnes:
implications of probiotics in acne vulgaris. Appl. Microbiol. Biotechnol. 98,
411–424 (2014).
29. Schloissnig, S. et al. Genomic variation landscape of the human gut microbiome.
Nature 493, 45–50 (2013).
30. Kapral, F. A., Smith, S. & Lal, D. The esterification of fatty acids by
Staphylococcus aureus fatty acid modifying enzyme (FAME) and its inhibition by
glycerides. J. Med. Microbiol. 37, 235–237 (1992).
31. Lowe, A. M., Beattie, D. T. & Deresiewicz, R. L. Identification of novel
staphylococcal virulence genes by in vivo expression technology. Mol. Microbiol.
27, 967–976 (1998).
32. Mlckova, P., Cechova, D., Chalupna, P., Novotna, O. & Prokesova, L. Enhanced
systemic and mucosal antibody responses to a model protein antigen after
intranasal and intratracheal immunisation using Bacillus firmus as an adjuvant.
Immunol. Lett. 77, 39–45 (2001).
33. Lutz, M. B. et al. An advanced culture method for generating large quantities of
highly pure dendritic cells from mouse bone marrow. J. Immunol. Methods 223,
77–92 (1999).
34. Mitsui, H. et al. Differential expression and function of Toll-like receptors in
Langerhans cells: comparison with splenic dendritic cells. J. Invest. Dermatol.
122, 95–102 (2004).
35. Malissen, B., Tamoutounour, S. & Henri, S. The origins and functions of dendritic
cells and macrophages in the skin. Nature Rev. Immunol. 14, 417–428 (2014).
36. Lubbe, J. Secondary infections in patients with atopic dermatitis. Am. J. Clin.
Dermatol. 4, 641–654 (2003).
37. Darabi, K., Hostetler, S. G., Bechtel, M. A. & Zirwas, M. The role of Malassezia in
atopic dermatitis affecting the head and neck of adults. J. Am. Acad. Dermatol.
60, 125–136 (2009).
38. Wu, G. et al. Genus-wide comparative genomics of malassezia delineates its
phylogeny, physiology, and niche adaptation on human skin. PLoS Genet. 11,
e1005614 (2015).
39. Casagrande, B. F. et al. Sensitization to the yeast Malassezia sympodialis is
specific for extrinsic and intrinsic atopic eczema. J. Invest. Dermatol. 126,
2414–2421 (2006).
40. Cork, M. J. et al. Epidermal barrier dysfunction in atopic dermatitis. J. Invest.
Dermatol. 129, 1892–1908 (2009).
41. Moffett, J. R. & Namboodiri, M. A. Tryptophan and the immune response.
Immunol. Cell Biol. 81, 247–265 (2003).
42. Lee, V. T. & Schneewind, O. Protein secretion and the pathogenesis of bacterial
infections. Genes Dev. 15, 1725–1752 (2001).
43. Zgur-Bertok, D. DNA damage repair and bacterial pathogens. PLoS Pathogens 9,
e1003711 (2013).
44. Cusumano, Z. T., Watson, M. E.Jr & Caparon, M. G. Streptococcus pyogenes
arginine and citrulline catabolism promotes infection and modulates innate
immunity. Infect. Immun. 82, 233–242 (2014).
45. Grice, E. A. & Segre, J. A. The skin microbiome. Nature Rev. Microbiol. 9,
244–253 (2011).
46. Dekio, I. et al. Characterization of skin microbiota in patients with atopic
dermatitis and in normal subjects using 16S rRNA gene-based comprehensive
analysis. J. Med. Microbiol. 56, 1675–1683 (2007).
47. Garcia-Garcera, M., Garcia-Etxebarria, K., Coscolla, M., Latorre, A. & Calafell, F.
A new method for extracting skin microbes allows metagenomic analysis of
whole-deep skin. PLoS One 8, e74914 (2013).
48. Oh, J., Conlan, S., Polley, E. C., Segre, J. A. & Kong, H. H. Shifts in human skin
and nares microbiota of healthy children and adults. Genome Med. 4, 77 (2012).
49. Conlan, S. et al. Staphylococcus epidermidis pan-genome sequence analysis
reveals diversity of skin commensal and hospital infection-associated isolates.
Genome Biol. 13, R64 (2012).
50. Zhang, E. et al. Anti-malassezia-specific IgE antibodies production in Japanese
patients with head and neck atopic dermatitis: relationship between the level of
specific IgE antibody and the colonization frequency of cutaneous Malassezia
species and clinical severity. J. Allergy (Cairo) 2011, 645670 (2011).
51. Andiappan, A. K. et al. Allergic airway diseases in a tropical urban environment
are driven by dominant mono-specific sensitization against house dust mites.
Allergy 69, 501–509 (2014).
52. Broccardo, C. J., Mahaffey, S. B., Strand, M., Reisdorph, N. A. & Leung, D. Y.
Peeling off the layers: skin taping and a novel proteomics approach to study
atopic dermatitis. J. Allergy Clin. Immunol. 124, 1113–1115 (2009).
53. Kezic, S., Kammeyer, A., Calkoen, F., Fluhr, J. W. & Bos, J. D. Natural
moisturizing factor components in the stratum corneum as biomarkers of
filaggrin genotype: evaluation of minimally invasive methods. Br. J. Dermatol.
161, 1098–1104 (2009).
54. Williamson, P. & Kligman, A. M. A new method for the quantitative
investigation of cutaneous bacteria. J. Invest. Dermatol. 45, 498–503 (1965).
55. Pachtman, E. A., Vicher, E. E. & Brunner, M. J. The bacteriologic flora in
seborrheic dermatitis. J. Invest. Dermatol. 22, 389–396 (1954).
56. Miller, C. S., Baker, B. J., Thomas, B. C., Singer, S. W. & Banfield, J. F. EMIRGE:
reconstruction of full-length ribosomal genes from microbial community short
read sequencing data. Genome Biol. 12, R44 (2011).
57. Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows–
Wheeler transform. Bioinformatics 25, 1754–1760 (2009).
58. Ong, S. H. et al. Species identification and profiling of complex microbial
communities using shotgun Illumina sequencing of 16S rRNA amplicon
sequences. PLoS One 8, e60811 (2013).
59. Segata, N. et al. Metagenomic microbial community profiling using unique
clade-specific marker genes. Nature Methods 9, 811–814 (2012).
60. Chen, L. et al. VFDB: a reference database for bacterial virulence factors. Nucleic
Acids Res. 33, D325–D328 (2005).
61. Wilm, A. et al. LoFreq: a sequence-quality aware, ultra-sensitive variant caller for
uncovering cell-population heterogeneity from high-throughput sequencing
datasets. Nucleic Acids Res. 40, 11189–11201 (2012).
62. Dickson, M. A. et al. Human keratinocytes that express hTERT and also bypass a
p16(INK4a)-enforced mechanism that limits life span become immortal yet
retain normal growth and differentiation characteristics. Mol. Cell. Biol. 20,
1436–1447 (2000).
63. Malinarich, F. et al. High mitochondrial respiration and glycolytic capacity
represent a metabolic phenotype of human tolerogenic dendritic cells.
J. Immunol. 194, 5174–5186 (2015).
64. Haniffa, M. et al. Human tissues contain CD141
hi
cross-presenting dendritic
cells with functional homology to mouse CD103
+
nonlymphoid dendritic cells.
Immunity 37, 60–73 (2012).
65. Ginhoux, F. et al. Blood-derived dermal langerin
+
dendritic cells survey the skin
in the steady state. J. Exp. Med. 204, 3133–3146 (2007).
66. Hong, C. et al. PathoScope 2.0: a complete computational framework for strain
identification in environmental or clinical sequencing samples. Microbiome 2,
33 (2014).
67. Langmead, B. & Salzberg, S. L. Fast gapped-read alignment with Bowtie 2.
Nature Methods 9, 357–359 (2012).
68. Abubucker, S. et al. Metabolic reconstruction for metagenomic data and its
application to the human microbiome. PLoS Comput. Biol. 8, e1002358 (2012).
69. Segata, N. et al. Metagenomic biomarker discovery and explanation. Genome
Biol. 12, R60 (2011).
Acknowledgements
The authors thank all volunteers for donating their skin microbiome samples, the
Biological Resource Centre at A*STAR for assistancewith animal work, the Flow Cytometry
Facility at the Singapore Immunology Network, A*STAR, for assistance with cell sorting,
and G. Low, H. Chan and M. Un naamalai for assistance with the isolation of epidermal LCs.
This workwas funded by an SRGNational Healthcare Groupgrant to J.C., N.N.and M.B.Y.T.
and A*STAR SPF funding for basic and translational skin research.
Author contributions
K.R.C., M.B.Y.T., F.T.C., J.E.A.C. and N.N. planned the study. A.S.L.T., A.H.Q.N., J.W.,
N.M., B.J., X.F.C.C.W. and B.V.A. designed and conducted all experiments under the
guidance of J.E.A.C., K.R.C., P.F.D.S., F.G. and J.E.C. Metagenomic data analysis was
performed by K.R.C. and C.L. with supervision from N.N. Human skin tissue samples were
collected by T.C.L. and A.W. provided technical computational assistance. B.K.S., S.A.M.,
Y.Y.S., F.T.C. and J.E.A.C. organized volunteer recruitment and sampling. K.R.C., A.S.L.T.,
E.B.L., J.E.A.C. and N.N. wrote the manuscript, with input from all authors.
Additional information
Supplementary information is available online.
Reprints and permissions information is
available online at www.nature.com/reprints. Correspondence and requestsfor materials should
be addressed to J.E.A.C. and N.N.
Competing interests
The authors declare no competing financial interests.
ARTICLES NATURE MICROBIOLOGY DOI: 10.1038/NMICROBIOL.2016.106
NATURE MICROBIOLOGY |www.nature.com/naturemicrobiology10
© 2016 Macmillan Publishers Limited. All rights reserved
