Genome-wide expression profiling of five mouse models identifies similarities and differences with human psoriasis.

Abstract

Development of a suitable mouse model would facilitate the investigation of pathomechanisms underlying human psoriasis and would also assist in development of therapeutic treatments. However, while many psoriasis mouse models have been proposed, no single model recapitulates all features of the human disease, and standardized validation criteria for psoriasis mouse models have not been widely applied. In this study, whole-genome transcriptional profiling is used to compare gene expression patterns manifested by human psoriatic skin lesions with those that occur in five psoriasis mouse models (K5-Tie2, imiquimod, K14-AREG, K5-Stat3C and K5-TGFbeta1). While the cutaneous gene expression profiles associated with each mouse phenotype exhibited statistically significant similarity to the expression profile of psoriasis in humans, each model displayed distinctive sets of similarities and differences in comparison to human psoriasis. For all five models, correspondence to the human disease was strong with respect to genes involved in epidermal development and keratinization. Immune and inflammation-associated gene expression, in contrast, was more variable between models as compared to the human disease. These findings support the value of all five models as research tools, each with identifiable areas of convergence to and divergence from the human disease. Additionally, the approach used in this paper provides an objective and quantitative method for evaluation of proposed mouse models of psoriasis, which can be strategically applied in future studies to score strengths of mouse phenotypes relative to specific aspects of human psoriasis.

Full text

Page 1

Genome-Wide Expression Profiling of Five Mouse Models
Identifies Similarities and Differences with Human
Psoriasis
William R. Swindell1*, Andrew Johnston2, Steve Carbajal3, Gangwen Han4, Christian Wohn5, Jun Lu6,
Xianying Xing2, Rajan P. Nair2, John J. Voorhees2, James T. Elder2,7, Xiao-Jing Wang4, Shigetoshi Sano8,
Errol P. Prens9, John DiGiovanni10,11, Mark R. Pittelkow6, Nicole L. Ward12, Johann E. Gudjonsson2*
1Department of Genetics, Harvard Medical School, Boston, Massachusetts, United States of America, 2Department of Dermatology, University of Michigan Medical
School, Ann Arbor, Michigan, United States of America, 3Division of Pharmacology & Toxicology, College of Pharmacy, The University of Texas at Austin, Austin, Texas,
United States of America, 4Departments of Pathology, Otolaryngology and Dermatology, University of Colorado, Denver, Colorado, United States of America,
5Departments of Immunology, Erasmus MC, University Medical Center, Rotterdam, The Netherlands, 6Department of Dermatology, Mayo Clinic, Rochester, Minnesota,
United States of America, 7Ann Arbor Veterans Affairs Hospital, Ann Arbor, Michigan, United States of America, 8Department of Dermatology, Kochi Medical School,
Kochi University, Okocho, Nankoku, Japan, 9Departments of Dermatology and Rheumatology, Erasmus MC, University Medical Center, Rotterdam, The Netherlands,
10Department of Nutritional Sciences, Dell Pediatric Research Institute, The University of Texas at Austin, Austin, Texas, United States of America, 11Division of
Pharmacology & Toxicology, Dell Pediatric Research Institute, The University of Texas at Austin, Austin, Texas, United States of America, 12Department of Dermatology
and the Murdough Family Center for Psoriasis, Case Western Reserve University and University Hospitals, Case Medical Center, Cleveland, Ohio, United States of America
Abstract
Development of a suitable mouse model would facilitate the investigation of pathomechanisms underlying human psoriasis
and would also assist in development of therapeutic treatments.
proposed, no single model recapitulates all features of the human disease, and standardized validation criteria for psoriasis
mouse models have not been widely applied. In this study, whole-genome transcriptional profiling is used to compare gene
expression patterns manifested by human psoriatic skin lesions with those that occur in five psoriasis mouse models (K5-
Tie2, imiquimod, K14-AREG, K5-Stat3C and K5-TGFbeta1). While the cutaneous gene expression profiles associated with
each mouse phenotype exhibited statistically significant similarity to the expression profile of psoriasis in humans, each
model displayed distinctive sets of similarities and differences in comparison to human psoriasis. For all five models,
correspondence to the human disease was strong with respect to genes involved in epidermal development and
keratinization. Immune and inflammation-associated gene expression, in contrast, was more variable between models as
compared to the human disease. These findings support the value of all five models as research tools, each with identifiable
areas of convergence to and divergence from the human disease. Additionally, the approach used in this paper provides an
objective and quantitative method for evaluation of proposed mouse models of psoriasis, which can be strategically applied
in future studies to score strengths of mouse phenotypes relative to specific aspects of human psoriasis.
However, while many psoriasis mouse models have been
Citation: Swindell WR, Johnston A, Carbajal S, Han G, Wohn C, et al. (2011) Genome-Wide Expression Profiling of Five Mouse Models Identifies Similarities and
Differences with Human Psoriasis. PLoS ONE 6(4): e18266. doi:10.1371/journal.pone.0018266
Editor: Stefan Bereswill, Charite ´-University Medicine Berlin, Germany
Received February 11, 2011; Accepted February 23, 2011; Published April 4, 2011
Copyright: ? 2011 Swindell et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by the Babcock Endowment to Dr. Gudjonsson and Dr. Johnston, the Taubman Medical Research Institute support (Frances
and Kenneth Eisenberg Emerging Scholar) and the Dermatology Foundation to Dr. Gudjonsson. The project described was supported by NIH grant number
CA76520 to Dr. DiGiovanni, AR052889 and AR054966 to Dr. Elder, GM70966, CA79998 and CA89849 to Dr. Wang and P30AR39750 and P50AR05508, National
Psoriasis Foundation and the Murdough Family Center for Psoriasis to Dr. Ward. The funders had no role in study design, data collection and analysis, decision to
publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: wswindell@genetics.med.harvard.edu (WRS); johanng@med.umich.edu (JEG)
Introduction
Psoriasis is a chronic inflammatory disease that leads to
widespread development of erythematous plaques with adherent
silvery scales. The disease is believed to be primarily mediated by
T-cells, which release cytokines that stimulate keratinocyte (KC)
hyperproliferation and altered differentiation [1,2]. Development
of a single suitable animal model would greatly facilitate research
on the mechanism(s) of action that drive inflammatory and
autoimmune processes associated with psoriasis [3–5]. Such an
animal model, for example, would permit experiments using
genetically uniform subjects within a controlled environment, as
well as high-throughput screening of potential therapeutic agents.
Most clinical features of psoriasis, however, arise spontaneously
only in humans and closely related primate species [6,7]. The
laboratory mouse offers the most flexible experimental system for
development of new psoriasiform phenotypes and previous studies
have described transgenic or deletion mutants with skin conditions
that resemble human clinical psoriasis [3–5]. Given the many
disparities between human and mouse skin, it cannot be expected
that psoriasiform phenotypes in mice will mirror the human
disease in every respect. For instance, relative to human skin,
mouse skin has a denser distribution of hair follicles, thinner
epidermis, and an underlying cutaneous muscle layer that is
PLoS ONE | www.plosone.org1April 2011 | Volume 6 | Issue 4 | e18266

Page 2

generally absent in humans [4,8,9]. Additionally, mice possess
subsets of inflammatory cells that are absent in humans [10,11].
Despite these challenges, psoriasis mouse models have already
provided mechanistic insights into inflammatory skin diseases [12–
16].
An ideal model would realistically recapitulate a marked
epidermal hyperproliferation, thickening and altered differentia-
tion of the epidermis, an inflammatory infiltrate that includes T-
cells, altered vascularity, and responsiveness to current anti-
psoriatic therapies [3–5]. Mouse phenotypes that satisfy a majority
of these characteristics include; overexpression of the endothelial-
specific receptor tyrosine kinase in basal KCs (K5-Tie2) [12],
topical application of the toll-like receptor (TLR) agonist
imiquimod (IMQ) [13], overexpression of human amphiregulin
in the basal epidermal layer (K14-AREG) [14], basal KC-specific
overexpression of a constitutively active mutant of signal
transducer and activator of transcription 3 (K5-Stat3C) [15],
and overexpression of the latent form of transforming growth
factor beta 1 in basal KCs (K5-TGFb1) [16]. The phenotypes of
these models recapitulate key features of human psoriasis, but
differ with respect to the initiating (biochemical) events. The K14-
AREG and K5-Stat3C models involve epidermal overexpression
of a growth factor and signaling component, respectively, which
directly perturbs KC homeostasis, leading to elevated cytokine or
chemokine production and secondary inflammatory responses
[14,15]. The K5-Tie2 and K5-TGFb1phenotypes may also arise,
in part, from a direct perturbation of KC homeostasis; but in these
models, a role has also been postulated for growth factor release
and certain concurrent processes (e.g., angiogenesis, oxidative
stress accumulation, and/or basement membrane degradation),
which may also contribute to KC proliferation and initiation of
inflammatory cascades [12,16]. By comparison, IMQ-treated mice
develop a psoriasiform phenotype that may differ in fundamental
ways, since the phenotype arises from direct (over)stimulation of
the immune system. IMQ is a TLR7 agonist that drives skin
inflammation and immune cell infiltration, followed by KC
proliferation and enhanced dermal vascularity [13]. In mice,
these effects of IMQ produce red, scaly skin similar to human
psoriasis, and indeed, clinical observations have indicated that
IMQ can exacerbate psoriasis in patients [17–20].
The five psoriasiform phenotypes discussed above represent
potentially useful tools for studying human psoriasis. However,
further characterization of each model at the biochemical level is
needed for refinement and development of mouse models with
stronger similarity to clinical disease [3–5]. Along these lines, it has
been challenging to align separate analyses performed by different
laboratories, and no side-by-side evaluation has previously assessed
each model using a common methodological strategy. Some
investigators have closely followed a set of proposed guidelines for
validation of psoriasis mouse models, which are based on
phenotypic measures (e.g., acanthosis, vascularity, absence of
papillomatosis) and experimental demonstration (e.g., T-cell
dependence and drug response) [13]. This strategy has been
valuable, when applied, but is at best semi-quantitative and does
not clearly differentiate mouse phenotypes that satisfy most or all
of the proposed standards (e.g., K5-Tie2 and K5-Stat3C). For this
reason, phenotypic and experimental approaches should be
complemented, but not replaced, by additional strategies.
Genome-wide transcriptional profiling is an example of one such
complementary approach. Microarray analyses have identified
dramatic differences between psoriatic and normal skin from
patients, which involve altered expression of hundreds of genes
[21–24], and several recent large-scale genome-wide association
studies have confidently identified 25 psoriasis susceptibility loci
[25–30]. These findings underscore the polygenic basis of psoriasis
and suggest that pathogenesis involves numerous biochemical
pathways. An expression profiling approach can provide a ‘‘big
picture’’ characterization of this process and support development
of quantifiable metrics for evaluating similarity between human
psoriasis and mouse phenotypes, allowing investigators to
determine whether correspondence between human and mouse
model psoriatic phenotypes is larger than expected on the basis of
chance alone, and to decompose the global transcriptional
correspondence into finer parts and evaluate correspondence with
respect to specific pathways and pre-defined gene categories.
Furthermore, microarray-based evaluation facilitates objective
evaluation of mouse phenotypes, without ‘‘dependent variable
selection bias’’ or over-emphasis of characteristics that best
correspond between mouse phenotypes and human psoriasis.
In this study, global transcriptional profiling was utilized to
evaluate the similarity between human psoriasis and the psoriasis-
like phenotypes that develop in five mouse models (K5-Tie2,
IMQ, K14-AREG, K5-Stat3C, K5-TGFb1). A broad comparison
is made between each mouse phenotype and clinical psoriasis on
the basis of global gene expression patterns, along with more fine-
grained comparisons that are specific to key psoriasis-associated
biological processes and biochemical pathways. Additionally, for
each mouse phenotype, we gauge the intensity of inflammation
and composition of inflammatory infiltrate by analysis of leukocyte
infiltration signatures embedded within genome-wide transcrip-
tional response patterns [31]. These analyses provide the first
transcriptomics-based assessment of correspondence between
human psoriasis and multiple mouse phenotypes, and we suggest
that similar analytic strategies can be adopted in future work to
evaluate existing and new mouse models of psoriasis and other skin
diseases.
Results
Gene expression patterns in psoriasis mouse models
have broad and statistically significant resemblance to
those of clinical psoriasis
Whole-genome microarray analysis was used to identify
transcripts altered in human psoriasis and each of five mouse
psoriasiform phenotypes (back skin of K5-Tie2 transgenic mice,
back skin of IMQ-treated mice, both ear and tail skin of K14-
AREG transgenic mice, back skin of K5-Stat3C mice, and back
skin of K5-TGFb1 mice; see representative images of each
phenotype in Figure 1). Expression patterns associated with
human psoriasis were evaluated by comparing psoriatic skin from
patients (n=58) with normal skin from control subjects with no
history of psoriasis (n=64). Expression patterns associated with
mouse phenotypes were evaluated by comparing lesional skin from
transgenic or IMQ-treated mice (n=2–3) with normal skin
obtained from control mice (n=2–3). Of 54,675 transcripts
represented on the Affymetrix Human Genome U133 Plus 2.0
Array, approximately 61% (33,322) were matched with at least
one transcript derived from an orthologous gene represented on
the Affymetrix Mouse Genome 430 2.0 Array. With respect to
these human-mouse transcript
associated with human psoriasis (psoriasis/control) were positively
correlated with corresponding estimates in mouse phenotypes
(psoriasiform/control) (Figures 2A–2F; 0.18# r #0.25). A cluster
analysis grouped the K5-Tie2, IMQ-treated and K5-TGFb1 mice
apart from the K14-AREG and K5-Stat3C models, with the K14-
AREG and K5-Stat3C phenotypes exhibiting slightly greater
similarity to the expression pattern of human psoriasis (Figure 2G).
pairs, fold-changeestimates
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org2April 2011 | Volume 6 | Issue 4 | e18266

Page 3

The global similarity between expression patterns in human
psoriasis and mouse phenotypes was statistically significant for
each mouse model evaluated. This was demonstrated by five
different analytical approaches (Figures 3, 4, S1, S2 and S3).
First, we asked whether mouse transcripts orthologous to genes
with increased expression in human psoriasis were dispropor-
tionately elevated in psoriasiform phenotypes, and conversely,
whether mouse orthologues of psoriasis-decreased genes were
disproportionately decreased in psoriasiform phenotypes. This
expectation was validated in both cases and with respect to each
mouse phenotype considered (Figure 3; P,0.001 for each mouse
model). Secondly, we evaluated whether overlap among ranked
gene lists was statistically significant (Figure 4) [32]. Human
transcripts were ranked according to the estimated fold-change
difference between lesional and normal skin, and likewise for each
mouse model, transcripts were ranked according to the estimated
fold-change difference between psoriasiform and normal skin.
This analysis revealed that ranked transcript lists associated with
human psoriasis significantly overlapped with corresponding lists
associated with each psoriasiform phenotype, regardless of
whether the top N psoriasis-increased or psoriasis-decreased
human transcripts were considered (P,0.05; 10# N #5000; see
Figure 4). These conclusions were further supported by three
additional statistical methods, including analysis of adjusted
residuals (Figure S1) [33], ‘‘detection rate’’ ROC curves (Figure
S2) [34], and the gene set enrichment score statistic proposed by
Subramanian et al. [35] (i.e., ‘‘GSEA analysis’’, see Figure S3).
We note that, for psoriasis-decreased genes, GSEA-generated p-
values were non-significant with respect to the K5-TGFb1 model
(P=0.318; Figure S3).
Correspondence between human psoriasis and mouse pheno-
types was also evident from inspection of the ‘‘trademark’’
expression patterns of human psoriasis, which we identified as the
most consistent and pronounced gene expression differences
between psoriatic plaques and normal human skin (Figure 5). In
all five mouse phenotypes, there was increased expression of
S100a9, Lcn2, S100a8, Sprr1b, Mpzl2, Has3 (Figure 5A), as well as
decreased expression of Tppp, Stxbp6, and Cldn23 (Figure 5B), and
each of these effects is consistent with characteristic expression
patterns in clinical psoriasis. Additionally, we identified a set of 27
mouse genes (with human orthologues) that were significantly
increased in all five mouse phenotypes (P,0.05), and in 23 of
these cases, the orthologous human gene was elevated in skin
from psoriasis patients (Figure S4). Likewise, we identified a set of
44 mouse genes (with human orthologues) that were significantly
decreased in all five mouse phenotypes (P,0.05), and in 32 of 44
cases, associated human genes were correspondingly decreased in
human psoriasis (Figure S5). There were also examples in which
psoriasiform phenotypes failed to recapitulate a robust gene
expression indicator of human psoriasis. For instance, expression
of protocadherin 21 (PCDH21) is decreased by 68.4% in human
psoriasis (FDR-adjusted P =6.62610237), but the mouse
ortholog Pcdh21 was in fact elevated 1.5-fold in K5-Tie2 lesions
(P=0.025), 54.2-fold in IMQ lesions (P=2.4361027), 5.2-fold in
K14-AREG ear lesions (P=1.161023), 24.9-fold in K14-AREG
tail lesions (P=1.4861026), 7.4-fold in K5-Stat3C lesions
(P=3.7361026), and22.5-fold
(P=4.861024) (Figure 5B and Figure S4). Another robust feature
of human psoriasis was increased expression of C-type lectin
domain family 7 member A (CLEC7A/DECTIN-1), which
encodes a membrane receptor that mediates Th1/Th17 immune
responses [36]. In human psoriasis, expression of CLEC7A was
elevated 8.7-fold (FDR-adjusted P,1.59610258), but expression
of the mouse ortholog Clec7a was not altered in the K5-Tie2
phenotype (P=0.843) and was decreased by 43% in the IMQ
phenotype (P=0.046) (Figure 5A).
in K5-TGFb1lesions
Each mouse phenotype exhibits a shift in epidermis- and
keratinization- associated expression patterns that
parallels clinical psoriasis
Gene Ontology (GO) biological processes significantly overrep-
resented among transcripts increased or decreased in human
psoriasis were identified (P,0.05), representing the most salient
features of the human psoriasis gene expression signature (Figures 6
and 7). For each overrepresented process, we identified the
associated human genes altered in clinical psoriasis, and
determined whether orthologous mouse genes were correspond-
ingly altered in mouse psoriasiform phenotypes. This analysis
revealed that many key aspects of the human psoriasis gene
expression signature were mirrored by mouse psoriasiform
phenotypes. The most striking point of correspondence, shared
among the five mouse models, was elevated expression of
transcripts involved in epidermal development and keratinization
(see Figure 6). For instance, among transcripts increased
significantly in human psoriasis, we identified 26 transcripts
associated with the ‘‘epidermis development’’ (GO:0008544) gene
ontology term (e.g., KRT16, KRT17, ELF3). Based on the human-
mouse orthology, there were 28 mouse transcripts associated with
Figure 1. Psoriasiform phenotypes in the laboratory mouse. The figure shows representative images of the five psoriasis-like phenotypes
evaluated in this study (i.e., K5-Tie2, IMQ, K14-AREG, K5-Stat3C and K5-TGFb1). Each mouse model exhibits red, scaly skin with macroscopic features
suggestive of and consistent with clinical psoriasis in humans.
doi:10.1371/journal.pone.0018266.g001
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org3April 2011 | Volume 6 | Issue 4 | e18266

Page 4

Figure 2. Global correspondence of gene expression between human psoriasis and mouse psoriasiform phenotypes. The global
correlation was evaluated between gene expression patterns in human psoriasis and those in psoriasiform plaques obtained from (A) K5-Tie2
transgenic mice (back skin), (B) IMQ-treated mice (back skin), (C) K14-AREG mice (ear skin) (D) K14-AREG mice (tail skin), (E) K5-Stat3C mice (back skin)
and (F) K5-TGFb1 mice (back skin). Scatterplots shown in (A) - (F) are each based upon 33322 matching transcripts associated with orthologous
human and mouse genes. For each transcript, the difference was calculated between its average expression across psoriatic skin samples from n = 58
patients and its average expression across normal skin samples from a group of n = 64 healthy subjects. The horizontal axis corresponds to the ratio
of gene expression in psoriatic skin relative to skin from control subjects (log2scale), with values larger than one indicating increased expression in
psoriatic skin. The vertical axis corresponds to the ratio of gene expression in mouse psoriasiform skin relative to skin samples from control mice (log2
scale). The intensity of the blue shading represents the empirical density of 33322 points within the bivariate space, and the dotted red line was
generated by least-squares regression. The red circle shown in each figure outlines the set of transcripts (75% of all transcripts) that are closest to the
bivariate centroid (based upon Mahalanobis distance). The Pearson correlation coefficient (r) calculated from each scatterplot is shown in the lower-
right corner of (A) - (F). In panel (G), the human-mouse correspondence was evaluated with respect to a subset of 2617 transcripts elevated in human
psoriasis and 3540 transcripts decreased in human psoriasis (FDR-adjusted P,0.05, log2-transformed fold-change estimate greater than 0.50 in
absolute value). Colors denote the fold-change estimate associated with human or mouse transcripts (see scale), with red regions indicating elevated
expression in psoriatic skin or mouse phenotypes (i.e., fold-change greater than one), and green indicating decreased expression in psoriatic skin or
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org4 April 2011 | Volume 6 | Issue 4 | e18266

Page 5

these human transcripts, and of these, 92.9% were elevated in K5-
Tie2 (26 of 28; P=1.0861027), 89.3% in IMQ (25 of 28;
P=1.5261026),71.4% inK14-AREG
P=6.2761023),92.9% inK14-AREG
P=1.0861027), 75% in K5-Stat3C (21 of 28; P=1.8661023),
and 82.1% in K5-TGFb1 (23 of 28; P=9.0061025). Among
transcripts decreased in human psoriasis, there was even stronger
similarity among mouse models, involving a broad range of
biological processes for which expression patterns were correspon-
dent between psoriasiform phenotypes and human (e.g., response
to insulin stimulus, development, transcription; see Figure 7).
These analyses were repeated based upon the KEGG
pathways overrepresented among transcripts increased or
decreased in human psoriasis (Figures S6 and S7). As an
alternative analysis method, we also extracted sets of human-
mouse gene pairs with correspondent expression shifts in mouse
phenotypes and human psoriasis, and identified GO biological
process terms overrepresented within these gene sets. This
strategy identified gene categories that, while not necessarily the
most salient features of the human psoriasis expression
signature, were nonetheless descriptive of points at which
human psoriasis and mouse phenotypes correspond (Figure S8
and S9).
ear
tail
(20
(26
of
of
28;
28;
The mouse phenotypes diverge with respect to
immune-associated gene expression patterns and
mitosis-related transcription
While key features of human psoriasis were recapitulated by
mouse model phenotypes, points of non-correspondence relative
to each other and to human psoriasis were also identified. Most
notably, there was disparity between the K5-Tie2 and IMQ
phenotypes relative to other mouse models, with respect to
psoriasis-increased genes with immune-associated gene ontology
terms, such as response to virus, response to lipopolysaccharide,
response to cytokine stimulus, cellular defense response, innate
immune response and positive regulation of ab T-cell prolifer-
ation (Figures 6 and 8). This disparity between K5-Tie2 and
IMQ phenotypes relative to other mouse models was also evident
based upon inspection of KEGG pathway terms associated with
transcripts elevated in human psoriasis (e.g., primary immuno-
deficiency, natural killer cell mediated cytotoxicity, toll-like
receptor signaling pathway, cytokine-cytokine receptor interac-
tion and leukocyte adhesion; see Figure S6). This trend was
further illustrated by GO analysis of human-mouse transcript
pairs, with conflicting expression patterns in clinical psoriasis and
mouse phenotypes (Figures S10 and S11). This analysis revealed
that pairs involving a psoriasis-increased transcript in humans
Figure 3. Statistically significant correspondence between human psoriasis and mouse psoriasiform phenotypes: Proportion of
psoriasiform-increased to psoriasiform-decreased transcripts. Genes significantly increased or decreased in human psoriasis were identified
and the expression of orthologous genes was studied in the K5-Tie2 phenotype (A and B), the IMQ phenotype (C and D), the K14-AREG phenotype on
ear skin (E and F), the K14-AREG phenotype on tail skin (G and H), the K5-Stat3C phenotype (I and J), and the K5-TGFb1 phenotype (K and L). We
identified 793 transcripts with significantly elevated expression in human psoriasis (FDR-adjusted P,0.05 and log2-transformed fold-change estimate
greater than 1.00), and these transcripts were associated with 981 transcripts derived from orthologous mouse genes. For these 981 mouse
transcripts, the fold-change ratio was calculated between psoriasiform and normal mouse skin, and in figures A, C, E, G, I and K fold-change estimates
have been ranked from smallest (left) to largest (right). Likewise, we identified 533 transcripts with significantly decreased expression in human
psoriasis (FDR-adjusted P,0.05 and log2-transformed fold-change estimate less than 1.00 in absolute value), and these transcripts were associated
with 709 transcripts derived from orthologous mouse genes. For these 709 transcripts, the fold-change ratio was calculated between psoriasiform
skin and normal skin from control mice, and in figures B, D, F, H, J and L fold-change estimates have been ranked from smallest (left) to largest (right).
In each figure, red symbols denote transcripts increased in psoriasiform mouse skin and green symbols denote transcripts decreased in psoriasiform
skin. The dotted vertical line is equal to the number of psoriasiform-decreased transcripts, and the grey region corresponds to the random
expectation, representing the 95% confidence limits associated with the null (hypergeometric) distribution. In figures A, C, E, G, I and K, there is a
significant overabundance of psoriasiform-increased transcripts, because the dotted vertical line is left of the gray region. In figures B, D, F, H, J and L,
there is a significant overabundance of psoriasiform-decreased transcripts, because the dotted vertical line is right of the gray region. Green and red
percentage values shown in each figure indicate the percentage of psoriasiform-decreased and psoriasiform-increased transcripts, respectively.
doi:10.1371/journal.pone.0018266.g003
mouse phenotypes (i.e., fold-change less than one). The expression profiles shown in the heatmap have been clustered according to the Pearson
correlation coefficient (see dendrogram on left).
doi:10.1371/journal.pone.0018266.g002
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org5 April 2011 | Volume 6 | Issue 4 | e18266

Page 6

and an orthologous transcript decreased in K5-Tie2 or IMQ
lesions were often associated with immunity (e.g., response to
virus, innate immune response; see Figure S10). The overall
pattern is well-illustrated by the 129 transcripts increased in
human psoriasis that were associated with the ‘‘immune
response’’ (GO:0006955) GO term (see Figure 8). Of the 128
Figure 4. Statistically significant correspondence between human psoriasis and mouse psoriasiform phenotypes: Analysis of
ranked gene lists. The5000transcriptswithexpressionmost stronglyelevated in humanpsoriasis were identified, along withthe 5000transcripts with
expression most stronglydecreasedin human psoriasis. These transcripts wererankedaccording tothe estimated fold-change expression ratio (psoriasis/
control), with lower ranks assigned to transcripts most strongly increased or decreased in human psoriasis. For any rank N, where N=1, …, 5000, we
isolated the top N human transcripts and identified orthologous mouse transcripts, and then determined whether these mouse transcripts overlapped
significantly with the top N mouse transcripts increased or decreased in the (A) K5-Tie2 phenotype, (B) IMQ phenotype, (C) K14-AREG phenotype on ear
skin,(D) K14-AREG phenotypeontailskin,(E)K5-Stat3Cphenotypeand(F) K5-TGFb1phenotype.Ineachfigure,theredlinecorrespondstotheoverlap,at
a given rankN, between the top N psoriasiform-increasedmouse transcripts andthe set ofmouse transcripts orthologous tothetop N psoriasis-increased
human transcripts. Similarly, the green line corresponds to the overlap, at a given rank N, between the top N psoriasiform-decreased mouse transcripts
and the set of mouse transcripts orthologous to the top N psoriasis-decreased human transcripts. The grey region outlines the level of overlap expected
by chance for any given rank N (i.e., the 95% confidence region of the null hypergeometric distribution). A significant level of overlap is indicated for each
psoriasiform phenotype because red and green lines lie above the grey region that spans the null expectation.
doi:10.1371/journal.pone.0018266.g004
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org6 April 2011 | Volume 6 | Issue 4 | e18266

Page 7

Figure 5. Trademark gene expression patterns of human psoriasis and expression of orthologous genes in mouse psoriasiform
phenotypes. Genome wide expression data from human psoriatic skin samples (n=58 patients) and normal skin (n=64 subjects) was analyzed to
identify (A) the 50 genes most strongly increased in human psoriasis and (B) the 50 genes most strongly decreased in human psoriasis (both lists
exclude any psoriasis-increased or psoriasis-decreased human gene that lacks an orthologous mouse gene). For each human gene, a matching
transcript associated with an orthologous mouse gene was identified, and the expression of this mouse transcript was compared in psoriasiform
(n=3) and normal skin (n=3). In part (A), mouse genes are listed in descending order according to the fold-change estimate calculated with respect
to the human orthologue (psoriasis/control), such that transcripts orthologous to genes most strongly increased in human psoriasis are positioned
near the top of the figure. In part (B), mouse genes are listed in ascending order according to the fold-change estimate calculated with respect to the
human orthologue (psoriasis/control), such that transcripts orthologous to genes most strongly decreased in human psoriasis are positioned near the
top of the figure. The colors in (A) and (B) correspond to the observed fold-change difference between expression in psoriasiform mouse skin and
normal skin obtained from control mice, with red indicating elevated expression in psoriasiform skin and green indicating decreased expression (see
scale). Filled up-triangles denote transcripts with significantly increased expression in psoriasiform mouse skin (P,0.05) and filled down-triangles
denote transcripts with significantly decreased expression in psoriasiform mouse skin (P,0.05). Unfilled up or down triangles denote transcripts for
which the expression difference between psoriasiform and control mouse skin was marginally significant (0.05# P ,0.10).
doi:10.1371/journal.pone.0018266.g005
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org7April 2011 | Volume 6 | Issue 4 | e18266

Page 8

mouse transcripts that could be matched to the 129 psoriasis-
upregulated human transcripts mapping to this term, more than
77% (.99/128) were correspondingly elevated in the K14-
AREG, K5-Stat3C and K5-TGFb1 phenotypes, but only 44%
(56/128) and 59% (75/128) were correspondingly elevated in the
K5-Tie2 and IMQ phenotypes. These observations suggest that,
as compared to the IMQ and K5-Tie2 models, the K14-AREG,
K5-Stat3C and K5-TGFb1 phenotypes better recapitulate
immune-associated gene expression patterns characteristic of
clinical psoriasis.
Human psoriasis is associated with increased expression of
transcripts involved in cellular proliferation and KC differentiation
and this was reflected in each mouse gene expression phenotype
(Figure 6). However, IMQ-treated mice were less correspondent
with this aspect of the human psoriasis expression pattern, while
the K5-Tie2 and K5-TGFb1 models were most similar to the
clinical disease in this respect (e.g., see mitosis, DNA replication,
DNA replication initiation, KC proliferation, cell division from
Figure 6; see cell cycle and DNA replication from Figure S6; see
mitotic anaphase from Figure S10). This trend is exemplified by
Figure 6. Gene ontology biological processes overrepresented among genes exhibiting increased expression in human psoriasis. A
total of 2617 transcripts were identified as significantly elevated in psoriatic plaques obtained from human patients relative to normal skin obtained
from control subjects (FDR-adjusted P,0.05 and log2-transformed fold-change greater than 0.50). These psoriasis-increased transcripts were analyzed
to identify gene ontology biological process terms significantly over-represented (P,0.05). For each over-represented term, we determined which of
the 2617 transcripts were annotated with the term, and for these human transcripts, we identified a set of mouse transcripts associated with
orthologous genes. For each mouse transcript within this set, we calculated the expression ratio between psoriasiform and normal mouse skin, and
determined the average value of this ratio among all transcripts in the set. The analysis was repeated with respect to each of the mouse skin
phenotypes. Colors in the chart reflect the average fold-change ratio (psoriasiform/control) of the set of mouse transcripts associated with the gene
ontology term listed in each row (see scale). Triangle symbols indicate whether the average fold-change estimate associated with mouse transcripts is
significantly different from one (see legend; two-sample t-test). The number of transcripts associated with each GO term is indicated in brackets (e.g.,
[x/y], where x denotes the number of human transcripts and y represents the number of mouse transcripts derived from orthologous genes). Since
mouse transcripts included in this analysis were orthologous to genes exhibiting elevated expression in human psoriasis, correspondence between
human psoriasis and mouse phenotypes is indicated by red colors and up-triangle symbols.
doi:10.1371/journal.pone.0018266.g006
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org8 April 2011 | Volume 6 | Issue 4 | e18266

Page 9

Figure 7. Gene ontology biological processes overrepresented among genes exhibiting decreased expression in human psoriasis. A
total of 3540 transcripts were identified as significantly decreased in psoriatic plaques obtained from human patients relative to normal skin obtained
from control subjects (FDR-adjusted P,0.05 and log2-transformed fold-change less than -0.50). These psoriasis-decreased transcripts were analyzed
to identify gene ontology biological process terms significantly over-represented (P,0.05). For each over-represented term, we determined which of
the 3540 transcripts were annotated with the term, and for these human transcripts, we identified a set of mouse transcripts associated with
orthologous genes. For each mouse transcript within this set, we calculated the expression ratio between psoriasiform and normal mouse skin, and
determined the average value of this ratio among all transcripts in the set. The analysis was repeated with respect to the K5-Tie2 phenotype, IMQ
phenotype, K14-AREG phenotype on ear skin, K14-AREG phenotype on tail skin, K5-Stat3C phenotype, and K5-TGFb1 phenotype. Colors in the chart
reflect the average fold-change ratio (psoriasiform/control) of the set of mouse transcripts associated with the gene ontology term listed in each row
(see scale). Triangle symbols indicate whether the average fold-change estimate associated with mouse transcripts is significantly different from one
(see legend; two-sample t-test). The number of transcripts associated with each GO term is indicated in brackets (e.g., [x/y], where x denotes the
number of human transcripts and y represents the number of mouse transcripts derived from orthologous genes). Since mouse transcripts included
in this analysis were orthologous to genes exhibiting decreased expression in human psoriasis, correspondence between human psoriasis and mouse
phenotypes is indicated by green colors and down-triangle symbols.
doi:10.1371/journal.pone.0018266.g007
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org9April 2011 | Volume 6 | Issue 4 | e18266

Page 10

Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org10April 2011 | Volume 6 | Issue 4 | e18266

Page 11

128 transcripts associated with mouse genes orthologous to the
human ‘‘mitosis’’ genes with elevated expression in human
psoriasis (e.g., GO:0007067; see Figure S12). Among these
transcripts, more than 87% are increased in the K5-Tie2 and
K5-TGFb1 phenotypes ($112 of 128), while 60.2% (77/128),
43.7% (56/128), 69.5% (89/128) and 64.1% (82/128) are
increased in the IMQ, K14-AREG (ear), K14-AREG (tail) and
K5-Stat3C phenotypes, respectively.
The K14-AREG, K5-Stat3C and K5-TGFb1 phenotypes
exhibit a heightened inflammation signature relative to
that of K5-Tie2 and IMQ psoriasiform lesions
We identified strong differences among mouse phenotypes with
respect to the immune-associated gene expression patterns
characteristic of human psoriasis, with the K14-AREG, K5-
Stat3C and K5-TGFb1 phenotypes exhibiting a closer correspon-
dences to human psoriasis (Figure 8). One possibility is that models
differ in terms of the type and abundance of leukocytes present
within psoriasiform plaques, leading to dissimilar expression
patterns among genes annotated with immune-related GO terms
or KEGG pathways. To evaluate this possibility, we characterized
each mouse phenotype using a microarray-based immunopheno-
typing algorithm, which estimates overall inflammation intensity
and also identifies leukocyte subsets underlying an inflammation
signature within microarray data (Figure 9) [31] (see also Haider et
al. [37] for a similar approach). In brief, the algorithm utilizes gene
expression profiles of cell populations harvested from mouse tissues
(mostly leukocytes; e.g., T-cells, DCs, macrophages), and for each
population, a set of ‘‘signature transcripts’’ is identified, which
consists of transcripts highly expressed in that population relative
to normal mouse skin. If a given leukocyte population is part of the
infiltrate in lesional skin, it is expected that signature transcripts
associated with that population will be disproportionately elevated
in lesional skin relative to normal skin from control mice, and this
information is used to establish an ‘‘inflammation profile’’ [31]. In
the present context, we have also adapted this approach to
estimate an inflammation profile for human psoriasis, based upon
expression patterns of human genes orthologous to signature
transcripts associated with individual mouse leukocyte populations
(Figure 9).
The immunophenotype of human psoriatic plaques was
consistent with elevated abundance of Th1 T-cell, CD4+ T-cell,
CD8+ T-cell, DC and macrophage signatures (P,0.05; see
Figure 9), which was an expected based upon prior clinical studies
of psoriatic lesions [38]. Most of these trends, however, were
absent from the IMQ phenotype, and in general, the intensity of
inflammation was low in IMQ-generated lesions, with significant
evidence for invasion by macrophages and granulocytes (P,0.05),
but no other leukocyte subsets (Figure 9). To some degree,
inflammation intensity was also relatively low in lesions from K5-
Tie2 mice, and surprisingly, there was no significant evidence for
macrophage or DC infiltration, although there was evidence for
infiltration by CD4+ and regulatory T-cells (P,0.05; Figure 9).
Inflammation intensity was strongest in the K14-AREG, K5-
Stat3C and K5-TGFb1 models, with the strongest patterns
associated with K14-AREG lesions (ear and tail), for which there
was significant evidence of infiltration by CD4+ T-cells, Th1 T-
cells, Th17 T-cells, CD8+ T-cells, regulatory T-cells, NK cells,
DCs, macrophages and monocytes (P,0.05; Figure 9). The
immunophenotypes of K5-Stat3C and K5-TGFb1 models were
comparable in most respects, although for K5-TGFb1 lesions,
there was stronger indication of T-cell infiltration, with significant
CD4+ T-cell, CD8+ T-cell, helper T-cell and regulatory T-cell
signatures (P,0.05; Figure 9). Taken together, these analyses
support a ranking of mouse phenotypes with regard to overall
inflammation intensity, with the strongest inflammation signature
in K14-AREG lesions, moderately strong inflammation in K5-
Stat3C and K5-TGFb1 lesions, weak-to-moderate inflammation
in K5-Tie2 lesions, and weak inflammation in the phenotype
generated by topical application of IMQ.
The K14-AREG, K5-Stat3C and K5-TGFb1 lesions exhibit
stronger increases in TNF-a, IFN-c and IL-6 expression
with heightened TNF-a and IFN-c-dependent KC
responses
The above results indicated that mouse models differed with
respect to immune-associated gene expression patterns (Figures 6
and S6) and also with respect to their microarray-based
inflammation profiles (Figure 9). We therefore evaluated the
expression of selected cytokines and chemokines thought to
contribute to initiation or maintenance of the inflammatory
cascade in human psoriasis (e.g., TNF, IL22, IL6, CXCL1) (Figure
S13). This analysis pointed to key expression responses that were
more prominent in the K14-AREG (ear and tail), K5-Stat3C and
K5-TGFb1 models as compared to the K5-Tie2 and IMQ
models, such as increased expression of tumor necrosis factor-a
(Tnf), interferon (IFN)-c (Ifng), interleukin (IL)-6 (Il6) CXC-
chemokine ligand 10 (Cxcl10), and integrin b2 (Itgb2) (Figure
S13). For instance, we noted little change in Tnf expression for K5-
Tie2 or IMQ lesions (6% increase for K5-Tie2 and 13% decrease
for IMQ; P.0.43), while expression of Tnf was elevated 31% in
K14-AREG ear lesions (P=0.063), 48% in K14-AREG tail lesions
(P=0.047), 34% in K5-Stat3C lesions (P=0.042) and 116% in
K5-TGFb1 lesions (P=0.01). These results are consistent with
decreased accumulation of key cytokines in psoriasiform lesions of
K5-Tie2 and IMQ mice, which may account for lower intensity
inflammation within these models (Figure 9).
Gene expression shifts in both human psoriasis and mouse
psoriasiform phenotypes are, in part, a consequence of KC
responses to the local cytokine and chemokine environment. While
mouse phenotypes differed with respect to localized abundance of
mRNAs encoding key cytokines and chemokines (Figure S13),
these disparities at the RNA level may or may not influence
steady-state cytokine abundance or KC responses. We therefore
identified sets of genes associated with such in vitro KC responses to
cytokine treatment and evaluated how these gene sets were altered
in both psoriatic skin from humans and lesional skin from mouse
model phenotypes (Figure 10). This indicated that K14-AREG
Figure 8. Mouse orthologs of immune response genes increased in human psoriasis. A total of 129 transcripts associated with ‘‘immune
response’’ were significantly elevated in human psoriasis (GO:0006966). With respect to these 129 psoriasis-increased transcripts, we identified 128
corresponding mouse transcripts derived from orthologous mouse genes. A subset of these 128 transcripts is listed in the figure (left margin). The
heat map image describes the response patterns of these transcripts in psoriasiform phenotypes relative to normal skin in control mice. Red colors
correspond to increased expression in psoriasiform phenotypes and green colors correspond to decreased expression (see scale; right margin). Up-
and down-triangles denote transcripts for which the fold-change difference between lesion and normal mouse skin is significant or marginally
significant. Since listed transcripts are orthologous to human genes exhibiting increased expression in clinical psoriasis, correspondence to the
human disease is denoted by red colors (i.e., increased expression in psoriasiform phenotypes).
doi:10.1371/journal.pone.0018266.g008
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org11 April 2011 | Volume 6 | Issue 4 | e18266

Page 12

Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org 12April 2011 | Volume 6 | Issue 4 | e18266

Page 13

(ear), K14-AREG (tail), K5-Stat3C and K5-TGFb1 phenotypes
exhibited increased expression of transcripts that are induced by in
vitro treatment of KCs with TNF-a, IFN-a and IFN-c, while such
responses were not observed with respect to the K5-Tie2 and IMQ
phenotypes (Figure 10). Additionally, as transcriptional markers of
KC differentiation, we identified gene sets associated with calcium
Figure 10. Gene expression signatures associated with cytokine stimulation and differentiation of keratinocytes (KCs). Microarray
data was used to identify human transcripts exhibiting increased or decreased expression in KCs treated with cytokines (e.g., TNF, IFN-a, IFN-c),
agents that induce differentiation (calcium (CA), KC growth factor (KGF)), and agents that inhibit differentiation (retinoic acid (RA)). Each row in the
chart corresponds to a set of transcripts, with the number of human and associated mouse transcripts indicated in brackets (n = x/y denotes x
human transcripts and y mouse transcripts associated with orthologous mouse genes). For human transcripts in each set, the average expression
ratio between psoriasis and normal skin samples (psoriasis/normal) was calculated, and the average ratio obtained for each set is denoted in the first
column according to the color code described in the legend. Likewise, for mouse transcripts within each set, the average expression ratio between
psoriasiform and normal skin (psoriasiform/normal) was calculated with respect to each mouse model, and the average ratio value for each set and
mouse model is indicated according to the color code (see legend). Triangle symbols indicate whether, on average, fold change estimates differ
significantly from one (see legend; two-tailed t-test). In some cases (rows), gene sets were defined based upon data from Gene Expression Omnibus
(GEO) and the GEO series identifier is given in parentheses (e.g., GSE7216).
doi:10.1371/journal.pone.0018266.g010
Figure 9. Leukocyte infiltration signatures of human psoriasis and mouse psoriasiform lesions. The immunophenotyping algorithm
developed by Swindell et al. (31) was used to generate inflammation profiles for human psoriasis and the K5-Tie2, IMQ, K14-AREG, K5-Stat3C and K5-
TGFb1 psoriasiform lesions. Signature transcripts highly expressed within different cell populations were identified, where most cell populations were
leukocyte subsets isolated from mice (e.g., T-cells, B cells, macrophages; see Methods for details). For each set of n signature transcripts associated
with a given cell population, and for each psoriasiform phenotype, we identified the number of transcripts with increased expression in mouse
psoriasiform lesions (n1) and the number of transcripts with decreased expression in mouse psoriasiform lesions (n2). For each set and each mouse
phenotype, the ratio of psoriasiform-increased to psorisiform-decreased transcripts was calculated (i.e., ratio = n1/n2, where n1+ n2= n). Colors in
the chart corresponds to this calculated ratio (see scale), where darker red colors indicate that the signature transcripts of cell populations (rows)
tended to be elevated in the mouse psoriasiform phenotype (columns). Filled symbols are used to indicate cell populations for which the proportion
of signature transcripts elevated in psoriasiform phenotypes (i.e., the n1/n2ratio) was significantly large (see Methods for significance criteria). Darker
colors (larger n1/n2ratios) thus indicate cell populations that, based upon the observed gene expression patterns, appear likely to comprise the
inflammatory infiltrate associated with a given mouse psoriasiform phenotype. In the first column, a similar methodology was applied with respect to
human psoriasis. However, for each cell population, we identified human transcripts associated with genes orthologous to the n signature transcripts,
and among these human transcripts, we evaluated the ratio of psoriasis-increased to psoriasis-decreased transcripts.
doi:10.1371/journal.pone.0018266.g009
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org 13April 2011 | Volume 6 | Issue 4 | e18266

Page 14

supplementation of cultured KCs, and these gene sets were more
strongly elevated in K5-Tie2 and K5-TGFb1 lesions (Figure 10).
Discussion
The development of a ‘‘high-fidelity’’ mouse model of psoriasis
would provide a valuable experimental system for investigating
mechanisms of psoriasis pathogenesis and anti-psoriatic therapies.
Despite differences that exist between humans and mice, mouse
disease models have made important contributions to the
investigation of major disease processes, including for example,
heart disease [39], Alzheimer’s disease [40], diabetes [41], cancer
[42], as well as autoimmune syndromes such as lupus erythema-
tosus and rheumatoid arthritis [43,44]. Because the primary
disease symptoms of psoriasis occur at the skin surface, direct
comparisons can be made between affected skin from patients and
putative mouse models. We have followed this approach and have
applied functional genomics methods to compare global tran-
scriptional signatures of psoriasis phenotypes in humans and five
psoriasis mouse models. With respect to an unbiased choice of
psoriasis-associated GO and KEGG pathway terms, we identified
specific similarities and differences between human psoriasis and
each mouse model, and we suggest that these differences can be
translated into strengths and weaknesses of each mouse model. For
each model, gene expression patterns associated with epidermis
development and keratinization mirrored those of clinical
psoriasis. However, we noted divergence among models with
respect to immune-associated gene expression, with a heightened
inflammatory signature in lesions from K14-AREG, K5-Stat3C
and K5-TGFb1 mice, along with stronger elevation of mRNAs
coding key cytokines and chemokines (e.g., Tnf, Ifng, IL-6) and
more pronounced TNF-a and IFN-c-driven expression patterns.
Nevertheless, at a global level, there is strong and statistically
significant similarity between expression patterns in clinical
psoriasis and those of each mouse model evaluated by our study.
Psoriasis results from an interaction between activated immu-
nocytes and KCs, in which KCs exhibit abnormal maturation and
proliferation in response to a complex cytokine network that
mediates disease maintenance and progression [38]. The key role
of the immune response and inflammatory process has been
convincingly supported by the identification of major histocom-
patibility complex and cytokine-associated loci in genome-wide
association analyses [45], the effectiveness of immunosuppressant
therapies in treatment of the disease [46], and the observation that
psoriasis can be transferred between donor and recipient with
bone marrow transplantation [47]. For these reasons, a consensus
point among investigators is that a realistic psoriasis mouse model
should exhibit an inflammatory infiltrate resembling that found in
clinical psoriasis, including T-cells, DCs and neutrophils, and that
the psoriasiform phenotype should be both T-cell dependent and
responsive to drug treatments targeting the immune system [3–5].
It was therefore significant, in the present study, to observe that
gene expression signatures of mouse psoriasiform phenotypes
diverged with respect to immune-associated transcripts, with three
phenotypes more closely mirroring clinical psoriasis (i.e., K14-
AREG, K5-Stat3C and K5-TGFb1) as compared to two others
(i.e., K5-Tie2 and IMQ). This disparity likely involves differences
in the relative abundance of certain immune cell subsets within
lesional skin of psoriasiform phenotypes, which would have
consequent effects on the cytokine and chemokine milieu and
development of the inflammatory cascade. For instance, the K14-
AREG, K5-Stat3C and K5-TGFb1 phenotypes showed a stronger
inflammation signature overall, with significant evidence for
infiltration by T-cells, monocytes, DC and macrophages (sure 9).
These phenotypes contrast with that of the K5-Tie2 model, where
our immunophenotyping algorithm supported increased expres-
sion of transcripts associated with CD4+ T-cells, but did not
provide evidence for increased expression of transcripts highly
expressed in cells from the monocyte-DC/macrophage lineage
(Figure 9). Similarly, for IMQ-generated lesions, evidence of
macrophage infiltration was obtained, but there was little or no
indication of T-cell infiltration (Figure 9). Previously, immunohis-
tochemical methods have detected increased abundance of CD8+
T cells, F4/80+ macrophages and CD11b and CD11c+ myeloid
cells in K5-Tie2 lesions [12,48], as well as increased abundance of
CD4+ and CD8+ T-cells in IMQ-generated lesions [13]. In these
cases, however, the number of infiltrating immunocytes may not
be sufficient to have a discernable impact on gene expression
patterns, and this is likely to at least partially account for the
observed disparity among models with respect to immune-
associated transcripts (Figure 8).
The environment of cytokines and chemokines in mouse
psoriasiform lesions is largely shaped by early-stage inflammation
responses. Along these lines, mouse phenotypes also differed in
their relative expression level of the proinflammatory cytokines
TNF-a, IFN-c and IL-6, where in each case expression was more
strongly elevated in K14-AREG, K5-Stat3C and K5-TGFb1
lesions as compared to K5-Tie2 or IMQ lesions (Figure S13).
Moreover, in vitro KC transcriptional responses to TNF-a, IFN-c
and IFN-a stimulation were not discernable with respect to the
K5-Tie2 or IMQ expression signatures, but were discernable with
respect to K14-AREG, K5-Stat3C and K5-TGFb1 expression
signatures (Figure 10). TNF-a is generated by NK-T and ab-T-
cells, macrophages and KCs, and is known to reinforce
inflammatory processes. In psoriasis, the pathogenic role of
TNF-a has been demonstrated by the effective treatment of
patients with anti-TNF therapies [49]. TNF-a is also a key
component of local cytokine networks, and indeed, the quantita-
tively lower increase in TNF-a expression that we observed in K5-
Tie2 and IMQ-generated lesions may explain the similar trends
observed for IFN-c and IL-6. In particular, a previous study using
peripheral blood mononuclear cells has shown that TNF blockade
suppresses IFN-c expression, while in turn, IFN-c blockade
suppresses IL-6 expression [50]. Our findings thus suggest that
psoriasis mouse models can be differentiated with respect to a
TNF-a/IFN-c/IL-6 axis, which ultimately, may be reflective of
disparities in the abundance of certain TNF-generating immuno-
cytes (e.g., macrophages, DCs or T-cells).
KC proliferation and consequent hyperplasia of the epidermal
layer is a hallmark feature of clinical psoriasis [38]. It is therefore
noteworthy that our analysis detected differences among models
related to mitosis-associated gene expression patterns. Mitosis-
associated transcripts were most strongly elevated in K5-TGFb1
lesions, and a comparable trend was evident for K5-Tie2 lesions,
and in this regard, the K5-TGFb1 and K5-Tie2 models were
generally most similar to the mitotic gene expression signature of
clinical psoriasis (see Figure S12). This trend was supported, for
example, by expression patterns of many transcripts involved in
DNA replication and the transitions between mitotic phases,
including cyclin A2 (Ccna2), cyclin F (Ccnf) and budding
uninhibited by benzimidazoles 1 homolog (Bub1b) (Figure S12).
The psoriasis phenotypes evaluated in this study were generated
on outbred, inbred and hybrid genetic backgrounds, including
outbred CD1 mice (K5-Tie2), the inbred strain C57BL/6 (IMQ),
as well as hybrid FVB/NCrIBR (K14-AREG), FVB/NHsd (K5-
Stat3C), and ICR/B6D2 mice (K5-TGFb1). While our analysis
has identified disparities among models, it is possible that these
disparities are, in part, driven by characteristics of individual
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org14 April 2011 | Volume 6 | Issue 4 | e18266

Page 15

mouse strains, which might in turn modulate the direct effects of
manipulations we have considered. For this reason, we suggest that
any psoriasis mouse model should be considered jointly with the
genetic background with which it is associated (e.g., the ‘‘K5-
Tie2/CD1 model’’, the ‘‘B6/IMQ model’’, etc.). Along these lines,
an important avenue for future investigation is to evaluate
background-dependence of manipulations that give rise to
psoriasis-like phenotypes in mouse.
The approach used in this study to evaluate psoriasis mouse
models represents a general strategy that can be applied to score
alternative mouse phenotypes with resemblance to specific human
diseases. Potentially, the same methodology could be applied to
determine whether psoriasiform phenotypes we evaluated are
equally or better-suited as models for other inflammatory skin
conditions, such as atopic dermatitis or allergic contact dermatitis.
The microarray-based analytical approach we have implemented
does not replace conventional validation criteria for psoriasis
mouse models, but extends these criteria by providing a
quantitative ‘‘yardstick’’ that can be used to clearly judge progress
towards more realistic mouse phenotypes. This can guide
development of new psoriasis mouse models by identifying
combinations of genetic manipulations that complement each
other well and also provide guidance to scientists in choosing the
most appropriate mouse model to use when testing specific
mechanistic and therapeutic hypotheses. Ultimately, this may lead
to development of psoriasis mouse models with stronger
biochemical similarity to the human disease.
Materials and Methods
Ethics Statement
This study was conducted in compliance with good clinical
practice and according to the Declaration of Helsinki principles.
Informed written consent was obtained from all human subjects,
under protocols approved by the institutional review board of the
University of Michigan (HUM00037994). All animal protocols
were approved by animal welfare committees at each participating
institution; Case Western Reserve University Institutional Animal
Care and Use Committee (IACUC, #2009-0193), Erasmus
Medical Center Animal Ethics Committee (Advies DEC Nr.
EUR 1846 (EMCnr. 128-09-09), Institutional Animal Care and
Use Committee of the University of Texas, Austin (AUP-2010-
00029), Institutional Animal Care and Use Committee of the
University of Colorado (B-850008(08)2E and the Institutional
Animal Care and Use Committee of Mayo Clinic (A1009-
Rochester).
Human and Mouse Expression Data
The patient population and sample processing methods for the
collection of human microarray data has been described in a
recent report [51]. In brief, the study involved 58 psoriasis patients
(ages 21- 69) and 64 normal healthy control subjects (ages 18 - 45)
recruited from areas surrounding Detroit, Michigan. The 58
psoriasis patients were chosen on the basis of having one or more
psoriatic plaques not limited to the scalp region. If the patient had
only one plaque, the single plaque was large in size (greater than
1% of total body area). No systemic anti-psoriatic treatments were
used for 2 weeks prior to biopsy, and no topical treatments were
used for 1 week prior to biopsy. In control subjects, biopsies were
always taken from the buttocks or upper thighs. RNA samples
were hybridized to the Affymetrix Human Genome U133 Plus 2.0
Array, which includes probesets corresponding to 54675 human
transcripts. All the microarray data are MIAME compliant. Raw
microarray data from the psoriasis cohort has been deposited in
the NCBI Gene Expression Omnibus (GEO, http://www.ncbi.
nlm.nih.gov/geo) and is accessible through GEO Series accession
number GSE13355.
Generation of transgenic mice bearing psoriasiform phenotypes
has been described in previous reports [12–16]. The K5-Tie2 mice
were generated by crossing a KC-specific K5-tTA driver line
(Tg(KRT5-tTA)1216Glk) with a TetOSTek/Tie2 responder line
(Tg(TetOS-Tek)1Dmt) that had been generated on an outbred CD1
background [52,53]. The generation of K14-AREG (FVB/
NCrIBR-Tg(KRT14-AREG)3Pwc), K5-Stat3C (FVB-Tg(KRT5-Sta-
t3*A661C*N663C)1Jdg) and K5-TGFb1 ((ICRxB6D2)F1-Tg(KRT5
-TGF-B1)F2020Xjw) transgenic mice is described in the reports of
Cook et al. [14], Sano et al. [15] and Li et al. [16], respectively.
For all experiments involving transgenic mice, experimental and
control animals were genotyped at weaning using PCR, and
development of psoriasiform lesions was spontaneous and not
induced by wounding (e.g., tape stripping). For the IMQ-
generated phenotype, the treatment protocol was to apply a daily
topical dose of 62.5 mg IMQ cream (5% Aldara; 3M
Pharmaceuticals) to the shaved back region of 8–10 week old
C57BL/6 mice. This treatment was carried out for six consecutive
days, and during this time, control animals were treated with a
vehicle cream (Vaseline Lanette cream; Fagron). In all experi-
ments, adult mice were euthanized at 10–20 weeks of age at which
time ear, tail or back skin samples were flash frozen and stored at
280uC prior to the isolation of total RNA. We have evaluated
back skin samples from the K5-Tie2, IMQ, K5-Stat3C and K5-
TGFb1. In K14-AREG mice, however, we have focused on tail
and ear skin phenotypes to be consistent with previous studies
[14], and because focusing on hairless regions prevented analyses
from being influenced by K14-driven overexpression of AREG in
the outer root sheath [54]. For all samples, total RNA was
extracted using the RNeasy mini kit (Qiagen, Valencia, CA), and
after further processing, cDNA was hybridized to Affymetrix 430
2.0 arrays (45,101 probesets).
Statistical methods
The human and mouse datasets were normalized using the
Robust Multichip Average (RMA) algorithm. Following RMA
normalization, human data were further processed to calculate
expression scores adjusted for gender and batch effects as
described by Gudjonsson et al. (51). Analyses were based upon a
between-chip mapping of transcripts represented on the Human
Genome U133 Plus 2.0 and transcripts on the Affymetrix Mouse
Genome 430 2.0 array. This map was downloaded as a single
CSV file from the NetAffx analysis center in July 2009 [55], and is
based upon reference-sequence similarity from the HomoloGene
database [56].
Differential expression of transcripts between human psoriatic
skin samples and control skin samples was evaluated using the
Limma linear modeling package [57], with P-value adjustment
using the Benjamini-Hochberg method [58]. All p-values gener-
ated from differential expression analyses were derived from two-
tailed hypothesis tests. The over-representation of KEGG and
gene ontology terms was evaluated based upon the conditional
hypergeometric test procedure implemented in the GOstats
package [59], which is available as part of the R Bioconductor
software suite [60]. Among transcripts increased or decreased in
human psoriasis, a large number of Gene Ontology biological
process or KEGG pathway terms were significantly over-
represented. Of these, we have focused on that were most
frequently associated with the transcripts significantly altered in
human psoriasis (i.e., at least 10–12 transcripts per term in
Figures 6 and 7; at least 3 transcripts per term in Figures S6 and
Transcriptomic Assessment of Psoriasiform Mice
PLoS ONE | www.plosone.org15April 2011 | Volume 6 | Issue 4 | e18266