Combining theoretical analysis and experimental datageneration reveals IRF9 as a crucial factor for accelerating interferon a-induced early antiviral signalling Tim Maiwald1,*, Annette S
Trang 1Combining theoretical analysis and experimental data
generation reveals IRF9 as a crucial factor for accelerating interferon a-induced early antiviral signalling
Tim Maiwald1,*, Annette Schneider2,*, Hauke Busch3, Sven Sahle1, Norbert Gretz4,
Thomas S Weiss5, Ursula Kummer1and Ursula Klingmu¨ller2
1 Heidelberg University, Department Modeling of Biological Processes, BIOQUANT ⁄ Institute of Zoology, Germany
2 German Cancer Research Center, Division Systems Biology of Signal Transduction ⁄ BIOQUANT, DKFZ-ZMBH Alliance, Heidelberg, Germany
3 Freiburg Institute for Advanced Studies (FRIAS), Albertstraße 19 and Center for Biosystems Analysis (ZBSA), University of Freiburg, Germany
4 Heidelberg University, Medical Faculty Mannheim, Germany
5 University Medical Center Regensburg, Center for Liver Cell Research, Germany
Introduction
Invading pathogens such as viruses elicit complex
responses in host cells, and the outcome of an infection
is critically determined by the dynamics of host defence mechanisms Important mediators of antiviral
Keywords
antiviral signalling; interferon a; IRF9; kinetic
model; signal transduction
Correspondence
U Klingmu¨ller, Division Systems Biology of
Signal Transduction, German Cancer
Research Center, Im Neuenheimer Feld
280, Heidelberg 69120, Germany
Fax: +49 6221 42 4488
Tel: +49 6221 42 4481
E-mail: u.klingmueller@dkfz.de
Database
The mathematical model described here has
been submitted to the Online Cellular
Systems Modelling Database and can be
accessed at http://jjj.biochem.sun.ac.za/
database/maiwald/index.html free of charge
*These authors contributed equally to this
work
(Received 19 June 2010, revised 20 August
2010, accepted 13 September 2010)
doi:10.1111/j.1742-4658.2010.07880.x
Type I interferons (IFN) are important components of the innate antiviral response A key signalling pathway activated by IFNa is the Janus kinase⁄ signal transducer and activator of transcription (JAK⁄ STAT) pathway Major components of the pathway have been identified However, critical kinetic properties that facilitate accelerated initiation of intracellular antiviral signalling and thereby promote virus elimination remain to be determined By combining mathematical modelling with experimental analysis, we show that control of dynamic behaviour is not distributed among several pathway components but can be primarily attributed to interferon regulatory factor 9 (IRF9), constituting a positive feedback loop Model simulations revealed that increasing the initial IRF9 concentration reduced the time to peak, increased the amplitude and enhanced termina-tion of pathway activatermina-tion These model predictermina-tions were experimentally verified by IRF9 over-expression studies Furthermore, acceleration of signal processing was linked to more rapid and enhanced expression of IFNa target genes Thus, the amount of cellular IRF9 is a crucial determinant for amplification of early dynamics of IFNa-mediated signal transduction
Abbreviations
C ⁄ EBP-b, CCAAT-enhancer-binding protein b; IFN, interferon; IRF9, interferon regulatory factor 9; ISGF3, interferon-stimulated gene factor 3; ISG, interferon-stimulated gene; JAK, Janus kinase; PIAS, protein inhibitor of activated STATs; PKR, protein kinase R; SOCS, suppressor
of cytokine signalling; STAT, signal transducer and activator of transcription; SHP-2, SH2-containing phosphatases; TYK2, tyrosine kinase 2; TFBS, transcription factor binding sites; USP18, ubiquitin specific peptidase 18.
Trang 2responses are type I interferons (IFN), which are used
in the treatment of hepatitis B and C virus infections
However, success of the treatment is highly
patient-dependent [1] As a rapid IFN response may be
deci-sive for a viral infection [2], it is important to identify
factors that regulate IFN signalling kinetics and that
may be a reason for patient-to-patient variations
In general, several potential mechanisms are possible
to expedite signal transduction As shown for the
epidermal growth factor receptor system, increasing
ligand concentrations result in earlier maximal pathway
activation and an increase in signal amplitude [3,4]
Furthermore, alterations in the activity of a kinase or
phosphatase may affect the speed of signalling pathway
activation However, theoretical studies by Heinrich
et al.[5] indicated that kinase activity primarily regulates
signal amplitude rather than signalling time, whereas
phosphatases enhance the signalling time, but lead to a
decrease in signal amplitude Currently, very little is
known regarding specific mechanisms that could be
exploited to accelerate IFN signalling
A key signalling pathway activated by type I IFN is
the JAK⁄ STAT pathway The Janus kinases JAK1
and tyrosine kinase 2 (TYK2) are activated in response
to ligand binding to the receptor, and these kinases
phosphorylate signal transducers and activators of
transcription STAT1 and STAT2 In contrast to other
JAK⁄ STAT signalling pathways, type I IFN signalling
additionally involves interferon regulatory factor 9
(IRF9⁄ p48), which, together with phosphorylated
STAT1⁄ STAT2 dimers, forms the
interferon-stimu-lated gene factor 3 (ISGF3) complex ISGF3 is
translo-cated to the nucleus and activates transcription of
interferon-stimulated genes [6] Amongst others, this
leads to induction of suppressor of cytokine signalling
(SOCS) proteins that modulate termination of pathway
activation Furthermore, the expression of IRF9 is
induced, constituting a positive feedback loop IRF9
plays an important role in IFNa signalling [7]
Increas-ing the amount of IRF9 by over-expression or
pre-stimulation of cells with IFNc or interleukin-6 results
in higher levels of transcription of IFN-stimulated
genes [8–11] and an augmented antiviral response [11–
13] However, the specific impact of IRF9 on the
dynamics of pathway activation, such as signalling
speed and extent, remains to be identified
To unravel the highly non-linear relations
determin-ing the timdetermin-ing and extent of signalldetermin-ing pathway
activa-tion, establishment of a dynamic pathway model is
required Previous modelling approaches analysing the
JAK⁄ STAT pathway focussed on the impact of
phos-phatases and induced negative inhibitors [14–16] that
primarily influence pathway termination Here, we
have developed a mathematical model of IFNa signal transduction, including the known key players and feedback mechanisms, to identify systems properties that facilitate accelerated IFNa signalling Using this approach, IRF9 was predicted to not only increase the amount of active ISGF3, but also to accelerate signal transduction into the nucleus, as verified experimen-tally by IRF9 over-expression studies Moreover, the accelerated signal processing also resulted in faster and increased expression of target genes Thus, we identified IRF9 as a pivotal player for the speed and efficiency
of IFNa signal transduction
Results
A data-based mathematical model of IFNa signalling with predictive power
To investigate the dynamic properties of IFNa-medi-ated JAK⁄ STAT activation, we developed a mathemat-ical model comprising the known key components, feedback responses and constitutive regulatory mecha-nisms (Fig 1A, Table S1 and Appendix S1) Specifi-cally, we included constitutive negative regulations by general phosphatases and protein inhibitor of activated STATs (PIAS), as well as constitutive degradation of receptors, IRF9 and mRNA Receptor dephosphoryla-tion by SH2-containing phosphatase 2 (SHP-2) was represented by a constant kinetic parameter indepen-dent of SHP-2 concentration, as changes in SHP-2 concentration were assumed to be negligible during the measured timescale Furthermore, the negative feed-back loop of ISGF3-mediated SOCS induction was incorporated IRF9 synthesis was included as a positive feedback mechanism as IFN-dependent expression of IRF9 was experimentally observed within the relevant timescale (Fig 1B) Further features of the model were based on literature evidence: (i) IRF9 is constitutively bound to unphosphorylated STAT2 in the unstimu-lated system [17], (ii) unphosphoryunstimu-lated STAT1 and STAT2 shuttle constantly between the cytoplasm and nucleus, and nuclear import of STAT2 is increased by IRF9 binding [18], (iii) free IRF9 is mainly localized to the nucleus [19], and (iv) phosphorylated STAT1⁄ STAT2 heterodimers require IRF9 to bind IFN-stimu-lated response elements [20] IRF9 independent DNA-binding of STAT heterodimers was not considered, as these complexes bind to different DNA elements, the
c activation sites that are involved in IFNc signalling Thus, in the model, no gene expression occurs in the absence of IRF9 (Table S1 and Appendix S1)
For model calibration, kinetic parameters were taken from the literature (Table S2) or trained against
Trang 3B
Fig 1 Kinetic model of the IFNa signalling pathway (A) Simplified view of the model architecture is shown, with the activation of STAT proteins summarized as one reaction and omitting receptor endocytosis, constitutive IRF9 degradation and nuclear translocation of phosphor-ylated STAT1 ⁄ STAT2 heterodimers For details, see Table S1 Circle-headed lines, reaction catalysis; lines with perpendicular bars, reaction inhibition; single-dotted arrows, transcription mRNA to IRF9 ⁄ SOCS; TFBS, transcription factor binding site The scheme was generated using CellDesigner [53] (B) Dynamic behaviour of IFNa signalling Activation of cytoplasmic pJAK1, pSTAT1 and nuclear IRF9 measured by quanti-tative immunoblotting after stimulating Huh7.5 cells with 500 UÆmL)1IFNa To facilitate direct comparison, the same scale was used on the
y axis for the experimental data as used for the model simulation Minor levels of basal phosphorylation could not be eliminated by starva-tion for 3 h For phosphorylated JAK1, the background signal (defined as the signal for the immunoblot in areas other than the protein bands) was subtracted to better distinguish background noise from the actual signal The background level in the data for phosphorylated STAT pro-teins was low, so background corrections did not alter the results A representative plot is shown in each case, and the experiment was repeated at least three times (see Fig S3A for additional data) The error bars represent a technical relative error of 18%, derived from multi-ple measurements (Fig S1) Filled circles, experimental data; dashed lines, smoothing splines; a.u., arbitrary units The model simulation (line) for pJAK1, pSTAT1 and IRF9 was performed using COPASI [23] The simulations are within the range of data reproducibility.
Trang 4experimental data (Fig 1B), with all kinetic
parame-ters being estimated within a physiologically
meaning-ful range Model simulations start in the steady state,
without any IFNa-dependent phosphorylation of
signalling components The minor amounts of basal
phosphorylation of STAT1 shown in the data
(Fig 1B) were not considered for calibration This
phosphorylation was not affected by 3 h starvation, a
time period that is sufficient for decay of IFNa
sig-nalling in Huh7.5 cells, and therefore appeared to be
independent of a major IFNa stimulus The initial
concentrations of STAT1, STAT2, JAK1, TYK2 and
IRF9 were experimentally determined (Fig S2 and
Table S3) Experiments were performed in Huh7.5
human hepatocarcinoma cells, which show dynamic
behaviour comparable to that of primary human
hepatocytes (Fig S3A,C) Previous studies suggested
that approximately 30% of the total amount of
STAT molecules is phosphorylated after IFNa
stimu-lation [21] Finally, the major signalling peak was
assumed to occur between 20 and 60 min after
IFNa stimulation The inclusion of various feedback
mechanisms is necessary for analysis of their specific
impact, but leads to an underdetermined system due
to the number of unknown kinetic parameters
How-ever, the established model is consistent with the
experimental data (Fig 1B), and permits qualitative
predictions
Identification of IRF9 as an accelerator of IFNa signalling
To systematically identify components that control the timing and extent of IFNa signalling, a sensitivity analysis was performed with initial protein concentra-tions as input (Fig 2A) As output, both the peak time and the integrated response of the DNA-bound pSTAT1–pSTAT2–IRF9 (ISGF3) complex were analy-sed These system quantities that represent the speed and the extent of signal transduction were selected as they are likely to be crucial for an efficient antiviral response
In contrast to other systems, for which control is widely distributed [22], only a few molecules controlled the systems behaviour of IFNa signalling Among the identified proteins, nuclear phosphatases had a pro-nounced effect, positively influencing the peak time, but greatly decreasing the integrated response, in line with previous theoretical studies [5] A higher ligand dose resulted in increased signal amplitude, but had only a minor effect on signal duration, as confirmed experimentally (Fig S3B) Of the signal transducers, STAT1 and IRF9 exerted the greatest control STAT1 had a positive effect on the integrated response, but negatively influenced the peak time IRF9 was the only factor that had a substantial positive effect on the peak time and also increased the integrated response
A
B
Fig 2 Sensitivity analysis for peak time and integrated responses Initial concentrations of all players were varied to calculate their control coefficient on the kinetic behaviour of the system (A) Sensitivity analysis using the original parameter set (B) Global sensitivity analysis using 998 parameter sets.
Trang 5To confirm that the results derived by the sensitivity
analysis were not restricted to the original parameter
set, the same approach was repeated using diverse
parameter sets For this purpose, a random search
implemented in the optimization task of the simulation
software copasi [23] was used to vary all model
parameters between ±50% of their original value As
fitting constraints, the resulting kinetic behaviours had
to reproduce the experimental data (Fig 1B) Using
this process, 998 parameter sets matching the given
criteria were obtained Further analysis of these data
sets showed that the kinetic parameters could vary
quite substantially and still reproduce the experimental
data Therefore, it is important to not only examine
parameter sensitivities at a single point in parameter
space, but also to use a more global approach The
obtained parameter sets were used for a global
sensi-tivity analysis As shown in Fig 2B, the most sensitive
component in both analyses was IRF9, supporting
its central role The influence of IFNa on the time
of the signalling peak differed: in contrast to the
previous analysis, increasing IFN concentrations led to
a delayed peak for most parameter sets (Fig 2B)
However, in the experimental data the peak time for
different interferon doses was comparable (Fig S3B),
and thus it was reasonable to retain the original
parameter set for further analysis In conclusion, major
sensitivities were conserved throughout the parameter
space, confirming that IRF9 has an important impact
on the kinetic behaviour of the system, independent of
specific parameter sets
We performed additional model simulations to
quali-tatively examine the impact of large variations in IRF9
expression levels on the dynamic behaviour of IFNa
signalling, as sensitivity analyses describe only small
changes at single points within the parameter space
Indeed, a major increase in IRF9 levels accelerated
signal transduction from the cytoplasm to the nucleus,
resulting in a greater amount of active ISGF3 in the
nucleus at earlier time points (Fig 3A) Furthermore,
our model predicted a steeper signalling decline after
the peak for cells with elevated IRF9 levels To
deter-mine whether this effect is the result of up-regulated
transcription of negative inhibitors (SOCS proteins),
we removed SOCS1 induction in silico Without this
negative feedback, signal termination was attenuated
in the IRF9 over-expressing cells, and de novo IRF9
synthesis in wild-type cells accounted for enhanced
signalling during the analysed timescale (Fig S4A)
To experimentally validate the model predictions,
IRF9 was stably over-expressed in Huh7.5 cells by
lentiviral transduction (Fig S5) The phosphorylation
kinetics of nuclear STAT1 and STAT2 in response to
stimulation with 500 UÆmL)1 IFNa were determined
by quantitative immunoblotting (Fig 3B) In line with the model analysis, cells over-expressing IRF9 showed
a higher and earlier activation peak in the nucleus as well as a steeper peak decline compared to wild-type cells To determine whether different IRF9 induction rates have similar effects, we varied the parameter for IRF9 synthesis in silico More rapid IRF9 synthesis resulted in enhanced IFNa signalling, and eliminating the positive feedback dampened the response (Fig S4B) Complete absence of IRF9 was predicted
to lead to a reduction in the amounts of phosphory-lated STAT proteins (Fig S4C)
In principle, the effects of IRF9 could be achieved
by two mechanisms IRF9 could decelerate dephos-phorylation of activated STAT1⁄ 2, as phosphorylated STAT1⁄ 2 complexes can only bind specifically to DNA in combination with IRF9, and DNA-bound STAT proteins are protected from nuclear phosphatase activity [24] This mechanism was implemented in the model As a potential alternative mechanism, nuclear import of phosphorylated STAT1⁄ 2 could be increased upon interaction with IRF9 This is based on the observation that IRF9 possesses a strong constitutive nuclear localization signal recognized by a variety of importins, whereas the nuclear localization signal of phosphorylated STAT1⁄ 2 heterodimers is only recog-nized by importin a-5 [25] Therefore, complexes harbouring both types of nuclear localization signal would have an increased chance of interacting with a matching importin, resulting in enhanced nuclear translocation kinetics
We performed model simulations to assess the impact of both effects In silico analysis indicated that increasing IRF9-dependent nuclear import kinetics while neglecting IRF9-mediated phosphatase protec-tion could not represent the experimental data However, a model describing the observed dynamics solely on the basis of IRF9-dependent phosphatase protection of DNA-bound ISGF3 was necessary and sufficient to reproduce the observed kinetic data (Fig 3C)
Hence, our analysis identified IRF9 as crucial for both rapid and efficient IFNa-mediated signal trans-duction, and suggests an increased probability of DNA-binding of ISGF3 as the underlying mechanism
Over-expression of IRF9 accelerates and increases IFNa-stimulated gene expression
To test whether the accelerated and enhanced nuclear presence of phosphorylated STAT1⁄ 2 proteins upon IRF9 over-expression resulted in altered gene activation
Trang 6kinetics, we analysed the expression of IFNa-stimulated
genes by quantitative real-time PCR RNA levels of
the antiviral genes protein kinase R (PKR) [26] and
interferon stimulated gene 56 (ISG56) [27], as well as the
genes encoding negative inhibitors SOCS1 [28] and
ubi-quitin specific peptidase (USP18), were determined at
various time points for up to 24 h USP18 is a protease
that cleaves the IFN-induced, ubiquitin-like modifier
ISG15 from its target proteins [29], and was also
reported recently to block phosphorylation of JAK1
[30]
The examined genes were strongly induced by IFNa (Fig 4A) Interestingly, each gene analysed displayed different expression kinetics SOCS1 exhibited very fast induction followed by rapid repression USP18, on the other hand, displayed increased expression for up to
24 h Similar to USP18, the antiviral genes ISG56 and PKR showed prolonged up-regulation Interestingly, for all genes investigated, induction of gene expression was faster when IRF9 levels were elevated, consistent with the general mRNA induction predicted by the model (Fig S4D) For ISG56, SOCS1 and USP18, a
C
Fig 3 IRF9 controls the dynamics of IFNa signalling (A) Model prediction of IFNa-dependent pSTAT1 ⁄ pSTAT2 accumulation in the nucleus, which is equivalent to the kinetics of ISGF3 (pSTAT1-pSTAT2-IRF9) Simulations (lines) were performed for wild-type cells (wt) and for cells with 32-fold IRF9 over-expression (IRF9oe) (B) Experimental validation of the model prediction Wild-type Huh7.5 cells (wt) or Huh7.5 cells stably over-expressing IRF9 32-fold (IRF9oe) were stimulated with 500 UÆmL)1IFNa, and phosphorylation of nuclear STAT proteins was mea-sured by quantitative immunoblotting (see Fig S5) To facilitate direct comparison, the same scale was used on the y axis for the experi-mental data as used for the model simulation Over-expression of a control protein (GFP) had no effect on the dynamic behaviour (Fig S6) The error bars represent a technical relative error of 18%, derived from multiple measurements (Fig S1) Filled circles, experimental data; dashed lines, smoothing splines; a.u., arbitrary units (C) In silico analysis of two potential mechanisms underlying the effect of IRF9 Simula-tion of DNA-bound ISGF3 and pSTAT1–pSTAT2 heterodimers in the nucleus, representing situaSimula-tions where IRF9 leads to increased nuclear import of pSTAT1–pSTAT2 or provides protection from phosphatase degradation.
Trang 7high IRF9 level resulted in an increased peak
ampli-tude, but the peak amplitude was unaltered for PKR
The integrated response was larger for each of the four
genes when IRF9 was overexpressed, with a more
pro-nounced difference during the first 4 h (Fig 4A)
To confirm that the observed effect was not restricted to the tested genes, we investigated the glo-bal induction of IFNa-stimulated genes using a time-resolved microarray (Fig 4B,C) For data analysis, we selected genes that showed an increased relative
α α
α α
A
–2
–2
Fig 4 IRF9 controls the dynamics of IFNa-induced gene expression Huh7.5 wild-type cells stably transduced with an empty vector (wt) or IRF9-over-expressing cells (IRF9oe) were stimulated with 500 UÆmL)1IFNa, and RNA was extracted at the indicated time points (A) Quanti-tative real-time PCR analysis of four sample genes For each gene, the integrated response was calculated for early (4 h) and late (24 h) time points (B,C) Time-resolved microarray analysis performed with one replicate per time point (B) Kinetics of representative genes in Huh7.5 wild-type cells stably transduced with an empty vector (wt) or in IRF9-over-expressing cells (IRF9oe) (C) Scatter plot showing the difference
in gene induction time and mean fold expression in control or IRF9-over-expressing cells Positive values indicate accelerated and augmented gene expression in IRF9oe cells The genes indicated show an increased relative expression upon IFNa stimulation in either wild-type or IRF9-over-expressing cells and have a difference in gene induction time of less than 6 h (257 genes) SOCS1 and IRF9 are also included There is a clear trend for faster and augmented gene expression in the IRF9-over-expressing cells, as demonstrated by the positive slope of
a linear model y = b + m*x, which was fitted to the data points The variables x and y indicate the time difference of activation and the mean differential expression, respectively The slope (m = 0.08; P value = 0.00006, t test) and intercept (b = 0.25; P value < 2e-16, t test) were estimated using the lm() function in R, version 2.11.1 (http://www.R-project.org).
Trang 8expression in both wild-type and IRF9-over-expressing
cells upon stimulation with IFNa A gene ontology
analysis using DAVID [31] showed that these 284
genes are related to immune and virus response as well
as antigen processing and presentation, as expected
(Table S4) Gene expression time series were
character-ized with respect to the differences in mean fold
expression and temporal regulation (see Experimental
procedures) There was an overall positive correlation
between the level of gene expression and the expression
kinetics: genes that were more strongly up-regulated in
the IRF9-over-expressing cells were also induced
ear-lier Remarkably, this was true for the majority of the
genes in the IRF9-over-expressing cells compared to
wild-type cells (160 out of 257) One exception was
IRF9 itself, as it could not be induced much beyond
the already high expression level in over-expressing
cells Taken together, these data demonstrate that an
elevated amount of IRF9 not only results in higher
levels of transcription, but also in accelerated
expres-sion of IFNa target genes
Discussion
Here we describe the development of a comprehensive
model of IFNa signalling and its experimental
valida-tion The aim of our modelling approach was to
qualitatively predict the interplay between various
molecules and feedback mechanisms, requiring
consid-eration of all known pathway components Obviously,
a mathematical model that includes all known
nega-tive and posinega-tive feedbacks represents an
underdeter-mined system as it contains too many parameters that
cannot be reliably estimated from the experimental
data To verify the predictive power of our model, a
sensitivity analysis of 998 parameter sets describing
the experimental data was performed, and the results
obtained were compared to those for the original
parameter set (Fig 2A,B) The observations were
comparable, indicating that they are intrinsic
proper-ties of the model structure The robustness of
sensitiv-ity against single parameter changes has been
described by Gutenkunst et al [32], suggesting that
model predictions are reasonable when they are
derived from collective fits and can only be improved
by precise and complete measurements of all kinetic
parameters According to the sensitivity analysis,
IRF9 is a decisive factor in IFNa signalling as it
rep-resents the only component that both augments and
accelerates antiviral gene expression This was
con-firmed by in silico analysis simulating IRF9
over-expression and subsequently experimentally validated
(Fig 3A,B) Our approach focused on the qualitative
analysis of mechanisms that determine signalling speed and the extent of pathway activation The model was used to design experiments that would be most infor-mative, and the experimental data were in qualitative agreement with the model predictions, although some deviations in quantitative terms were observed, such
as smaller differences between peaks in experimental data compared to the model prediction Importantly, the main characteristics of signalling kinetics could be validated
Increasing the initial IRF9 concentration by over-expression resulted in higher levels of phosphorylated STAT proteins in the nucleus, and consequently augmented expression of IFNa target genes This is consistent with previous reports describing the impact of IRF9 on the amount of active ISGF3 [8–11] However, in contrast to previous studies, we analysed the IFNa response in a time-resolved manner We showed that enhanced IFNa-induced gene expression not only applies at isolated time points but also for the overall integrated response In addition, we demon-strated that IRF9 is also crucial for the speed of the IFNa response, with higher IRF9 levels accelerating signal transduction and gene expression Theoretically, these effects of IRF9 could be achieved by two mechanisms: by increased nuclear import of the signal transducers or by IRF9-mediated protection from nuclear phosphatases Model analysis excluded acceler-ated nuclear import and indicacceler-ated protection from nuclear phosphatases as the underlying mechanism These mechanisms are difficult to address experimen-tally, but, by disentangling the involved processes, the mathematical modelling approach provides important insights for further studies
Both in wildtype and IRF9 overexpressing cells, the analysed genes showed different expression kinetics Possible mechanisms explaining this behaviour are varying production rates and differences in mRNA stability Furthermore, IFN-stimulated transcription factors may account for the sustained activation of certain genes, constituting positive feedback loops Regulatory networks in which individual genes are regulated by a cascade of multiple transcription fac-tors were recently shown to play an important role in the antiviral response [33] Here, SOCS1 expression was rapidly activated and repressed, whereas the acti-vation of USP18 was sustained These obseracti-vations are in agreement with a recent report stating that SOCS1 is responsible for early inhibition of IFNa sig-nalling, whereas USP18 mediates late inhibition [34] The sustained response could be explained by an addi-tional positive feedback As shown in previous studies, expression of the IRF9 gene is regulated through a
Trang 9positive feedback loop by the IFN-stimulated
tran-scription factor CCAAT-enhancer-binding protein b
(C⁄ EBP-b) [35] The prolonged up-regulation of other
IFN-stimulated genes could be mediated by IRF7,
which is produced in response to IFNa and is able to
bind promoters of IFN-inducible genes [36] The
exis-tence of positive feedback mechanisms could be a
gen-eral design principle in IFN signalling to enhance the
antiviral response In contrast to oncogenic pathways,
augmented IFN signalling is not detrimental to an
organism, as it does not lead to uncontrolled cell
pro-liferation, but rather to apoptosis [37]
While we were performing experiments in Huh7.5
cells to validate the model predictions, the enhanced
IFN response as an effect of increased IRF9 levels was
also demonstrated in various cell types, suggesting a
general mechanism [11,12] Individual cells in a cell
population may elicit different responses [38]
Never-theless, we aimed to develop a population-based
model, as the IFNa response primarily occurs at the
tissue level, comprising a population of individual cells
Additionally, regulation of gene expression is likely to
differ between individuals Therefore, variations in
either IRF9 initial concentrations or the IRF9
induc-tion rate may be one reason for patient-to-patient
vari-ations in responses to IFNa therapy As demonstrated
by model simulations, not only higher IRF9 levels, but
also faster IRF9 synthesis, significantly augment early
IFN signalling Consequently, the balance between
positive and negative feedback loops (e.g IRF9⁄
SOCS) may be decisive
As a rapid IFN response could be crucial for a viral
infection [2], the IRF9 level in a cell may play a
pivotal role In line with this, IRF9 was shown to be
targeted by several viruses in order to interfere with
the cellular antiviral response, as demonstrated for
human papillomavirus [39,40], reovirus [41],
adeno-virus [12], hepatitis B adeno-virus [42] and human
cytomega-lovirus [43] Moreover, it was shown that elevated
IRF9 levels increase the antiviral response [10,12]
Additionally, it was recently reported that IRF9 is
necessary for the anti-proliferative activity of IFNa, as
only RNAi against IRF9, but not against STAT1,
inhibited IFNa-mediated apoptosis [44]
In conclusion, our modelling approach, in
combi-nation with experimental analysis, confirmed that
ele-vated IRF9 starting levels are a crucial determinant
for amplified IFNa-mediated antiviral signalling,
and additionally identified the IRF9 level to be vital
for a rapid response As a key regulator shaping
the early phase of IFNa signalling, IRF9
repre-sents an appealing target for innovative therapeutic
approaches
Experimental procedures
Cells and time-course experiments
Huh7.5 cells (a kind gift from C M Rice, Laboratory of Virology and Infectious Disease, Rockefeller University, NY) were cultivated in Dulbecco’s modified Eagle’s medium (DMEM; Invitrogen, Carlsbad, CA, USA) supple-mented with 10% fetal bovine serum (Invitrogen) and 1% penicillin⁄ streptomycin (Invitrogen) One day before com-mencement of a time-course experiment, 1.7· 106 cells were seeded into 6 cm dishes Prior to stimulation with IFNa, the cells were washed three times by removing the culture medium and replacing it with DMEM, and after-wards cultivated in starvation medium for 3 h [DMEM supplemented with 1 mgÆmL)1BSA (Sigma-Aldrich, St Louis,
MO, USA) and 25 mm Hepes pH 7.0 (Invitrogen)]
To stimulate cells, human leukocyte IFNa (R&D Systems, Minneapolis, MN, USA) was added to the medium to a final concentration of 500 UÆmL)1 For each time point, the contents of one dish were lysed using 1% Nonidet P-40 lysis buffer (1% Nonidet P-40, 150 mm NaCl, 20 mm Tris
pH 7.4, 10 mm NaF, 1 mm EDTA pH 8.0, 1 mm ZnCl2
pH 4.0, 1 mm MgCl2, 1 mm Na3VO4, 10% glycerol, freshly supplemented with 2 lgÆmL)1 aprotinin and 200 lgÆmL)1 4-(2-aminoethyl)-benzensulfonylfluoride), and the lysates were used for immunoprecipitation or directly analysed by SDS⁄ PAGE For cell fractionation, cells were lysed using 0.4% Nonidet P-40 cytoplasmic lysis buffer (0.4% Noni-det P-40, 10 mm Hepes pH 7.9, 10 mm KCl, 0.1 mm EDTA, 0.1 mm EGTA, freshly supplemented with
2 lgÆmL)1 aprotinin, 200 lgÆmL)1 4-(2-aminoethyl)-benzen-sulfonylfluoride, 1 mm dithiothreitol, 1 mm NaF, 0.1 mm
Na3VO4), and vortexed for 10 s After centrifugation (1 min at 17 900 g, 4C), supernatants were used as the cytoplasmic fraction and the nuclear pellet was resus-pended in nuclear lysis buffer (20 mm Hepes pH 7.9, 25% glycerin, 400 mm NaCl, 1 mm EDTA, 1 mm EGTA, freshly supplemented with 2 lgÆmL)1 aprotinin, 200 lgÆmL)1 4-(2-aminoethyl)-benzensulfonylfluoride, 1 mm dithiothreitol,
1 mm NaF, 0.1 mm Na3VO4) by repeated vortexing The suitability of the procedure was verified by confirming the presence of the nuclear marker protein poly [ADP-ribose] polymerase 1 and the cytoplasmic marker protein Eps15 in the corresponding fractions
Primary human hepatocytes were isolated and cultivated
in serum-free Williams’ Medium E (Biochrom AG, Berlin, Germany) [45] The viability of isolated hepatocytes was determined by trypan blue exclusion Only cell preparations with a viability > 80% were used for experiments The iso-lated cells were seeded on collagen type I-coated culture dishes at a density of 1.2· 105
cells per cm2 Tissue sam-ples from human liver resection were obtained from patients undergoing partial hepatectomy for metastatic liver tumor secondary to colorectal cancer Experimental
Trang 10procedures were performed according to the guidelines of
the charitable state-controlled foundation Human Tissue
and Cell Research, with the patient’s informed consent [46],
as approved by the local ethical committee
The day after isolation, the primary hepatocytes were
cul-tivated for 2 days in Williams Medium E supplemented with
2 mm l-glutamine (Invitrogen), 100 nm dexomethasone
(Sigma) and 1% penicillin⁄ streptomycin (Invitrogen) Prior
to stimulation with IFNa, the cells were washed three times
by removing the culture medium and replacing it with
Williams Medium E and afterwards cultivated in starvation
medium for 3 h (Williams Medium E supplemented with
2 mm l-glutamine) The time-course experiment was
per-formed according to the protocol for Huh7.5 cells
Quantitative immunoblotting
For immunoprecipitation, the lysates were incubated with
anti-JAK1 serum (Upstate Millipore, Billerica, MA, USA)
and anti-TYK2 polyclonal IgGs (Upstate Millipore) and
protein A–Sepharose beads (GE Healthcare, Chalfont, NJ,
United Kingdom) For cellular lysates, protein
concentra-tions were measured using the BCA assay (Pierce, Thermo
Fisher Scientific Inc., Waltham, MA, USA)
Immunoprecipi-tated proteins, cytoplasmic (70–80 lg) or nuclear lysates
(45 lg) were loaded in a randomized manner on a 10%
SDS⁄ polyacrylamide gel as described previously [47],
sepa-rated by electrophoresis and transferred to poly(vinylidene
difluoride) (STATs, IRF9) or nitrocellulose membranes
(JAK1, TYK2) Proteins were immobilized using Ponceau S
solution (Sigma-Aldrich) followed by immunoblotting
anal-ysis using anti-phosphotyrosine monoclonal IgG 4G10
(Upstate Millipore) for the phosphorylation signal of
im-munoprecipitated JAK1 and TYK2, anti-phospho-STAT1
IgG (Cell Signaling Technologies, Danvers, MA, USA),
anti-phospho-STAT2 IgG (Cell Signaling Technologies)
and anti-IRF9 IgG (BD Bioscience, Franklin Lakes, NJ,
USA) Antibodies were removed by treating the blots with
b-mercaptoethanol and SDS Reprobing was performed
using anti-JAK1 (Cell Signaling Technologies), anti-TYK2
(Upstate Millipore), anti-STAT1 and anti-STAT2 (Upstate
Millipore) For normalization, IgGs against calnexin
(Stressgen, Enzo Life Sciences, Plymouth Meeting, PA,
USA) and poly [ADP-ribose] polymerase 1 (Roche, Basel,
Switzerland) were used Secondary horseradish
peroxidase-conjugated IgGs (anti-rabbit HRP, anti-goat HRP,
protein A HRP) were purchased from GE Healthcare
Immunoblots were incubated with enhanced
chemilumines-cence (ECL) or ECL Advance substrate (Amersham), and
signals were detected using a CCD camera (LumiImager F1
workstation; Roche) This ensured measurements were in
the linear range, avoiding saturation effects Data were
quantified using lumianalyst 3.1 software (Roche)
Quantitative immunoblotting data were processed using
gelinspectorsoftware [48] The following normalizers were
used: GST-TYK2DC or GST-JAK1DN for pJAK1, JAK1, pTYK2 and TYK2, calnexin for pSTAT1, STAT1, pSTAT2, STAT2 and IRF9, in the cytoplasm and poly [ADP-ribose] polymerase 1 for pSTAT1, STAT1, pSTAT2, STAT2 and IRF9 in the nucleus To smooth spline estimates of the data, MATLAB (http://www.mathworks.com) csaps-splines with a smoothness between 0.7 and 0.9 were used
Plasmids, recombinant proteins and lentiviral transduction
Recombinant proteins were used as normalizers and as refer-ences to determine the number of molecules per cell To generate N-terminally GST- and SBP-tagged constructs, the pGEX system (GE Healthcare) and the derived pSBPEX sys-tem, in which glutathione S-transferase (GST) was replaced
by strepatavidin binding tag (SBP), were used To construct GST-TYK2DC, the N-terminal FERM and SH2 domain of TYK2 (amino acid 1–586) were amplified by PCR, using human TYK2 cDNA (Open Biosystems, Huntsville, AL, USA, cDNA number 4591726) as template The resulting fragment was cloned into the BamHI–EcoRI site of pGEX-2T GST-JAK1DN was generated by amplifying human JAK1 cDNA (a kind gift from I Behrmann, Life Sciences Research Unit, University of Luxembourg) from the SH2 domain to the C-terminus (amino acids 421–1150) The resulting fragment was cloned into the BamHI–EcoRI site of pGEX-2T SBP-STAT1DN was generated by amplifying human STAT1 cDNA (a kind gift from H Hauser, Helmholtz Centre for Infection Research, Braunschweig, Germany), to yield a product corresponding to amino acids 131–750 The resulting fragment was cloned into the BamHI–EcoRI site of pSBPEX-2T SBP-STAT2DN was gen-erated by amplifying human STAT2 cDNA (a kind gift from
H Hauser, Helmholtz Centre for Infection Research, Braun-schweig Germany), resulting in a product corresponding to amino acids 133–851 The resulting fragment was cloned into the BamHI–EcoRI site of pSBPEX-2T To express the recombinant proteins, the expression plasmids were trans-formed into competent Escherichia coli BL21(DE3) Codon-PlusRIL (Stratagene, Agilent, Santa Clara, CA, USA), and proteins were purified using glutathione–Sepharose beads for GST-tagged proteins, or streptavidin–Sepharose beads for SBP-tagged proteins GST-tagged IRF9 was kindly provided
by Rainer Zawatzky (German Cancer Research Center, Divi-sion of Viral Transformation Mechanisms, Germany) For over-expression studies, IRF9 cDNA was cloned into the lentiviral expression vector pRRLSIN.cPPT.PGK-GFP.WPRE (deposited in the non-profit plasmid repository Addgene, number 12252) by PCR amplification of pCMV-Sport6-IRF9 (Open Biosystems) and digestion with BamHI and SalI, replacing the gene encoding GFP (green fluorescent protein) and resulting in pRRLSIN.cPPT.PGK-IRF9.WPRE
To generate pRRLSIN.cPPT.PGK-MCS.WPRE, a multiple cloning site with the restriction sites BmtI, PacI, SmaI, PstI,