Quantitative Analysis of TLR-4-Mediated Cell.
Responses in Murine Macrophages
by
Rongcong Wu
M.S., Biology (2004)
B.E., Chemical Engineering and Technology (2001)
Tsinghua University, China
SUBMITTED TO THE DEPARTMENT OF BIOLOGICAL ENGINEEING IN
PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE IN BIOLOGICAL ENGINEERING
AT THE
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
JUNE 2008
© 2008 Massachusetts Institute of Technology
All rights reserved
.r
11-
Signature ofAuthor:
w
A
In/
(I
-/
Department of Biological bngineering
'May
9, 2008
Certified by
auffenburger
ii/6 gl
Professor of Biological Engineering, Chemica'l ngmiering, and Biology
Thesis Supervisor
Certified by:
I
David B. Schauer
Medicine
Comparative
/pessor of Biological EBmeering and
Thesis Supervisor
'0
Z~l';~c
Accepted by:
-
' -- -----
I----
Alan J. Grodzinsky
Professor of Biological Eng
MASSACHLSE~TS INSTn".
g, and Mechanical Engineering
ering, Electrica-ngin
Chairman, Department Committee on Graduate Students
OF TEOHNOLOGY
JUL 2 8 2008
LIBRARIES
eCHNVEe
Quantitative Analysis of TLR-4-Mediated Cell
Responses in Murine Macrophages
by
Rongcong Wu
Submitted to the Department of Biological Engineering
on May 9, 2008 in Partial Fulfillment of the
Requirements for the Degree of Master of Science in
Biological Engineering
ABSTRACT
TLR-4 is essential in host defense against bacterial infection. By recognition
of specific pathogen-associated molecular patterns such as lipopolysaccharide (LPS),
TLR-4 can in tandem initiate a pair of downstream signaling pathways to regulate
cytokine/chemokine release, endotoxin tolerance and apoptosis, which have been
suggested to directly or indirectly participate in the regulation of innate and adaptive
immune responses. However, little is known about their detailed signal-response
relationships.
In this thesis, we sought to identify these potential signal-response
relationships in RAW264.7 cells through systematic analysis. We first measured LPS
stimulated dynamic signaling profiles over a range of an inhibitor of p38 MAPK,
SB202190 concentrations for a distribution of kinases centrally involved in TLR-4
signaling network. We then applied quantitative analytical approaches to determine
the most important signals or signal combinations contributing to induction of either
IL-6 and TNF-ao secretion or apoptosis and construct their corresponding predictive
mathematical models.
Particularly, we found that the partial least squares regression (PLSR) models
built using the ratio of phosphorylated Jun N-terminal kinase (JNK) and extracellular
signal regulated kinase (ERK) predicted LPS plus SB202190-induced apoptosis
accurately even following perturbation with pharmacological inhibitors of JNK and
ERK. Thus, by combining experimental and computational approaches, this thesis has
proposed two new potential targets, JNK and ERK, for development of drug therapies
against bacterial infection.
Thesis Supervisor: Douglas A. Lauffenburger
Title: Professor of Biological Engineering, Chemical Engineering, and Biology
Thesis Supervisor: David B. Schauer
Title: Professor of Biological Engineering and Comparative Medicine
BIBLIOGRAPHIC NOTE
Rongcong Wu received a bachelor of engineering in Chemical Engineering
and Technology in 2001 and a master of science in Biology in 2004 from Tsinghua
University in Beijing, China. While at Tsinghua University, Rongcong worked in the
laboratory of Dr. Zhao Wang on process optimization for fermentation of polyactins
by Hemolytic streptococcus and on molecular and cellular pharmacology of natural
products. During the period, Rongcong published several articles on the molecular
mechanisms of apoptosis induced by cycloheximide, tannins or saponins. Upon
graduation, he received the university-level "Outstanding Master's Thesis" award and
"Outstanding Master's Student" award. Following receipt of his degrees, he worked
as a consultant for one year at Center for Information & Consultation of the Chinese
Academy of Sciences in Beijing, China.
Rongcong started his graduate work in the Biological Engineering Division at
MIT in the fall of 2005. Under the joint supervision and financial support of Drs.
Douglas Lauffenburger and David Schauer, Rongcong completed his thesis entitled,
"Quantitative Analysis of TLR-4-Mediated Cell Responses in Murine Macrophages".
After graduation, Rongcong will take a development scientist position at
Agencourt Bioscience, a Beckman Coulter Company.
ACKNOWLEGEMENTS
I would like to thank all the people who have contributed to the successful
ending of all this work. They are my colleagues, advisors, family and friends.
Foremost, I thank my two advisors, Drs. Douglas Lauffenburger and David Schauer,
for their invaluable professional guidance throughout my thesis work.
For particular aspects of this work, I think Arthur Goldsipe for help with PLSR
modeling; Ming Chen, Suzanne Gaudet, Pamela Kreeger, Nancy Guillen and Paul
Huang for help with quantitative western blot; Megan McBee for help with ELISA
and isolation of BMDMs; and Katie Schlieper for help with reagent order and mouse
preparation. And along with all those already mentioned, I also thank many other
members of the Schauer group for helpful conversations and for making the lab a fun
place to work.
Finally, I think my parents for their unconditional support of my work and my
continued education in addition to offering me the love. And I think my wife,
Shanshan Feng, who has given me the love and confidence to continue to tackle the
challenge of scientific research and life, and prepares delicious Chinese foods for me.
TABLE OF CONTENTS
1. Introduction ...............................................................................
9
1.1. Complexity of Toll-like receptor 4 signaling network............................. 9
1.1.1.
Activated pathways in response to TLR-4 engagement.................9
1.1.2.
Negative regulators..........................................12
1.2. TLR-4-mediated cell responses in macrophages.............................. 12
1.2.1.
Induction of cytokines/chemokines...................................13
1.2.2.
Endotoxin tolerance..........................................14
1.2.3.
Apoptosis................................................................. 14
......16
Challenges in TLR-4 signaling network analysis ............
1.2.4.
1.3. Data-driven approaches for systematic analysis of biological networks......17
1.3.1.
Network-function modeling by partial least squares regression.....18
Challenges in quantitative analysis of TLR-4-mediated cell
1.3.2.
responses ................................................................................ .20
2. Experimental techniques for quantitative TLR-4 signaling network analysis......21
2.1. Introduction................................................... 21
2.2. Repurification of commercial LPS ............................................... 21
2.3. Optimization of protocols for the measurement of various cellular events...23
2.3.1.
Measurement of apoptosis ................................................ 23
2.3.2.
Quantification of cytokine release by ELISA .......................... 30
2.3.3.
Quantitative western blots measurements .............................. 33
3. Quantitative analysis and data-driven modeling of TLR-4 signaling-mediated cell
responses ........................................................................................ 39
3.1. Introduction ......................................................................... . 39
40
3.2. Directed data collection and preliminary data analysis ..................
The critical role of p38 MAPK in LPS-stimulated cell responses...40
3.2.1.
Proteomic compendium of the LPS plus SB202190-stimulated
3.2.2.
signaling network.....................................................................44
3.3. PLS model construction and interpretation ..................................... 48
3.3.1.
Clustering through principal component analysis ................. 49
3.3.2.
PLS modeling of TLR-4 signaling-mediated secretion of IL-6 and
TN F-ac .................................................................................. 51
3.3.3.
PLS modeling of TLR-4 signaling-mediated apoptosis ............... 56
3.4. Model testing through prediction......................................63
3.4.1.
Construction of new independent datasets .............................64
3.4.2.
Comparison of model predictivity on independent datasets........68
3.4.3.
Other biological considerations ......................................... 69
4. Conclusions and future directions ...................................................... 73
73
4.1. Dynamic behavior of TLR-4 signaling network ..................
4.2. Looking into constructed models for potential applications and future
74
directions ..................................................................
5. Appendices................................................................................. 77
.... 77
5.1. Materials .................................................................
5.1.1.
Cell culture ............................................................... 77
5.1.2.
Drug treatments .......
...........
......................................
.... 77
5.2. Experimental protocols.............................................. 77
5.2.1.
Repurification of raw LPS ............................................. 77
5.2.2.
MTT assay................................................................78
5.2.3.
7-AAD & Annexin V-PE double staining...........................78
5.2.4.
Assessment of apoptosis based on active caspase-3 & cleaved
PARP.. ........................................................... 79
5.2.5.
Sandwich ELISA ......................................................... 79
5.2.6.
Quantitative western blot ................................................. 80
5.3. PCA and PLSR model construction and refinement............................81
5.4. References ...................................................... ..........
82
LIST OF FIGURES AND TABLES
Figure 1-1. Schematic of TLR-4 signaling network ...................................... 11
Figure 2-1. The measurement of apoptosis by two double staining assays..............25
Figure 2-2. Effect of LPS treatment on RAW264.7 cell size...........................26
Figure 2-3. SB202190 has no effect on RAW264.7 cell size............................27
Figure 2-4. Death profiles of RAW264.7 cells in response to increasing doses of
............ 28
SB202190 and LPS in combination............................
Figure 2-5. General schematic of the sandwich ELISA protocol........................ 31
Figure 2-6. The release of IL-6, IL-12 and TNF-a in LPS-stimulated RAW264.7 cells
32
.......................................................................
Figure 2-7. Phosphorylation of kinases in response to LPS insults in RAW264.7 cells
......................... 34
....... ......................
................................
Figure 2-8. The linear ranges of various proteins for quantitative western blots
................................ 36
A nalysis .....................................................
Table 2-1. The dilution ratios of antibodies and the linear ranges of various proteins
For quantitative western blot analysis ................................................ 36
Figure 2-9. Phosphorylation of JNK2 in response to multiple combinations of LPS
and SB202190 measured by quantitative western blot.............................38
Figure 3-1. Schematic of data collection plan............................................. 39
.......... 40
Table 3-1. The list of 13 cues.......................................
Figure 3-2. Inhibition of p38 MAPK strongly reduced the secretion of IL-6 and
TNF-a in LPS-stimulated RAW264.7 cells....................................41
Figure 3-3. High doses of SB20210 were required for induction of apoptosis
Synergically with LPS in RAW264.7 cells............................................ 41
Figure 3-4. A proteomic compendium of LPS plus SB202190-stimulate signaling in
... 44
RAW264.7 cells ..................................................................
Figure 3-5. Comparison of the effects of SB202190, SB203580 and U0126 on LPS
stimulated kinase phosphorylation ..................................................... 46
Figure 3-6. Both PI-3K and p38 MAPK were required for LPS-stimulated
Phosphorylation of Akt in RAW264.7 cells..............................................47
Figure 3-7. PCA mapping of input treatments............................................. 49
Figure 3-8. Principal component analysis distinguished distinct time-resolved
activities of specific kinases.............................................................50
Figure 3-9. PLSR modeling indicates Akt is the primary informer for IL-6 and TNF-a
secretion in RAW264.7 cells..............................................53
Figure 3-10. Different effects of two PI-3K inhibitors on LPS-induced IL-6 and
.... 54
TNF-a release .................................................................
Figure 3-11. Iterations of modeling on apoptosis using subsets of signal variables
Indicate that no model outperforms others for all descriptors ........................ 57
Figure 3-12. Inhibitors screen indicates SB202190 plus LPS-induced RAW264.7
apoptosis can be regulated by a few kinases ........................................... 58
Figure 3-13. PLSR models built on kinase ratio and response could be another good
choice description of TLR-4 signaling mediated apoptosis........................... 59
Figure 3-14. The comparison of four kinase ratio PLSR models.......................61
Figure 3-15. Effects of the timing of U0126 addition on LPS plus SB202190-induced
apoptosis ........ ..................
..
....................................... 62
Figure 3-16. Comparison of model predictivity on the trained data.....................64
Figure 3-17. Opposite effects of U0126 and SP600125 on LPS plus SB202190
induced apoptosis ................................................. .... ...........
65
Table 3-2. The list of new treatments for data collection.................................66
Figure 3-18. Quantitation of JNK inhibition based on the linear relationship between
phospho-JNK and phospho-ATF-2 ..................................................... 66
Figure 3-19. A priori prediction using independent datasets .............................. 67
Figure 3-20. Inhibition of c-Jun phosphorylation by various inhibitors appeared to
correlate with enhanced LPS plus SB202190-induced apoptosis..................70
Figure 4-1. A putative model for sequential activation of survival and death programs
............. 76
triggered by LPS plus SB202190.............................
Table 5-1. Lysis buffer for quantitative western blot ...................................... 80
CHAPTER 1
Introduction
1.1. Complexity of Toll-like receptor 4 signaling network
Vertebrates are constantly threatened by the invasion of microorganisms and
have evolved systems of immunity to eliminate infectious pathogens in the body.
Initial sensing of microbial agents is mediated by the recognition of
pathogen-associated molecular patterns (PAMPs), which are conserved structures
expressed uniquely by microbes of the same class. Toll-like receptors (TLRs), a
family of type I transmembrane receptors, have been well documented to recognize
PAMPs through a highly variable extracellular region containing a leucine-rich repeat
(LRR) domain and play an essential role in the host defense against microbial
pathogens. To date, more than 11 human TLRs and 13 mouse TLRs have been
identified, and their homodimers or heterodimers can recognize a variety of PAMPs
ranging from bacterial and viral components to fungal and protozoal molecules [1].
For example, lipopolysaccharide (LPS) that is uniquely expressed in the outer
membrane of cell wall by gram-negative bacteria is specifically recognized by TLR-4
and has been broadly applied to initiate TLR-4 signaling as we did in this study.
1.1.1. Activated pathways in response to TLR-4
engagement
Generally, after recognition of PAMPs TLRs activate a cascade of intracellular
signaling events through highly conserved TIR homology domains localized in their
intracellular tails [2]. Differing from other TLRs, TLR-4 recruits two sets of adaptors
(MyD88-MAL and TRAM-TRIF), although these adaptors are also shared with other
TLRs [2], and initiates two different downstream signaling cascades (Figure 1-1). The
MyD88-MAL-dependent pathway recruits IRAK-1 and IRAK-4, which then
phosphorylate TRAF-6 leading to the activation of MKK complexes and IKK
complex by binding to and activating TAKI. Subsequently, activated MKK
complexes and IKK complex phosphorylate several pivotal downstream kinases,
including ERK1/2, p38 MAPK, JNK/SAPK and IKKcp3, which then co-regulate
many transcription factors, such as ATF-2, c-Jun, Elk, NF-cB, and so on, to control
various cell responses, including pro- or anti-inflammatory cytokine/chemokine
release and apoptosis. In contrast, the TRAM-TRIF-dependent pathway recruits and
activates TBK1 and IKKE [3], which then phosphorylates IRF-3 to control the
transcription of IFN-inducible genes. Moreover, the TRAM-TRIF-dependent pathway
also associates with TRAF-6 [4, 5] and RIPI [2, 5] to crosstalk with the
MyD88-MAL-dependent pathway. Additionally, LPS-stimulated activation of PI3K is
also observed and might involve TLR-4 and RAC, although the precise mechanism
remains to be determined [6, 7]. The activation of PI3K and its downstream kinase,
Akt/PKB are thought to act as the negative regulators of the MyD88-MAL-dependent
pathway and cytokine expression by inactivating ERK1/2, p38 MAPK, JNK and
IKKaf through as-yet-unknown mechanisms [2, 7, 8, 9] and probably also by
regulating the activity of BTK, which phosphorylates MAL and then interacts with
SOCS-1 resulting in MAL polyubiquitination and subsequent degradation [6, 10].
However, some conflicting results were reported. For example, PI3K was observed to
mediate TLR-4-induced activation of NF-KB in endothelial cells [11] and two PI3K
inhibitors, wortmannin and LY294002, were shown to have opposite effects [12]. The
same situation also exists in activation of PKR. Although how dsRNA from virus
binds to PKR and induces autophosphorylation and activation of PKR which
subsequently phosphorylates eIF2a to inhibit protein synthesis has been well
documented [13, 14], the precise mechanism resulting in activation of PKR in
LPS-stimulated signaling cascades still remains obscure. Moreover, whether the
engagement of TLR-4 will even be able to activate PKR is controversial. Some
groups have reported that they didn't detect increased phosphorylation of PKR or
eIF2a in LPS-treated primary human macrophages [15], ANA-1 murine macrophages
and fresh macrophages obtained from C57BL/6 mice [16], or rat gastrocnemius and
..
....
.. . ...
....
..
.....
..
..........
.......
.... . ...
... .............
. .....
.
...
. ..
....
[.. . ..
....
... .
.. . .. . ..
. .........
.......
...
......
....
. .... . .....
heart [17] while some groups did in LPS-treated bone marrow derived macrophages
(BMDMs) [18], or rat brain glial cells [19], RBL-2H3 cells [20] and rat liver [17].
These inconsistent results might be attributable to different cell types used, but
whether there are different mechanisms adapted by different cell types still needs to
be testified.
0
0 LPS
L s-7e proteins (Bcl-2--family,
-- ----FADD,
/ ----FLIP...);
Apoptosis-related
- ---- ...
--- - --
Figure 1-1. Schematic of TLR-4 signaling network
Dashed lines with a red question mark denote unconfirmed relationships. Lines terminating in a
perpendicular line represent inhibition, whereas arrows represent activation. The individual steps
and proteins are described in more detail in the text.
.
......
1.1.2. Negative regulators
Besides the above forward pathways, there are also many constitutively
expressed or inducible endogenous negative regulators in the TLR-4 signaling
network which target different stages of the pathways and participate in the precise
regulation of TLR-4-stimulated cell responses ensuring that appropriate reaction is
undertaken. Liew et al. have given us a good summary in their review article [2] on a
major group of them, including soluble TLR-4, MyD88s, SOCS1, IRAK-M, TOLLIP,
ST2L, A20, SIGIRR, TRAILR, and TRIAD3A, which decoy TLR-4, adaptors or
kinases, and/or even promote their ubiquitylation and degradation resulting in
turndown of proximal signal transduction of MyD88-MAL-dependent pathways. In
contrast, dual-specificity protein phosphatases (DUSPs), also referred to as MAPK
phosphatases (MKPs) in mammalian cells, target the mitogen-activated protein
kinases (MAPKs). They selectively dephosphorylate and inactivate ERK1/2, p38
MAPK and/or JNK (Figure 1-1) and thereby alter cell responses regulated by these
kinases. Up to now, there are more than 10 MKPs identified in mammalian cells, but
how exactly they are induced and what their substrate specificities are are still
undefined. For instance, MKP-1 was initially thought to be a specific phosphatase of
ERK1/2, but more and more emerging evidences support that MKP-1 can inactivate
p38 MAPK and JNK preferentially [21, 22]. In addition, the induction of MKP-l was
recently demonstrated to be controlled by p38 MAPK and MK2 [23], but some groups
also argued that ERK1/2 was more important in this process [21, 23]. Since negative
regulators are usually activated by some pathways and then target other pathways, the
TLR-4 signaling network acts globally which makes it difficult to use single-pathway
investigations to successfully predict cell responses?
1.2. TLR-4-mediated cell responses in macrophages
As shown in Figure 1.1, the distal target of TLR-4 signaling is the regulation
of gene transcription and translation. The induction of a large number of various
proteins will ultimately determine the major cell responses, though they might vary in
different cell types. In this study, we just focused on macrophages, a type of white
blood cells and a key player in the immune system, which can ingest foreign materials
by phagocytosis and present their antigens to T cells.
1.2.1. Induction of cytokines/chemokines
As an important cell response to microbial infection, the release of cytokines
and chemokines is under precise control of TLRs-stimulated signaling networks [1].
Released cytokines and chemokines can directly suppress or kill microbial pathogens
or drive recruitment and activation of additional inflammatory cells. This system has
enormous capacity for synergy and redundancy, and blocking these mediators can in
turn inhibit or attenuate the inflammatory response. For example, a highly critical
mediator appears to be TNF-a, now established as an important therapeutic target in a
range of chronic inflammatory diseases [24]. Although induced cytokine/chemokine
release is indispensable for counteracting the growth and dissemination of microbial
pathogens, overproduction of pro-inflammatory cytokines can also lead to severe
inflammatory diseases, such as sepsis syndrome, aka endotoxic shock, which is
characterized by fever, myocardial dysfunction, acute respiratory failure, hypotension,
multiple organ failure, and in a large number of cases, death. Therefore, under normal
circumstances the release of cytokines and chemokines is usually self-limiting and it
is achieved by highly concerted cooperation of all components in TLR-4-mediated
signaling pathways [2]. The essential functions of each cytokine or chemokine are
quite diverse, necessitating their control patterns are distinct as well and sometimes
even unique. The investigation of Pradervand et al. [25] has strongly supported this
possibility, but we still have a long way to fully understand these patterns and then to
benefit from that knowledge, including the rational development of drugs that can
effectively and specifically regulate the release of one or more cytokines and
chemokines and rescue patients suffering from abnormal expression of cytokines
and/or chemokines.
1.2.2. Endotoxin tolerance
Particularly, in macrophages a prior exposure to a low dose of LPS,
lipoteichoic acid (LTA), CpG or other PAMPs can induce a transient state of cell
refractoriness to subsequent LPS, LTA, CpG exposure, a phenomenon known as
"endotoxin tolerance", characterized by strongly decreased pro-inflammatory
cytokine release. Induction of tolerance is thought to be an effective way to protect the
host from cellular damage caused by hyperactivation of macrophages and other
immune cells and likely presents a means of immune cell adaptation to a persistent
microbial infection. Repression of transcription and rapid degradation of
pro-inflammatory cytokine mRNAs is observed in LPS-tolerant cells, but it doesn't
mean that tolerant cells are completely unresponsive to further LPS challenge.
LPS-tolerized cells can continue to produce several anti-inflammatory proteins, such
as secretory IL-1R antagonist (sIL-1RA) [26] and IL-10 [27]. Sometimes,
asymmetrical cross-tolerance also occurs. For instance, LPS-tolerized THP-1 cells do
not respond to LTA-induced IL-13I/TNFa production while cells tolerized with LTA
are only refractory to subsequent LTA challenge, but not LPS treatment [28]. Under
some specific conditions endotoxin-induced tolerance can even be blocked
(neutralizing antibodies to TGFS [29], okadaic acid [30], etc) or partially reversed
(pretreatment with PMA [31], etc). The strong induction of various negative
regulators and anti-inflammatory cytokines has been observed in many tolerized cells
and was indicated to play an important role in the occurrence of endotoxin tolerance
[23, 28-30]. However, they still can't well explain why some anti-inflammatory
cytokines continue being secreted and what causes the asymmetrical cross-tolerance.
Accordingly, more complicated mechanisms might exist in these processes.
1.2.3. Apoptosis
Besides cytokine/chemokine release and endotoxin tolerance, apoptosis also
plays a crucial role in some microbial infections [32]. The strategies applied by
pathogens to activate or inhibit apoptosis are probably necessary to subvert normal
host defense responses to protect them from being cleared from the host [32, 33]. A
number of pathogens have been reported to be armed with an array of virulence
determinants which interacts with key components of cell death pathways of the host
or interfere with the regulation of transcription factors monitoring cell survival [34].
The impact of the former virulence determinants is usually direct, highly specific and,
potentially reversible, especially as the major pathways of apoptosis have been well
documented. For example, Legionellapneumophila,as well as Legionella micdadei,
were shown to induce apoptosis in macrophages via the direct activation of caspase-3
independently of de novo protein synthesis, and Legionella-inducedapoptosis was
completely blocked by caspase-3 inhibitors [34]. In contrast, the latter virulence
determinants may just inhibit some key kinases in TLR-4 signaling network resulting
in alteration of protein synthesis that participates in the regulation of cell death.
Bacteria Yersinia and Bacillus anthracisare two good examples. They respectively
inject Yersinia outer protein YopJ (in Y pestis and Y pseudotuberculosis)or YopP (in Y
enterocolitica)and lethal factor (LF) into the host through a type III secretion system,
and protective antigen (PA) that binds to a cellular receptor and mediates the delivery
[35, 36]. YopJ or YopP subsequently inactivates MKKs and IKKP probably by
removing an ubiquitin or an Ubl modification off them [37, 38] while LF inactivates
MKKs by cleaving their N-terminal extensions [39]. As a result, both of them can
successfully switch cell responses to rapid apoptosis [35, 39]. Since the inhibition of
MKKs by YopJ or YopP and LF can thereby prevent the activation of several
downstream MAPKs including ERK1/2, JNK and p38 MAPK simultaneously [37, 39],
whether all these MAPKs and IKK3 contribute to Yersinia and Bacillus
anthracis-inducedapoptosis has become a new area of interest. Some previous reports
have indicated that the engagement of TLR-4 and the inhibition of p38 MAPK both
are indispensable for Yersinia-induced apoptosis [40, 41], but whether the inhibition
of IKK3 or NF-riB just strongly promotes it [41] or is also required to trigger it [42]
remains controversial and the roles of ERKl/2 and JNK in this process aren't clear
either. Particularly, the investigation using the specific inhibitor of p38 MAPK,
SB202190, to partially mimic the functions of YopJ or YopP and LF has further
revealed that IRF-3 and PKR in the TRAM-TRIF-dependent pathways (Figure 1) are
also important mediators of LPS plus SB202190-induced apoptosis in BMDMs [18].
The ablation of IRF-3 or PKR was found to strongly or even completely block LPS
plus SB202190 induced BMDMs apoptosis. Obviously, the pathways initiated by the
engagement of TLR-4 and subsequently mediated by these kinases/proteins should be
well organized to co-determine the cell death, but how exactly they do that is unclear
yet though the activation of PKR has been reported to enhance the expression of
members of TNFR family, including Fas, and pro-apoptotic members of Bcl-2 family,
including Bax and Bad, and reduce the expression of anti-apoptotic members of Bcl-2
family, including Al/Bfll [18, 43], suggesting that the alteration of protein synthesis
might be a critical cause. Moreover, we also can't exclude the possibility that the
activities of these kinases or pathways are overlapped or redundant since the increased
phosphorylation of PKR wasn't observed in some types of macrophages incubated
with LPS.
1.2.4. Challenges in TLR-4 signaling network analysis
In light of the above recapitulation, so far we have had a good understanding
on the operation of TLR-4 signaling network, but how its individual pathways
cooperate to determine its outputs is still far from being clarified. That might be, at
least partially, ascribed to the fact that most previous research just focused on how the
abnormal performance of a single pathway or a specific kinase in TLR-4 signaling
network would perturb cell responses of interest and hardly considered the impact of
dynamic couplings and potential crosstalk between activated pathways or kinases due
to the constraints of traditional biological assays. Without being comprehensive these
studies might even draw some wrong or conflicting conclusions as exemplified
previously, and thereby mislead the discovery of new drugs and the development of
new diagnostic methods. In fact, several independent signaling pathways or proteins
are usually required to accurately predict a specific cell output though it is really
possible that sometimes only one or two of them are dominant [44-47], and the
inducible autocrine, paracrine and other feedback signalings occurring in a
time-course manner could be very important as well [48]. Hence, here we also
hypothesized that an intricate working pattern might be exhibited by TLR-4 signaling
network in determining its outputs, and sought to obtain some new insights that might
be helpful to address the unsolved biological questions and applied to develop new
drugs against infectious diseases through systematic analysis and quantitative
modeling.
1.3. Data-driven approaches for systematic analysis of
biological networks
With the fast development and breakthrough of experimental techniques, the
large-scale data collection via increasingly quantitative and high-throughput assays
has become much more reliable, reproducible, accurate and easier. That also means
the quantitative and systematic analysis of biological signaling networks is technically
and economically feasible now by monitoring the kinetics and dynamics of the
activation of multiple intracellular signaling pathways, biochemical reactions or
kinases/proteins and an array of cell responses simultaneously or independently. Some
detailed quantitative assays will be introduced in the following chapter, but firstly we
may raise concern about how to deal with a huge number of produced raw data since
the data usually looks messy and are informationless till they are well reorganized,
normalized and carefully analyzed using special mathematical tools. In order to
accommodate such requirements, a spectrum of computational approaches, which
vary in their level of abstraction and specificity, ranging from detailed mechanistic
models (consisting of biochemical reactions based ordinary differential equations) to
probabilistic models (e.g., Bayesian networks and Markov chains) to statistical
models (e.g., principal component analysis, partial least squares regression, decision
tree analysis, and clustering) [49, 50], has been developed and is being applied to the
interpretation and systematic analysis of various cell signaling networks increasingly
[25, 45, 50-53]. Obviously, these approaches are not equally valid for all biological
questions, although they are all able to generate their own types of computational
models which can be useful for assembling and analyzing quantitative data and
formalize a complex biological or experimental process mathematically. The choice of
computational approaches or models is flexible and can be based on the specifics of
the biological system, including its current extent of understanding and the data
available [50]. As mentioned previously, many details of TLR-4 signaling network
remain obscure and even controversial, so some statistical approaches like PCA and
PLSR would be the most proper choice so far for systematically analyzing this
complex system since their modeling process per se is independent on the codification
of existing knowledge [45, 50].
1.3.1. Network-function modeling by partial least squares
regression
Partial least squares regression (PLSR) is a multivariate analysis technique
that generalizes and combines features from principal component analysis (PCA) and
multiple regression. Based on a lot of iterative computations that implement linear
transition from a large number of original descriptors to a new variable space called
principal component space formed by small number of orthogonal factors or latent
variables, PLSR can availably reduce the data complexity [45, 54]. Unlike some
similar approaches (e.g. PCA), latent variables in PLSR are chosen in such a way to
provide maximum correlation with dependent variables, and their information
captured by each latent variable can be estimated using resulting scores and loading
matrices or the formulated regression coefficient vector [54]. As a good PLSR model,
it not only should show a strong correlation between latent and dependent variables,
but also usually needs to have good predictivity that will decrease as model starts to
represent not just the true pattern of their relation but also random noise and
individual features of the training set. The predictivity of a constructed PLSR model
can be assessed by the parameter Q2 that is computed via cross-validation approach
and has been broadly used to select optimal number of PLS components [55, 56].
PLSR originated from economics and became popular first in chemometrics
and sensory evaluation. Since it provides a rapid analysis of large amounts of high
dimensional data, has the ability to handle "fat data", where there are more variables
than observations, and leads to stable, correct and highly predictive models that often
allows for an interpretation and a better understanding of the different sources of
variation, nowadays PLSR has been one of the most prevalent regression techniques
applied in various areas, of course, including biological sciences and biotechnology.
The application of PLSR in genetics and molecular biology, such as the analysis of
microarray gene expression data [57] and the estimation of relative genotype in cell
samples from mixed microbial populations via analysis of complex DNA sequence
electropherograms [58], has been demonstrated to be effective and accurate, and its
use in analysis of proteomic networks is boosting up as well especially when it was
confirmed that the involvement of large-scale heterogeneous data on a wide variety of
biochemical processes measured by very different experimental techniques in the
modeling is actually compatible [45, 59]. More importantly, network-function
modeling by PLSR also takes on many special advantages in systematic analysis of
diverse extracellular stimuli-triggered cellular responses, such as the amount of
secreted cytokines, or binary cellular decisions, such as death versus survival. For
example, it can extract highly informative signals that are usually helpful for the
interpretation of specific biological events [44-46], can identify potential time points
of kinases that can convey opposite messages depending on the timing and
mechanism of activation [44, 47], and can disclose the common effector processing in
a given cell type that might be applied to develop cell-specific drug therapies [47].
Obviously, PLSR isn't a technique that is just able to simply repeat and validate the
results or conclusions already addressed by traditional biological methods, but rather
an important supplement to them. Hence, we also expected that quantitative analysis
and modeling of TLR-4-mediated cell responses by PLSR would give us some new
valuable insights on macrophages in response to bacterial infection that might be
helpful for the development of safer and more effective drugs or agents for related
infectious diseases.
1.3.2. Challenges in quantitative analysis of
TLR-4-mediated cell responses
An unabridged quantitative analysis of an indicated system usually requires
several procedures, including variable selection, data collection, model construction,
information mining, and conclusion validation. Each of them can be a big challenge
and there is no common rule available for reference so far, though several biological
signaling networks were analyzed and characterized [44-46]. For example, a typical
mammalian cell consists of thousands of molecules and it is unpractical to involve all
of them in the models. Even if considering TLR-4-stimulated signaling pathways only,
there are also a large number of proteins involved and many of them are even
dependent due to significant crosstalk (Figure 1-1). Inclusion of excessive variables,
especially noisy and unimportant variables, can influence PLSR models and decrease
their predictivity notwithstanding the accuracy of description may increase [60].
Moreover, the determination of variables will tamper with the choice of methods for
and the results of data collection, model construction, information mining and
conclusion validation, and the availability and feasibility of these procedures will
inversely impact the determination of variables. Therefore, this is an integrated
process. The remaining chapters of this thesis will show up how these potential
challenges were handled, and summarize the results of this study.
CHAPTER 2
Experimental techniques for quantitative TLR-4
signaling network analysis
2.1. Introduction
All along, experimental data or outcomes are the most important part of
biological research, though some pure mathematics-based methods have already been
developed and applied in DNA/RNA sequence and genome analysis. Different
experimental techniques have their own features, pros and cons, and also scopes of
application in data collection, thereby the selection of approximate experimental
assays is a crucial step leading to robust, reproducible and valid results.
The objective of this thesis was to quantitatively address how TLR-4
stimulated individual intracellular signaling pathways cooperate to determine various
cell responses in murine macrophages. To do this, we treated RAW264.7 cells, a
murine macrophage-like cell line, purchased from ATCC with LPS to specifically
activate TLR-4 signaling cascades, which were then selectively perturbed by a
specific inhibitor of p38 MAPK, SB202190. This system has broadly been used in
previous studies, and has well-understood biological significance [18, 25, 39, 61, 62],
and also is easy to handle. Obviously, treatment with different combinations of LPS
and SB202190 will generate different kinetics of cellular events, including
phosphorylation and degradation of kinases or proteins, the amounts of cytokines
released into media, cell death (e.g. apoptosis), and so on. We have eventually
developed and/or optimized a couple of experimental conditions and approaches in
RAW264.7 cells as discussed in the remaining Chapter 2 to fast, effectively and
accurately quantify these inducible cellular events that were subsequently involved in
systematical analysis by PLSR modeling.
2.2. Repurification of commerciai LPS
LPSs are characteristic components of the cell wall of Grath negative bacteria.
They are localized in the outer leaflet of the membrane and contribute to the integrity
of the outer membrane and as well protect the cell against the action of bile salts and
lipophilic antibiotics [63]. LPS consists of a glucosamine-based phospholipid, lipid A,
which is responsible for the endotoxic properties of the molecule, a hydrophilic core
polysaccharide chain, and a hydrophilic O-antigenic polysaccharide side chain usually
made up of repeating units of oligosaccharides, which determine the serological O
specificity of respective strains based on the sugar constituents, their sequence and
their mode of linkage [65]. Bacteria with different serotypes might generate distinct
antibody responses in vivo, but most of their LPSs can bind to TLR-4 associated with
MD-2 [64] and initiate innate immunity. In this study, we utilized LPSs extracted from
E. coli serotype 055:B5 as many other groups did [18, 61].
There are a lot of vendors providing commercial LPS powders. For instance,
we purchased our LPS 055:B5 from Sigma-Aldrich. However, it is noticeable that
investigators have documented historically that established protocols for isolating
LPS result in the copurification of varying amount of other endotoxin proteins, and
these contaminants in commercial LPS preparations possess extremely potent
bioactivity and might account for the discrepancy concerning whether TLR-2 also
participated in LPS signaling [66-73]. Therefore, in order to avoid the potential effects
on our data of other endotoxins, we also repurified our LPS 055:B5 before they were
applied to treat cells. The repurification method was made up of several step-by-step
procedures, briefly including dissolution in endotoxin-free water with 0.2%
triethylamine (TEA) and 0.5% deoxycholate (DOC), extraction with equal-volume
water-saturated phenol, precipitation in 75% ethanol and 30 mM sodium acetate
solution at -20 0 C for 1 h, and then air-dry at 40 C after precipitates were washed with
cold 100% ethanol. The recovery was almost one hundred percent, and this method
has been demonstrated to effectively eliminate signaling through TLR-2 without loss
of TLR-4 activation [73]. On the other hand, since LPS can bind to plastics and
certain types of glass and adsorption to the sides of the vial is negligible only when
the concentration is > lmg/ml, we finally resuspended our phenol re-extracted LPS in
endotoxin-free 0.2% TEA at 5 mg/ml stock concentration and stored the aliquots at
-200 C for long-term use (stable up to 2 years) or at 40 C for up to one month.
2.3. Optimization of protocols for the measurement of
various cellular events
As mentioned previously, in response to LPS insults, macrophages will
trigger a discrete series of cellular events in tandem, including sensing of the insult
and signaling by phosphorylation cascades of kinases or proteins, synthesis and
release of cytokines and chemokines, and initiation of suicide program. All of these
cellular events were what we were interested in and going to measure and analyze in
this study.
2.3.1. Measurement of apoptosis
The balance of cell proliferation and cell death is a pivotal mechanism for
maintaining homeostasis in metazoans. Deregulation of cell death is the crucial cause
of many severe diseases (e.g. cancer, AIDS and Alzheimer's disease) [74]. Up to now,
three major morphologies of cell death have been defined: apoptosis (type I), cell
death associated with autophagy (type II) and necrosis (type III) [75]. Apoptosis may
be the most important one of them since it is a highly regulated and programmed
process and can occur under normal physiological and certain pathological conditions,
while a special or extreme circumstance is usually required for occurrence of the other
two types. Apoptosis can be distinguished from the other two types of cell death by its
characteristic morphological and biochemical changes, including cell shrinkage,
membrane blebbing, chromatin condensation, DNA fragmentation and a proteolytic
cleavage cascade by caspases [75-77], and these features provide ideal hallmarks by
which to identify apoptotic cells in a population and quantify their amount and/or
percentage.
Double staining is one of the most popular assays employed in apoptosis
analysis. It is a flow cytometry-based method and can quickly distinguish and count
cells bearing different fluorescence emission spectra individually and automatically,
which are subsequently divided into four groups representing four different biological
conditions of cells. The interpretation of the status relies on what kind of paired dyes
are used. In this study, we used two pairs of dyes, 7-Amino Actinomycin D (7-AAD)
vs. Annexin V-PE and anti-active caspase-3-Alexa 488 vs. anti-cleaved
poly(ADP-ribose) polymerase (PARP)-Alexa 647. Since 7-AAD is a cell impermeant
dye binding to DNA, only dead cells can be stained because of loss in plasma
membrane integrity. In contrast, Annexin V-PE recognizes phosphatidylserine which
resides on the inner cell membrane of healthy cells but will be externalized to the
outer membrane layer when in the early stage of apoptosis. Hence, the assay of double
staining consisting of 7-AAD and Annexin V-FITC requires unfixed cells and has the
capacity to measure the percentage of live cells, early apoptotic cells, late apoptotic,
and dead cells. On the other hand, as the name implies, double staining consisting of
anti-active caspase-3-Alexa 488 and anti-cleaved PARP-Alexa 647 will determine the
percentage of apoptotic cells only in which caspase-3 is activated and cleaves its
substrate, PARP [77, 78].
For 7-AAD & Annexin V-PE double staining analysis, we used Annexin
V-PE apoptosis detection kit I purchased from BD PharMingenTM and strictly
followed the instruction provided by the manufacturer to prepare the samples. And for
anti-active caspase-3-Alexa 488 & anti-cleaved PARP-Alexa 647 double staining
analysis, we purchased primary antibodies, purified rabbit anti-active caspase-3 and
purified mouse anti-cleaved PARP (Asp214), from BD PharMingenTM and secondary
antibodies, Alexa 488-linked goat anti-rabbit IgG (H+L) and Alexa 647-linked
anti-mouse IgGi (yl), from Invitrogen TM separately and modified the protocol that
previously developed to understand the TNF-growth factor-mediated cell death
decision [52] to properly label the samples collected from 24-well plates. The process
included the optimization of working dilutions of all antibodies and washing times.
Additionally, the parameter setting of flow cytometry machines was also optimized
using unlabeled cells and single dye or antibody labeled cells under the supervision of
Glenn Paradis at MIT flow cytometry core facility. As shown in Figure 2-1, after
I
;--~~
~.~iiiiiii~~~~i-iii'i-iil'~i-iiiiii-ii~
optimization both double staining assays can be used to measure apoptosis induced in
RAW264.7 cells. The percentages of apoptosis induced by 1 ng/ml LPS plus 75 pM
SB202190 are respectively 31.80% for 7-AAD &Annexin V-PE double staining
(Regions II +IV) and 27.47% for anti-active caspase-3-Alexa 488 &anti-cleaved
PARP-Alexa 647 double staining (Regions I+II+IV) compared to control (6.17% vs.
a
LPS & SB202190
Control
V_
3.18%
2.74%
0-
I
II
r
·
B-
2.99%
•"
0-
IV
0-
lqA
100
10 ]
10 1
L.
1
10
10
PE ANNEXIN V
;
1
104'
103
b
lot
1.24%
0.37%
SII
L-
1.32%
o
01
0.
Iwvwý
jW
W
1Wv·
WWI
-
10'
10V
S
10
--
I
10"
w
10
1ALEXA 488
4
14
Figure 2-1. The measurement of apoptosis by two double staining assays
RAW264.7 cells were pretreated with either 75 pM SB202190 or a drug vehicle (DMSO, Control)
for 30 min and then stimulated with either 1 ng/ml LPS or carrier (0.2% TEA, Control) only. Cells
were collected 8 hours after LPS addition, stained for either active caspase-3 (Alexa 488) and
cleaved PARP (Alexa 647) (a) or 7-AAD and Annexin V-PE (b), and analyzed by flow cytometry.
The numbers are the percentage of cells in corresponding regions I-IV calculated by the analysis
software.
Control
4 hours
8 hours
18 hours
FSC-H
PEANNEXIN V
Figure 2-2. Effect of LPS treatment on RAW264.7 cell size
RAW264.7 cells were treated with either 0.2% TEA (Control) or 100 ng/ml LPS for 4 hours, 8
hours or 18 hours. Cells were collected at indicated time points, stained for 7-AAD and Annexin
V-PE and analyzed by flow cytometry. Their generated SSC-H vs. FSC-H and 7-AAD vs. Annexin
V-PE plots were aligned together for easy comparison.
~~
-------------------~----------------~.~
;..~.,~~ ;;;;;~;;-·;;~
It
Control
T-
R1
!
cn
0T0-
oo
)
0
S
0
200
40000
FSC-H
400
600
100
1
800
1000
FSC-H
Figure 2-3. SB202190 has no effect on RAW264.7 cell size
RAW264.7 cells were treated with a drug vehicle (DMSO, Control) or 10 piM SB202190. After 18
hours, cells were collected, stained for 7-AAD and Annexin V-PE and analyzed by flow
cytometry.
2.93%). Moreover, 7-AAD & Annexin V-PE double staining also showed that there
were few dead cells in control (2.74%) or LPS plus SB202190-treated cells (5.09%)
(Region I).
Besides optimization of assays, another important factor that should be
seriously considered is the effect of variation of cell condition on the assay accuracy.
Although we had tried to keep our RAW264.7 cells consistent all along by purchasing
new cells from ATCC, using the same culture conditions, and discarding cells that had
already been passaged more than 20 times, agent-induced changes of cell properties
sometimes are inevitable once the treatment conditions are selected. For example, we
observed that treatment with 100 ng/ml LPS for 18 hours markedly increased
RAW264.7 cell size that was correlated with increase of forward light scatter (FSC-H)
. . .
. . . .. ..... .. .. . . .......
• m /~ i¸ T........ •........ . . . . ..
Tl i i'
•
.. ....
.
.
....
..
values (Figure 2-2). Saxena et al. have suggested it might be caused by LPS-induced
differentiation of RAW264.7 cells into dendritic like cells in a dose-dependent manner
[79]. Such explanation also makes sense in our case as incubation with SB202190 up
to 18 hours or 100 ng/ml LPS up to 8 hours didn't induce significantly increased cell
size of RAW264.7 cells (Figure 2-2 & Figure 2-3). Increased cell size will probably
a
b
07T
60-
-a-
100 pM SB202190
-A-
50 pM SB202190
--- 100 pM SB202190
75 pM SB202190
60-4 --A- 50 pM SB202190
50 -
50-
-4
40 -40
L.3030
S20-
20-
10-
10-t
0
0-
*
*
Ctrl 10o
C
·
Ctrl 100 0
0.1 0.5 1 5 10 20 50 100
LPS (nglml)
0
I
·
I
I
-
I
·
I
·
0.1 0.5 1 5 10 20 50 100
LPS (nglml)
L.U
1.8
+, ý
--
1.6----e
-A
*--
1.4-
I
1.2 -
.
0
100 pM SB20,
75 pM SB2021
50 jM SB2021
100 ng/ml LPS
Control
4,
0.1
0.5
1
,
5
10
.fI
20
I
50
100
LPS (ng/mi)
Figure 2-4. Death profiles of RAW264.7 cells in response to increasing doses of SB202190
and LPS in combination
RAW264.7 cells were treated with either a drug vehicle (DMSO, Control) or indicated doses of
SB202190 and/or LPS. Particularly, DMSO and SB202190 were added 30 minutes prior to LPS. 8
hours after LPS addition, cells were collected, stained for either 7-AAD and Annexin V-PE (a) or
active caspase-3 (Alexa 488) (b), and analyzed by flow cytometry (a & b), or subjected to do MTT
assay and subsequently the absorbance at the wavelength of 550 nm was measured (c). Values are
plotted as the mean of three biological replicates in (b) or of three technical replicates in (c) +
s.e.m
. . . ...-----.............
.... ...
absorb more dyes and may interfere with the measurement of apoptosis by double
staining assays (e.g. a little bit shifted to the right in Figure 2-2). Therefore, in order to
eliminate the effects of altered cell size, no more than 8-hour incubation time and 100
ng/ml LPS were used in our following experiments.
Subsequently, we compared the difference of these two flow cytometry-based
assays in measurement of apoptosis induced by multiple combinations of
SB202190and LPS. Both of them confirmed that up to 100 ng/ml LPS alone could not
significantly enhance the occurrence of apoptosis in RAW264.7 cells compared to
control and showed similar profiles indicating that the induction of apoptosis was both
LPS- and SB202190-dose dependent before the plateau was reached in which the
concentrations of LPS were more than 10 ng/ml (Figure 2-4a & b). However, the
percentages computed by these two techniques for the same treatment were quite
different. The values given by the 7-AAD &Annexin V-PE assay were much higher
especially when the fraction of apoptotic cells was low (Figure 2-4a &b). It might be
due to the fact that Annexin V-PE can detect PS-externalized early apoptotic cells
while the activation of caspase-3 occurs late, but we can't exclude the possibility that
other changes in cells may also contribute to this discrepancy. For instance, we
observed that LPS and SB202190 were able to synergistically induce the formation of
vacuoles in RAW264.7 cells as Hassan et al. reported [62]. The number and size of
vacuoles increased in a LPS- and SB202190-dose dependent manner, and these
induced vacuoles might interfere with the uptake or binding of 7-AAD and Annexin
V-PE since unfixed cells are employed in 7-AAD and Annexin V-PE double staining
assay. Additionally, Figure 2-4a &b also showed anti-active caspase-3-Alexa 488 and
anti-cleaved PARP-Alexa 647 double staining assay was more sensitive to small
changes of apoptosis. In light of all these concerns, we thought that double staining
consisting of anti-active caspase-3-Alexa 488 and anti-cleaved PARP-Alexa 647 was
more suitable to accurately quantify apoptosis stimulated by LPS plus SB202190 in
RAW264.7 cells.
As shown in Figure 2-4b, when the concentration of LPS was more than 10
ng/ml, LPS plus SB202190-induced apoptosis reached a plateau. Interestingly, it
occurred at all three tested doses of SB202190, suggesting that TLR-4-initiated
signaling might have been saturated by >10 ng/ml LPS. This is supported by our
observation of unchanged phosphorylation levels of several key kinases in TLR-4
signaling network (Figure 2-7a). Another interesting thing was that >10 ng/ml LPS
plus 75 .M SB202190-induced apoptosis was a little bit higher than that induced by
the same concentrations of LPS plus 100 pM SB202190 (Figure 2-4b). Since many
chemicals at different dosage ranges can mainly induce either apoptosis (low doses)
or necrosis (high doses) [75, 80], we speculated that necrosis was induced by high
doses of LPS plus 100 pM SB202190 and thereby reduced the induction of apoptosis.
If true, we would see total cell viability that can be easily determined by a developed
MTT assay [81], in LPS plus 100 p.M SB202190-treated cells is lower than that of
LPS plus 75 ptM SB202190-treated cells. As expected, the reduction of total viable
cells was SB202190-dose dependent in the whole dosage range of LPS (Figure 2-4c).
These results also indicated that >5 ng/ml LPS and >75 pM SB202190 might not be a
good choice for quantitative analysis of TLR-4-mediated cell responses.
2.3.2. Quantification of cytokine release by ELISA
Even as it is common for a single cytokine to act on several different cell
types, a specific cell type usually secretes several different cytokines concurrently in
response to a simple stimulus. Cytokines that can be secreted by macrophages include
G-CSF, IL-lo, IL-6, IL-8, IL-10, IL-12, MIP-la, RANTES, TNF-a, and so on. They
are probably regulated by distinct signaling patterns [25]. In this study, we were really
interested in three of them: IL-6, which plays an important role in triggering the acute
phase response of the body to injury or inflammation [82], IL-12, which mediates
enhancement of the cytotoxic activity of NK cells and CD8+ cytotoxic T lymphocytes
[83], and TNF-a, a pleiotropic cytokine that regulates a broad range of biological
activities, including cell differentiation, proliferation and death, as well as
inflammation, innate and adaptive immune responses, and tissue development [84].
Since many cytokines must be produced de novo after stimulation and then
.....
...
..
.......
..
...
released into the cell culture media, they can be easily monitored by using
enzyme-linked immunosorbent assay (ELISA). There are many types of ELISA
already developed for different detection aims and conditions. In this study we
selected sandwich ELISA that is widely used to detect various sample antigens
including cytokines. A general schematic of the protocol that we used is illustrated in
Figure 2-5. The paired unlabeled and biotinylated antibodies were all purchased from
BD PharMingen TM , and they were respectively purified rat anti-mouse IL-6 (clone:
MP5-20F3) vs. biotin rat anti-mouse IL-6 (clone: MP5-32C11), which detect mouse
IL-6 proteins, purified rat anti-mouse IL-12 p40/p70 (clone: C15.6) vs. Biotin rat
anti-mouse IL-12 p40/p70 (clone: C17.8), which reacts with both free and complexed
Cell culture media containing
various substances
.Biotin-conjugated
i
InfiMl
Unlabeled a
Diluted with 0.1M
Na2HPO 4 pH 9.0
Diluted with PB!
}LibJ·
Washing
Blocking with PBS-B At 370 C for I hour At RT for 45 min
Extravadin-Peroxidase
Washing
at 370 C for 1 hour
Coating covernight
at 40C
H20 2 1:1 TMB
_
Blue
Clear--Afl
lnA
N
Diluted with PBS-B/
I
Washing
1M PA
#4,1
Measure atbsorption
at 450 nm
At 37PC for 10 min
At RT for 30 min
Figure 2-5. General schematic of the sandwich ELISA protocol
The washing buffer was PBS-T (PBS + 0.05% Tween 20). PBS-B: PBS + 1%BSA; B, biotin; RT:
room temperature; PA: phosphoric acid.
.I
...........
..
..
............
-000ý ý
I
80-
40
1
60-
E
30
E
C)
0)
Q3.
(9 40-
20
0.
-J
U-
-J
I--
10
20-
0-
huh
LPS
,
.
-
0
,
+
LPS
0
0.5
-
LPS
+
1
5
20
-
100
SLPS (ng/ml)
Figure 2-6. The release of IL-6, IL-12 and TNF-a in LPS-stimulated RAW264.7 cells
(a) RAW264.7 cells were treated with either 0.2% TEA (-) or 5 ng/ml LPS (+). After 2 hours, the
amounts of IL-6, IL-12 and TNF-a in their media were analyzed by ELISA. (b) RAW264.7 cells
were pretreated with either a drug vehicle (DMSO) or SB202190 at 10 or 75 pM for 30 min before
indicated concentrations of LPS were added. Then the amounts of IL-6 in their media were
analyzed by ELISA another 2 hours after LPS addition. Values are plotted as the mean of three
biological replicates ± s.e.m.
p40 and p70 subunits of IL-12, and purified hamster ant-mouse/rat TNF (clone:
TN3-19.12) vs. Biotin rabbit anti-mouse/rat TNF (clone: C1150-14), which are useful
for measuring both free TNF-a as well as TNF-a bound by soluble TNF receptor. The
optimal concentrations of these antibodies have been determined previously in our lab
by titration and were also adopted here. Subsequently, we used standard curve-based
method to quantitate the absolute amounts of secreted IL-6, IL-12 and TNF-a in the
media. The standard curve was generated for each individual plate using serially
diluted purified recombinant mouse IL-6, IL-12 or TNF-a of known concentrations
which should encompass the levels in the experimental samples and stay within the
linear range. If necessary, the experimental samples might be diluted as well.
Our measurements using developed sandwich ELISA showed that LPS can
strongly induce the release of IL-6 and TNF-a in RAW264.7 cells (Figure 2-6). The
consistent results were also reported previously in various types of murine
macrophages [12, 29, 85, 86]. By contrast, we didn't see that IL-12 was significantly
secreted by LPS-stimulated RAW264.7 cells (Figure 2-6a). Saito et al. also observed
the same phenomena and further indicated that it was caused by activation of
repressor element GA-12 through hyperactivation of the ERK pathway in RAW264.7
cells [86]. From another point of view RAW264.7 cells might have some special
properties (e.g. high level of ERK) distinguished from other murine macrophages that
can alter its response to the same insult. On the other hand, as we saw in Figure 2-4b,
>10 ng/ml LPS also induced a plateau of IL-6 secretion (Figure 2-6b), recapitulating
the possibility that TLR-4-initiated signaling transduction was saturated.
2.3.3. Quantitative western blots measurements
Western blot is a mature technique to detect a specific protein in a given
sample of tissue homogenate or extract. Nowadays, it has become one of the most
popular assays utilized in biological research particularly as a huge number of various
antibodies are provided commercially here and there. In this study, we employed
western blot to indirectly monitor the activation of several kinases in TLR-4 signaling
network by examining the phosphorylation of serine (Ser), threonine (Thr), and/or
tyrosine (Tyr) residuals at their specific sites. The phospho-specific antibodies that we
used included anti-phospho-p38 MAPK(Thrl 80/Tyr182), anti-phospho-JNK(Thrl 83/
Tyr185), anti-phospho-ERK1/2(Thr202/Tyr204), anti-phospho-MK2(Thr334),
anti-phospho-IKKa/P(Ser176/180), anti-phospho-Akt(Ser473), anti-phospho-
1ý
elF2a(Ser51), anti-phospho-c-Jun(Ser73), and anti-phospho-ATF-2(Thr69/7 1), and
the phosphorylation of indicated residuals in the parentheses has been demonstrated to
be indispensable for full activities of corresponding kinases in response to LPS insults
[8, 13, 22, 23, 41]. As shown in Figure 2-7a, LPS actually induced dose-dependent
phosphorylation of p38 MAPK, JNK and MK2 and ERK1/2 but reaching a plateau
after 10 ng/ml. This result strongly supported our hypothesis that saturation of TLR-4
signaling transduction might be achieved and account for the plateaus observed in the
measurements of apoptosis (Figure 2-4b) and IL-6 secretion (Figure 2-6b). In addition,
we also examined the effects of SB202190 and/or LPS treatment on the expression of
total p38 MAPK and JNK proteins and saw there was no significant change (Figure
2-7b).
a
LPS (ng/ml)
-
0.1
1
0.5
5
20 100
p-p38 MAPK
p-MK2
p-JNK2
p-JNK1
------
p-ERK1
p-ERK2
b
+
SB202190
LPS
-
+
+
+
Total p38 MAPK
Total JNK2
Total JNK1
Figure 2-7. Phosphorylation of kinases in response to LPS insults in RAW264.7 cells
(a) RAW264.7 cells were treated with either 0.2% TEA (-) or increasing concentrations of LPS. (b)
RAW264.7 cells were pretreated with either a drug vehicle (DMSO, -) or 10 WiM SB202190 for 30
min before either 0.2% TEA (-) or 100 ng/ml LPS was added subsequently. Lysates were all
collected 30 min after LPS addition, and the indicated proteins were measured by western blot. All
the lanes were loaded with the same amount of samples, 37.5 lag.
.iiiii_~illl*yl~·l····-X
1~~1___~___;~~~,.r.*.l.l;~·;~111·--~;~~1~.....
~~___llil~!liiiiiiii·ii···~··~~lllll
phospho-IKKa/p
120000-
R'= 0.98
120000-
•
"
R= 0.9•9
90000-
IA
6000030000-
.
W
n I . . .
4
2
.
6
Total proteins (pg)
.
.
18
20
.
.
8 10 12 14 16
Total proteins (pg)
22
C
z
Total proteins (pg)
Total proteins (pg)
phospho-ERK1
1200000-
2=
R
0.99
2 = 0.
9000gO-
60004
z
3000
.
20- .0
.
.
.
1.0
0.5
0
.
!
!
2.0
1-5
"
|
2.5
Total proteins (pg)
Total proteins (pg)
phosho-MK2 (Upstate)
120004
phosho'MK21Upsta•!
F = 0.98
9000402
R = 0.99
6000
z
a
3000 0o
n.
5
20
15
10
Total proteins (Plg)
25
0
5
20
15
10
Total proteins (pig)
25
---- _ _ -___
........
- 111~ ~ ~ ~
_ ---.___
phosho-c-Jun
400000
e= 00999
300000-
300000
2000001
.s 200000
z
100000.
100000
R2 0.99
||
E
0
0
5
10
20
15
Total proteins (pg)
25
30
0
•
I
5
"
I
"
I
20
15
10
Total proteins (ig)
25
Figure 2-8.The linear ranges of various proteins for quantitative western blot analysis
SD1 were loaded in each gel lane at the range of 0-30 glg or SD2 were loaded in each gel lane at
the range of 0-50 jig and then resolved in 10% SDS-PAGE. The gels were subsequently examined
phospho-Akt, phospho-ERK1/2, phospho-MK2, phosphofor phospho-JNK2, phospho-IKKa/l,
ATF-2, phospho-c-Jun (SDI) and phospho-elF2a (SD2) using quantitative western blot. Finally,
the net intensity of bands was plotted versus the amount of total loaded proteins when the point
2
fell into the range in which the R value of linear regression was no less than 0.95. For each
protein, at least two independent experiments (black and red) were performed. The detailed
dilution ratios of antibodies and identified linear ranges were summarized in Table 2-1.
Table 2-1. The dilution ratios of antibodies and the linear ranges of various proteins for
quantitative western blot analysis
Linear range
Second antibody
(Upstate: 12-348)
(1tg)
N/A
1:5000
1:1000 (CST: 9215)
p-p38 MAPK (Thr180/Tyr182)
2.5-20
1:5000
1:2000 (CST: 9251)
p-JNK2 (Thrl83/ Tyr185)
2.5-20
1:5000
1:2000 (CST: 2697)
p-IKKa/P (Serl76/180)
2.5-20
1:5000
1:2000 (CST: 9271)
p-Akt (Ser473)
5-50
1:5000
1:2000 (CST: 3597)
p-eIF2a (Ser51)
0.25-2.5
1:10000
1:5000 (CST: 9101)
p-ERK1 (Thr202/Tyr204)
0.25-2.0
1:5000
1:5000 (CST: 9101)
p-ERK2 (Thr202/Tyr204)
2.5-25
1:5000
1:4000 (Upstate: 7-851)
p-MK2 (Thr334)
2.5-25
1:5000
1:4000 (CST: 3007)
p-MK2 (Thr334)
2.5-30
1:5000
1:3000 (CST: 9225)
p-ATF-2 (Thr69/71)
2.5-25
1:5000
1:3000 (CST: 9164)
p-c-Jun (Ser73)
Note: p, phosphorylated; CST, Cell Signaling Technology, Inc.; Upstate, Upstate Biotechnology, Inc.;
Protein name
Primay antibody
However, traditional western blot is only qualitative or semi-quantitative, and
~I
its accuracy isn't good enough for quantitative analysis. Hence, we have developed a
new protocol for quantitative western blots measurements, including using Amersham
ECLTM advance western blotting detection kit, which has much higher sensitivity (1 pg in model system), and directly developing and quantitating the blots using Kodak
Image Station 1000 with Kodak ID software instead of burning medical X-ray films
in a dark room. Particularly, before the protocol was applied to measure the proteins
in unknown samples, we also needed to figure out the linear ranges for these proteins
so that we could rationally determine the proper amounts to be loaded since
over-loaded samples might produce saturated signals and it would become impossible
to compute the amount of specific protein based on its linear relationship with
detected signal intensities [45]. In order to do that, we prepared two different standard
samples. Standard sample 1 (SD1) was the lysate of RAW264.7 cells treated with I
gig/ml LPS for 30 min and standard sample 2 (SD2) was the lysate of RAW264.7 cells
treated with 5 gLg/ml tunicamycin from Streptomyces sp. for 6 hours [87]. Then, SD2
was utilized to determine the linear range of phospho-eIF2a (Ser51) while SDI for
the others as detailedly described in Figure 2-8. The acquired linear ranges were also
summarized in Table 2-1. Interestingly, as shown in Figure 2-8 two independent
experiments performed on the same range of standard samples might produce
different regressed linear relationships (slope and intercept), but their linear ranges
appeared to be quite consistent and reproducible, confirming that it is feasible using
identified linear ranges to monitor loading amounts of unknown samples. For example,
we always ran a standard sample at upper-limit with unknown samples in one gel, and
if the generated signals of unknown samples were stronger than that of the upper-limit,
we reduced their loading amounts until all of them fell into the linear range. However,
it doesn't mean small loading amounts are recommended here since low signals will
introduce high experimental errors. Accordingly, we had adjusted the loading amounts
of unknown samples in each gel lane individually, if possible, to obtain proper signal
intensities and thereby improve the accuracy of measurements. Furthermore, since the
linear relationships (slope and intercept) might vary from gel to gel (Figure 2-8), we
had to determine the linear relationship for each individual gel. That required at least
two standard samples at different amounts. Therefore, besides the upper-limit one, we
also ran another standard sample at lower amount in each gel, and subsequently used
the linear relationship produced by these two standard samples to calculate the
relative amounts of unknown samples in the same gel to standard samples, which
were then normalized to a fixed amount of standard sample and converted into
non-dimensional data so that the inter-gel results became comparable finally (Figure
2-9).
3.0-
10000 0-
2.5-
8000 0-
--
='
2.0
6000 0-
1.54000 0
1.02000
0.5
0.0
i
2 3
4 5
6 7
8 9 10 11 12 13 14 15
Lane
1 2 3 4 5 6 7 8 9 10 11 12 13
Lane
Figure 2-9. Phosphorylation of JNK2 in response to multiple combinations of LPS and
SB202190 measured by quantitative western blot
The lysates of RAW264.7 cells treated with a drug vehicle (DMSO) plus 0.2% TEA (lane 1,
12.5 jlg), DMSO plus 0.3 ng/ml LPS (lane 2, 12.5 jig), DMSO plus 1ng/ml LPS (lane 3, 12.5 jig),
DMSO plus 5 ng/ml LPS (lane 4, 12.5 jig), 10 pLM SB202190 plus 0.3 ng/ml LPS (lane 5, 12.5
jig), 10 pIM SB202190 plus 1 ng/ml LPS (lane 6, 10 jig), 10 pM SB202190 plus 5 ng/ml LPS
(lane 7, 7.5 jig), 50 jiM SB202190 plus 0.3 ng/ml LPS (lane 8, 10 pig), 50 PiM SB202190 plus I
ng/ml LPS (lane 9, 7.5 jig), 50 piM SB202190 plus 5 ng/ml LPS (lane 10, 10 gig), 75 pM
SB202190 plus 0.3 ng/ml LPS (lane 11, 10 jig), 75 IpM SB202190 plus 1 ng/ml LPS (lane 12, 7.5
jig), 75 piM SB202190 plus 5 ng/ml LPS (lane 12, 7.5 jig), and SD1 (lane 14, 5 tig; lane 15, 20 jig)
were fractioned by 10% SDS-PAGE, blotted for phospho-JNK2 (Thrl83/ Tyr185), and analyzed
by quantitative western blot. The raw data were plotted in left panel. Then, the relative amounts of
lane 1-13 to SDI were calculated based on the linear relationship produced by lane 14 and lane 15,
and normalized to 10 tig SD1. The normalized data were then re-plotted in right panel.
Si
CHAPTER 3
Quantitative analysis and data-driven modeling of
TLR-4 signaling-mediated cell responses
3.1. Introduction
Signal
Cue
I
I
LPS wlo SB202190
I
I
Cell
ysate
I
I
Cell lysates
I
I
(15min. 30min. 1h, 2h. 4h, 6h)i
I
I
p ~mI
AX
=r
C
I
I
I
32(p
Supernatants
(2h, 4h)
o
I
Quantitative WB
;(Phosphorylation of kinases)
Y)
(1)
(4)
I
I
I
I
I
y
Y
!
Y
Y
Y
Y
Sandwich ELISA
(Secretion of Cytokines)
Double staining assay
(Apoptosis measurement)
Response
Figure 3-1. Schematic of data collection plan
2
5
RAW264.7 cells were plated at 2x10 cells/cm and cultured for 24 hours before various
combinations of SB202190 and LPS were added. (Note: all small-molecule inhibitors were always
added 30 min prior to LPS in this study, unless indicated otherwise.) At indicated time points after
LPS was added, methanol-fixed cells, supernatants and cell lysates were prepared or collected and
then respectively analyzed by anti-active caspase-3-Alexa 488 & anti-cleaved PARP-Alexa 647
double staining assay, ELISA, and quantitative western blot (WB). In particular, the datasets
constructed by quantitative WB and by double staining assay and ELISA were then respectively
named as "Signal" block and "Response" block while the dataset consisting of various treatments
was called as "Cue" block.
In the previous chapter, we successfully developed and optimized several
assays that enabled us to quantitatively monitor and measure signaling transduction
~;;~_~;;
--
;::1_1.:.1~~1~.-.....~
~II1 -;~~~i~~~
and different cell responses in RAW264.7 cells incubated with LPS and SB202190. In
particular, these preliminary experiments also provided valuable references for the
subsequent determination of proper treatment concentrations and time points. Based
on these information, we then formalized our data collection plan as shown in Figure
3-1. However, generated large datasets do not directly equate to biological
understanding [88]. Hence, in this chapter we also described how we used PLSR
algorithm to reduce data complexity, build up data-driven models and discover new
mechanisms for how TLR-4 signaling network determines its outputs.
3.2. Directed data collection and preliminary data
analysis
As shown in Figure 3-1, the first step we needed to do as the start of large
scale data collection was the determination of various cues or treatments, since these
inputs directly worked on our following measurements and also decided how many
valuable information the collected data would carry about. Generally, each cue should
be able to provoke significant changes of cell status. Based on this principle, we had
picked up 13 different combinations of LPS and SB202190 as listed in Table 3-1 as
our cues to stimulate our system of RAW264.7 cells.
Table 3-1. The list of 13 cues
1
2
3
4
5
6
7
8
9
10
11
12
13
0
0.3
0.3
0.3
0.3
1
1
1
1
5
5
5
5
0
0
10
50
75
0
10
50
75
0
10
50
75
Note: The red numbers were the doses of LPS (pg/ml), and the blue numbers were the doses of
SB202190 (pM). All the cues contained the same amounts of drug vehicles (0.2% TEA and DMSO).
DMSO and SB202190 were added 30 min prior to 0.2% TEA and LPS.
3.2.1. The critical role of p38 MAPK in LPS-stimulated
cell responses
14(
12(
10(
86
6(
41
2(
0.3
1.0
5.0
0.3
80-
1.0
5.0
10, 50, 75 pM SB)
(ng/ml LPS)
4 hours
7060E 50o2 hours
40-
TI
-r
U. 3020-
rnnr nnri n1,r
10-
C
1.0
0.3
n
5.0
Lb
0.3
4
T
I
rln
1.0
FK~
1
5.0
(0,10, 50, 75
(ng/ml LPS)
pM SB)
Figure 3-2. Inhibition of p38 MAPK strongly reduced the secretion of IL-6 and TNF-a in
LPS-stimulated RAW264.7 cells
RAW264.7 cells were treated with 13 cues listed in Table 3-1. Their supernatants were collected 2
hours or 4 hours after LPS was added, and the amounts of IL-6 and TNF-a in these supernatants
were then determined by ELISA. Values are plotted as the mean of three biological replicates ±
s.e.m.
LA
0o
0.
<0.
Ctrl
-
0
--------------
10
----
50
-----
75
(0.3,1, 5 nglml LPS)
(pM SB202190)
Figure 3-3. High doses of SB202190 were required for induction of apoptosis synergically
with LPS in RAW264.7 cells
RAW264.7 cells were treated with 13 cues listed in Table 3-1. 4 hour or 8 hours after LPS addition,
cells were collected, stained for active caspase-3 (Alexa 488) and cleaved PARP (Alexa 647) and
analyzed by flow cytometry. Values are plotted as the mean of three biological replicates + s.e.m.
First, we measured how RAW264.7 cells responded to the above selected 13
treatments and further analyzed the impact of each individual component, LPS or
SB202190, on these cell responses. As shown in Figure 3-2, LPS strongly induced the
secretion of both IL-6 and TNF-at within 4 hours in a dose-dependent manner, but the
amount of TNF-a was shown to be much higher than that of IL-6. Particularly, the
amount of LL-6 was low at 2 hours even if 5 ng/ml LPS was used and sharply
increased after 4 hours, suggesting that IL-6 was a late-response cytokine compared to
TNF-a. This result went along with the previous report that TNF-a regulated IL-6
production in normal B cells via an autocrine pathway [89]. On the other hand,
LPS-induced secretion of IL-6 and TNF-a also appeared to have different sensitivity
to the inhibition of p38 MAPK by SB202190 (Figure 3-2). For example, 10 iM
SB202190 inhibited > 80% secretion of IL-6 but only <50% secretion of TNF-ca at 4
hours, and the secretion of IL-6 was completely blocked by 50 pM SB202190
whereas SB202190 up to 75 jM was unable to completely block TNF-a release. All
of these implied that LPS-induced secretion of IL-6 and TNF-a should be controlled
by different signaling patterns notwithstanding activation of p38 MAPK held the
balance for both of them.
Unlike the secretion of IL-6 and TNF-a, SB202190, by contrast, played a
positive role in induction of apoptosis in LPS-stimulated RAW264.7 cells (Figure 3-3).
Interestingly, apoptosis likely occurred only when high doses (> 50 PM) of SB202190
were administered and appeared to be significantly both LPS- and SB202190-dose
dependent as well as time-dependent. Since 50 WpM is an extremely high concentration
compared to 10 pIM, a commonly-used concentration and we saw that 50 jiM and 75
jiM of SB202190 had no significant difference of impact on the secretion of IL-6 and
TNF-oa (Figure 3-2), a question might be raised that nonspecific targets of SB202190
rather than p38 MAPK played a right role in this process. However, SB20290 is
thought to be highly selective and has no effect observed on a lot of protein kinases at
concentrations up to 100 jiM [90]. Hence, we would like to let our systematic analysis
of TLR-4 signaling network tell us the answer.
........
.
- a-9-A-r-5
-+-4-o-
Ctri
0.3 ngmnl LPS
1 ng/ml LPS
ng/ml LPS
0.3 ng/nl LPS&10 pM SB
1 ng/ml LPS&10 LM SB
5 nglml LPS&10 IrM SB
0.3 ng/nl LPS&50 pM SB
-*-a-
1 ng/ml LPS&50 IpM SB
5 ng/ml LPS&50 trM SB
---x-
0.3 ng/mnl LPS&75 pM SB
1 ng/ml LPS&75 IpM SB
5 ng/ml LPS&75 IpM SB
phospho-JNK2
x
i-
\
0
-' ' i ' '
{----a--
~---·-------------U
I mi w I
' '- ' i '• I •
'
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
Time (h)
phospho-ERK2
phospho-ERK1
1.00.8-
._X
T
. ;T-0.6-
J-
9
*__
0.40.20.0
-.
U
I
II
.
I
-
I
"
J,
I
0.0
I
0.5
I
1.0
I
I
1.5
2.0
0
2
.
.
.
.
.
.
.
.
.
.
.
phospho-Akt
TI S
I
.
L
I
0..0
.
.
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
Time (h)
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
Time (h)
0
0
Time (h)
l
___
I
-I
I
_____I
I
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
Time (h)
1
1.00.80.60.40.2-
]
ni
V•'
Time (h)
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
Time (h)
Figure 3-4. Aproteomic compendium of LPS plus SB202190-stimulated signaling in
RAW264.7 cells
RAW264.7 cells were treated with 13 cues listed in Table 3-1. Then their lysates were all collected
at 6 time points (15 min, 30 min, 1hour, 2 hours, 4 hours and 6 hours), resolved in 10%
SDS-PAGEs, and measured for indicated proteins by quantitative western blot as Figure 2-10
described. The information of antibodies was available in Table 2-1 (Upstate antibody was used
for phospho-MK2), and for easy reading generated data were finally normalized to the maximum
value in each kinase dataset and plotted as the mean of three biological replicates + s.e.m.
3.2.2. Proteomic compendium of the LPS plus
SB202190-stimulated signaling network
Next, we wanted to know the dynamic process of TLR-4 signaling network
stimulated by the above 13 cues. To do that, we measured the phosphorylation levels
of six kinases at a series of time points by quantitative western blot. These six kinases
were JNK, ERK1/2, MK2, Akt, IKKa/3, and eIF2a. They are distributed in six
relatively independent branched downstream pathways of the TLR-4 signaling
network (Figure 1-1) and thereby might have a good representation of this network.
Here we didn't consider p38 MAPK but selected its direct substrate MK2, because
SB202190 will suppress the activity of p38 MAPK but can't prevent the
phosphorylation of p38 MAPK by its upstream kinases so that measurement of
phosphorylated p38 MAPK would no longer represent the real changes of its activity.
Furthermore, we also had to miss the IRF-3 branch pathway since we didn't find a
suitable antibody to quantify it. On the other hand, we also picked up six time points
spanning 15 min to 6 hours to prepare triplicate cell lysates and subsequently collect
six measurements of the above phosphorylated kinases from each lysate (Figure 3-4
and Table 2-1). Since many kinase signals exhibit rapid changes between 0 and 1 hour
and relax after 2 hours, time points were distributed unevenly, with four time points
concentrated within the first two hours (Figure 3-1).
As shown in Figure 3-4, the profiles of these six phosphorylated kinases were
quite different from each other. Even for the two isotypes of ERK, their profiles also
appeared to be a little bit different. For example, the phosphorylation of ERKI was
more inducible than that of ERK2 and had steeper response curves. Thus, sometimes
we would like to consider ERK1 and ERK2 separately. However, if we checked the
plots carefully, we might find that the most difference of these profiles consisted in
the part of SB202190 addition. In contrast, LPS congruously enhanced
phosphorylation of all these kinases except eIF2a, which peaked at around 30 min
and then dropped sharply. Such transient activation is thought to limit cell responses
and protect the host from overreaction-caused damage [2]. Consistently with some
previous observations using other cell types [15-17], LPS alone failed to promote the
phosphorylation level of eIF2a in RAW264.7 cells, but LPS plus SB202190 actually
did that (Figure 3-4). Since LPS plus SB202190-induced transient increase of eIF2c
phosphorylation was strictly SB202190-dose-dependent, we speculated that LPS
might have no contribution in this process. As expected, SB202190 with or without
LPS induced the comparable levels of phosphorylated eIF2aC (Figure 3-5). Moreover,
the same results were also observed with another specific inhibitor of p38 MAPK,
SB203580 [90], but not with the MEK1/2 inhibitor, U0126 [90] (Figure 3-5),
suggesting that p38 MAPK is another potential regulator of eIF2a phosphorylation.
However, how the suppression of p38 MAPK induces eIF2a phosphorylation hasn't
been reported so far.
Besides inducing a transient action of eIF2a phosphorylation, SB202190 also
had different effects on the other six kinases. For example, it extended LPS-stimulated
phosphorylation of JNK, ERK1/2, and IKKa/P3 in a roughly dose-dependent manner
(Figure 3-4). The same phenomenon was also reported previously [23, 91] and has
been demonstrated to be mediated by the inhibition of MKP-1 expression [23] though
other members of the MKP family might be involved as well [21, 92]. Since MKP-1
regulated by p38 MAPK pathway [23] preferentially inhibits p38 MAPK and JNK
[22], SB202190 had the most strongly extended LPS-stimulated JNK phosphorylation
but had only moderate or even negligible effects on the extension of LPS-stimulated
ERKi, ERK2 and IKKa/3 phosphorylation as shown in Figure 3-4. However, it was
unusual for us to see that the phosphorylation of MK2 was strongly extended as well
since SB202190 blocked the activity of p38 MAPK and thereby should prevent the
'
I
LPS (ng/ml) SB202190 (pM) -
5
-
- 5
10 10
SB203580 (pM)
U0126 (pM)
- 5 - 5 - 5 75 75 - - - - - - .- 10 10 75 75 10
.
5 5
- 75
- 10 10
p-IKKa/B
p-Akt
'•-
p-JNK2
p-JNK1
30 min
p-MK2 (Upstate)
-.0-p-MK2 (CST)
* *aw~*
p-eIF2a
- p-lKKa/P
4 hours
.
*-a- p-Akt
.
000
4W
o
-- p-JNK2
-..
• ... p-JNKI
Figure 3-5. Comparison of the effects of SB202190, SB203580 and U0126 on LPS-stimulated
kinase phosphorylation
RAW264.7 cells were treated with various combinations of LPS, SB202190, SB203580 and
U0126 as indicated. 30 min or 4 hours after LPS addition, cell lysates were collected, resolved on
10% SDS-PAGE, and then analyzed for indicated phosphorylated kinases (Table 2-1) by
quantitative western blot. The amount of each sample loaded in the same gel was equal.
phosphorylation of its direct substrate, MK2 [23, 92-94]. On the other hand, we also
found a literature describing that not only p38 MAPK but also ERK phosphorylated
MK2 in human neutrophils [95]. Accordingly, we then tested whether the same thing
occurred in RAW264.7 cells. As expected, LPS-induced MK2 phosphorylation was
dramatically prevented by 10 pM U0126, which is a specific inhibitor ofMEK1/2 and
inhibits the phosphorylation of ERK1/2 by MEK1/2 [90] (Figure 3-5), and completely
blocked by U0126 synergically with SB202190 (Figure 3-5). Such outcome couldn't
be caused by nonspecific targets of U0126 since another MEK1/2-ERK1/2 inhibito,:
PD98059 [90], also exhibited the same capacity (data not shown). Given that
RAW264.7 cells have an extremely high level of ERK1/2 (Figure 2-8 and reference
86), the result might make sense. However, when we ran the same experiment using
another anti-phospho-MK2 (Thr334) antibody purchased from Cell Signaling
Technology Inc., we got a completely different result. The new antibody was unable
to detect phospho-MK2 proteins in LPS plus SB202190 or SB203580-treated samples,
and instead detected comparable levels of phospho-MK2 proteins in both LPS only
and LPS plus U0126-treated samples (Figure 3-5). Based on these new data, we might
conclude that only p38 MAPK but not ERK1/2 can phosphorylate MK2 in RAW264.7
cells. Since both antibodies detected phospho-MK2 proteins at the right molecular
weight, we didn't think the antibodies per se had any problem but rather speculated
that MK2 phosphorylated by ERK1/2 and by p38 MAPK might have different
additional modifications so that the new antibody could only recognize the latter one.
Such difference did exist in JNK and ERK1/2 phosphorylated c-Jun,
LPS(ng/ml) SB202190 (pM) LY294002 (pM)
Wortmannin (tM) -
5
5
5 5
mhUursU
5
5
-
-
- - 10 75
- - 10 30 10 30 - - - - - - - 0.1 0.3 0.1 0.3
-
30..
p-Akt
---
.
30 min
5
m-
-
p-MK2 (CST)
-
p-Akt
-- p-MK2 (CST)
Figure 3-6. Both PI-3K and p38 MAPK were required for LPS-stimulated phosphorylation
of Akt in RAW264.7 cells
RAW264.7 cells were treated with various combinations of LPS, SB202190, LY294002 and
Wortmannin as indicated. 30 min or 4 hours after LPS addition, Lysates were collected, resolved
on 10% SDS-PAGE, and then analyzed for phospho-Akt (Ser473) and phospho-MK2 (Thr334) by
quantitative western blot. The amount of each sample loaded in the same gel was equal.
though they also have common phosphorylated sites, Ser63 and Ser73 [96].
Considering that SB202190 can inhibit Akt and MK2 (CST) phosphorylation in a
similar manner (Figure 3-4), we finally decided to use the antibody purchased from
Upstate to collect our data of phosphorylated MK2 so that we would be able to get
more information from the modeling.
Unlike the other six kinases, LPS-stimulated phosphorylation of Akt was
almost completely blocked by both p38 MAPK inhibitors, SB202190 and SB203580
even at 10 jpM, but wasn't affected by U0126 (Figure 3-4 and Figure 3-5), suggesting
that p38 MAPK is required for Akt activation. The similar result was also observed
previously in erbB-2-overexpressing human mammary epithelial (HME) cells [97],
fMLP-treated human neutrophils [98] and polymorphonuclear leukocytes [99], and
stress-stimulated L929 cells [100]. The process has been demonstrated to be
implemented by the signal complex consisting of p38 MAPK, MK2, Akt, and Hsp27
[98-100]. However, it may not be perfect since two PI-3K inhibitors, LY294002 and
wortmannin [90], were also shown to dramatically prevent LPS-stimulated
phosphorylation of Akt, but had no significant impact on phosphorylated MK2
(Figure 3-6). So, a new mechanism involving both p38 MAPK and PI-3K might be
proposed in the future.
3.3. PLS model construction and interpretation
So far, we had successfully collected more than 1800 data of various cell
responses and dynamics of kinase phosphorylation using different quantitative assays.
After careful compilation, these data could become quite readable and informative for
the analysis of each individual cell response or kinase (see Section 3.2). However, for
the mining of their inter-relationships the employment of a proper mathematical tool
would be much helpful and effective [88]. Hence, our next step had focused on using
computational approaches to reduce the complexity of our data and do systematical
data analysis.
To do that, we cast our data in three blocks, respectively called cue, signal
·_
.·__i_
11_______
· ~··· __ ·_ ; :i·__ 1·_··_11··~~
~iir~·
~
~ _·I~_·_I___~_
and response as described in Figure 3-1. The cue block was a description matrix
containing the detailed information of 13 treatments (Table 3-1), while the signal
block and the response block respectively consisted of our compiled phosphorylation
measurements (Figure 3-4) and cell response measurements (Figure 3-2 and Figure
3-3) (error bars weren't included.). Since the different time points were collected from
separate tissue culture plates, the individual time points from 13 treatments were
blindly considered as independent samples for the purpose of the multivariate analysis.
Moreover, all the measurements were made on samples prepared under identical
conditions, so we can safely include these in the same sample row. Accordingly, the
final dimensions of the signal block and the response block were respectively 13 x 42
and 13 x 4 without empty elements.
3.3.1. Clustering through principal component analysis
Firstly, we sought to explore the potential biological significances hiding in
each individual data block by clustering their relevant intra-variables through
4
3
0
CL
C
-2
Principal component #1
Figure 3-7. PCA mapping of input treatments
Scores plot for two-component PCA model built on the data of cell responses (Figure 3-2 and
Figure 3-3), showing the projection of each sample along principal components #1and #2
identified by the PCA model. L, LPS (ng/ml); SB, SB202190 (pM).
0.8-
phospho-JNK2
0.6-
15min
phospho-IKKal•
15 min 3rno
0.6-
0.4
30 min
lh
0.4-
0.2-
0.2-
15
0.02h
0.0-0.2 -
-0.2-
12 h
· 4h
h
-0.40.0
0.8
0.1
-0.6
I
•~
0.2
phospho-ERK1
0.3
-0.4-
0.4
15min
0.0
0.1
0.2
phospho-ERK2
0.8
0.6
0.6
0.4-
0.4
0.3
0.4
i15min
30min
30 min
0.2
0.2
1h
0.0 -
0.0 -0.2
2
-0.2
-0.4 .
-0.4
0.0
0.1
0.2
0.3
0.0
0.4
phoshpho-MK2
0.1
0.2
0.3
0.4
0.2
0.3
0.4
phospho-Akt
30 min
0.4 -
0.3-
15min
0.2-
1-I
2h
how
0.1
0.2
0.3
014
0.0,1
0.0
0.1
Principal component #1
Figure 3-8. Principal component analysis distinguished distinct time-resolved activities of
specific kinases
Loadings plots for two-component PCA model constructed based on the sub-dataset of each
individual phosphorylated kinase in the signal block (Figure 3-4), showing the projection of each
signal along principal components #1 and #2 identified by the PCA model. The phospho-elF2a
loadings plot was unavailable since there was no significant PCA model constructed.
principal component analysis (PCA). PCA is a nondirected multivariate analysis
technique that employs singular value decomposition to separate linearly mixed raw
data into a structure part and noise part in a block-based manner [101]. The model can
be displayed by the following equation: X = TPT+E (T: scores matrix, PT: loadings
matrix, E: error matrix). The fundamental idea is to use fewer dimensions to replace a
complex multidimensional dataset, but still be considered a good approximation to fit
the original data closely enough. In particular, the resulting scores matrix and loadings
matrix can also help us to find out in what respect one sample is different from
another, which variables contribute most to this difference, and whether the variables
contribute in the same way or are independent of one another. As shown in Figure 3-7,
13 samples or treatments were mapped almost linearly in two distinct directions
which respectively represented increasing apoptosis and increasing secretion of IL-6
and TNF-a in scores plot. Interestingly, these two directions appeared to be closely
orthogonal, suggesting that the occurrence of these two cell responses was probably
independent. In other words, LPS-induced secretion of IL-6 and TNF-a might have
no impact on LPS plus SB202190-induced apoptosis through autocrine while the
activation of caspase-3 in apoptotic cells might not interfere with the secretion of IL-6
and TNF-a either. Furthermore, we also successfully constructed two-component
PCA models for each individual phosphorylated kinase except phospho-eIF2a, and
from their corresponding loadings plots we saw the phosphorylation of JNK, IKKa/p3,
ERK1/2 and MK2 but not Akt contributed differently to the two principal components
identified by their PCA models depending upon the time point (Figure 3-8), indicating
these kinases might have different activities at different time points even in response
to the same stimuli. Exceptionally, the signals collected after 1 hour tended to cluster
together (circled) (Figure 3-8). Given the extension of JNK, IKKa/13, ERK1/2 and
MK2 phosphorylation after 1 hour was attributable to the inhibition of p38 MAPK by
SB202190 (Figure 3-4), these PCA models had revealed the same kinase can convey
diverse messages depending on the timing and the status of other kinases or proteins.
3.3.2. PLS modeling of TLR-4 signaling-mediated
secretion of IL-6 and TNF-a
Our next step in data analysis focused on attempts to explore the potential
relationships between TLR-4 signaling and cell outputs. To attain this aim, we used
PLSR, since it is designed to generate data-driven predictive models that relate a
matrix, or block, of independent measurements to a block of dependent measurements
or classifications.
As mentioned previously, our data had been cast in three blocks respectively
representing cues, signals and responses. Any one of them can act as independent or
dependent blocks in PLSR modeling, but it is conceptually helpful to organize them
according to the cause-effect relationship. Therefore, we set the signal block as the
independent block and set the response block as the dependent block. Unlike PCA,
which extracts latent variables to maximize the information captured from a single
dataset, PLSR executes a simultaneous and interdependent PCA decomposition in
both independent (X) and dependent (Y) blocks in such a way that the information in
the Y block is used directly as a guide for optimal decomposition of the X block, and
subsequently performs regression of Y [54]. The model equation can be written as: Y
= XW*CT+F (W*=W(PTW) -1, W: X weights matrix, CT: Y weights matrix, F: residuals
matrix) [56]. Hence, the resulting w*c vs. w*c plots can show the correlation structure
between X and Y, including how the X and Y variables combine in the projections and
how the X variables relate to Y. In addition, three parameters, respectively called
R2Y(cum), Q2 (cum) and JDmodY, have been proposed to evaluate the quality of
constructed PLSR models [46, 56]. R2Y(cum) is a measure of how well the model fits
the data and a value close to 1 is desired for a good model. Q2(cum) indicates how
well the model predicts new data and a large value (> 0.5) indicates good predictivity.
YDModY is the distance of the observation in the training set to the Y model plane or
hyper plane. No critical limits are shown for 2DModY, but the smaller value is better.
Furthermore, there is also a parameter, called variable importance for the projection
(VIP), proposed to evaluate the importance of the variables [56]. VIP values larger
than 1 indicate important X variables, and values lower than 0.5 indicate unimportant
X variables. A good PLSR model usually is powerful to identify the determinative
changes in a particular response, even when the characteristics of many other proteins
are varying simultaneously, since it can highlight the measurements that strongly
co-vary with the known outcome and deemphasizing those that do not.
In order to make the interpretation easier, we divided the response block into
two sub-blocks, one containing the measurements of IL-6 and TNF-a release and the
other containing the measurements of apoptosis, especially when these two cell
mi
MI1
R2Y(cum)
Q2 (cum)
60-
40 -
20-
/
r --T--
NIA
L
B
-
NIA
H
01
01I
0.4
10
P
SSB
1L751+8
S 8i
+
0.2-
* Akt
0.1-
All other kinases
A IL-6
* TNF-o
SB
L 1,S
1U
s
0-
0.3-
00 3 L*75B
-SI -3/-
0.0
I
-0.1
Ui
·
I
-6
-4
-2
0
2
4
Princinpal component #2
6
-02
-0.1
0.0
0.1
0.2
PCI weights
Figure 3-9. PLSR modeling indicates Akt is the primary informer for IH-6 and TNF-a
secretion in RAW264.7 cells
(a) The plot of R2Y(cum) and Q2(cum) values for full model and single kinase models across all
time points. (b) The corresponding deviation from regression line as calculated by the root square
distance summed across all 13 samples. (c) Scores plot for two-component PLSR full model
projects each sample along PCl and PC2 identified by the PLSR model. Circle colors are IL-6 and
TNF-a at high (red), median (blue) and low (green) levels. (d) The loading plot of weights for
two-component full model projects the response variables, IL-6 and TNF-a, on the same plot as
the contributions of signal variables for seven kinases: JNK2, ERKI, ERK2, IKKa/p, eIF2a, Akt
and MK2. N/A, not applicable; PC, principal component; L, LPS (ng/ml); SB, SB202190 (pM).
n'"
ZO
20(
S15(
10)
(9
-1 10(
5(
8-
60
E
-- 40
Z
P
20
0
LPS(ng/ml)
SB202190 (pM)
LY294002 (pM)
Wortmannin (pM)
- 5 5 5
- - 10 75
- - - - - - -
5 5
- 10 30
- -
5
10
10
-
5
10
30
-
10
-
- 5 5 5 5 - - - - 10 10 - 30 - - - - - - 0.10.3 0.10.30.10.3
Figure 3-10. Different effects of two PI-3K inhibitors on LPS-induced IL-6 and TNF-a
release
RAW264.7 cells were treated with various combinations of LPS, SB202190, LY294002 and
wortmannin as indicated. 4 hours after LPS addition, the amounts ofIL-6 and TNF-a in their
media were analyzed by ELISA. Values are plotted as the mean of three biological replicates +
s.e.m.
responses were suggested to be probably independent (Figure 3-7). Subsequently, we
performed PLSR modeling using each of them separately, and in this section we
focused on the analysis of TLR-4-signaling mediated 11-6 and TNF-oc release. As
shown in Figure 3-9, the optimal full model built on the whole signal block exhibited
a value of 0.96 for R2 Y(cum), 0.77 for Q2(cum) and 22.96 for YDModY.
Theoretically this is a good model, and it discriminated the samples that secreted IL-6
and TNF-a at low, median and high levels well in scores plot (Figure 3-9c and Figure
3-2). Moreover, in its loading plot of weights we even saw that two cytokines were
clustered and mapped closely to Akt signals across all time points (Figure 3-9d) which
had much higher VIP values (> 1.35) compared to other signals (< 1.08), suggesting
Akt was the dominant informer of LPS-induced IL-6 and TNF-a release in the full
model. Accordingly, we then examined whether a single kinase model of Akt also had
the same capacity. As expected, the Akt-only model had slightly improved Q2(cum) of
0.82 and the same R2 Y(cum) of 0.96 but slightly worse 2DModY of 24.13 compared
to the full model, whereas all other single-kinase models performed poorly (Figure
3-9a and b). This result indicated that the complex TLR-4 signaling network could be
fully superseded by dynamic measurements of the phosphorylation of a single kinase,
Akt in capturing and predicting the information of IL-6 and TNF-a secretion induced
by LPS and then perturbed by SB202190, and JNK2, ERK1/2, IKKa/03, eIF2a, and
MK2 appeared to be redundant and less effective in this process.
It is noticeable that what PLSR generates is a statistic model rather than a
mechanistic model. Thus, the relationship identified by a PLSR model doesn't have to
represent a real biological mechanism. Critical experimental validation is usually
required. To confirm whether the dominant informer, Akt can practically control IL-6
and TNF-oa release or not, two specific inhibitors of PI-3K, LY294002 and
wortmannin, were used to block LPS-induced Akt phosphorylation independently of
p38 MAPK (Figure 3-6). Interestingly, these two inhibitors showed opposite effects
on LPS-induced IL-6 secretion. LY294002 strongly suppressed it while wortmannin
mildly promoted it (Figure 3-10). The same result was already reported in human
macrophages [12]. In particular, dominant-negative mutants of PI-3K behaved
similarly with wortmannin rather than with LY294002 [12], suggesting augmentation
of IL-6 secretion by wortmannin might be mediated actually through PI-3K inhibition,
whereas the effect of LY294002 might be caused by its nonspecific targeting. If this is
true, the inhibition ofAkt phosphorylation should be able to promote IL-6 secretion as
wortmannin does. By contrast, both LY294002 and wortmannin sightly, but not
significantly, inhibited TNF-a secretion (Figure 3-10). Obviously, all these results
were inconsistent with the outcome of PLSR modeling that indicated Akt
phosphorylation should be highly positively correlated with both IL-6 and TNF-a
secretion (Figure 3-9). However, it doesn't mean there was any problem with our
previously constructed PLSR models, but rather suggested that an upstream kinase or
protein of Akt might control IL-6 and TNF-a secretion through an Akt-independent
mechanism so that Akt could act as a dominant informer but not a real participator. In
our system, this upstream kinase is p38 MAPK. Since phosphorylated Akt was treated
as indirect quantification of p38 MAPK activation in our modeling, our constructed
PLSR models might not be applicable for the analysis of IL-6 and TNF-a secretion in
which the activation of PI-3K was inhibited. To eliminate such restraint, more diverse
treatments might need to be involved in the construction of PLSR models.
3.3.3. PLS modeling of TLR-4 signaling-mediated
apoptosis
Next, we sought to analyze the relationship between TLR-4 signaling and
apoptosis by PLSR modeling. As we did in the previous section, we also constructed
full model using the whole signal block and single kinase models using its individual
kinase subsets. As shown in Figure 3-11a and b, although jDModY of each model
was comparable, the optimal full model had the highest value of 0.96 for R2Y(cum)
and 0.54 for Q2 (cum). All single kinase models performed poorly, suggesting that a
satisfactory model relating signals downstream of TLR-4 activation to induction of
apoptosis in RAW264.7 cells requires a quantitative combination of measurements
across multiple branch pathways. The fact that in the full model most kinases,
including eIF2a, IKKa/13, JNK2, MK2 and ERK1/2, had >1 VIP at some time points
(Figure 3-11 c) also supported such deduction. To confirm if all these kinases actually
participated in regulation of LPS-induced apoptosis, LPS-stimulated RAW264.7 cells
were co-incubated with various combinations of small molecule inhibitors. The
inhibitors that we tested included SB2023580 and SB202190, p3 8 MAPK inhibitors,
PD980059 and U0126, MEKI/2 inhibitors, wortmannin and LY294002, PI-3K
b
a
^^
"'
1.0
0.8
60-
0.6
2
40-
0.4
200.2
0-
0.0
HHHFHF7F1.F7
a~;·4"B~d
0-
g
I
g
w i
R
I
I
-
B
-
C
3
2
0
U-
z z z•
A"
IxNe
LL •"•-
zR
L
Figure 3-11. Iterations of modeling on apoptosis using subsets of signal variables indicate
that no model outperforms others for all descriptors
(a) The plot of R2Y(cum) and Q2(cum) values for full model and single kinase models across all
time points. (b) The corresponding deviation from regression line as calculated by the root square
distance summed across all 13 samples. (c) The plot of the VIP values of the last component with
95% confidence intervals for each signal variable in full model.
inhibitors, PS1145, an IKKo/j3 inhibitor, and SP600125, a JNK inhibitor [102]. As a
result, except SB2023580 and SB202190 all the other inhibitors alone at tested
concentrations or with 5 ng/ml LPS were unable to trigger apoptosis in RAW264.7
cells within 8 hours (Figure 3-12). Accordingly, we can conclude that both the
inhibition of p38 MAPK and the engagement of TLR-4 play an indispensable role in
this process. However, all these inhibitors except wortmannin could either promote or
suppress SB2023580 or SB202190 plus LPS-induced RAW264.7 cells significantly
(Figure 3-12), to some extent consistently with the description of our full model that
except Akt all other kinases are probably involved in the regulation of SB202190 plus
LPS-induced apoptosis at some, if not all, time points (Figure 3-11). Nevertheless,
how good this full model would be might be evaluated more pertinently based on its
real prediction ability that we were going to test in the next section.
Particularly, we had noticed that JNK and ERKI/2 had opposite effects on
LPS plus SB202190-induced apoptosis conveyed by the data of SP600125, PD980059
and U0126-treated samples in Figure 3-12. Since some microtubule inhibitors (e.g.
Taxol, vinblastine, vincristine, and colchicine) [103] and lipids (e.g. ceramide and
sphingosine) [104] induced apoptosis was reported to correlate with the imbalance of
JNK and ERK activation, we was wondering if the ratio of phosphorylated JNK and
(0
Figure 3-12. Inhibitor screen indicates SB202190 plus LPS-induced RAW264.7 apoptosis can
be regulated by a few kinases
RAW264.7 cells were incubated with indicated treatments. 8 hours after LPS addition, cells were
collected, stained for active caspase-3 (Alexa 488) and cleaved PARP (Alexa 647) and analyzed
by flow cytometry. Values are expressed as the mean of no less than two independent experiments.
If> 2 independent experiments were done, error bars (s.e.m.) were calculated and plotted as well.
The unit for LPS is ng/ml, and the unit for others is pM.
2J
I
I~
-.
~--. -----
-
I~-
---.
fAAS
5L+75SB '~
AD.3L+75SB
/
AOAL+75SB
r
I
"
I
o AlB
AIJ3L+SB
A.3LI
I L
"s.OS
6
-6
__ _
I
ALN
_
-4
__
_
__
_
_
2
0
-2
Principal component #1
_
_
_
4
_
_
.I
6
A
II
-
I
R2Y(cum)
u(CUm)
0.8-
0.60.4
0.2 -
0.0.C~3~.--v
Si
Figure 3-13. PLSR models built on kinase ratio and response could be another good choice
for description of TLR-4 signaling mediated apoptosis
(a) Scores plot for two-component PLSR built on the ratio of phosphorylated JNK2 and ERK1
across all time points and apoptosis projects each sample along principal component #1 and
principal component #2 identified by the PLSR model. The blue ellipse is Hotelling T2 (0.95). L:
2
2
LPS (ng/ml); SB: SB202190 (pM). (b) and (c) were the plots of R Y(cum) and Q (cum) values for
indicated PLSR models that were built with the data of the control sample excluded.
ERK1/2 will determine LPS plus SB202190 induced RAW264.7 apoptosis. To test
this hypothesis, we subsequently performed PLSR modeling for each possible pair
(about 28) of phosphorylated kinases (here ERKI, ERK2 and total ERK were treated
as three individual kinases) using datasets consisting of their ratios across all 13
samples and 6 time points. During this process, we found that the control sample was
frequently identified as an outlier which is projected outside the confidence ellipse of
Hotelling T2 (0.95) [56] in the scores plot (Figure 3-13a). It does mean that the control
sample might not belong to these models (less than 5% probability) and should be
excluded. We didn't know why the control sample was so special here, but the high
noise level caused by the low signal level in the control sample that can be amplified
during the calculation of kinase ratios might be an important reason. Thus, all our
kinase ratio PLSR models were built with the data of the control sample excluded, and
interestingly the removal of the control sample data was also shown to increase the
Q2(cum) value of the full model from 0.54 to 0.73 but have slight effects on single
kinase models (Figure 3-13b). By comparison, we had found four kinase ratio PLSR
models had both good R2Y(cum) and Q2(cum) values (> 0.5) while others exemplified
by the JNK2/IKKa•/
model performed poorly (Figure 3-13c). Since we had already
demonstrated that MK2 were mainly detected to be phosphorylated by ERK1/2 in our
system when the Upstate antibody was used (Figure 3-5), all these four decent PLSR
models just implied that dynamic measurements of the ratio of activated JNK and
ERK might also be able to monitor TLR-4 signaling mediated apoptosis in RAW264.7
cells effectively as we hypothesized.
More importantly, since the X variable datasets used to construct kinase ratio
PLSR models were much simpler and just consisted of the ratios of phosphorylated
paired kinases across 12 observations and 6 time points, the above four models might
provide a good opportunity to understand how LPS plus SB202190-induced apoptosis
progresses and is regulated in timing. Hence, we subsequently compared their loading
plots of weights, VIP values and coefficients (Figure 3-14). Actually, all of these four
models indicated that 4-hour and 6-hour kinase-ratio variables were informative since
they were mapping closely to the apoptosis variables in scores plots and exhibited > 1
VIP values and high positive coefficients, suggesting that the apoptotic process should
be initiated or determined before 4 hours. However, as for earlier time points, the
results of these four models were inconsistent. Both the JNK2/E1 and JNK2/ERK
models further extended the possible apoptosis starting time point to < 1 hour, as
1-hour and 2-hour variables had comparable VIP values and coefficients with 4-hour
11~~~1·····-~111~~11
1·····(·--·1···-·--·~1~~111
ni
AII,
JNK2/ERK1
4h
0.so. JNK2/ERK2
Z
0.2-
£2h
£5h
0..45
4h
0.0-
M8h
0o.
-0.2-
min
,30
£15
-0.4
0.0.35
04
-0.4-
0.30-
-0.8-1.0
0.
n*
0.35
0.30
0 0.25
-~--
.30 min
0.25
£15 min
0.40
0.
0.45
0.25
0.45
0.40
0.35
0.30
0.50
PC1 weights
PC1 weights
0.6
4 ,,h 5
JNK2/MK2
I
0.6- JNK2/ERK
4h
Sh A
0.4*
0.2-
4h
A2h
Ash
S0.0" -0.2-
A2h
.15mn
L -0.4-
-0.6-0.8-
1
A30mMin
omin
30min
I
0.30
0.PC25
0.35
0weights
0.
.50
.I
"
.V.
V
3o
V
A
na
U
PC1 weights
PCI weights
ERK1
* JNK2
* JNK2IERK2
* JNK2NMK2
15
oU.
16min
15min
30min
1h
2h
4h
6h
n
6
i
6h
Figure 3-14. The comparison of four kinase ratio PLSR models
(a) The loading plots of weights for indicated kinase ratio PLSR models. Apoptosis variables are
red square while kinase ratio variables are black triangle. (b) The plot of their VIP values at six
time points. (c) The plots of their coefficients at six time points for apoptosis at 4 h and at 8 h.
61
70
60
u, 50
C 40
0
< 30
20
10
0
Ctrl
/
-0.5
0
0.25 0.5
1
2
75 pM SB202190+1 ng/ml LPS
4
6h
U0126
(500 nM)
Figure 3-15. Effects of the timing of U0126 addition on LPS plus SB202190-induced
apoptosis
RAW264.7 cells were pretreated with either a drug vehicle (DMSO, Ctrl) or 75 pM SB202190 for
30 min before either 0.2% TEA (Ctrl) or 1 ng/ml LPS was added. 500 nM U0126 was also added
to some samples at indicated time points relative to LPS addition. 4 hours or 8 hours after LPS
addition, cells were collected, stained for active caspase-3 (Alexa 488) and cleaved PARP (Alexa
647) and analyzed by flow cytometry. Values are plotted as the mean of three biological replicates
± s.e.m. *, p < 0.05, and **,p < 0.01 (t-test) compared with the sample collected at the same time
points (4 hours or 8 hours) and in which U0126 was administered at -0.5 h.
and 6-hour variables and 15-minute and 30-minute variables might had no significant
contributions due to their coefficients close to zero. In contrast, the JNK2/MK2 model
had less informative variables at 15 minutes and 2 hours and even a negative
coefficient at 30 minute predicting that LPS plus SB202190-induced apoptosis could
be an unstable process, whereas the JNK2/ERK2 model predicted a relatively stable
one starting as early as within 15 minutes. In order to figure out which mechanism
conveyed by these four kinase-ratio PLSR models is valid, we investigated the effects
of the timing of U0126 addition on LPS plus SB202190-induced apoptosis. Our
hypothesis was that if apoptosis is determined by the ratio at a specific time point, the
addition of U0126 before that time point should promote LPS plus SB202190 induced
apoptosis to the similar extent and instead to a smaller extent if apoptosis is initiated
after that time point. As shown in Figure 3-15, such time point exactly exists in our
system and is around I hour after LPS addition. Moreover, LPS plus
SB202190-induced apoptosis should progress stably as well since apoptosis decreased
steadily with the timing increase of U0126 addition. All results were consistent with
the prediction of the JNK2/ERK1 and JNK2/ERK models, suggesting that these two
models might perform better in presenting TLR-4 signaling-mediated apoptosis
though they had a little bit lower R2Y(cum) and Q2(cum) values than those of the
JNK2/MK2 model (Figure 3-13b).
3.4. Model testing through prediction
As we saw in the previous section, for a simple cell response you might be
able to find a couple of PLSR models with decent R2Y(cum) and Q2(cum) to describe
it, and these descriptions can then be translated into distinct biological mechanisms.
Therefore, model validation is usually required even if only one PLSR model was
applicable. Besides designing critical experiments, testing through prediction is
another powerful way frequently utilized to validate and evaluate constructed PLSR
models. We might do prediction respectively using the existing training dataset and
other relevant independent datasets. However, the former is usually less useful since it
has already been done to compute Q2(cum) during the model construction so that we
can evaluate such model predictivity directly by Q2(cum) [56]. Congruously, our
PLSR models that exhibited higher Q2(cum) values were shown to have lower root
mean square errors (RMSE) of prediction done using the trained data (Figure 3-16).
Yet, from these plots we also found that the full model and the Akt-only model had
different predictivity on IL-6 and TNF-a secretion especially when their amounts
were small though both of them had close Q2(cum) values, suggesting that Q2(cum)
values might just represent the average model predictivity if multiple cell responses
are included and recalculation of individual RMSEs might be helpful to distinguish
them. Howsoever, to evaluate our constructed PLSR models more convectively, we
still needed to do prediction using an independent dataset.
"""""`~--""""""-"-----------
-n'r-···I~
~..,~*ii.r~i~i^r*.r··-·--.·r~~i,~~r~
3.4.1 Construction of new independent datasets
OU,
Full model
r
I
0,
.-im
'30-
• ,.6
V
'
40-
r
u.
z
A
0
,
10
S4 h (RMSE 4.23)
0;0
40
30
2 h (RMSE= 1.90)
S.
O0
2 h (RMSE= 1.09)
4 h (RMSE = 3.13)
*
A
20-
50
40
30
20
10
Predicted TNF-a (ng/nl)
0
Predicted TNF-a (nglnl)
1250
1250u
Full model
*,.
1000-
1000-
E
750-
.750600-
,•
..
4h(RMSE=113.93)
.
_s_
•ll-
500-
.
338
2h RME
250-
2500-
* *
*
10
2 h (RMSE S3.38)
i
250
•
i
500
•
I
i
750
i
1000
-
•
1250
0
2h(RMSE2.71)
S4 h (RMSE=85.40)
250
1000
750
500
Predicted IL-6 (pg/mi)
Predicted IL-6 (pg/mi)
,
Full model wlo Crl
Full model
*
,'
JNK2/ERK1
40"
40-
30-
30-1
30-
20-
20-
S10-
,
10-
0
4h(RMSE =229)
S1,I·a
0-
S8
a
0
0
4 h (RMSE = 2.30)
*
0
0
50
40
30
20
10
Predicted % Apoptosis
0
10
x
A
o
-
h (RMSE = 2.22)
20
-
71
0
8 h (RMSE = 2.20)
40 "
"a3
10
Predicted % ADoDtoeis
10
o
4 h (RMSE =2.84)
Bh(RMSE =2.85)
SD
;40
30
Predicted % Acootoais
50
JNK2/MK2
JNK2/ERK
,
JNK2/ERK2
40.
•ifl,4
40
.U
o
0
*
.
10 _• .-- i
*
*
0
10-"
I-'.
i
20-
10
44~.1
20.
10
U -
30-
30so
0 20-
S30-
I.
20
4 h (RMSE= 4.98)
8 h (RMSE = 6.39)
i. 30
Predicted %Apoptosis
40
4 h (RMSE 3.63)
8h(RMSE 4.11)
1
0
50
*
-
..
-
,
•
Predicted %Apoptosis
4 h (RMSE = 2.16)
.
0
= 2.77)
* 8 h (RMSE
0
10.20.-
10
.
,
2D
.---
-.
30
40.
4o0
Predicted %Apoptosis
Figure 3-16. Comparison of model predictivity on the trained data
Observed cell responses (TNF-a, IL-6 and apoptosis) in the training dataset were plotted against
their corresponding predicted results of indicated PLSR models. The RMSE values for each model
were also calculated and shown in the plots.
64
.
4
W
I
80.
10 pMU0126 (-)
10M U0126 (+)
.L 60.
MA
0
0
.-t
OR
0-
0. I
U0126
5
r·-·
pM
7
SB202190+5
-··r····
n
10
-z
50
I
75 pM SB202190
-·
S
g
40
._8
.L 3
0.
0
0.
0
0
<2
00.
%
0
1
0
PI*
,UI
n
U
r
I
r
O
in
I,
75 pMSB202190+5 nglml LPS
I
~n · · ~~pMl~a
LU ,
"U U IO.....r-
0
.
pM
2
SP600125
.
LPS
250 nM U0126+75 pM SB202190+5 ngmli
Figure 3-17. Opposite effects of U0126 and SP600125 on LPS plus SB202190-induced
apoptosis
RAW264.7 cells were treated with indicated stimuli. 4 hours (top-left) or 8hours after LPS
addition, cells were collected, stained for active caspase-3 (Alexa 488) and cleaved PARP (Alexa
647) and analyzed by flow cytometry.
In order to make our new dataset really independent of the training dataset,
we had introduced new stimuli to perturb our system. These stimuli were wortmannin,
U0126 and SP600125. They were selected because they can inhibit the activity of the
kinases that were identified to highly correlate with secretion of IL-6 and TNF-a or
apoptosis in our constructed PLSR models so that we can interpret the result more
pursuantly. Unfortunately, as we demonstrated previously (Figure 3-12), wortmannin
can not significantly perturb LPS-induced IL-6 and TNF-a release, we finally gave up
doing prediction on them and just focused on the prediction on apoptosis. As expected,
U0126 and SP600125 had opposite effects on LPS plus SB202190-induced apoptosis
(Figure 3-17). Based on a lot of preliminary experiments and careful comparison, we
then picked up 6 new treatments as listed in Table 3-2 to treat cells and collected new
3
Table 3-2. The list of new treatments for data collection
Treatments
1
2
3
4
5
6
LPS (ng/ml)
5
1
5
5
5
1
SB202190 (pM)
75
75
75
75
75
75
250
500
-
-
250
500
10
10
U0126 (nM)
10
5
-
-
SP600125 (lM)
Note: All the treatments contained the same amounts of drug vehicles (0.2% TEA and DMSO). DMSO,
SB202190, U0126 and SP600125 were added 30 min prior to 0.2% TEA and LPS.
c'
z
0.
to
0
CL
0.0
0.5
1.0
1.5
2.0
2.5
3.0
phospho-ATF-2
Figure 3-18. Quantitation of JNK inhibition based on the linear relationship between
phospho-JNK and phospho-ATF-2
RAW264.7 cells were treated with 75 pM SB203580+5 ng/ml LPS (1), 75 pM SB202190+1
ng/ml LPS (2), 75 pM SB202190+5 ng/ml LPS (3), 250 nM U0126+75 jpM SB202190+5 ng/ml
LPS (4), 500 nM U0126+75 pM SB202190+1 ng/ml LPS (5), 10 pM PS1145+75 p.M SB202190+
5 ng/ml LPS (6), 5 pM SP600125+75 pM SB203580+5 ng/ml LPS (7), 10 pM SP600125+75 pM
SB203580+5 ng/ml LPS (8), 10 pM SP600125+250 nM U0126+75 pM SB202190+5 ng/ml LPS
(9), 10 pM SP600125+500 nM U0126+75 pM SB202190+1 ng/ml LPS (10) or 10 pM JNKi+75
pM SB203580+5 ng/ml LPS (11). 30 min after LPS addition, cell lysates were collected, resolved
on 10% SDS-PAGE, and then analyzed for phospho-JNK2 and phospho-ATF-2 (Table 2-1) by
quantitative western blot. Subsequently, the through-zero linear fit was performed on samples 1-6
the
(black square, R2 = 0.95) and the generated linear relationship will allow the determination of
values.
phospho-ATF-2
activity-correlated phospho-JNK2 levels of all samples based on their own
Finally, through comparison of the activity-correlated phospho-JNK2 levels of samples with (blue
triangle) and without (black square) JNK inhibitors (SP600125 or JNKi VIII), we would be able to
a
compute the percentage of JNK inhibition by SP600125 and JNKi at indicated concentrations for
specific treatment.
" imil~~
......
...........
.........
EZ]4 hours
8 hours
60.
60
i20
10
0
123458
IKK
eF2
Akt
time
M
ER
ERK
JK2
250U+75SB+5LPS
500U+75SB+1LPS
5SP+75SB+5LPS
10SP+75SB+5LPS
10SP+250U+75SB+5LPS
10SP+500U+75SB+1LPS
I~
0.00
Min = -1.03
Max = 1.03
JNK2/ERK1
Full model wlo Ctrl
Full model
ru-
lo
60.
.12
00-
50
S50-
40-
40
•40
30-
30-
20-
20,
o
10.
4 h (RMSE=232.47,R=0.24)
R=0 27)
S8 h (RMSE=502.04,
0ý
10
·-
20
30
40
50
60
S20-
o
a 4h(RMSE=3.98, #R=0.49)
8h(RMSE=797.80.
Ro0.49)
10-
)
0ý
10-
10
20
30
40
50
60
70
R'=0.96)
• 4 h(RMSE=5.92,
-
v
1
S
S8 h (RMSE=I0.24,R=0.96)
7 ·50
60
40
1i
20
30
10
20
30
50
40
Predicted %Apoptoeis
Predicted %Apoptosis
Predicted %Apoptosis
JNK2/ERK2
JNK2/ERK
JNK2/MK2
70
60.
60-
70
60-
60
S
V
0,<
10
45h(RMSE=34.91,R=0.91)
10-
4 h(RMSE=17.86, R=0.91)
,•
10
S'
U
1U
20
U
30
u
Predicted %Apoptosis
u
20
30
40
50
Predicted % Apoplosis
60
Re=0.91)
4h (RMS.E=30.23
10a
8 h (RMSE-38 18, R=0.98)
S*•
8 h (RMSE=8.23,R=0.99)
h (RMSE=34.73.8R=0.98)
S8
h
70
1
10
.
,
20
.
•
30
I
40
7
-
I
so50
60
Predicted %Apoptos
Figure 3-19. A priori prediction using independent datasets
RAW264.7 cells were incubated with the treatments listed in Table 3-2. (a) 4 hours or 8 hours after
LPS addition, cells were collected, stained for active caspase-3 (Alexa 488) and cleaved PARP
(Alexa 647) and analyzed by flow cytometry. Values are plotted as the mean of three biological
replicates ± s.e.m. (b) Heat map of the dynamic phosphorylation levels of the indicated kinases
from new observations measured, corrected and normalized as described in Figure 3-4 and Figure
3-18. Values are expressed as the mean of two independent experiments. (c) Prediction was done
for indicated PLSR models using the data of part (a) and (b), and observed vs. predicted apoptosis
67
70
were plotted. The RMSE and R2 values for each model were also calculated and shown in the
plots. Data are presented as the mean of three biological replicates ± s.e.m, and model
uncertainties were estimated by jack-knifing [56].
data to do prediction. Apoptosis induced by these treatments were then determined
and shown in Figure 3-19a.
Once the treatments were determined, we next needed to measure the
dynamic phosphorylation levels of the 6 kinases across 6 time points in RAW264.7
cells in response to these new treatments by quantitative western blot. Since U0126 is
an inhibitor of MEK1/2, an upstream kinase of ERKI/2, U0126 addition will inhibit
the activity and the phosphorylation of ERK1/2 simultaneously and might not change
their interrelationship. So, we could still use the data of phosphorylated ERK1/2 to do
prediction. By contrast, the interrelationship between the activity and phosphorylation
of JNK will be changed by the addition of SP600125 since SP600125 tends to inhibit
the activity of JNK rather than its phosphorylation [105]. Therefore, we corrected our
data ofphosphorylated JNK2 if SP600125 was added so that they could transmit the
real activity of JNK2 well by multiplying an inhibition coefficient (0.81 and 0.66 for 5
iM and 10 jpM SP600125) that was generated based on a linear relationship between
phospho-JNK2 and phospho-ATF-2 as detailedly described in Figure 3-18. ATF-2 is a
shared substrate of JNK and p38 MAPK (Figure 3-20a and reference 106), so we were
not surprised to see such linear relationship when 75 p.M SB202190 in our treatments
blocked the activity of p38 MAPK. Finally, these corrected and normalized signal
measurements (Figure 3-19b) and apoptosis measurements (Figure 3-19a)
respectively forming our new independent signal and response blocks were used to do
prediction subsequently.
3.4.2. Comparison of model predictivity on independent
datasets
As shown in Figure 3-19c, the predictivities of our constructed PLSR models
were compared using the above new independent datasets. Obviously, they were quite
different and no longer correlated with the Q2(cum) values as we saw in Figure 3-16.
For instance, the full model was shown to have far smaller RMSE values than that of
the full model built with the control sample excluded, though the latter had a higher
Q2(cum) value. Notwithstanding, both of them performed poorly. That might suggest
we had gotten overfitted models that just represent the pattern of random noise or
individual features of the training dataset, and instead other kinases, proteins or
pathways that play a critical role in regulation of LPS plus SB202190-induced
apoptosis were not involved in our full models. It is possible since we actually missed
IRF-3 in our models, which were demonstrated to be required for LPS plus
SB202190-induced BMDM apoptosis [18]. However, when we zoomed in on the full
model plot (green rectangle), we found the full model, in fact, predicted U0126-free
samples quite well (RMSE = 6.38 and R2 = 0.90), thereby raising another possibility
that a nonlinear correlation between kinases and apoptosis might exist in RAW264.7
cells. Actually, all our four kinase-ratio PLSR models exhibited excellent predictivity
on these independent datasets (Figure 3-19c), though the JNK2/ERK2 and JNK/MK2
models performed a little bit worse and were likely to underpredict the samples. All
these results were consistent with our previous experimental validation, further
confirming that TLR-4 signaling mediated apoptosis can be well described and
predicted by a simple two-kinase PLSR model, the JNK2/ERK1 model or JNK2/ERK
model. In other words, the inhibition of p38 MAPK might regulate LPS-induced
apoptosis by simply increasing the ratio of activated JNK and ERK.
3.4.3. Other biological considerations
Actually when we were trying to figure out a way to accurately quantitate
JNK inhibition by SP600125, we also considered another substrate of JNK, c-Jun. We
found that c-Jun phosphorylation could be significantly induced by LPS especially
after 1 hour and increased sharply till no less than 4 hours (Figure 3-20a and data not
shown). In contrast, LPS-induced phosphorylation of ATF-2 boosted within 30min
and then rapidly decreased after I hour (Figure 3-20a). Phosphorylation of c-Jun is a
I
I
Cs\ý~;I
p-c-Jun
p-ATF-2
p-JNK2
p-JNK1
Ip-ERK2
p-ERK1
4 hours
30 min
[1
0
,-
4 0-
HnHnHHHH
S3 0-
2
0-
0
L
.I i1,.1I1 LIt 1I,pMJNKIVII:
Ji I iiL
20
0
0
0
1
0
10
5
0
10 20
5
2
0 250 250 250
0
0
0
75 mM SB202190+5 ng/ml LPS
20
250nMU0126
il
I I I
II
I
*ii
0
125
250
500
0
125
250
500
I
II
II
1000nMN I
I
1000 nM JNKi VIII
75 SB202190+5 nqrlmlLPS
Figure 3-20. Inhibition of c-Jun phosphorylation by various inhibitors appeared to correlate
with enhanced LPS plus SB202190-induced apoptosis
RAW264.7 cells were treated with indicated stimuli. (a) 30 min or 4 hours after LPS addition, cell
lysates were collected, resolved on 10% SDS-PAGE, and then analyzed for p-c-Jun, p-ATF-2,
p-JNK, and p-ERK1/2 (Table 2-1) by quantitative western blot. All lanes were loaded with the
same amount of samples for p-c-Jun, p-ATF-2 and p-ERKl/2 blotting, but for p-JNK blotting
samples collected at 30 min and 4 hours had the different loading amounts. The units are ng/ml for
LPS, nM for U0126, and pLM for all the others. (b) 8 hours after LPS addition, cells were collected,
stained for active caspase-3 (Alexa 488) and cleaved PARP (Alexa 647) and analyzed by flow
cytometry.
late response because the expression of c-Jun is inducible and regulated by the ERK
pathway [107], whereas ATF-2 is constitutively expressed and can be degraded via the
ubiquitin-proteasome pathway dependently on its phosphorylation and dimerization
[108]. More importantly, various inhibitors including U0126, PS 1145, and JNKi VIII
were observed to strongly prevent c-Jun phosphorylation at 4 hours in LPS plus
SB202190-treated RAW264.7 cells (Figure 3-20a). Since ERK has been indicated to
act as a dominant activator of c-Jun phosphorylation (around 90%) in phorbol ester
stimulated U937 cells, a human leukemic monocyte lymphoma cell line [109], and
also contribute to the phosphorylation of c-Jun in PC12 cells, a cancer cell line
derived from a pheochromocytoma of the rat adrenal medulla [107], we were not
surprised to see that U0126 even at 250 nM could suppress c-Jun phosphorylation
dramatically while SP600125 just did it slightly in our system especially considering
RAW264.7 cells express a high level of ERK. However, we were confused about the
almost complete inhibition of c-Jun phosphorylation by JNKi VIII, another JNK
inhibitor developed in 2006 by Trevillyan's group and shown to have over-1000 fold
selective for JNK1/2 over a panel of 74 kinases [110]. Since SP600125 and JNKi VIII
both strongly inhibited ATF-2 phosphorylation, but had different effects on c-Jun
phosphorylation (Figure 3-20c), we then tested whether JNKi VIII could suppress
LPS plus SB202190-induced apoptosis as SP600125 did. As shown in Figure 3-20b,
JNKi VIII inhibited LPS plus SB202190-induced apoptosis only at < 1 pM and
promoted it at > 20 pM while had slight effects at middle concentrations, suggesting
that higher concentrations of JNKi VIII might have some unexpected functions, such
as interference with the phosphorylation of c-Jun by ERK. Actually, no observed
phosphorylation of c-Jun by ERK in vitro had been explained by a requirement for
further factors, probably including JNK [107]. However, another comprehensive study
might be needed to address this question. Besides U0126 and JNKi VIII, the IKK
inhibitor, PS1145, was also observed to sharply prevent c-Jun phosphorylation at 4
hours in our system (Figure 3-20a). We had no idea so far how PS1145 did that and
whether such inhibition was attributable to its nonspecific targeting or not. However,
all these three inhibitors conformably enhanced LPS plus SB202190-induced
apoptosis (Figure 3-12 and Figure 3-20b), suggesting phosphorylation of c-Jun might
play a crucial anti-apoptotic role in our system as it did in other systems [107, 111].
In light of all the above results, we not only confirmed that ATF-2 rather than
c-Jun was a better choice for our JNK inhibition measurements, but also got some
hints about how the ratio of activated JNK and ERK determines LPS plus SB202190induced apoptosis as our model promised. The balance of ERK-phosphorylated c-Jun
and JNK-phosphorylated ATF-2 and/or other transcription factors might be the next
step.
CHAPTER 4
Conclusions and future directions
4.1. Dynamic behavior of TLR-4 signaling network
As an important sensor for bacterial infection, TLR-4 is the most intensively
studied member of the TLR family since it was first identified in humans in 1997
[112]. Nowadays, a relatively well-rounded TLR-4 signaling network (Figure 1-1) has
already been constructed under the contributions of scientists all over the world,
though some details still remain controversial and missing, providing the fundamental
for us to do systematic analysis of TLR-4 signaling mediated cell responses. Through
the measurements of the phosphorylation levels of 6 kinases distributed in different
branched downstream pathways of TLR-4 signaling network at 6 time points in
response to various combinations of LPS and SB202190 by developed quantitative
western blot, we find that the inhibition of p38 MAPK can extend the phosphorylation
of several kinases including JNK, ERK1/2 and IKKa/p3 to different extent, and our
PCA models further reveal that such extension will probably alter the activities of
these kinases on the timing. In fact, we observed LPS plus SB202190-induced
apoptosis was initiated after 1 hour exactly when the phosphorylation of JNK and
ERK1/2 was extended. Our results also indirectly highlight that negative feedbacks,
such as MKPs [23], in TLR-4 signaling network might regulate cell death as well as
cytokine release. Furthermore, we find both p38 MAPK and PI-3K are required for
the activation of Akt in response to LPS insult. So far we don't know how and why
these two kinases converge on Akt, though the mechanisms of p38 MAPK-mediated
and PI-3K-mediated phosphorylation of Akt have respectively been reported
previously [98, 113]. However, the inhibition ofAkt by wortmannin promoting the
release of some cytokines (e.g. IL-6) but having no significant effect on the secretion
of some other cytokines (e.g. TNF-o) might imply the importance of such complex
regulation ofAkt in deliberately controlling specific cytokine release.
In addition, we also analyze the secretion of IL-6 and TNF-oc and apoptosis
stimulated by the same treatments, and find inhibition of p38 MAPK is required for
induction of apoptosis in LPS-stimulated RAW264.7 cells but will also block
secretion of IL-6 and TNF-a, indicating the critical role of p38 MAPK in these two
processes. Moreover, our PCA model shows that these two cell responses are almost
orthogonal in the loadings plot, suggesting that they might occur independently. In
other words, IL-6 and TNF-a might not be able to affect LPS plus SB202190-induced
apoptosis through autocrine or paracrine pathways within 8 hours, while active
caspase-3 and cleaved PARP would not interfere with the secretion of IL-6 and
TNF-a within 4 hours either.
4.2. Looking into constructed models for potential
applications and future directions
The most important thing that we have done in this study is the successful
construction of several predictive PLSR models and through these models we have
obtained some new insights on TLR-4-signaling mediated cytokine secretion and
apoptosis. Firstly, we saw that the Akt only model can well capture the information of
secretion of IL-6 and TNF-a induced by LPS with or without SB202190 as well as
the full model, but they failed to predict PI-3K inhibitor-treated samples accurately.
This result tells us that the involvement of a signal that is regulated by more than one
pathway simultaneously may seriously reduce the application scope of constructed
models especially when the treatments are not diverse enough to cross all the involved
pathways. Therefore, we might re-build the model using two upstream kinases of Akt,
p38 MAPK and PI-3K instead of Akt in future to get a better predictivity.
Besides, we also find that PLSR models built using the ratio of activated
JNK2 and ERKs, especially ERK , can capture the information of apoptosis induced
by LPS plus SB202190 well. Unlike the Akt only model, these kinase-ratio models
also possess an excellent predictivity on independent datasets, suggesting that LPS
plus SB202190 might really trigger apoptosis in RAW264.7 cells through increase of
the ratio of activated JNK and ERK. Obviously, these two kinases could be developed
as new drug targets to protect the host from being killed by some bacteria, such as
Yersinia and Bacillus anthracis,which can kill macrophages in a manner dependent
on the inhibition of p38 MAPK by their virulence determinants (YopP or YopJ and LT)
[35, 39]. For example, we might use small molecules to inhibit the activation of JNK
or/and promote the activation of ERK through some extracellular signals. Moreover,
Since JNK, ERK and p38 MAPK pathways are shared with other TLRs, it might be
possible to use this mechanism to selectively and rapidly kill some virus-infected cells
in case the activation of p38 MAPK is blocked.
On the other hand, as validated by experiments the JNK2/ERK models
further indicate that LPS plus SB202190-induced RAW264.7 apoptosis is initiated by
1 hour and progresses steadily. That might well explain why the inhibition of p38
MAPK is required for induction of apoptosis, because the inhibition of p38 MAPK
can extend the phosphorylation of JNK and ERK through MKP-1 with different
preferences [22] and accordingly increase their ratio to trigger apoptosis. Both the
extension and apoptosis are shown to be SB202190 dose-dependent (Figure 3-3 and
Figure 3-4). By contrast, the inhibitor of ERK only doesn't have such ability (Figure
3-5) and thereby can not cooperate with LPS to induce apoptosis, though it can
dramatically enhance LPS plus SB202190-induced apoptosis. Probably just because
of this, the importance of ERK in LPS-induced apoptosis was highly undervalued
before. Moreover, these models might also be able to properly answer our question
about why 10 WM SB202190 is enough to induce apoptosis in LPS-stimulated
BMDMs [18], but > 50 pM is required for RAW264.7 and J774.1 cells (Figure 2-5
and reference 61), since the last two cell lines have very high levels of ERK. From
another point of view, it would be great if we can use these PLSR models to predict
the outcome in different types of cells.
In particular, we also discover that the inhibition of ERK can significantly
block c-Jun phosphorylation as well (Figure 3-20a). Considering that phosphorylated
c-Jun is an anti-apoptotic transcription factor, we have proposed a putative model
(Figure 4-1) to describe how the ratio of phosphorylated JNK and ERK determine cell
apoptosis. Simply, activated ERK might phosphorylate c-Jun while activated JNK
might phosphorylate other transcription factors, and the balance of these transcription
factors eventually determine what kind of proteins, anti-apoptotic or pro-apoptotic, to
be expressed. However, we didn't have strong evidences to support this hypothesis so
far. In future, some critical experiments might be designed to test it, including
constructing a c-Jun constitutively activated cell lines to see if it will be resistant to
LPS plus SB202190-induced apoptosis. Studies on these downstream pathways might
also be helpful to address other questions. For example, our kinase-ratio PLSR models
do not include the information of IKK, but inhibition of IKK has been reported to
strongly promote LPS and Yersinia induced apoptosis [41]. Actually, PS1145 sharply
enhanced LPS plus SB202190-induced apoptosis (Figure 3-12), but we also see that
PSI1145 strongly suppress c-Jun phosphorylation (Figure 3-20a). Therefore, we were
wondering if inhibition of IKK promotes apoptosis by suppressing LPS-induced c-Jun
phosphorylation. In addition, do other kinases (e.g. PKR and IRF-3 [18]) that were
reported previously to be required for LPS plus SB202190-induced apoptosis use the
similar mechanisms?
LPS
SB202190
Survival
Figure 4-1. Aputative model for sequential activation of survival and death programs
triggered by LPS plus SB202190
The dashed line indicates activation of MK2 by phosphorylated ERK identified using UpstateTM
anti-phospho-MK2 antibody. Lines terminating in a perpendicular line represent inhibition,
whereas arrows represent activation.
CHAPTER 5
Appendices
5.1. Materials
5.1.1. Cell culture
RAW264.7 cells were purchased from ATCC and grown in Dulbecco's
Modified Eagle's Medium (DMEM, Invitrogen) freshly prepared using endotoxin-free
water and supplemented with 10% heat-inactivated fetal bovine serum, 1.5 g/l sodium
bicarbonate, 100 mg/ml streptomycin and 100 U/ml penicillin. The cultures were
maintained at 37 (C in a humidified 5%CO2 atmosphere.
5.1.2. Drug treatments
The following reagents, unless indicated otherwise, were all purchased from
Sigma-Aldrich. SB202190, SB203580, PD980065, U0126 (Calbiochem), wortmannin,
LY294002, JNKi VIII (Calbiochem), SP600125 and PS 1145 were all dissolved in
dimethyl sulfoxide (DMSO), stored at -20 TC and protected from light. In all
experiments, exponentially growing cells were seeded at an initial cell density of
2x 105/cm2 for the following 24 h of culture and exposed to treatment for the times
indicated.
5.2. Experimental protocols
5.2.1. Repurification of raw LPS [731
At room temperature, 10 mg raw LPS (from Escherichiacoli 055:B5,
Purchased from Sigma-Aldrich, L2880-10MG) was resolved in 2 ml of endotoxin-free
water containing 0.2% triethylamine (TEA) and 0.5% sodium deoxycholate (DOC).
After split into four 500 pl aliquots, 500 pl of water-saturated phenol were added. The
samples were vortexed intermittently for 5 min, and then the phases were allowed to
separate at room temperature for another 5 min. Samples were placed on ice for 5 min,
followed by centrifugation at 40 C for 2 min at 10,000xg. The top aqueous layer was
transferred to a new tube and the left phenol phase was subjected to re-extraction with
500 gl of 0.2% TEA/0.5% DOC. The aqueous phases were pooled and re-extracted
with 1 pl water-saturated phenol again. Subsequently, the pooled aqueous phases were
moved to another new tubes, adjusted to 75% ethanol and 30 mM sodium acetate and
were allowed to precipitate at -20 0 C for I hour. The precipitates were centrifuged at
4TC for 10min at 10,000xg, washed with 1 ml of cold 100% ethanol, and air-dried at
4°C for several hours. After the recovery was calculated (approximate 100% recovery
is usually achieved), the precipitates were resuspended in 0.25% TEA at stock
concentrations of 5 mg/ml or 1 mg/ml and stored at -20 0 C (stable for 2 years) or 4 TC
(stable for 1 month). Repeated freeze/thaw cycles were not recommended.
5.2.2. MTT assay
RAW264.7 cells were seeded into 96-well microtiter plates. 4 hours before
the end of the treatment, 0.5 mg/ml methylthiazolyldiphenyl-tetrazolium bromide
(MTT) was added. At the end of the treatment, the medium was removed with needle
and syringe gently, and then the formazan precipitate was dissolved in 200 gpl DMSO.
After incubation at 37TC for 5 min, the absorbance at 550 nm was measured with
VersaMax TM microplate reader (Molecular Devices Corporation).
5.2.3. 7-AAD & Annexin V-PE double staining
RAW264.7 cells were seeded into 24-well plates. After treatment, cells were
harvested and washed twice ice-cold phosphate-buffered saline (PBS). Cells were
then stained with 7-AAD and Annexin V-PE according to the manufacturer's
instruction (Annexin V-PE apoptosis detection kit I, BD PharmingenTM) and analyzed
by flow cytometry.
5.2.4. Assessment of apoptosis based on active caspase-3
& cleaved PARP
RAW264.7 cells were seeded into 24-well plates. After treatment, cells were
harvested at 40C for 10 min at 3,000rpm and then fixed in100% methanol at -200 C for
at least 12 hours. After methanol was removed and the pellets were washed twice with
ice-cold PBS-T (PBS+0.1% Tween) at 3,000rpm for 5 min, cells were incubated with
primary antibodies, purified rabbit anti-active caspase-3 and mouse anti-cleaved
PARP (Asp214) (1:250 dilution in PBS-T plus 1%BSA, BD PharMingenM), for I
hour at room temperature. Subsequently, cells were washed three times with ice-cold
PBS-T at 3,000rpm for 5 min again and incubated with second antibodies purchased
from Invitrogen TM , Alexa 488-linked goat anti-rabbit IgG (H+L) (1:500 dilution in
PBS-T plus 1%BSA) and Alexa 647-linked anti-mouse IgGI (yl) (1:1000 dilution in
PBS-T plus 1%BSA), for 1 hour at room temperature. After washed four times with
ice-cold PBS-T at 3,000rpm for 5 min, cells were finally resuspended in 250 jl
PBS-T and analyzed by flow cytometry.
5.2.5. Sandwich ELISA
RAW264.7 cells were seeded into 6-well plates. After treatment, their
supernatants were collected at 40C for 5 min at 800xg and stored at -800C for later use.
96-well microtiter plates (Coming 3590, EIA high binding plates) were coated with 50
gl I1 g/ml (diluted in coating buffer 0.1 MNa 2HPO 4, pH 9.0) purified rat anti-mouse
IL-6 (clone: MP5-20F3), rat anti-mouse IL-12 p40/p70 (clone: C15.6) or hamster
ant-mouse/rat TNF (clone: TN3-19.12) (All purchased from BD PharMingen TM ) at
40C overnight. After washed three times with PBS-T (0.05% Tween), the coated plates
were blocked with 200 til PBS plus 1%BSA at 37 oC for 1 hour. After the plates were
washed three times with PBS-T, 50 gl diluted unknown samples (no dilution for IL-6
and IL-12, but 1:20 dilution in PBS plus 1%BSA for TNF-a), blank (PBS plus 1%
BSA) or cytokine standards in 2-fold serial dilutions in PBS plus 1%BSA starting
with: 2 ng/ml (mouse IL-6), I ng/ml (mouse IL-12) and 5 ng/ml (mouse TNF-a) were
added and then incubated at 37 TC for I hour. After washed three times with PBS-T,
the plates were then incubated with biotin rat anti-mouse IL-6 (clone: MP5-32C11),
Biotin rat anti-mouse IL-12 p40/p70 (clone: C 17.8) or Biotin rabbit anti-mouse/rat
TNF (clone: C1150-14) (All purchased from BD PharMingen T ') for 45 min at room
temperature. After washed three times with PBS-T, the plates were then incubated
with 100 l1of extravidin-peroxidase (1:2000 dilution in PBS plus 1%BSA) for 30
min at room temperature. After the plates were washed three times with PBS-T, 100
pM TMB (reagent A 1:1 reagent B) were added to each well. After 10 min, the
reaction was stopped by 100 dl IM phosphoric acid and the absorbance at 450 nm
was measured with VersaMaxTM microplate reader (Molecular Devices Corporation).
5.2.6. Quantitative western blot
Table 5-1. Lysis buffer for quantitative western blot
Total volumes (ml)
Solutions stored at 4 "C
10
ml
5
ml
2
ml
ddH20
5.755
2.877
1.141
0.5 M 0-glycerophosphate, pH 7.3 (50 mM)
1.000
0.500
0.200
0.1 M NaPP (10 mM)
1.000
0.500
0.200
0.50 M NaF (30 mM)
0.600
0.300
0.150
1 M Tris, pH 7.5 (50 mM)
0.500
0.250
0.100
20% Triton X-100 (1%)
5 M NaCl (150 mM)
100 mM benzamidine
500 mM EGTA (2 mM)
.1 M PMSF (1 mM, Note: add < 5 min before use)
Solutions storedat -20 'C
40 mM sodium orthovanadate Na 3VO4 (100 pM)
500 mM DTT (1 mM)
5 mg/ml aprotinin (10 plg/ml)
5 mg/ml leupeptin (10 gig/ml)
1 mg/ml pepstatin (1 jpg/ml)
1 mg/ml Microcystin-LR (1lg/ml)
0.500
0.300
0.100
0.040
0.100
0.250
0.150
0.050
0.020
0.050
0.100
0.060
0.020
0.008
0.020
0.025
0.020
0.020
0.020
0.010
0.010
0.013
0.010
0.010
0.010
0.005
0.005
0.005
0.004
0.004
0.004
0.002
0.002
Note: Final concentrations in the lysis buffer are in parentheses; kindly provided by Nancy Guilllen.
RAW264.7 cells were seeded into 6-well plates. After treatment, cells were
washed twice with ice-cold, lysized with 60 tl/well lysis buffer (Table 5-1) and stored
at -80'C for later use. Protein levels were quantified using Micro BCA TM protein kit
(Pierce). The amounts of loaded proteins for each lane were determined according to
experimental aims and didn't have to be the same. Loaded proteins were then
fractioned by 10% SDS-PAGE, and subsequently transferred to PVDF membranes.
After blocking with TTBS (20 mM Tris-HCI pH 7.4, 0.14 M NaCl, 0.1% Tween 20)
containing 2% ECL Advance TM Blocking Agent (GE Healthcare, UK), the membranes
were incubated with primary antibodies (Table 2-1) at 4 oC overnight. Blots were then
T Advance Western Blotting Detection kit (GE
developed using Amersham ECLM
Healthcare, UK). The net intensities of sample bands were determined using Kodak
Image Station 1000 and analyzed with Kodak ID software.
5.3. PCA and PLSR model construction and refinement
Prior to all analysis, the signaling and cell responses matrices were variance
scaled to non-dimensionalize the different measurements. Both PCA and PLSR
models were constructed in the SIMCA-P 11.5 (Umetrics) commercial software suite
with default settings.
5.4 References
1. Akira S, Uematsu S, and Takeuchi O. Pathogen recognition and innate immunity.
Cell. 124: 783-801. (2006)
2. Liew FY, Xu D, Brint, EK, and O'Neill LAJ. Negative regulation of Toll-like
receptor-mediated immune responses. Nat. Rew. Immunol. 5: 446-458. (2005)
3. Sharma S, tenOever BR, Grandvaux N, Zhou GP, Lin R, and Hiscott J. Triggering
the interferon antiviral response through an IKK-related pathway. Science. 300:
1148-1151. (2003)
4. Sato S, Sugiyama M, Yamamoto M, Watanabe Y, Kawai T, Takeda K, and Akira S.
Toll/IL-1 receptor domain-containing adaptor inducing IFN-P (TRIF) associates
with TNF receptor-associated factor 6 and TANK-binding kinase 1, and activates
two distinct transcription factors, NF-icB and IFN-regulatory factor-3, in the Tolllike receptor signaling. J. Immunol. 171: 4304-4310. (2003)
5. Kawai T, and Akira S. TLR signaling. Cell Death Differ 13: 816-825. (2006)
6. Okkenhaug K, and Vanhaesebroeck B. PI3K in lymphocyte development,
differentiation and activation. Nat. Rev. Immunol. 3: 317-330. (2003)
7. Fukao T, and Koyasu S. PI3K and negative regulation of TLR signaling. Trends
Immunol. 24(7): 358-363. (2003)
8. Guha M, and Mackman N. The phosphatidylinositol 3-kinase-Akt pathway limits
lipopolysaccharide activation of signaling pathways and expression of
inflammatory mediators in human monocytic cells. J. Biol. Chem. 277(35):
22124-32132. (2002)
9. Ojaniemi M, GlumoffV, Harju K, Liljeroos M, Vuori K, and Hallman M.
Phosphatidylinositol 3-kinase is involved in Toll-like receptor-4-mediated
cytokine expression in mouse macrophages. Eur.J. Immunol. 33: 597-605. (2003)
10. Mansell A, Smith R, Doyle SL, Gray P, Fenner JE, Crack PJ, Nicholson SE,
Hilton DJ, O'Neill LAJ, and Hertzog PJ. Suppressor of cytokine signaling 1
negatively regulates Toll-like receptor signaling by mediating Mal degradation.
Nat. Immunol. 7(2): 148-155. (2006).
11. Li X, Tupper JC, Bannerman DD, Winn RK, Rhodes CJ, and Harlan JM.
Phosphoinositide 3 Kinase mediates Toll-like receptor 4-induced activation of
NF-icB in endothelial cells. Infect. Immun. 71(8): 4414-4420. (2003)
12. Hazeki K, Kinoshita S, Matsumura T, Nigorikawa K, Kubo H, and Hazeki O.
Opposite effects of wortmannin and LY294002 on TLR-mediated nitric oxide
production: negative regulation of NF-rB by phosphoinositide 3-kinase. Mol.
Pharmacol.69(5): 1717-1724. (2006)
13. Dey M, Cao C, Dar AC, Tamura T, Ozato K, Sicheri F, and Dever TE. Mechanistic
link between PKR dimerization, autophosphorylation, and eIF2a substrate
recognition. Cell. 122: 901-913. (2005)
14. Garcia MA, Gil J, Ventoso I, Guerra S, Domingo E, Rivas C, and Esteban M.
Impact of protein kinase PKR in cell biology: from antiviral to antiproliferative
action. Microbial.Mol. Biol. Rev. 70(4): 1032-1060. (2006)
15. Andreakos E, Sacre SM, Smith C, Lundberg A, Kiriakidis S, Stonehouse T,
Monaco C, Feldmann M, and Foxwell BM. Distinct pathways of LPS-induced
NF-riB activation and cytokine production in human myeloid and nonmyeloid
cells defined by selective utilization of MyD88 and Mal/TIRAP. Blood 103(6):
2229-2237. (2004)
16. Gusella G.L., Musso T., Rottschafer S.E., Pulkki K., and Varesio L. Potential
requirement of a functional double-stranded RNA-dependent protein kinase (PKR)
for the tumoricidal activation of macrophages by lipopolysaccharide or IFN-a3,
but not IFN-y. J. Immunol. 154: 345-354. (1995)
17. Lang CH, Frost RA, Jefferson LS, Kimball SR, and Vary TC. Endotoxin-induced
decrease in muscle protein synthesis is associated with changes in eIF2B, eIF4E,
and IGF-I. Am JPhysiolEndocrinolMetab. 278: El1133-E1143. (2000)
18. Hsu L-C, Park JM, Zhang K, Luo J-L, Maeda S, Kaufman RJ, Eckmann L,
Guiney DG, and Karin M. The protein kinase PKR is required for macrophage
apoptosis after activation of Toll-like receptor 4. Nature. 428: 341-345. (2004)
19. Lee JH, Park EJ, Kim OS, Kim HY, Joe E-H, and Jou I. Double-stranded
RNA-activated protein kinase is required for the LPS-induced activation of Stat-i
inflammatory signaling in rat brain glial cells. Glia. 50: 66-79. (2005)
20. Gon Y, Nunomura S, and Ra C. Common and distinct signalling cascades in the
production of tumour necrosis factor-a and interleukin-13 induced by LPS in RBL
-2H3 cells. Clin. Exp. Allergy. 35(5): 635-642. (2005)
21. Wang X, and Liu Y. Regulation of innate immune response by MAP kinase
phosphatase-1. Cell. Signal. 19: 1372-1382. (2007)
22. Franklin CC, and Kraft AS. Conditional expression of the mitogen-activated
protein kinase (MAPK) phosphatase MKP-1 preferentially inhibits p38 MAPK
and stress-activated protein kinase in U937 cells. J. Biol. Chem. 272(27):
16917-16923. (1997)
23. Hu J-H, Chen T, Zhuang Z-H, Kong L, Yu M-C, and Liu Y. Feedback control of
MKP-1 expression by p38. Cell. Signal. 19: 393-400. (2007)
24. Feldmann M, and Maini RN. Lasker clinical medical research award. TNF defined
as a therapeutic target for rheumatoid arthritis and other autoimmune diseases. Nat.
Med. 9: 1245-1250. (2003)
25. Pradervand S, Maurya MR, and Subramaniam S. Identification of signaling
components required for the prediction of cytokine release in RAW264.7
macrophages. Genome Biology. 7: R11. (2006)
26. Learn CA, Boger MS, Li L, and McCall CE. The phosphatidylinositol 3-kinase
pathway selectively control sIL-1RA not interleukin-103 production in the septic
leukocytes. J. Biol. Chem. 276: 20234-20239. (2001)
27. Flohe S, Dominguez Fernandez E, Ackermann M, Hirsch T, Borgermann J, and
Schade FU. Endotoxin tolerance in rats: expression of TNF-a, IL-6, IL-10,
VCAM-1 and Hsp70 in lung and liver during endotoxin shock. Cytokine. 11(10):
796-804. (1999)
28. Jacinto R, Hartung T, McCall C, and Li L. Lipopolysaccharide- and lipoteichoic
acid-induced tolerance and cross-tolerance: distinct alterations in IL-I receptor-
associated kinase. J. Immunol. 168: 6136-6141. (2002)
29. Sly LM, Rauh MJ, KalesnikoffJ, Song CH, and Krystal G LPS-induced
upregulation of SHIP is essential for endotoxin tolerance. Immunity. 21: 227-239.
(2004)
30. Ropert C, Closel M, Chaves ACL, and Gazzinelli RT. Inhibition of a p38/stressactivated protein kinase-2-dependent phosphatase restores function of IL-1
receptor-associated kinase-1 and reverses Toll-like receptor 2- and 4-dependent
tolerance of macrophages. J. Immunol. 171:1456-1465. (2003)
31. Kraatz J, Clair L, Rodriguez JL, and West MA. Macrophage TNF secretion in
endotoxin tolerance: role of SAPK, p38, and MAPK. J. Surg. Res. 83(2): 158-164.
(1999)
32. Weinrauch Y, and Zychlinsky A. The induction of apoptosis by bacterial
pathogens. Annu. Rev. Microbiol. 53: 155-187. (1999)
33. Hacker QG
and Fischer SF. Bacterial anti-apoptotic activities. FEMS Microbiol.
Lett. 211: 1-6. (2002)
34. Gao L-Y, and Kwalk YA. The modulation of host cell apoptosis by intracellular
bacterial pathogens. Trends Microbiol. 8(7): 306-313. (2000)
35. Viboud GI, and Bliska JB. Yersinia outer proteins: role in modulation of host cell
signaling responses and pathogenesis. Annu. Rev. Microbiol. 59: 69-89. (2005)
36. Bradley KA, Mogridge J, Mourez M, Collier RJ, and Young JA. Identification of
the cellular receptor for anthrax toxin. Nature. 424: 225-229. (2001)
37. Orth K, Palmer LE, Bao Z-Q, Stewart S, Rudolph AE, Bliska JB, and Dixon JE.
Inhibition of the mitogen-activated protein kinase kinase superfamily by a Yersinia
effector. Science. 285: 1920-1923. (1999)
38. Orth K, Xu Z, Mudgett MB, Bao Z-Q, Palmer LE, Bliska JB, Mangel WF,
Staskawicz B, and Dixon JE. Disruption of signaling by Yersinia effector YopJ, a
ubiquitin-like protein protease. Science. 290: 1594-1597. (2000)
39. Park JM, Greten FR, Li Z-W, and Karin M. Macrophage apoptosis by anthrax
lethal factor through p38 MAP kinase inhibition. Science. 297: 2048-2051. (2002)
40. Zhang Y, and Bliska JB. Role of Toll-like receptor signaling in the apoptotic
response of macrophages to Yersinia infection. Infect. Immun. 71(3): 1513-1519.
(2003)
41. Zhang Y, Ting AT, Marcu KB, and Bliska JB. Inhibition of MAPK and NF-rB
pathways is necessary for rapid apoptosis in macrophages infected with Yersinia. J
Immunol. 174: 7939-7949. (2005)
42. Ruckdeschel K, Harb S, Roggenkamp A, HornefM, Zumbihl R, Kohler S,
Heesemann J, and Rouot B. Yersinia enterocoliticaimpairs activation of
transcription factor NF-rB: involvement in the induction of programmed cell
death and in the suppression of the macrophage tumor necrosis factor a
production. J. Exp. Med 187(7): 1069-1079. (1998)
43. Balachandran S, Kim CN, Yeh W-C, Mak TW, Bhalla K, and Barber GN.
Activation of the dsRNA-dependent protein kinase PKR, induces apoptosis
through FADD-mediated death signaling. EMBO J. 17(23): 6888-6902. (1998)
44. Janes KA, Albeck JG, Gaudet S, Sorger PK, Lauffenburger DA, and Yaffe MB. A
systems model of signaling identifies a molecular basis set for cytokine-induced
apoptosis. Science. 310: 1646-1653. (2005)
45. Janes KA, Kelly JR, Gaudet S, Albeck JQ Sorger PK, and Lauffenburger DA.
Cue-signal-response analysis of TNF-induced apoptosis by partial least squares
regression of dynamic multivariate data. J. Comput. Biol. 11(4): 544-561. (2004)
46. Kemp ML, Wille L, Lewis CL, Nicholson LB, and Lauffenburger DA.
Quantitative network signal combinations downstream of TCR activation can
predict IL-2 production response. J. Immunol. 178: 4984-4992. (2007)
47. Miller-Jensen K, Janes KA, Brugge JS, and Lauffenburger DA. Common effector
processing mediates cell-specific responses to stimuli. Nature. 448: 604-608.
(2007)
48. Janes KA, Gaudet S,Albeck JQ Nielsen UB, Lauffenburger DA, and Sorger PK.
The response of human epithelial cells to TNF involves an inducible autocrine
cascade. Cell. 124: 1225-1239. (2006)
49. Ideker T, and Lauffenburger DA. Building with a scaffold: emerging strategies for
high- to low-level cellular modeling. Trends Biotechnol. 21: 255-262. (2003)
50. Janes KA and Lauffenburger DA. A biological approach to computational models
of proteomic networks. Curr Opin. Chem. Biol. 10(1): 73-80. (2006)
51. Schoeberl B, Eichler-Jonsson C, Gilles ED, and Muller G Computational
modeling of the dynamics of the MAP kinase cascade activated by surface and
internalized EGF receptors. Nat. Biotechnol. 20: 370-375 (2002)
52. Sachs K, Perez O, Pe'er D, Lauffenburger DA, and Nolan GP. Causal protein
-signaling networks derived from multiparameter single-cell data. Science. 308:
523-529. (2005)
53. Hautaniemi S, Kharait S, Iwabu A, Wells A, and Lauffenburger DA. Modeling of
signal-response cascades using decision tree analysis. Bioinformatics.21(9):
2027-2035. (2005)
54. Geladi P, Martens H, Hadjiiski L, and Hopke P. A calibration tutorial for spectral
data. Part 2. Partial least squares regression using Matlab and some neural
network. J. Near InfraredSpectrosc. 4: 243-255. (1996)
55. Wold, S. Cross-validatory estimation of number of components in factor and
principal components models. Technometrics. 20(4): 397-405. (1978)
56. Umetrics AB. User's guide to SIMCA-P, SIMCA-P+ Version 11.0. (2005)
57. Nguyen DV, and Rocke DM. Tumor classification by partial least squares using
microarray gene expression data. Bioinformatics. 18: 39-50. (2002)
58. Trosvik P, Skanseng B, Jakobsen KS, Stenseth NC, Nas T, and Rudi K.
Multivariate analysis of complex DNA sequence electropherograms for
high-throughput quantitative analysis of mixed microbial populations. Appl.
Environ. Microbiol. 73(15): 4975-4983. (2007)
59. Gaudet S, Janes KA, Albeck JG Pace EA, Lauffenburger DA, and Sorger PK. A
compendium of signals and responses triggered by prodeath and prosurvival
cytokines. Mol. Cell. Proteomics.4: 1569-1590. (2005)
60. Westad F, and Martens H. Variable selection in NIR based on significance testing
in partial least squares regression. J Near InfraredSpectrosc. 8: 117-124. (2000)
61. Karahashi H, Nagata K, Ishii K, and Amano F. A selective inhibitor of p38 MAP
kinase, SB202190, induced apoptotic cell death of a lipopolysaccharide-treated
macrophage-like cell line, J744.1. Biochim. Biophys. Acta. 1502: 207-223. (2000)
62. Hassan F, Islam S, Koide N, Mu, MM, Ito H, Mori I, Yoshida T, and Yokochi T.
Role of p38 mitogen-activated protein kinase (MAPK) for vacuole formation in
lipopolysaccharide (LPS)-stimulated macrophages. Microbiol.Immunol. 48(11):
807-815. (2004)
63. Mayer H, Tharanathan RN, and Weckesser J. Analysis of lipopolysaccharides of
Gram-negative bacteria. Methods Microbiol. 18: 157-207. (1985)
64. Nagai Y,Akashi S, Nagafuku M, Ogata M, Iwakura Y, Akira S, Kitamura T,
Kosugi A, Kimoto M, and Miyake K. Essential role of MD-2 in LPS
responsiveness and TLR4 distribution. Nat. Immunol. 3(7): 667-672. (2002)
65. Beutler B, and Rietschel ET. Innate immune sensing and its roots: the story of
endotoxin. Nat. Rew. Immunol. 3: 169-176. (2003)
66. Manthey CL, and Vogel SN. Elimination of trace endotoxin protein from rough
chemotype LPS. J Endotoxin Res. 1: 84-91. (1994)
67. Morrison DC, Betz SJ, and Jacobs DM. Isolation of a lipid A bound polypeptide
responsible for "LPS-initiated" mitogenesis of C3H/HeJ spleen cells. J. Exp. Med
144(3): 840-846. (1976)
68. Manthey C, Perera PY, Henricson BE, Hamilton TA, and Qureshi N, and Vogel
SN. Endotoxin-induced early gene expression in C3H/HeJ (Lpsd) macrophages. J.
Immunol. 153: 2653-2663. (1994)
69. Hogan MM, and Vogel SN. Production of tumor necrosis factor by rIFN-y-primed
C3H/HeJ (Lpsd) macrophages requires the presence of lipid A-associated proteins.
J. Immunol. 141: 4196-4202. (1988)
70. Kirschning CJ, Wesche H, Ayres TM, and Rothe M. Human Toll-like receptor 2
confers responsiveness to bacterial lipopolysaccharide. J. Exp. Med 188:
2091-2097. (1998)
71. Yang RB, Mark MR, Gray A, Huang A, Xie MH, Zhang M, Goddard A, Wood WI,
Gurney AL, and Godowski PJ. Toll-like receptor-2 mediates lipopolysaccharideinduced cellular signalling. Nature. 395: 284-288. (1998)
72. Chow JC, Young DW, Golenbock DT, Christ WJ, and Gusovsky F. Toll-like
receptor-4 mediates lipopolysaccharide-induce signal transduction. J. Biol. Chem.
274: 10689-10692. (1999)
73. Hirschfeld M, Ma Y, Weis JH, Vogel SN, and Weis JJ. Cutting edge: repurification
of lipopolysaccharide eliminates signaling through both human and murine
Toll-like receptor 2. J. Immunol. 165: 618-622. (2000)
74. Meier P, Finch A, and Evan G. Apoptosis in development. Nature. 407: 796-801.
(2000)
75. Krysko DV, Berghe TV, D'Herde K, and Vandenabeele P. Apoptosis and necrosis:
Detection, discrimination and phagocytosis. Methods. 44: 205-221. (2008)
76. Jacobson MD, Weil M, and Raff MC. Programmed cell death in animal
development. Cell. 88: 347-354. (1997)
77. Hengartner MO. The biochemistry of apoptosis. Nature. 407: 770-776. (2000)
78. Tewari M, Quan LT, O'Rourke K, Desnoyers S, Zeng Z, Beidler DR, Poirier GQ
Salvesen GS, and Dixit VM. Yama/CPP32 beta, a mammalian homolog of CED-3
is a CrmA-inhibitable protease that cleaves the death substrate poly(ADP-ribose)
polymerase. Cell. 81: 801-809. (1995)
79. Saxena RK, Vallyathan V, and Lewis DM. Evidence for lipopolysaccharideinduced differentiation of RAW264.7 murine macrophage cell line into dendritic
like cells. J Biosci. 28: 129-134. (2003)
80. Lieberthal W, Triaca V, and Levine JS. Mechanisms of death induced by cisplatin
in proximal tubular epithelial cells: apoptosis vs. necrosis. Am JPhysiol.270:
F700-708. (1996)
81. Liu M-J, Wang Z, Li H-X, Wu R-C, Liu Y-Z, and Wu Q-Y. Mitochondrial
dysfunction as an early event in the process of apoptosis induced by woodfordin I
in human leukemia K562 cells. Toxicol. Appl. Pharmacol. 194(2): 141-155. (2004)
82. Akira S, and Kishimoto T. IL-6 and NF-IL6 in acute-phase response and viral
infection. Immunol. Rev. 127: 25-50. (1992)
83. Yun CH, Lundgren A, Azem J, Sjoling A, Holmgren J, Svennerholm, A-M, and
Lundin BS. Natural killer cells and Helicobacterpyloriinfection: bacterial
antigens and interleukin-12 act synergistically to induce gamma interferon
production. Infect. Immun. 73(3): 1482-1490. (2005)
84. Locksley RM, Killen N, and Lenardo MJ. The TNF and TNF receptor
superfamilies: integrating mammalian biology. Cell. 104(4): 487-501. (2001)
85. van den Blink B, Juffermans NP, Hove T, Schultz MJ, van Deventer SJH, van der
Poll T, and Peppelenbosch MP. p38 mitogen-activated protein kinase inhibition
increases cytokine release by macrophages in vitro and during infection in vivo. J
Immunol. 166: 582-587. (2001)
86. Saito S, Matsuura M, and Hirai Y. Regulation of lipopolysaccharide-induced
interleukin-12 production by activation of repressor element GA-12 through
hyperactivation of the ERK pathway. Clin. Vaccine Immunol. 13(8): 876-883.
(2006)
87. Novoa I, Zeng H, Harding HP, and Ron D. Feedback inhibition of the unfolded
protein response by GADD34-meidated dephosphorylation of eIF2a. J. Cell Biol.
153(5): 1011-1021. (2001)
88. Pennisi E. Systems biology. Tracing life's circuitry. Science. 302: 1646-1649.
(2003)
89. Rieckmann P, Alessandro FD, Nordan RP, Fauci AS, and Kehrl JH. IL-6 and
tumor necrosis factor-alpha. Autocrine and paracrine cytokines involved in B cell
function. J. Immunol. 146(10): 3462-3468. (1991)
90. Davies SP, Reddy H, Caivano M, and Cohen P. Specificity and mechanism of
action of some commonly used protein kinase inhibitors. Biochem. J. 351: 95-105.
(2000)
91. Hall JP, and Davis, RJ. Inhibition of the p38 pathway upregulates macrophage
JNK and ERK activities, and the ERK, JNK, and p38 MAP kinase pathways are
reprogrammed during differentiation of the murine myeloid Ml cell line. J. Cell.
Biochem. 86: 1-11 (2002)
92. Tanoue T, and Nishida E. Molecular recognitions in the MAP kinase cascades.
Cell. Signal. 15: 455-462. (2003)
93. ter Haar E, Prabakhar P, Liu X, and Lepre C. Crystal structure of the
p38a0-MAPKAP kinase 2 heterodimer. J Biol. Chem. 282(13): 9733-9739. (2007)
94. Stokoe D, Campbell DG, Nakielny S, Hidaka H, Leevers SJ, Marshall C, and
Cohen P. MAPKAP kinse-2; a novel protein kinase activated by mitogen-activated
protein kinase. EMBO J. 11(11): 3985-3994. (1992)
95. Coxon PY, Rane MJ, Uriarte S, Powell DW, Singh S, Butt W, Chen Q, and
McLeish KR. MAPK-activated protein kinase-2 participates in p38 MAPKdependent and ERK-dependent functions in human neutrophils. Cell. Signal. 15:
993-1001. (2003)
96. Morton S, Davis RJ, McLaren A, and Cohen P. A reinvestigation of the multisite
phosphorylation of the transcription factor c-Jun. EMBO J. 22(15): 3876-3886.
(2003)
97. Ignatoski KMW, Livant DL, Markwart S, and Grewal NK. The role of
phosphatidylinositol 3'-kinase and its downstream signals in erbB-2-meidated
transformation. Mol. CancerRes. 1: 551-560. (2003)
98. Rane MJ, Coxon RY, Powell DW, Webster R, Klein JB, Pierce W, Ping P, and
McLeish KR. p38 kinase-dependent MAPKAPK-2 activation functions as
3-phosphoinositide-dependent kinase-2 for Akt in human neutrophils. J. Bio.
Chem. 276(5): 3517-3523. (2001)
99. Wu R, Kausar H, Johnson P, Montoya-Durango DE, Merchant M, and Rane MJ.
Hsp27 regulates Akt activation and polymorphonuclear leukocyte apoptosis by
scaffolding MK2 to Akt signal complex. J. Biol. Chem. 282(30): 21598-21698.
(2007)
100.
Zheng C, Lin Z, Zhao ZJ, Yang Y, Niu H, and Shen X. MAPK-activated
protein kinase-2 (MK2)-mediated formation and phosphorylation-regulated
dissociation of the signal complex consisting of p38, MK2, Akt, and Hsp27. J.
Biol. Chem. 281(48): 37215-37226. (2006)
101.
Geladi P, and Martens H. A calibration tutorial for spectral data. Part 1. Data
pretreatment and principal component regression using Matlab. J Near Infrared
Spectrosc. 4: 243-255. (1996)
Bain J, Plater L, Elliott M, Shpiro N, Hastie CJ, Mclauchlan H, Klevernic I,
102.
Arthur JSC, Alessi DR, and Cohen P. The selectivity of protein kinase inhibitors: a
further update. Biochem. J. 408: 297-315. (2007)
Stone AA, and Chambers TC. Microtubule inhibitors elicit differential effects
103.
on MAP kinase (JNK, ERK and p38) signaling pathways in human KB-3
carcinoma cells. Exp. Cell Res. 254: 110-119. (2000)
104.
Jarvis WD, Fornari FA, Auer KL, Freemerman AJ, Szabo E, Birrer MJ,
Johnson CR, Barbour SE, Dent P, and Grant S. Coordinate regulation of stressand mitogen-activated protein kinases in the apoptotic actions of ceramide and
sphingosine. Mol. Pharmacol. 52: 935-947. (1997)
Bennett BL, Sasaki DT, Murray BW, O'Leary EC, Sakata ST, Xu W, Leisten
105.
JC, Motiwala A, Pierce S, Satoh Y, Bhagwat SS, Manning AM, and Anderson DW.
SP600125, an anthrapyrazolone inhibitor of Jun N-terminal kinase. Proc.Natl.
Acad. Sci. USA. 13681-13686. (2001)
106.
Vlahopoulos SA, Logotheti S, Mikas D,Giarika A, Gorgoulis V, and
Zoumpourlis V. The role ofATF-2 in oncogenesis. BioEssays. 30: 314-327. (2008)
107.
Leppa S, Saffrich R, Ansorge W, and Bohmann D. Differential regulation of
c-Jun by ERK and JNK during PC12 cell differentiation. EMBO J. 17(15):
4404-4413. (1998)
108.
Fuchs SY, and Ronai Z. Ubiquitination and degradation of ATF-2 are
dimerization dependent. Mol. Cell. Biol. 19(5): 3289-3298. (1999)
109.
Pulverer BJ, Kyriakis JM, Avruch J,Nikolakaki E,and Woodgett JR.
Phosphorylation of c-Jun mediated by MAP kinases. Nature. 353: 670-674. (1991)
110.
Szczepankiewicz BQ Kosogof C, Nelson LTJ, Liu G Liu B, Zhao H, Serby
MD, Xin Z, Liu M, Gum RJ, Haasch DL, Wang S, Clampit JE, Johnson EF,
Lubben TH, Stashko MA, Olejniczak ET, Sun C, Dorwin SA, Haskins K,
Abad-Zapatero C, Fry EH, Hutchins CW, Sham HL, Rondinone CM, Trevillyan
JM. Aminopyridine-based c-Jun N-terminal kinase inhibitors with cellular activity
and minimal cross-kinase activity. J.Med Chem. 49: 3563-3580. (2006)
111.
Smeal T, Binetruy B, Mercola D, Grover-Bardwick A, Heidecker QG
Rapp
UR, and Karin M. Oncoprotein-mediated signalling cascade stimulates c-Jun
activity by phosphorylation of serines 63 and 73. Mol Cell. Biol. 12(8): 3507-3513.
(1992)
112.
Medzhitov R, Preston-Hurlburt P, and Janeway CA Jr. A human homologue
of the Drosophila Toll protein signals activation of adaptive immunity. Nature. 388:
394-397. (1997)
Song GQ Ouyang G,and Bao S. The activation of Akt/PKB signaling pathway
113.
and cell survival. J. Cell. Mol. Med. 9(1): 59-71. (2005)