Quantitative Analysis of Adenoviral Vector Modification of a CytokineMediated Cell Death Decision
by
Kathryn E. Miller
A.B., Engineering Sciences, 1997
B.E., Chemical Engineering, 1998
Dartmouth College
SUBMITTED TO THE DEPARTMENT OF CHEMICAL ENGINEERING IN
PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY IN CHEMICAL ENGINEERING
AT THE
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
JUNE 2006
© 2006 Massachusetts Institute of Technology. All rights reserved.
,t
Signature of Author:
k '
J
A
w
-
X.4
'-
.
,
,
( Department of Chemical Engineering
April 24, 2006
v
, X
Certified by:
g
/ I.-/'
Professor of Chemical Engineering, Bio
Douglas A. Lauffenburger
cal Engineering, and Biology
Thesis Supervisor
Accepted by:
William M. Deen
Professor of Chemical Engineering
Chairman, Committee for Graduate Students
MASSACHUSETTS INSTTUME
OF TECHNOLOGY
ARCHIVES
1
JUN 13 2006
LIBRARIES
Quantitative Analysis of Adenoviral Vector Modification of a CytokineMediated Cell Death Decision
By
Kathryn E. Miller
Submitted to the Department of Chemical Engineering
on April 24, 2006 in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy in
Chemical Engineering
ABSTRACT
Intracellular networks arise from complex interactions between proteins that relay
signals and control cellular responses. Viruses, with limited genetic material, can modify
network signals and change cell behavior. Replication-deficient viruses are used
extensively as delivery vectors in clinical gene therapy and in molecular biology, but
little is known about how the viral carrier itself contributes to cellular responses.
In this thesis, we explored the link between viral vector modifications of signaling
networks to changes in cellular phenotype. We approached this problem by studying a
therapeutically relevant model in which an adenoviral vector (Adv) sensitizes human
tumor epithelial cells to tumor necrosis factor (TNF)-induced apoptosis. We first
measured TNF-stimulated signaling profiles over a range of Adv infection levels for a
distribution of kinases centrally involved in the TNF signaling network. We then applied
quantitative analytical techniques to determine the most important signals contributing to
Adv-induced changes in TNF-mediated apoptosis.
We experimentally derived a mathematical equation describing the saturation of
anti-apoptotic Akt effector signaling in the presence of high levels of Adv infection,
which could predict TNF-induced apoptosis in HT-29 cells. However, the same equation
did not apply in HeLa cells, suggesting that one-signal models are insufficient to account
for complex network interactions. Therefore, we applied a systems-modeling approach
to our Adv-TNF system and mathematically identified a multivariate signal-processing
function sufficient to predict Adv-TNF induced apoptosis in both HT-29 cells and HeLa
cells. The common-processing model identified critical Adv-induced cell-specific
signaling modifications, and accurately predicted apoptosis following perturbation with
pharmacological inhibitors of Akt and IKK. Thus, by combining experimental and
computational approaches, this thesis has identified an important biological principle,
common signal processing, for studying cell-specific responses to viral infections and
rational drug therapies.
Thesis Supervisor: Douglas A. Lauffenburger
Title: Professor of Chemical Engineering, Biological Engineering, and Biology
2
BIBLIOGRAPHIC NOTE
Kathryn Miller received a bachelor of arts in Engineering Sciences in 1997 from
Dartmouth College in Hanover, NH. While at Dartmouth, Kathryn worked in the
laboratory of Dr. Lee Lynd studying recombinant microorganisms for converting
cellulosic biomass into ethanol. Upon graduation, she was awarded a fellowship to
complete a one-year bachelor's degree in Chemical Engineering at the Thayer School of
Engineering at Dartmouth College, which she was awarded in 1998. Following receipt
of her degrees, she worked for two years at the Monitor Consulting Group in Cambridge,
MA.
Kathryn started her graduate work in the Chemical Engineering Department at
MIT in the fall of 2000. Under the supervision of Dr. Douglas Lauffenburger, she
completed her thesis entitled "Quantitative Analysis of Adenoviral Vector Modification
of a Cytokine-Mediated Cell Death Decision." For three years, she was supported by an
NIH Biotechnology Training Grant. As part of this fellowship, she spent six months in
2004 at Merck & Co. under the supervision of Dr. Jean-Luc Bodmer, working on an
adenoviral HIV vaccine project. In January through March of 2006, Kathryn had the
opportunity to pursue her long-time interest in science and technology policy as a
Christine Mirzayan Science and Technology Fellow at the National Academy of Science.
Upon leaving MIT, Kathryn will take a postdoctoral position in the laboratory of
Dr. David V. Schaffer at the University of California at Berkeley.
3
ACKNOWLEDGEMENTS
Although many consider a doctoral thesis a solitary endeavor, it is usually
accomplished with the support of many, and mine is no exception. Family, friends,
teachers, collaborators, and lab members, by no means mutually exclusive categories, all
contributed to this accomplishment. Specifically, I thank my advisor, Dr. Douglas
Lauffenburger, for his unswerving encouragement and invaluable professional guidance
throughout my thesis work. In addition, I thank Dr. Kevin Janes, a collaborator and
mentor from whom I have learned a great deal about executing and communicating high
quality scientific research.
For particular aspects of this work, I think Yun-Ling Wong for data on the
infection protocol optimization; John Albeck for help with apoptosis assays; Nathan
Tedford for help with quantitative PCR; and Suzanne Gaudet for help with Western blots.
I thank Kevin Janes, my collaborator for the common effector-processing analysis. And
along with all those already mentioned, I also thank many other members of the
Lauffenburger group for helpful conversations and for making the lab a fun place to
work.
Finally, I thank my parents, Ronald and Joyce Miller, who in addition to offering
me their love and support, instilled in me an ethic of work and perseverance that was
indispensable in completing this thesis. And I thank my husband, Kyle Jensen, who has
given me the love and confidence to continue to tackle the challenge of scientific research
for the foreseeable future.
4
TABLE OF CONTENTS
1.
Introduction .....................................................................
9
1.1.
Adenoviruses as tools for biological discovery and gene therapy ...................... 9
1.1.1.
Introduction ..............................................................................................
9
1.1.2.
Adenovirus biology and vector design ...................................................... 9
1.1.3.
Gene therapy challenge: Adenovirus and the immune system ................ 13
1.2.
Signaling network interactions between virus and host .................................. 13
1.2.1.
Introduction ............................................................................................
14
14
1.2.2.
TNF-activated signaling and apoptosis .........................................
1.2.3.
Transcription-dependent mechanisms of wild-type Ad to counteract TNF
15
1.2.4.
The intersection of Adv-TNF signaling .................................................. 18
1.3.
A systems biology approach to understand how viruses perturb cell response
functions
...........................................................
20
1.3.1.
Introduction ..........................................................
20
Model of a cytokine-growth factor signaling network ............................ 21
1.3.2.
1.3.3.
Challenges in virus-host cell signaling analysis ...................................... 23
2. Experimental methods for quantitative Adv-TNF network analysis ....................... 24
2.1.
Introduction ...........................................................
24
2.2.
Optimization of the Adv-TNF sensitization protocol ..................................... 25
2.2.1.
Apoptosis ...............................................................................................
25
2.2.2.
Infection efficiency ...........................................................
28
2.3.
Large-scale signaling data collection ........................................
3...............31
2.3.1.
A high-throughput multiplex radioactivity-based kinase activity assay for
quantitative signaling analysis ............................................................................... 31
2.3.2.
Quantitative western blots measurements ............................................... 35
2.4.
Partial least squares regression analysis ........................................................ 37
2.4.1.
Measurement processing and metric extraction ....................................... 37
2.4.2.
Predictions with TNF-GF compendium ................................................. 38
2.5.
Other biological considerations ..........................................................
39
2.5.1.
Adenoviral replication in carcinoma cell lines ........................................ 40
2.5.2.
Transcriptional shedding of adenoviral backbone proteins ...................... 41
2.5.3.
Helper-dependent adenoviral vector infection to test effect of viral
transcription
. .........................................................42
3.
Insight into the mechanism of Adv-TNF synergy through quantitative signaling
measurements
...........................................................
45
3.1.
Introduction ...........................................................
45
Adv modification of the TNF-induced signaling network ............................... 45
3.2.
3.3.
Role of Akt in TNF sensitization ...........................................................
49
3.3.1.
Upregulation of endogenous Akt activity in Adv-infected cells by insulin
49
3.3.2.
Saturation of downstream Akt effectors by Adv ..................................... 51
3.4.
Construction of a global Akt-survival dose-response curve combining Adv- and
IFN-mediated sensitization ........................................................
53
3.5.
Further considerations .........................................
55
5
4. Testing a fundamental biological principle through data-driven modeling of
Adv-TNF synergy ...............................................................
58
4.1.
Introduction: cell-specific versus common processing ................................... 58
4.2.
Experimental and computational approach to test common effector-processing
mechanism ................................................................................................................
59
4.2.1.
Signaling inputs from multiple cell lines................................................. 59
4.2.2.
PLS model construction and testing through prediction .......................... 63
4.2.3.
Biological insights through principal component analysis ....................... 65
4.3.
Common processing model as a tool for designing rational drug therapies ..... 70
4.3.1.
Predictions of Akt inhibition test robustness of the common effector70
processing model ..............................................................
4.3.2.
Common effector-processing model predicts differential effects of IKK
inhibition in HeLa cells and HT-29 cells ..............................................................
73
4.4.
Further considerations ..............................................................
77
5. Conclusions and future directions ..............................................................
79
5.1.
Contributions to signaling network analysis ................................................... 79
5.2.
Application to the development of Adv-mediated cancer gene therapy ........... 80
81
5.3.
Ongoing and future directions ...............................................................
Adenoviral gene therapy in the liver ....................................................... 81
5.3.1.
5.3.2.
NF-KB signaling in HIV latency ............................................................. 82
84
.
......... ................................................
6. Appendices ........................
6.1.
Materials........................................................................................................84
6.1.1.
Adenoviral vectors ..............................................................
84
....................... 84
Experimental Protocols ........................................
6.2.
6.2.1.
Cell culture and infection .........................................
84
6.2.2.
Beta-galactosidase Assays .........................................
85
6.2.3.
Apoptosis assays .................................................................................... 85
6.2.4.
Kinase activity assays ............................................................................. 86
Quantitative polymerase chain reaction (PCR) ....................................... 87
6.2.5.
6.2.6.
RNA isolation and RT-PCR .........................................
87
6.2.7.
Western blots .........................................
88
6.2.8.
PLSR model construction and refinement .
.............................
89
6.2.9.
Statistical analysis ........................................
90
6.3.
References ........................................
91
6
LIST OF FIGURES AND TABLES
Fig. 1-1. Schematic of the adenovirus particle .............................................................
.. i
Fig. 1-2. Schematic map of the wild type adenovirus genome and adenovirus-derived
vectors .............................................................
12
Fig. 1-3. Schematic representation of the signaling pathways induced by TNF and Adv 16
Table 1-1. Evidence for TNF and Adv activation of the protein kinase network
controlling apoptosis ..............................................................................................
19
Fig. 2-1. General schematic of the Adv-TNF sensitization protocol .............................. 26
Fig. 2-2. Adv infection sensitizes human epithelial cells to TNF-a-mediated apoptosis 27
Fig. 2-3. High Adv concentrations are required for efficient infection of HT-29 cells ... 28
Fig. 2-4. Sensitization dose response ..............................................................
29
Fig. 2-5. Adv sensitizes multiple human cell lines to TNF-induced apoptosis ................ 30
Fig. 2-6. Kinase activities are information-rich measures of signaling networks ............31
Fig. 2-7. General schematic of the high-throughput kinase activity assay....................... 32
Fig. 2-8. Signaling time courses following TNF-activation in uninfected and Advinfected cells .............................................................
34
Fig. 2-9. Caspase 8 cleavage in Adv-TNF HT-29 cells ................................................. 36
Table 2-1. Signaling metrics extracted from dynamic network measurements ............... 38
Fig. 2-10. TNF-growth factor model predicts Adv-TNF-induced apoptosis but cannot
differentiate uninfected cells .............................................................
39
Fig. 2-11. HT-29 cells support Adv genome replication ................................................. 41
Fig. 2-12. Adv E4 shedding in HT-29 cells in consistent with increased Akt activity ..... 42
Fig. 2-13. Helper-dependent Adv reduces but does not eliminate sensitivity to TNF......43
Fig. 3-1. Adv infection upregulates TNF-induced pro- and anti-apoptotic signaling ....... 47
Fig. 3-2. Adv infection alone upregulates Akt but not other TNF-growth factor-activated
pathways
...........................................................
48
Fig. 3-3. Insulin increases Akt activity but does not reverse TNF-induced apoptosis or
reduce MK2 activity in Adv-infected cells ........................................
...............50
Fig. 3-4. Adv infection upregulates multiple Akt-pSubs before the addition of TNF and
saturates phosphorylation of downstream Akt effectors .
7
..............................
53
Fig. 3-5. Adv-infected are trapped near the plateau of a global Akt-survival doseresponse curve ..............................................................
54
Fig. 4-1. Hypothesis description: Cell-specific processing versus common processing. 59
Fig. 4-2. Cell-specific apoptotic responses to Adv-TNF stimulation of HT-29 and HeLa
cells ..........................................................
60
Fig. 4-3. Cell-specific signaling responses to Adv infection and TNF stimulation of HT29 and HeLa cells ...............................................................
61
Fig. 4-4. Colorgram of time-dependent kinase activities in HT-29 and HeLa ................. 62
Fig. 4-5. Test of a common effector-processing hypothesis in epithelial cells treated with
Adv and TNF ........................................
...............................................................64
Table 4-1. Top 20 kinase metrics contributing to the JNK1-IKK axis of the commonprocessing model ..............................................................
65
Table 4-2. Top 20 kinase metrics contributing to the Akt-MK2 axis of the commonprocessing model ...............................................................
66
Fig. 4-6. Reduced principal-component dimensions of the common-processing model.. 67
Fig. 4-7. Common-processing dimensions differentiate apoptotic role for early and late
IKK activity ........................................................................
68
Fig. 4-8. HPV E6 in HeLa cells may cooperate with Ad E40RF1 in mediating TNF
response .
.............................................................
69
Fig. 4-9. Common effector processing uniquely predicts resistance of Adv-infected HeLa
cells to pharmacological inhibition of PI3K ..........................................................
72
Fig. 4-10. Deduction of cell-specific sensitivity to early-phase IKK inhibition ............. 74
Fig. 4-11. HeLa cells are more sensitive to IKK-NF-KB inhibition than HT-29 cells ..... 75
Fig. 4-12. SC-514 inhibition in vivo causes hyperactivation of IKK after removal ......... 77
8
CHAPTER
1
1. Introduction
1.1. Adenoviruses as tools for biological discovery and gene therapy
1.1.1. Introduction
Recombinant adenoviruses are commonly used as delivery vectors in gene
therapy clinical trials and in basic bioscience studies for the delivery of transgenes. In
both of these areas, it is important to understand how the adenoviral vector (Adv) alters
cells during infection. Adv-induced cellular changes might synergize or antagonize the
effects of certain transgenes, confounding interpretation of their normal role in cells.
Deconstructing the contributions of Adv and the transgene is also relevant in vivo,
because gene products intrinsic to Adv may cause host cells to react differently to
proinflammatory cytokines that circulate during infection [1]. The virology of wild type
adenovirus has been studied extensively [2], but there is not the same molecular-level
understanding of the engineered Advs used in gene therapy applications. In this
Introduction, we describe the complexity of the interaction network between wild type
Ad and recombinant Adv and inflammatory cytokines of the immune system. We then
discuss quantitative systems biology approaches that could be useful in deconstructing
how these interactions contribute to changes in cellular response.
1.1.2. Adenovirus biology and vector design
Adenoviridae have been isolated from multiple species and tissue types, and have
been extensively studied and developed for gene therapy applications (for a full review of
adenovirus biology, see [2]). The human adenovirus (Ad) family consists of more than
9
50 serotypes that can infect and replicate in a wide range of tissues including the
respiratory tract, the gastrointestinal tract, the eye, and the liver. Adenoviruses have
nonenveloped capsids with a 30- to 40-kb linear double-stranded DNA genome that
encodes for more than 50 proteins through extensive splicing.
The adenovirus has an icosahedral capsid composed of hexon proteins coming
together at a point to join with the penton base (Fig. 1-1). The fiber shaft extends from
the penton base, ending in the fiber knob. The fiber knob binds to the coxsackie and
adenovirus receptor (CAR) found on most cell types [3], but entry of the Ad particle
proceeds via endocytosis mediated by the av
3 /5
integrins [4]. Subsequently, Ad capsids
are transported to the host nucleus where they insert their DNA and initiate the
replication cycle.
Ad binding and entry initiate a number of signaling pathways (Fig. 1-3, right
side). Internalization via av integrins requires association with the Crk-associated
substrate p 130CAS and activation of phosphotidylinositol-3-OH kinase (P13K) [5, 6]. In
addition, the Ad capsid stimulates activation of the MKK6-p38-MK2 pathway which
enhances microtubule-mediated viral nuclear targeting [7]. Inhibition of either of these
pathways significantly inhibits infection.
The Ad lifecycle occurs in an early and late phase, divided by the onset of viral
replication. This division also describes how Ad regulates cell viability, inhibiting
apoptosis in the early stages of infection such that viral replication is maximized, and
promoting apoptosis (and ultimately cell lysis) late in infection to release progeny virions
[8]. To execute this lifecycle, Ad genes are transcribed in a complex temporal manner
and are divided on this basis into three major groups: early (E1A, El B, E2, E3, and E4),
intermediate or delayed (IX and Iva2) and the major late transcription unit (processed in 5
mRNAs L1-L5) [2, 8] (Fig. 1-2). The EIA proteins function to trans-activate the other
Ad early transcription units (EIB, E2, E3, and E4) to induce the cell to enter S phase in
order to create an environment optimal for virus replication [9]. The E2 region encodes
proteins necessary for replication of the viral genome, including the DNA polymerase,
the preterminal protein, and a DNA-binding protein (for a more detailed review, see
[10]). Proteins of the E3 region function primarily to subvert the host immune response
and E4 proteins regulate the cell cycle, both contributing to a productive infection [1 ].
10
Fig. 1-1. Schematic of the adenovirus particle
Major proteins of the capsid (fiber, hexon, penton base) and core (protein V, protein VII, tenninal protein)
are shown. Adapted from [11].
Proteins of the late genome region are primarily involved in vector packaging.
Adenoviruses have been extensively studied, designed and redesigned for use as
gene therapy delivery vectors, due to several attractive features including their wide
tropism (range of cell types and tissues in which a virus can sustain a productive
infection), ease of production, and large gene carrying capacity [12]. Adenoviral vectors
comprise 25% of all past and current gene therapy clinical trials, second only to retroviral
vectorsl. The most commonly used Adv is the so-called first-generation
Adv (FG-Adv).
This virus has been deleted for the adenoviral early regions £ 1 and E3, eliminating or
greatly impairing viral replication.
Despite this genetic modification, £ l/E3-deleted
Advs are not biologically inert. Cells infected with £ 1!E3-deleted Adv still express low
levels of other wild-type gene products, which are known to cause potent immunogenic
responses [13]. In an attempt to mitigate this problem, £2 and/or £4 coding sequences
were removed from second-generation
Advs [13]. These vectors have shown reduced
toxicity in vivo [14, 15]. Construction and production of these multiply-deleted
I
Genetic Modification
Clinical Research Infonnation
System, htto://www.l!emcris.od.nih.l!ov.
11
viruses
MLP )
:1A
L1
L2
L3
.
L
1R
L4
L5
-
-
Wild type
I
I_
ITR
First
generation
E2
E4
ITR
PI
AEl
ITR
Second
p
generation
ITR
Helper-
AE3
AE1
AE2
AE4
I
ITR
'
dependent
AE3
ITR
I
ITR
ITR
Fig. 1-2. Schematic map of the wild type adenovirus genome and adenovirus-derived vectors
Adapted from [11]. The central, solid line represents the viral genome. Positions of the left and right ITRs,
the packaging sequence (P), early transcription units (EIA, E1B, E2, E3, and E4), and the major late
transcription unit (major late promoter [MLP], L1-L5) are shown. Arrows indicate the direction of
transcription. Below are genome structures of first-generation, second-generation, and helper-dependent
vectors. Open boxes indicate regions that have been deleted.
is difficult, however, due to the need for isolation of cell lines expressing the missing
functions in trans [1 ]. The development of helper-dependent Adv (HD-Adv), vectors
devoid of all viral genes, has resulted in the most significant improvements in Adv safety
and efficacy [16]. A direct in vivo comparison between an FG-Adv and HD-Adv
carrying the same expression cassette showed that TNF and IL-6 levels were upregulated
in response to the FG-Adv but not for the HD-Adv [ 17]. However, because these vectors
have no coding sequences, a helper-virus is required for their propagation, and the
resulting vector product is invariably contaminated with helper-virus [ 16]. In addition,
acute toxicity is still present after high-dose systemic injection of HD-Advs into
nonhuman primates [18], due to the intact innate immune response that is induced by the
Ad capsid (discussed below). While the development of helper-dependent Advs
continues to progress, to date this type of vector has been used only once in a clinical
trial 1 .
12
1.1.3. Gene therapy challenge: Adenovirus and the immune system
A major limitation in the application of Adv gene therapy is the activation of the
host immune response. Much work has been done to decouple the early innate immune
response induced by Ad binding and entry from the later adaptive immune response. The
adaptive immune response is largely directed to the newly synthesized viral proteins
produced at low levels in cells transduced by FG- (E1/E3-deleted) Adv, consequently
resulting in transient transgene expression and long-term toxicity [19]. In contrast, acute
inflammation occurs within 24 hours of transduction, is dose-dependent, and is
independent of viral transcription [20]. Following vector administration, resident
macrophages efficiently take up adenoviral vectors and release inflammatory cytokines
within the tissue to eliminate infected cells [21, 22]. In vivo evidence exists for induction
of tumor necrosis factor (TNF), interleukin 6 (IL-6), IL-1IP,interferon-y (IFN-y) and
numerous chemokines [23-25].
Ad proinflammatory gene expression is mediated through a number of signaling
events which usually peak within 30 minutes of the initial infection [26]. In epithelial
cells, ERK and p38 are activated within minutes of Adv entry, leading to induction of the
chemokine IP-10 [27]. Capsid activation of ERK, initiated by av integrin interaction,
also leads to the production of IL-8 [28]. Adv vectors induce translocation of NF-KiBto
the nucleus, presumably through activation of IKK [29], resulting in the transcriptional
activation of IP-10 and RANTES [30, 31]. Finally, all three major MAPK pathways
(ERK, p38 and JNK1) appear to cooperate with NF-iB in the induction of ICAM-1 [32].
Little work has been done to characterize altered cell responses in this complicated multistimulant environment.
1.2. Signaling network interactions between virus and host
13
1.2.1. Introduction
The inflammatory cytokine, tumor necrosis factor-a (TNF), plays a key role in
the early innate immune response and subsequent elimination of E1/E3-deleted Adv from
the host [21, 33, 34]. Consequently, Ads have evolved complex counterstrategies to
evade TNF inflammatory responses and complete the viral replication cycle [35]. A key
role of TNF is to regulate programmed cell death, or apoptosis. TNF-induced apoptosis
is controlled by a complex intracellular signaling network of kinases, proteases, and
transcription factors that is well characterized (for a complete review, see [36]).
In this section we describe the TNF-activated signaling network controlling
apoptosis and the mechanisms evolved by wild type Ad to block the cell death response.
We then consider signaling network interactions between Adv and TNF and describe how
we can take advantage of the TNF signaling network to understand underlying Adv
perturbations.
1.2.2. TNF-activated signaling and apoptosis
The TNF-activated signaling network is generally considered common across cell
types. Briefly, binding of the trimeric TNF ligand to the TNF receptor (TNFR1) causes
recruitment of the TNFRl-associated death domain (TRADD). TRADD subsequently
recruits downstream adaptors resulting in activation of IKB-a kinase (IKK), p38mitogen-activated protein kinase (p38), and c-Jun N-terminal kinase (JNKl) [37]. These
pathways initiate a variety of biological responses, both prodeath and prosurvival.
IKK mediates the classical NF-KB activation pathway. IKB sequesters the NF-KB
p50-p65 heterodimer in the nucleus by masking the nuclear localization sequence (NLS)
of p65. Activated IKK, a complex containing two kinase subunits (IKKa and IKKOI)and
a regulatory subunit (NEMO), phosphorylates IKB on two serine residues [29].
Phosphorylated IKBs are then polyubiquitinated and subsequently degraded by the
proteasome [38]. Because the NLS of the NF-KB heterodimer is then exposed, the dimer
14
freely translocates to the nucleus where it can bind to KB binding sites of promoter
regions and stimulate transcription. Phosphorylated JNK1 translocates directly to the
nucleus where it phosphorylates c-Jun and ATF2, which both bind to AP- 1sites in the cjun promoter, thus enhancing its transcriptional activity [39]. ERK and p38 have also
been shown to activate AP-1 genes [39].
TNF also stimulates caspase activation. Caspases (or cysteine proteases that
cleave after an aspartate residue in the substrate) are a conserved family of enzymes that
execute the apoptotic (cell death) process [40]. TRADD and the subsequent recruitment
of the Fas-associated death domain (FADD) and the initiator caspase 8 forms the deathinducing signaling complex (DISC) [40]. DISC activation leads to cleavage and
activation of the executor caspase 3 and induction of apoptosis [41]. This is known as the
extrinsic apoptotic pathway.
The intrinsic apoptotic pathway, also known as the mitochondrial pathway, can be
indirectly activated by TNF [42]. The Bcl-2 family of proteins, consisting of both antiapoptotic and pro-apoptotic members, regulate apoptosis by controlling the release of
cytochrome c from the mitochondria [43]. TNF is linked to the intrinsic pathway via
caspase 8-mediated activation of Bid, a pro-apoptotic Bcl-2 family protein [42].
Activated Bid is subsequently translocated to the mitochondria and induces cytochrome c
release, initiating formation of the apoptosome and activation of caspase 9 [44]. These
interactions are summarized in Fig. 1-3 (left side).
1.2.3. Transcription-dependent mechanisms of wild-type Ad to
counteract TNF
In addition to stimulating inflammatory cytokines, gene products of wild type Ad
have both pro- and anti-apoptotic functions, creating complex interactions with the TNF
network (Fig. 1-3). The Ad E1A protein promotes or induces cell death via multiple
mechanisms. EA induces apoptosis directly through the p53 pathway [8]. In addition,
E A stimulates the production of TNF [45] and then exacerbates the prodeath stimulus
by down-regulating expression of anti-apoptotic c-FLIP [46]. Rather than serving a
15
-'"
cti
~
<Jl
o
U
-0
ro
l/l
I
Q)
Q)
I
.c c
~Q)
00>
L.L.
c
.2
Iii
>
'uro
.E
~~j
O:::.c
wlii
a.
r
I
c..
--I
I
I
I
I
I
I
I
I
~
~~
~
~
N
L.L.
: l~
L
.2
Iii
>
I
I
I
I
I
I
I
"'~2
I
I
I
I
I
I
I
I
I
c
)
00
~~
.- L.L.
t
I
I
CD
CD~
-"'-"'C
CD ..
...... ::!!!:
-'0.
L.L.
Z
-to
L.L.
Q)
Z 0>
~ I
0:::
I
I
I
I I
I I
I I
I I
I I
I I
I I
I
I I
I I
I I
I I
I I
I
I
I I
I I
I I
noct
I
,
+
-g
CD
,,
,
I
.J
Q)
0>
ra
>
ro
111
13
.5
Iii
~
-'"
o
;;,
o
cd
M
III
l/l
ra
a.
l/l
ra
U
functional purpose, E1A promotion of cell death is most likely an unwanted consequence
of deregulating the cell cycle for transcription and subsequent replication [8]. Therefore,
Ad has evolved several mechanisms to inhibit the apoptotic response in order to complete
the replication cycle, including another El protein product, EIB-19K [48]. The E B19K protein is a member of the Bcl-2 family and is functionally interchangeable with the
anti-apoptotic protein Bcl-2 [49, 50]. E1B-19K blocks apoptosis by binding and
inhibiting Bax, after Bax has undergone a conformational change induced by TNF [51],
and thus preventing release of cytochrome c from the mitochondria. EB-55K
inactivates and degrades p53, allowing host cell replication to proceed without being
inhibited the by the p53-induced cell cycle check point [52]. E3 products primarily
function to inhibit death signals initiated by the host immune system [8] and are not
necessary for efficient viral replication. The E3-10.4/14.5K complex inhibits activation
of the IKK complex, thus preventing the release of NF-KB to the nucleus and blocking
TNF-induced NF-KB activation [53]. E3 proteins are also involved in the downregulation of TNF-family receptors from the cell surface, including Fas [54, 55] and
TRAIL [56], although there is no specific evidence for down-regulation of TNFR1 itself.
Retention of the E3 locus is a strategy for mitigating the Adv-mediated immune response
[57], however generally it is removed to take advantage of additional space for packaging
a transgene.
In contrast to Adv E3-deleted mutants, Adv deleted for the E4 locus, specifically
ORF 6 and ORF 3, are severely impaired for growth [58]. In the absence of the proteins
encoded by E40RF3 and E40RF6, the DNA of the Ad genome is joined into
concatemers too large to be packaged into the capsid [59]. The E4 proteins inactivate the
DNA repair complex and thus prevent end-joining of Ad genomes [60]. E40RF4 has
been associated with a pro-apoptotic role in infection, correlated with its binding to
protein phosphatase 2A (PP2A) [61], however this mechanism has not yet been
elucidated. More recently, Ad E40RF4 and Ad E40RF 1 have been shown to induce
proliferative signals in infected cells, presumably to induce cell (and thus viral)
replication [62]. Ad E40RF4 substitutes for glucose-mediated signaling to the
mammalian targert of rapamycin (mTOR) and Ad E40RF 1 mimics growth factor
signaling via activation of PI3K. Ad E40RF 1 selectively stimulates PI3K via interaction
17
with its PDZ domain, resulting in downstream activation of both Akt and p70S6-kinase
[63]. This explains recent evidence that the Ad E4 gene promotes endothelial cell
survival through activation of Akt [64]. While wild type Ad proteins operate together to
optimize infection, deletion of one or more of these gene loci, such as El and E3 deletion
in FG-Adv, could result in unforeseen effects.
1.2.4. The intersection of Adv-TNF signaling
Wild type Ad uses transcription-dependent mechanisms to promote cell survival and counteract TNF-induced apoptosis - long enough to complete the viral replication
cycle. However, gene therapy vectors are transcription-limited and replication-deficient
due to the absence of the El trans-factor and are deleted for all E3-mediated mechanisms.
Based on the overlap in the signaling networks activated by TNF and Adv infection
(summarized in Table 1-1), we hypothesized that Adv would still alter signaling and cellfate responses to TNF. If true, this finding would have important implications for
interpreting both gene therapy and basic bioscience studies in which Adv and TNF are
involved.
We also hypothesized that the study of Adv cellular responses would benefit from
a quantitative systems biology methodology. The intracellular signal transduction
network stimulated by TNF and altered by Adv infection acts as a processor that converts
multiple input cues into an output cell fate decision. Ultimately the cell fate decision
involves a balance between prodeath and prosurvival signals [65]. Although it is possible
to assign functions to individual viral-host protein-protein interactions, we do not yet
understand how all interactions in concert contribute to the binary decision of life and
death. However, by taking a more systems-oriented approach to the problem, we may be
able to understand signaling mechanisms without a detailed knowledge of specific
interactions. We therefore considered current methods in computational systems biology
that could be applied to our system.
18
bb
CC
0
a. ,
-e
0
oI E© ZC-~
£~~~~~~~
~,
~
·. , m
C
lu
C,.
CC
C OE
=
m,
~U~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~0
a ,
00
5<
0
g
~
C.
i
.,
000
~~-.O2~
"J0
C)
o
-~~~~~~~~~~~
u
CZ -o
e.° Z
C
5
-C
C-s
UN
n<C)
0
IE
.4
C
C
CO
C)
CO
-o
o
c
0c~
Cb
C)
C)~
0
*;- *-~~~c~*-.,.-*
0
CZ
-e
0
*,
~
0
'
0
0
0
0
S
.0
0o
C)
CO
CO
C.
c)
CO
00
Ca
o
C)
C)
>1
CZ.
~~~~~~
-o
&Z'
C
,_o ~
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~..~
C)O
~ ~ C~>
C)CCO
0 C~~~'
~
~ =
C)
CZ
)
~,~
Cl
~~
~ 0O,~ C O
~~
0
*-- .&
0
0~
C.~~~~c
Cl
.-
c:~,~
~
, o· ~~~~t o .
C4
~4
C)
0
0'
>
~~~
>= E
C 'z ,-
~
~.
0.
0
-E
Q4
o
CZ1
m
c
o
.P
C
4a
YS%
2
00O
C
>
1.3. A systems biology approach to understand how viruses perturb
cell response functions
1.3.1. Introduction
The fundamental components of many intracellular signaling pathways are
common to all cells [85]. Intracellular networks arise from complex modular interactions
between the receptors, adaptors, and enzymes that together relay signals [86].
Computational models are useful tools for organizing and analyzing individual
component interaction data obtained through extensive experimental measurements.
It is increasingly recognized that useful models of biological networks - models
that provide insight into unanswered biological questions - will require a spectrum of
computational approaches, varying in their level of abstraction and specificity [87, 88].
This spectrum ranges from detailed mechanistic models (consisting of transport- and
reaction-based equations) to probabilistic models (e.g., Bayesian networks and Markov
chains) to statistical models (e.g., partial least squares, decision tree analysis, and
clustering) [88, 89]. A range of these approaches have already been applied to questions
of viral-host interactions at the population and gene expression level. For example,
compartmentalized models of HIV replication dynamics in a T-cell population have
provided significant insight into HIV propagation [90]. Cluster analysis of gene
expression changes from multiple datasets, heterogeneous for both cell type and pathogen
species profiled, revealed a common host-transcriptional-program shared among hostpathogen pairs [91]. Both of these approaches contribute to our understanding and
treatment of infectious diseases.
This spectrum of computational approaches is increasingly being applied to the
analysis protein signaling networks. Models have been used to analyze data quality and
information content ([92], [47]) and to deduce signaling network structure ([93],[94]).
Dinh et al recently developed a spatial-temporal model of the adenoviral trafficking
20
network for the purpose of understanding which network parameters influence viral gene
delivery [95]. This model related molecular-level trafficking events to whole-cell
distribution of viruses and is one of the few examples of quantitative modeling of a viral
signaling network.
Models have also been applied to understanding how signaling networks control
functional responses. However, in the face of limited knowledge of the intracellular
network, most of these models are constructed by correlating experimental measurements
to measured cellular responses ([96],[97],[88]). Network-function models have not yet
been used as tools to deconstruct pathogen-host signaling networks, but there is growing
interest in this application. In a recent review on DNA viruses and cancer, O'Shea
describes how DNA viral proteins and tumor cell mutations functionally converge in
perturbing similar cellular pathways [98]. O'Shea suggests that a systems biology
approach to the study of DNA viruses could further our understanding of and ability to
treat network deregulation in tumor cells. Kitano and Oda hypothesize that the
interactions between bacterial flora and the host immune system optimize robustness
against pathogen attacks and nutrient perturbations in mammals [99]. Thus, applying
quantitative modeling approaches to understand how pathogens modify cell responses
may help us understand fundamental signaling network functions.
1.3.2. Model of a cytokine-growth factor signaling network
Partial least squares regression analysis (PLSR) has emerged as a powerful means
of uncovering biological network functions by relating heterogeneous, multivariate
measurements of cell signaling to measurements of cellular response for different stimuli.
In the PLSR algorithm, the relative emphasis of a particular measurement is adjusted
based upon its contribution to the outcome being modeled. In this way, the signals most
closely related to cell response are highlighted while less important signals are
deemphasized. In addition, PLS does not require detailed knowledge about the
21
interrelationship of network components, a significant advantage when studying
complex, open-ended signaling networks.
Recently, a PLSR model was applied to understand how cells process the
cytokine-growth factor-mediated cell death decision [97]. Specifically, the study
examined the signaling networks that control the apoptosis-survival decision in HT-29
human colon adenocarcinoma cells pretreated with interferon-y (IFN-y), followed by
treatment with combinations of the prodeath cytokine TNF and the prosurvival growth
factors epidermal growth factor (EGF) and insulin [47]. Heterogeneous measurements of
kinase activities, changes in protein phosphorylation, caspase cleavage, and changes in
protein abundance were collected for 19 signaling proteins distributed across the
TNF-growth factor networks, at 13 time points over a 24-hour period, under 10 different
treatment combinations. Cellular response for each treatment was characterized by four
different assays of apoptosis at three time points. The complete cytokine compendium
included 8,340 individual biological measurements [47].
PLSR analysis of the cytokine compendium identified a molecular basis set of
signals controlling cytokine-induced apoptosis [97]. This basis set was organized into
axes described by linear combinations of the input signals, including a stress-apoptosis
axis enriched for signals inducing apoptosis, and a survival signaling axis, enriched for
pro-life signals [97]. These axes permit predictive analysis by calculating where a new
set of signaling measurements induced by a different treatment combination would lie in
the stress-survival signaling space, and using this information to predict the resulting
level of apoptosis. The predictive capability of the cytokine compendium model was
tested by perturbing two regulated autocrine stimuli known to contribute to TNF-induced
molecular signals [69]. Specifically, TGF-a and IL-la autocrine circuits were
individually disrupted with antibodies or agonists, the set of resulting signals measured,
and the PLSR model used to convert these signals into a prediction of apoptosis (or the
apoptotic signature consisting of all four measurements at three time points.) There was
a 90% correlation between the measured apoptotic outputs and the model predictions,
confirming that the basis axes identified by PLSR capture the dynamic intracellular signal
22
processing of diverse stimuli [97]. Given that adenoviral infection perturbs the same
underlying signals mediating the TNF response, we hypothesized that this approach may
be useful in deconstructing underlying Adv signaling modifications.
1.3.3. Challenges in virus-host cell signaling analysis
Adenoviral gene therapy gives rise to complex patterns of signal transduction in
the host cell, caused by binding and entry of the Adv capsid, background transcription of
the viral backbone, and expression of the transgene. While there is an increased
understanding of how individual signaling pathways are upregulated in response to Adv
infection (Table 1-2), cells ultimately process cues through an interconnected network of
these pathways. Therefore, we hypothesized that Adv-induced signals alter the response
to TNF in a manner analogous to EGF, insulin, and the pretreatment cytokine INF-y
perturbation of the TNF response in HT-29 cells. In this case, quantitative measurement
of signals distributed throughout the TNF network for different combinations of Adv and
TNF and subsequent PLSR analysis of those datasets Adv would inform which signaling
pathways modified by Adv contribute most to altering the cytokine-mediated cell
response. The remaining chapters of this thesis summarize the results of this study.
23
CHAPTER 2
2. Experimental methods for quantitative Adv-TNF network
analysis
2.1. Introduction
The objective of this thesis was to quantitatively deconstruct the contributions of
Adv infection in the host cell response to the proinflammatory cytokine TNF that
circulates during infection in vivo [21]. To do this, it was necessary to develop a
quantitative experimental system with clear TNF-induced outcomes that were specific to
the state of Adv infection. Ad and Adv have been shown to synergize with TNF in many
cellular contexts [100-102]. We observed that infection with a first generation Adv
synergized with TNF to induce apoptosis in human epithelial cells. We chose to study a
first-generation Adv because these are the most common vectors used in both the
laboratory and the clinic. We initially developed our experimental approach in the HT-29
human adenocarcinoma cell line consistent with the cytokine compendium work.
Our methodology was to explore the mechanism of Adv-mediated sensitization to
TNF by measuring changes in the activity of signaling proteins centrally involved in the
Adv-TNF network. We hypothesized that the Adv perturbations controlling changes in
apoptotic response would be encoded in the upstream signals. We used statistical and
computational analysis to identify the most important network signals controlling
apoptosis. In Chapter 2, we describe our experimental methods for quantifying biological
metrics and our computational methodology to analyze those measurements. In the final
section of Chapter 2, we measure several biological responses specific to our Adv system
that should be taken into account when interpreting our results presented in the following
chapters.
24
2.2. Optimization of the Adv-TNF sensitization protocol
2.2.1. Apoptosis
Apoptosis, or programmed cell death, is central to metazoan homeostasis and
development [103]. Apoptosis is characterized by highly conserved morphological and
biochemical changes in the cell, including chromatin condensation, membrane
fragmentation, and ultimately the formation of apoptotic bodies that are rapidly
phagocytosed by neighboring cells [104]. These tightly regulated morphological and
biochemical changes provide ideal hallmarks by which to quantify the amount of
apoptosis occurring in a population.
We tested multiple combinations of Adv and TNF to determine a suitable
experimental protocol to quantify the extent to which Adv infection alters the cell-death
response to TNF stimulation. When designing our protocol, we sought to mimic the
cytokine compendium approach discussed above, a protocol previously developed in our
laboratory to understand the TNF-growth factor-mediated cell death decision [97].
Systematic collection of biological data under carefully matched experimental conditions
makes it possible to compare, with high reliability, data that were not collected
simultaneously in the same experiment [47]. Therefore, our objective here was to
maximize comparability of the data for future biological investigations.
A general schematic of our protocol is illustrated in Fig. 2-1. Briefly, we infected
HT-29 cells with either an E1/E3-deleted adenovirus carrying the CMV promoter and a
P-gal reporter gene (Adv.P3-gal)or an identical vector without a reporter gene
(Adv.empty). Mock infections were carried out with media containing vector buffer
only. The final Adv concentration in the infection media was 1.4 x 1010v.p./ml,
resulting in approximately 100% of the cell population positive for Adv infection (see
Fig. 2-3b). For control cells, buffer without Adv was added. TNF (100 ng/ml) was added
24 hours after the start of infection. Cells were collected 48 hours after TNF treatment
and apoptosis was verified by three independent methods: quantitatively by flow
cytometry with 1) an anti-cleaved caspase 3 antibody and the M30 antibody [105] against
25
•
24h
24h
24-48h
--.~
~
Plate Cells
Infect with Adv
for 6 hr
Measure
signaling and
apoptosis
AddTNF
Fig. 2-1. General schematic of the Adv-TNF sensitization protocol
Cells were plated at the density determined to be optimal for each individual cell line. 24 hours later the
media was replaced with a reduced media volume containing the adenoviral vector. After 6 hours of
exposure to the virus, the infection media was removed, cells were washed with PBS, and a full volume of
media was replaced. 24 hours after the start of the infection, 100 ng/ml TNF was added to the cells.
Signaling measurements were taken 0-24 hours after TNF addition. Apoptosis measurements were taken
24 or 48 hours after TNF addition. See Section 6.2 for additional detail.
caspase-cleaved
cytokeratin (Fig. 2-2, a-d) or 2) annexin V and propidium iodide (PI)
staining (Fig. 2-2, a-b); and verified visually by 3) DAPI-staining
for chromatin
condensation (Fig. 2-2, e-f). Adv infection alone caused only low levels of apoptosis
compared to control cells; however, all three assays showed that the apoptotic population
significantly increased upon TNF-treatment
in Adv-infected cells as compared to
uninfected cells. By annexin-PI and cleaved-caspase
staining, the increase in apoptosis
was two- to three-fold (Fig. 2-2, a-b; p < 0.0 I). Adv constructs with and without a I)-gal
reporter enzyme resulted in similar rates of cell death, indicating that the presence of the
I)-gal transgene did not influence Adv sensitization.
Furthermore,
TNF -induced
apoptosis required Adv preinfection in these cells, because TNF treatment by itself
caused minimal cell death (Fig. 2-2, a, c, e). Two-factor analysis of variance (ANOVA)
revealed a significant interaction effect between Adv infection and TNF treatment (p <
10-4), indicating that Adv infection synergistically
enhances TNF -induced apoptosis in
HT -29 cells. Therefore, apoptosis provides a clear, quantitative response measurement
for a system-level study. Measurements
of cleaved-caspase
products at 24 and/or 48
hours following TNF treatment were used for all subsequent studies.
26
a
b
WithTNF-a
No TNF-a
60
60
D Cleaved casp-cytoker
D Annexin/PI
£40
-S
ttI
ttI
QJ
40
QJ
0
0
~ 20
~
~D OoD
0
Adv.
Buffer
C
0
empty
j~
IV,
Uo
6iJ
.,,'",
-g~
.
:IV
'"
.~9
:' ~{.
~
.:f:
10:
.:.
:~.:
+ TNF-a
l-
Yo
~~
••••
10'
10'
10'
Anti-cle<r.ed cytokeratin
TNF only
Adv
"''"n>c.
'"n>,
Uo
{.;/.'::
~
empty
Me,
..
~
Adv.
~-gal
d
~
~
Adv.
Buffer
TNF-a only
~
o~
20
Adv.
~-gal
e
D Cleaved casp-cytoker
D Annexin/PI
10'
10'
f
10'
10'
10'
Anti-clea'o'edc~1ckeratin
10'
Adv+ TNF
Fig. 2-2. Adv infection sensitizes human epithelial cells to TNF-a-mediated
apoptosis
10
(a-b) HT-29 cells were infected with either Adv.13-gal or Adv.empty (I.4 x 10 vp/ml) or treated with
buffer only and then stimulated with either 100 nglml TNF or carrier only. Cells were collected 48 hours
after TNF addition, stained for either caspase-cleaved cytokeratin and cleaved (active) caspase 3 (white
bars) or annexin V/PI (gray bars) and analyzed by flow cytometry. Data are presented as the mean of 3
biological replicates:!: SE. Adv.13-gal was used for all subsequent experiments. (c-f) Visual comparisons
of apoptosis in Adv-sensitized and control cells treated with TNF. (c-d) Sample flow cytometry plots for
anti-caspase-cleaved
cytokeratin and anti-cleaved (active) caspase 3 and (e-f) Fluorescent microscopy
images of chromatin condensation (with DAPI-staining).
27
2.2.2. Infection efficiency
The 8-gal reporter transgene allowed us to characterize the efficiency of Adv
infection in the HT-29 cell line. HT-29 cells have been reported to have a low Adv
infection efficiency [106] when considered in comparison with other cell lines [107].
Many variables affect net Adv dosage, including differences in cell plating density,
infection time, and concentration-dependent rates of diffusion [108, 109]. To account for
this, we altered total viral uptake by changing infection time and concentration (see
Methods for details). Quantitative PCR of the Adv genome showed that infection with
Adv for six hours at a concentration of 1.4x 10°v.p./ml minimized exposure time with
the virus while maximizing the number of transgene copies per cell (Fig. 2-3a).
Therefore, we chose these infections conditions for our general protocol. The
concentration of 1.4xl100 v.p./ml, contains approximately 20 capsids per infectious
particle by plaque assay, corresponding to a multiplicity of infection (MOI) of
approximately 1000 infectious particles per cell. Although this MOI is high compared to
a
Viraluptakeper cell
s
b
Infectedpopulation
C
1500
80
aL
3 12
U
:E
'a 1000
o
NormalizedBgalactivity
oc 40
500
u
m
-
n
an
0
0
5
10
Infection
time
4
O
0
0
500
MOI
1000
0
200
400
MOI
600
Fig. 2-3. High Adv concentrations are required for efficient infection of HT-29 cells
(a) HT-29 cells were infected with two concentrations of Adv (0.7 x 100° and 1.4 x 101° v.p./ml) for 0, 1, 3,
6, and 12 hours. The total number of Adv copies per cell was quantified by PCR. Values represent mean
total -gal DNA normalized to mean total [5-actinDNA for duplicate samples. (b-c) HT-29 cells were
infected for six hours at varying MOIs (with MOI 1000 = 1.4 x 10 0° v.p./ml), collected 48 hours later, and
assayed for presence of ,8-gal. (b) Population positive for 5-gal activity is determined by microscopy. 400
cells were counted and binned for two biological samples. Data is presented as the mean of the positive
population (No. cells visibly expressing ,-gal/400 cells). (c) Total P-gal enzymatic activity was measured
and normalized to protein content of sample. Data are presented as the mean of 3 biological replicates +
SE.
28
other literature values, this was the lowest MOI tested that resulted in > 95% of the cell
population positive for Adv infection (Fig 2-3b). Finally, we measured enzyme activity
versus viral dose and found that ,-gal activity increases with MOI, as expected (Fig. 23c). Therefore, we confirmed that Adv.pgal productively infects HT-29 cells, and that an
MOI of approximately 1000 is necessary to infect almost 100% of the cell population.
It was possible that the high Adv titers used in the sensitization experiments had
exceeded a threshold, above which apoptosis occurred non-physiologically. Therefore,
we explored how Adv sensitization varied with Adv dosage. Quantitative PCR data of
the Adv genome revealed that TNF-mediated apoptosis correlated linearly with dosage
from 0 - 1400 total Adv copies per cell (Fig. 2-4; r = 0.95), demonstrating that Adv
sensitization to TNF is directly proportional to vector uptake in our in vitro experimental
system. Moreover, synergistic apoptosis was evident with as few as 400 copies per cell.
High infectivity levels are necessary to achieve efficient infection in vivo [20, 110],
arguing that Adv sensitization to TNF-induced apoptosis can occur at clinically relevant
infectivities.
40-
r = 0.95
o
0
20
0c)
._
o
0
>
°
* 7.x1' vp./m
* 07.x1' 0 v.p./m1
0
0
0
400
v.pml
1.4x10o
800
1200
Adv copies per cell
Fig. 2-4. Sensitization dose response
HT-29 cells were infected with two concentrations of Adv (0.7 x 10L ° and 1.4 x 10'0 v.p./ml) for 0, 1, 3, 6,
and 12 hours. The total number of Adv copies per cell was quantified by PCR. Values represent mean
total ,3-galDNA normalized to mean total 3-actin DNA for duplicate samples. Variance in the Adv copy
values ranged from 10-30% of the mean value. In parallel, HT-29 cells subject to the same infection
conditions were treated with 100 ng/ml TNF and collected after 48 hours. Cleaved caspase 3-cytokeratin
measurements are plotted as the mean apoptosis (above control level) of three biological replicates i s.e.m.
Pearson's correlation coefficient for the mean values was r = 0.95.
29
a
b
Advonly
60
Adv+
HT-29
HT-29
60
A549
.~
a.0
VI
.;;;
40
0
a.0
0.
OA
40
0.
<t
<t
~
A549
HeLa
HeLa
C3A
VI
TNF
~
20
0
20
0
10
lOa
1000
100
10
1000
MOl
MOl
Fig. 2-5. Adv sensitizes multiple human cell lines to TNF-induced apoptosis
(a-b) HT-29 (black), A549 (green), HeLa (red), and C3A (blue) cells were infected with increasing doses of
Adv (expressed as multiplicities of infection, p.f.u.lcell) and treated with (a) carrier or (b) 100 nglml TNF.
HT -29 and A549 cells were collected 48 hours post TNF treatment. HeLa and C3A cells were collected 24
hours post TNF treatment. Cleaved caspase 3-cytokeratin measurements are plotted as the mean of three
biological replicates :J:: SE.
To detennine whether Adv- TNF synergy existed in diverse tissue types, we
compared the sensitization of HT -29 cells to that of other human cell lines, including
HeLa cervical carcinoma cells, A549 lung carcinoma cells, and C3A human
hepatocarcinoma
cells over a range of infectivities (MOl: 30 -1000). Two-factor
ANOV A revealed a significant interaction effect between Adv infection and TNF
treatment in all cell types, even at the lowest MOls tested (p < 0.01), indicating that
Adv- TNF synergy was present to varying degrees in all cell types (Fig. 2-5, a-b).
Surprisingly, of the cell types tested, Adv-infected HT-29 cells were the most insensitive
to TNF-induced apoptosis.
Apoptosis in Adv-infected A549 (Fig. 2-5, green) cells
collected after 48 hours ofTNF treatment was approximately
two-fold greater than in
HT -29 cells (Fig. 2-5, black) following the same treatment conditions.
However, A549
cells were much more sensitive to Adv infection alone, in part accounting for this
difference (Fig. 2-5a). Apoptosis in HeLa cells and C3A cells after 24 hours ofTNF
exposure was consistently two-fold greater than cell death in HT-29 cells after 48 hours
ofTNF exposure (Fig 2-5, a-b). HeLa cells were more sensitive than C3A cells to Adv
infection alone at higher MOls. These results demonstrate that Adv synergizes with TNF
to induce apoptosis in many human cell types, and that these responses are cell-specific.
30
2.3.Large-scale
signaling data collection
2.3. t. A high-throughput multiplex radioactivity-based kinase activity
assay for quantitative signaling analysis
Once we had optimized an experimental system with which to quantitatively
evaluate how Adv infection sensitizes cells to TNF-induced
apoptosis, we sought to
optimize a similarly quantitative procedure for measuring cell signaling. Evaluating how
phenotypic information is encoded within cell-specific signaling requires dynamic and
quantitative measurements of the intracellular network [47, 97]. Previously, it has been
argued that measurements of kinase catalytic activity are more useful than measurements
of simple protein abundance for monitoring information flow through a signaling
1.6
c
.Q
1.4
U
QJ
'0'
a.
1.2
QJ
-5
'0
QJ
v
C
~
o
a.
.~
0.8
0.6
OJ
:0
.!!1
~
0.4
0.2
Kinase assay
Western blot
Antibody
array
Fig. 2-6. Kinase activities are information-rich measures of signaling networks
Notched box plot of variable importance from a systems model of cytokine-induced apoptosis [97].
Variable importance is normalized so that variables with average information content in the model are
assigned a value of one (see [97] for details). Measurements of kinase assays by kinase assay (left) were
compared to measurements of protein state or protein level by Western blotting (center) or antibody array
(right). Kinase measurements were significantly more informative on average than western blotting or
antibody array (p < 0.05 by signed rank test against a median value of I). Figure courtesy of K. Janes.
31
network [111]. Recent analysis of a large-scale signaling compendium
further suggested that kinase activity measurements
measures of protein phosphorylation
[47, 97] has
are more informative on average than
or cleavage state (Fig. 2-6).
A generalized assay in a 96-well format for the multiplex analysis of multiple
protein kinase activities has been previously developed in our laboratory [Ill,
procedure, which utilizes kinase-specific
quantitative high-throughput
immunopurification
activity measurements,
112]. This
steps followed by rapid
has been applied to measure the
activities of five kinases: ERK, Akt, IKK, JNKl, and MK2 [1 1I](for general assay
schematic, see Fig. 2-7). The assay was previously optimized for HT -29 cells and HeLa
cells to monitor information through the TNF, EOF, and insulin signaling networks in
HT-29 cells sensitized with IFN-y [47, 97, 111].
Using our general Adv- TNF sensitization protocol (Fig. 2-1) and the large-scale
data collection procedure developed for cytokine compendium measurements
(ERK, Akt,
JNKl, IKK, and MK2) [47], we collected the analogous set offive kinase signals
following TNF treatment in Adv-infected or uninfected HT -29 cells. Briefly, Advinfected or uninfected HT -29 cells were treated with or without 100 ng/ml TNF and cell
lysates were collected at 13 time points over 24 hours (see Chapter 6 for details).
1.
2.
wash
3.
4.
incubate
Fig. 2-7. General schematic ofthe high-throughput kinase activity assay
Reproduced from [III]. (I) Lysates are incubated with Protein A or G microtiter wells precoated with
anti-kinase antibodies. (2) After several washes, an appropriate substrate and [y _32p]_ATP are added to the
plate to initiate (3) an in vitro phosphorylation reaction. Signal-to-noise ratio is optimized by varying the
duration of the reaction for each kinase. (4) The reaction is terminated with H3P04 (for ERK, Akt and
JNKI assays) or EDTA (for IKK and MK2 assays), and a fraction of the reaction mix is transferred to a
phosphocellulose (PC) filter plate and washed to remove free 32p.
32
Our measurements show that kinase signaling time courses are quantitatively
different between uninfected and Adv-infected cells treated with 100 ng/ml TNF and
between Adv-infected cells with and without TNF treatment (Fig. 2-8). Measurements
were sampled more densely in the first four hours because much of the signaling induced
by TNF occurs shortly after cytokine addition [111]. Therefore, in addition to evaluating
the significance of the difference between each pair of signaling curves for the full time
course by two-factor analysis of variance (ANOVA), we also evaluated differences over
the first four hours (7 time points) and the last twenty hours (6 time points) by the same
method (Fig 2-8, table).
Akt, IKK, and MK2 time courses are significantly different between uninfected
and Adv-infected cells treated with 100 ng/ml TNF (see Chapter 3 for more details).
However, with the exception of Akt, signals during the first four hours of the time course
are not significantly different between these two conditions. This may be due to the
importance of instant, transcription-independent signals to TNF cell-fate decision, such as
the acute activation of MK2 and JNK1 [113]. However, signals during the later hours of
the time course are significantly different for all kinases except JNK1. Because we have
not inhibited gene expression in our studies (unlike other studies with TNF [83] ), both
TNF-induced and Adv-induced transcripts and proteins can affect the signaling network
measurements, and therefore the later differences are likely transcription-dependent. In
recent work growing out of the signaling compendium, the responses of HT-29 cells to
the nine multi-input stimuli specified in the signaling and apoptosis datasets were
transcriptionally profiled to interrogate the role of gene expression in the HT-29 system
[114]. Preliminary analyses of these measurements revealed a strong clustering of TNFtreated samples. The data here suggest that the mechanism by which Adv alters the TNFinduced cell fate decision depends on late, possibly transcription-dependent, signals.
All signaling time courses are significantly different between Adv-infected cells
with and without TNF treatment, as expected (Fig. 2-8, table), with the exception of early
(0-4 h) Akt activity and late (4-24 h) IKK activity. Akt is significantly upregulated in
Adv-infected cells prior to TNF treatment, consistent with literature linking Adv viral
backbone proteins to Akt activation [62, 64]. The biological implications of Adv-induced
Akt activity are explored in Chapter 3.
33
TNF only
:r
Adv + TNF
Advonly
0
Full time
course
~
c
c
0
o
.~ 10
10
.~
"0
"0
~
~
0
0
20
10
10
1
0
20
1\•
0
0
.~ 10
>
.~ 10
>
.,
.
.~
~
~
"0
~:-~
~ ..
~
4
0
"e
'"
~
~
1J
~
I
"
"0
~
~
~
0
4
0
Time (hr)
4
Time (hr)
Time (hr)
Probability: Adv + TNF
20
~
c:
c:
c
0
.~ 10
10
Time (hr)
Time (hr)
Time (hr)
First 4
hours
~
= TNF only
Probability:
Adv + TNF
= Adv
only
0-4
Akt
Full time
0-4 hours
4-24 hours
Full time
hours
4-24 hours
3.19E-11
1.37E-04
3.33E-10
7.78E-03
0.333
5.28E-Q4
IKK
9.65E-Q4
0.058
3.23E-03
4.46E-04
4.08E-03
0.065
MK2
7.74E-03
0.256
5.65E-05
1.53E-13
2.86E-08
6.82E-07
ERK
0.426
0.119
0.017
5.18E-05
1.04E-03
0.014
JNK1
0.707
0.673
0.962
4.79E-07
8.99E-05
2.45E-03
Fig. 2-8. Signaling
Unfected
time courses following
and Adv-infected
TNF-activation
cells were treated with
Adv in the absence ofTNF
treatment
and 90 minutes and 2, 4,8,
12, 16,20 and 24 hours and kinase activities
(blue), IKK
(purple)
normalized
to total protein content.
replicates
normalized
significance
and Adv
and MK2
+ TNF
conditions)
as a control.
Results are plotted
of the untreated control
at 0 minutes
was evaluated
by two-factor
TNF and Adv only versus Adv
+ TNF
differences
were compared.
(p
< 0.05)
P-values
(i.e., uninfected
Both differences
with
15,30,60,
assay and
of two biological
cells).
The
between
Adv only
in the full time course
between TNF only versus Adv
for each comparison
are shaded in gray.
34
kinase activity
activation
curves (e.g. ERK activity
ANOV A.
at 0,5,
for ERK (green), Akt (red), JNKI
as the mean relative
between each pair of signaling
cells.
Cells infected
Lysates were collected
and partial time courses (0 to 4 hours and 4 to 24 hours) and comparisons
tabular form and significant
and Adv-infected
(orange) were measured by a high throughput
to activity
of the difference
were included
in uninfected
100 nglml TNF at time t = 0 min.
are presented
in
+
Ad and Adv are also known to activate NF-KB in a variety of cell types [31, 32,
115] and recently an E 1/E3-deleted Adv has been shown to attenuate cancer cell
sensitivity to chemotherapeutic drugs via Adv-mediate NF-KB activation [ 116].
However, in all these studies, NF-KB was induced during the early phase of Adv
infection leading to a host inflammatory response including secretion of IL-6 [20]. The
cause of late IKK-NF-KB in Adv-infected cells with and without TNF is to our
knowledge unexplored. One hypothesis is that this signal occurs indirectly in response to
an autocrine signal from the IL-la autocrine loop identified in the cytokine compendium
study [69]. We discuss this hypothesis in more detail in Chapter 4.
2.3.2. Quantitative western blots measurements
Many critical signals for apoptosis cannot be measured via an enzymatic activity
assay but instead require a protein state assay. For example, caspases are produced in
cells as catalytically inactive protease precursors called zymogens. They are activated
via proteolytic cleavage during apoptosis [40] and are thus effectively measured via size
resolution and immunoblotting. Caspase 8 is an initiator caspase that is critical to
mediating apoptosis of the extrinsic pathway, initiated by the binding of an extracellular
death ligand such as TNF [ 117]. Cytokine-signal mapping of the TNF-growth factor
signaling compendium revealed that caspase 8, in addition to JNK1 and MK2, co-varied
most closely with TNF-treatment [47]. Therefore, we sought to augment our data set
with the addition of caspase 8 measurements via Western blot.
Cleaved caspase 8 is membrane-associated and therefore requires a whole-cell
homogenate (observations by S. Gaudet). To add caspase 8 measurements to our time
course, we repeated the three HT-29 conditions (see Fig. 2-8) on a single day. Caspase 8
cleavage does not occur until approximately 4-8 hours post TNF addition [47].
Therefore, we collected an abbreviated time course with samples taken at 0, 4, 8, 12, 16,
20, and 24 hours. In our experimental system, cleaved caspase 8 was undetectable in
Adv-TNF treated cells until 16 hours and it rose steadily as measured at 20 and 24 hours
(Fig. 2-9). There was no cleaved caspase 8 detectable in uninfected cells treated with
35
a
Pro casp 8
t>..I..
Jf ItIl
17'
Cleaved casp 8 _
24h
12 h
b
--.
-{)-
TNFonly
-Advonly
Adv + TNF
0.1
Caspase 8
fraction
(cleaved/pro)
0.05
o
15
20
Fig. 2-9. Caspase 8 cleavage in Adv- TNF HT -29 cells
Caspase 8 time abbreviated time course following TNF-activation in uninfected and Adv-infected cells.
Unfeeted and Adv-infected eells were treated with 100 ng/ml TNF at time t = 0 min. Cells infected with
Adv were included as a control. Lysates were collected at 12, 16,20, and 24 hours and caspase 8 (pro- and
cleaved-) were measured by quantitative western blot. (a) Pro- and c1eaved-caspase 8 at 12 hand 24 h. (b)
Fraction of cleaved caspase 8 (cleaved/pro) as a function of time. Approximately equal amounts of protein
content were loaded in each lane. Results are plotted as the mean ratio of three biological replicates:!: SE.
TNF or in Adv-infected cells without TNF treatment, consistent with apoptosis
measurements.
Because the caspases are direct effectors of apoptosis, adding these measurements
to a network model of apoptosis would likely increase the predictive power. However,
our final network model did not require the pro- or cleaved-caspase
8 signals to predict
apoptosis in Adv-infected epithelial cells (see Chapter 4). This is consistent with the
findings of the cytokine compendium that the caspase effector signals are encoded by the
upstream signaling network [97].
36
2.4. Partial least squares regression analysis
2.4.1. Measurement processing and metric extraction
One aim of this work was to combine our measurements with the cytokine
compendium for the purpose of predicting apoptotic response. If possible, this would
augment the findings of Gaudet, et al. [47] to show that data sets collected at different
times and by different pairs of hands can be compared with confidence. To do this, we
processed our measurements using the same method of signal-specific normalization: 1)
correcting each measurement for the total amount of cellular protein in the sample and 2)
adjusting kinase activity measurements for changes in the specific activity of [y-32 P]ATP
[47]. In the cytokine compendium, each signal was referenced to its own background (t =
0 min average) activity in the sample. However, while observations in the cytokine
compendium were all pretreated with IFN-y, the pretreatment in our measurements, i.e., +
Adv infection, was the variable of interest. Differences between baselines contained
biologically relevant information, creating a key difference in our data set. To explore
this, we processed our data in two ways: 1) signals for each pretreatment (i.e., Adv @
1000 MOI) were normalized to the baseline signals for the same time course pretreatment
(internal normalization) or 2) signals for each pretreatment were normalized to the
baseline signal for the uninfected control (control normalization). Measurements
processed via both methods were evaluated in subsequent PLS analysis.
For PLS analysis all measurements are combined into a single matrix; therefore
the temporal organization of the data is lost. It was unknown whether time-dependent
features such as the steady-state signal or the maximum activity contained important
information metrics were empirically derived from [97]. Therefore, we used the method
developed by Janes and Kelly [89] in which time-dependent signaling metrics are derived
empirically from a dynamic data set (Table 2-1). Each metric is then entered as a
separate variable into the data matrix. A total of 35 metrics were calculated for each
signal for a combined total of 175 measurements per observation.
37
Table 2-1. Signaling metrics extracted from dynamic network measurements
Adapted from [97].
Metric class
Local metrics
Metrics extracted
Time points
Point-to-point derivatives
Area under the curve
Mean signal
Maximum signal
Steady-state signal
Area under the curve
Activation slope
Summary metrics
Peak metrics
2.4.2. Predictions with TNF-GF compendium
We recalculated the cytokine-growth factor PLSR model using only the five
dynamic kinase-activity profiles from the nine cytokine-growth factor treatments fitted to
two apoptosis measurements (cleaved caspase products at 24 and 48 hours). We then
used this model to predict apoptosis for our combinations of TNF only, Adv only, and
Adv+TNF in HT-29 cells based on the analogous set of signaling measurements. While
the model accurately predicted apoptosis in cells infected with Adv and treated with TNF,
the model could not differentiate those cells that were not sensitized to apoptosis and
instead predicted significant levels of apoptosis for all treatments (Fig. 2-10a).
Although this result was disappointing, the conditions used to train the model
were designed to study the combination of growth factors and TNF on the cell fate
decision. We reasoned that this would not provide enough information to differentiate
between infected and uninfected conditions where TNF is the major stimulus; instead,
these conditions would correlate most closely with TNF conditions in the compendium,
resulting in a positive prediction of apoptosis.
The PLSR model is derived as a series of latent variables called principal
components, which contain linear combinations of the original kinase activity
measurements [97, 118]. These principal components form a set of intracellular basis
axes that define the dimensions of the signaling network that controls apoptosis. IFNy +
38
a
b
Model predictions
Treatment mapping
100 IFN+GF/rNF
U.
E
u -5-
24h
It +,
'
o48h
Notx
50
00
·
TNF
TNF
Adv+TNF
.
-
-10 -
0-
TNFonly
EGF
rO 0
E
50 -
0!
0,
a
'~ F
Ct
'a
* Adv only
1
Adv+ TNF
Advor TNF
.v,
I
100
Predicted % apoptosis
* Insulin
4
4
8
12
Principal component #1
Fig. 2-10. TNF-growth factor model predicts Adv-TNF-induced apoptosis but cannot differentiate
uninfected cells
(a) Correlation between measured apoptosis and predictions of the TNF/GF model. Model training with
cytokine compendium data (black) is compared to model predictions of Adv+TNF (blue) and Adv only or
TNF only apoptosis (green) at 24 hr (filled) and 48 hr (hollow). Data are presented as the mean of three
biological replicates ± S.E., and model uncertainties were estimated by jack-knifing[89]. (b) TNF and
insulin axes are recapitulated using only kinase activity data. Adv+TNF (blue), TNF only (green) and Adv
only conditions are shown.
TNF treatments projected strongly along principal component 1, while insulin
measurements projected along principal component 2, as was found in the full
compendium model [97]. (Interestingly, EGF treatmentdid not project as strongly as
found in the original compendium model). When the TNF treatments + Adv-infection
were plotted on this treatment mapping, they projected strongly onto the TNF axis (Fig.
2-1 Ob). The Adv only condition did not fall within the region outlined by the TNF and
growth factor conditions.
This supported our hypothesis that information on how Adv
changes apoptotic response to TNF was needed to constructa useful model. We pursue
this in Chapter 4.
2.5.Other biological considerations
39
2.5.1. Adenoviral replication in carcinoma cell lines
As described above, E1/E3-deleted Advs are not biologically inert. Therefore, by
using a first generation Adv (FG-Adv), our results are specific to cytokine-induced cell
signaling networks in the context of Adv background transcription. This system is
appropriate because this is the most common vector used; however, to understand its
limitations, we quantified aspects of Adv biology that may be contributing to the
sensitization response.
Ad EIA products function as transcription factors to promote viral replication by
trans-activating other Ad transcription units, namely the E2 and E4 regions, which
encode proteins for necessary for Ad DNA replication [2]. Thus, Adv that have been
deleted for the E 1A region are highly attenuated [ 119]. However, human tumor cell lines
derived from colon, liver, cervical, and lung cancer support replication of Adv deleted for
the E lA/E lB genes [107]. It has been hypothesized that cellular proteins can substitute
functionally for E1A in the trans-activation of viral proteins and cause this phenomenon
[107, 120].
Given the high MOI used in our experimental system, we suspected that Adv
genome replication was occurring. Using quantitative PCR, we confirmed nearly 100fold replication of the Adv genome in HT-29 cells over 72 hours following the start of
infection (Fig. 2-1 a). Exposure to TNF exacerbated this effect: the amount of Adv
genome replication in the 24 hours after TNF addition was almost 4-fold greater in the
presence of TNF (Fig. 2-1 lb). To our knowledge, no literature evidence exists
documenting this effect, although it is likely that TNF treatment results in an influx of
proteins to the nucleus that can act as trans-factors for Adv genome replication.
For most tumor cell lines, the cellular trans-complementing factors substituting
for E 1A and causing Adv DNA replication remain unclear. However, in human
papilloma virus (HPV)-associated cervical HeLa carcinoma cells, HPV 18 DNA is
integrated into the genome and is amplified [121]. One of the HPV 18 DNA products,
the E7 protein, structurally and functionally similar to Ad E1A [122]. In HeLa cells, the
HPV E6 and E7 proteins can substitute for adenoviral EIA and E1B, allowing for
replication of Adv [123]. Based on this information and the data for HT-29 cells, we
40
a
b
Genome increase with
Advgenome replication in HT-29cells
and without TNF
105
60 c
as
.
4
---
40 -
10
._
0
U
20103
NA- I
TNF
1-
0
20
40
60
[
I
I
-
+
Time(hr)
Fig. 2-11. HT-29 cells support Adv genome replication
(a) HT-29 cells were treated with 1.4 x 10 0° v.p./ml Adv and 100 ng/ml TNF () or mock (O) as
previously described and collected at 6, 24, and 48, and 72 hours post infection. The total number of Adv
copies per cell was quantified by PCR. Data are presented as the mean total fiber knob DNA normalized to
mean total -actin DNA for three biological replicates ± S.E. (b) TNF increases Adv genome replication.
Data are presented as [mean copies/cell @ 48 h _ TNF] / [mean copies/cell @ 24 h].
conclude that Adv genome replication is occurring in HeLa cell lines in our experimental
system, although we did not measure it directly.
2.5.2. Transcriptional shedding of adenoviral backbone proteins
Gene products of the Adv early transcription region 4 (E4) operate through a
complex network of protein interactions with cellular regulatory components of important
host cell functions, including transcription, apoptosis, and DNA repair [124], and
signaling. Commonly, vectors contain the E4 gene (Adv.E4 + ) and these vectors
selectively elicit phosphorylation of Akt in endothelial cells [64]. E40RF1 has also been
shown to mimic growth factor signaling via activation of PI3K [63]. In addition,
E40RF4 activates mTOR [62], and likely Akt as well through mTOR regulation [125].
Because our vector is positive for E4, we reasoned that E4 proteins are responsible for
Akt activation in Adv-infected HT-29 cells prior to TNF exposure (Fig. 2-8).
Construction of an Adv vector without a full deletion of the E4 region is difficult because
E40RF6 is critical for adenoviral replication [124]. Therefore, we tested our hypothesis
41
b
a
Increasein shedding
with and without TNF
Adv E4shedding in HT-29cells
I
-
,I
20,000 -
3
ac
* 10,000-
U,
C
0
u
U
u- ..I
0
I
20
Time (hr)
,
U-
I
40
I
TNF
,
I
I
+
Fig. 2-12. Adv E4 shedding in HT-29 cells in consistent with increased Akt activity
(a) HT-29 cells were infected with Adv as previously described and collected at 6, 24, and 48 hours post
infection. E4orf3 transcripts was measured via RT-PCR. Each sample contained an equivalent amount of
total RNA. Data are presented as the mean of 3 biological samples ± S.E. (b) E4orf3 transcription was
measured 24 hours after infection and again 24 hours later in the presence and absence of TNF (48 hours
after start of infection). Data are presented as [mean total E4orf3 @ 48 h - TNF] / [mean total E4orf3 @
24 h].
indirectly by measuring the shedding of E4 transcripts using RT-PCR.
We found evidence of significant E4 transcription 24 hours following Adv
infection (Fig. 2-12a). Adding TNF 24 hours after the initial infection enhanced E4
transcription more than 50% over 24 hours as compared to infected cells that were not
treated with TNF (Fig. 2-12b). Our system is consistent with the demonstrated
mechanisms linking Ad E4 and Akt activation and therefore we conclude Adv E4 has a
role in Adv-TNF synergy. We explore the implications of this high level of Akt
activation in Chapter 3.
2.5.3. Helper-dependent adenoviral vector infection to test effect of
viral transcription
Because of the known activity of Adv E4 and the presence of Adv genome
replication, it is likely that wild type adenoviral genes remaining in the vector backbone
account at least in part for Adv-TNF synergy. As described above (Section 1.1.2),
helper-dependent Adv (HD-Adv), deleted for all adenoviral genes, show the most
42
promise for eliminating vector-induced toxicity. Therefore, we sought to determine if
this new generation of vector would mitigate Adv-TNF synergy.
We compared our standard E1/E3-deleted Adv (Adv.dE1E3 with a CMV
promoter and ,-gal transgene) to an HD-Adv carrying the :-gal transgene under control
of the Rous sarcoma virus (RSV) promoter (Adv.HD). We matched MOI based on blueforming units (b.f.u.), a measure of infectivity in Adv.HD analogous to p.f.u. for FG-Adv.
TNF-induced apoptosis in cells infected with Adv.HD was reduced as compared to
Adv.dE1E3-infected cells but was still significantly above apoptosis in control
(uninfected) cells (p < 0.05; Fig. 2-13a). To confirm a productive infection, we measured
[-gal activity. Cells infected with Adv.dE1E3 exhibited strong ,6-galactivity while
enzyme activity in Adv.HD-infected cells was undetectable (Fig. 2-13b). This difference
could reflect difference in infectivity, thus raising questions about the comparability of
our apoptosis data. However, the presence of Adv genome replication in HT-29 cells
infected Adv.dE 1E3 may result in amplification of the P-gal gene that is unrelated to
infectivity. In addition, the different promoters in the vectors likely lead to different
amounts of activity. A more direct and sensitive measure of viral infection by Q-PCR
a
b
c
40
I-
1
8
60 O
.Ln
o
20
.0
E 40-
X-
o:
- 200
o-
-
40
too low
too low
I
I
o?
c
I
i
'W
I
·
O
LY
9
-
+
Adv.dElE3
+HD
Adv.HDI
Fig. 2-13. Helper-dependent Adv reduces but does not eliminate sensitivity to TNF.
HT-29 cells were infected with either Adv.dElE3 (6 x 108pfu/ml) or Adv.HD (6 x 108 bfu/ml) or treated
with buffer only and then stimulated with 100 ng/ml TNF. Cells were collected 48 hours after TNF
addition and (a) stained for caspase-cleaved cytokeratin and analyzed by flow cytometry and (b) assayed
for presence of ,-gal. Total P-gal enzymatic activity was measured and normalized to protein content of
sample. (c) Cells were infected as in a but LY294002 was added one hour prior to TNF addition. Cells
were collected 24 hours after TNF addition, stained for caspase-cleaved cytokeratin, and analyzed by flow
cytometry. In all cases, data is presented as the mean of 3 biological replications i SE.
43
would be necessary to compare infection efficiency. We also measured apoptosis in the
presence of PI3K inhibition, the upstream activator of Akt, using20 FM LY294002.
Surprisingly, Adv.HD were very sensitive to PI3K-Akt pathway inhibition, with
apoptosis increasing more than 5-fold to 40% and the size of the apoptotic population
rose to almost 30% (Fig. 2-13c). This data indicates that Adv-infected cells' heightened
dependence on Akt activity to confer protection against TNF-mediated apoptosis may not
be entirely caused by the presence of wild-type viral genes. However, another
convoluting factor is that propagation of Adv.HD vectors requires a helper virus, which
invariably contaminates the final vector product [16]. The most effective way to test the
effects of Adv wild-type gene products would be to inactivate transcription using UVpsoralen exposure [126]. This is an important control for future work.
44
CHAPTER 3
3. Insight into the mechanism of Adv-TNF synergy through
quantitative signaling measurements
3.1. Introduction
In this study we apply our quantitative methods of signaling and apoptosis
analysis to further the molecular-level understanding of the engineered Advs used in gene
therapy applications. Specifically, we deconstructed the contributions of Adv and the
transgene to understand how Adv-induced cellular changes synergize or antagonize the
effects of certain transgenes, confounding interpretation of their normal role in cells.
3.2. Adv modification of the TNF-induced signaling network
Because Adv infection sensitizes cells to TNF-induced cell death, we reasoned
that Adv might modulate one or more kinase signaling pathways centrally involved in the
TNF-activated network (Fig. 3-1A). Cell death responses to TNF are highly regulated by
a number of kinase signaling pathways [127-129]. In addition, Adv E4 is known to
mimic growth factor signaling, leading to activation of PI3K and Akt [62, 64].
Therefore, we measured three kinases (MK2, IKK, and JNK1) that are distributed within
the TNF signaling network (Fig. 3-1A, blue shading) as well as two growth-factor
activated signals (Akt and ERK; Fig. 3-1A, lavender shading). To measure Adv-induced
activity changes leading to sensitization, we treated Adv-infected and uninfected HT-29
cells with TNF (100 ng/ml) and collected cell lysates at 13 time points over 24 hours.
From these cell extracts, we measured MK2, IKK, JNKI, Akt, and ERK activities by
using a high-throughput quantitative multiplex kinase assay [111]. To account for
45
A
----------~
---ill ------,
I
I
I
- - -
TRADD
_
FADD
pathway
IRS1
PI3K
Gm2
Ras
Sos
l
RIP
--+
IKK
I
I
I
~
NFkB
J--t
:: r---r------- -----Gasp. _I
3
__
clAPs
----
Rac
--------'-----
S6K
1
1
mTOR
MEKK1
MEK~~
---
---
i
1
~se
E~RK
AP-1!l
HSP27
p38
pathway
XIAP ---------
I
.... --
Rheb
t
~::~:ene~---
L_
.-
:
Shc
ERK
RAF
T l ! [~!B I
BF
~
SFLIP
+
TRAF2 MAP3K
1
GyloG Apaf
+ Gasp.
I
'I
I
LFLIP
TRADD
•
I
AP-1
genes
I
I
~
JNKl
-.
Bcl-2
Bad .-------------------MIIO~~~~~1"al
!...
""
Pldlns
!
Akl
pathway
Piase
PDK1
I
---t
Rsk
TSG1
PTEN
-------t_+_.J
-~-------------------------
TSG2
JNKpKa~~:ay
:
Akt
I
,
FKHR
AFX
Piase
J-
I
Forkhead
genes
pathway
Gaspase-3& cytokeratincleavage
-
Activation
-
Inactivation -SlOw
--
Cross-talk
..
~r, 1Q .~.~40a
.~.~2°BSC
2°B
B
MK2
0
01
c_
~
1
.~.g
*
0
~ E.
.c 0
cr:v.u
QI--
III
V
a.
~
III
10
TNF
20
0
-40
-20
o
*
0'1'"
-.u
.!!1>
va.
~
LV
Akt
C10
~/o~
g-
.2:.p 10
R;.u
Qj.2:
cr:v.u
c
58
Adv +
o
TNF
10
TNF
20
Time (hr)
Time (hr)
uo
ERK
10
E.
01 '"
1ii
.c
V
~
0
0
*
*
QlC
.~.g
-.u
10
a::
III
TNF
Adv+
TNF
0'1'"
~ E.
is 8.
~ '"
~
o
0
-20
10
TNF
20
.~.~2°B' ?Y
SP
TNr
QlBlO
'~',ij
o
TNF
Adv
~ Q. 0
.c 0
v 0.
~~
QI'-
~
o
vra
'ClJ .-'"
Time (hr)
JNKl
cr:v.u
Adv + ~~:_
Q)'-
a.
20
Time (hr)
1.
c:
.!!1>
-20
o
~
III
-20
10
Adv +
TNF
20
TNF
Akt
MK2
Adv+
TNF
"
Time (hr)
Apoptosis
46
./
Adv +
TNF
Fig. 3-1. (Previous page) Adv infection upregulates TNF-induced pro- and anti-apoptotic signaling.
A, Schematic representation of the signaling pathways induced by TNF (shaded in blue) and insulin
(shaded in lavender). The kinases measured (MK2, IKK, JNK 1, Akt, and ERK) are highlighted in yellow.
Figure is modified from [47] with permission. B-C, Dynamic activation status and inhibitor response data
for (B) TNF-activated pathways and (C) growth factor-activated pathways in uninfected cells treated with
100 ng/ml TNF (TNF; blue) and Adv-infected cells treated with 100 ng/ml TNF (Adv + TNF; red).
Lysates were collected at 0, 5, 15, 30, 60, and 90 minutes and 2, 4, 8, 12, 16, 20 and 24 hours and kinase
activity was measured by a high throughput kinase activity assay and normalized to total protein content.
Results are plotted as the mean relative activation of two biological replicates normalized to activity of the
untreated control at 0 minutes (i.e., uninfected cells). The significance of the difference between each pair
of curves was calculated by two-factor ANOVA (MK2: p < 0.01; IKK: p < 0.001; JNK1: p = 0.71; Akt: p <
10-4; ERK: p = 0.43). Apoptosis in the presence of kinase inhibition was measured by flow cytometry for
cleaved caspase-cytokeratin. The following inhibitor concentrations were used: 10 [tM SB202190 (SB);
20 [tM SC-514 (SC); 10 [tM SP600125 (SP); 20 utM LY294002 (LY); 10 tM U0126 (UO). Cells were
collected at 24 hours, rather than 48 hours, to minimize baseline apoptosis resulting from inhibition.
Measurements are plotted as the mean % change in apoptosis resulting from inhibition (i.e., mean %
apoptosis in the presence of inhibitor - mean % apoptosis without inhibitor) of three biological replicates
(SC, SP, UO) or six biological replicates (SB, LY) ± s.e.m. Note that LY plot ranges from -40 to +40 (as
compared to -20 to +20 for other inhibitor plots). Changes are labeled as significant (*) if p < 0.05. D,
Adv-TNF synergy via pro-apoptotic p38-MK2 signaling and anti-apoptotic PI3K-Akt signaling. Synergy is
illustrated schematically as an operational amplifier, with the strength of the Adv or TNF input into either
the MK2 or Akt signaling time course indicated by line weight. Amplification of the inputs results in
activation (MK2) or inhibition (Akt) of apoptosis.
the contribution of Adv infection alone, kinase activities were also measured in Advinfected cells after mock stimulation. Finally, to determine the phenotypic consequences
of each kinase activity, we measured the change in TNF-stimulated apoptosis in the
presence of pathway-specific, small-molecule inhibitors.
We observed potent activation of the TNF signaling network when either
uninfected or Adv-infected cells were treated with TNF (Fig 3-1 B). Among TNF
pathways, Adv infection significantly increased TNF-induced MK2 and IKK signaling (p
< 0.01, two-factor ANOVA) but not JNKI (p = 0.71). These altered intracellular patterns
were a product of Adv-TNF synergy, because Adv infection alone contributed minimally
to JNK1, IKK, and MK2 activation (Fig 3-2). Small-molecule inhibition of the measured
pathways confirmed the importance of MK2 and IKK for Adv sensitization. Inhibition of
the upstream activator of MK2 with SB202 190 significantly decreased apoptosis (p <
0.01), suggesting a pro-apoptotic role. Conversely, IKK inhibition with SC-514 resulted
in a small but significant increase in apoptosis (p < 0.05), consistent with the recognized
anti-apoptotic functions of IKK [38]. JNK1 was transiently activated by TNF but was
unchanged by Adv infection; thus, as expected, JNKI inhibition with SP600125 did not
47
t~.---.I'
MK2
l~
10
~ 'c'"
1
~.----10
QJe:
.2:
QJ
Q)
10
",
.!!1>
QJ'-
"oW
O:v
20
e:
.==.g
...
.g
~ ~ 1
0: ~
'"
~
o
JNKl
IKK
o
10
o
20
Time (hr)
Time (hr)
1
10
20
Time (hr)
Akt
Adv
e: 10
Adv + TNF
.~.g
...
",
.!!1>
~'C
5
'"
o
10
20
Time (hr)
o
10
20
Time (hr)
Fig. 3-2. Adv infection alone upregulates Akt but not other TNF-growth factor-activated pathways
Dynamic activation status for TNF-activated pathways and growth factor-activated pathways in Advinfected cells following mock stimulation (Adv; green). Activation status of Adv-infected cells treated
with 100 ng/ml TNF (Adv + TNF; red) is plotted for reference (see also Fig. 3-1 B-C). Lysates were
collected at 0, 5, 15,30, and 60 minutes and 2, 4,8, 12, 16,20 and 24 hours and kinase activity was
measured by a high through kinase activity assay. Results are plotted as the mean relative activation of two
biological replicates normalized to activity of the untreated control at 0 minutes (i.e., uninfected cells). The
significance of the difference between each pair of curves was evaluated by two-factor ANOV A (MK2: p <
10.12; IKK: p < 0.00 I; JNK I:p < 10.6; Akt: p < 0.0 I; ERK: p < 10'\ Note that Akt activity is plotted on a
linear scale (as compared to a semi-log scale for the other kinases).
significantly affect Adv-sensitized
cell death. We conclude that, of the TNF-dependent
kinases measured, the p38-MK2 pathway is the most important pro-apoptotic signal
contributing to Adv- TNF synergy.
Multiple growth factor pathways were also strongly activated by Adv and TNF
(Fig. 3-1 C). Similarly to TNF -induced JNK I, ERK was transiently activated by TNF but
was not differentially affected by prior Adv infection.
Thus, inhibition of the upstream
ERK kinase with UOI26 did not significantly affect Adv-TNF synergy.
In contrast, Adv
infection alone induced a three- to five-fold increase in Akt activity for up to 24 hours (p
< 0.01; 3-2), consistent with the PI3K-activating
role ofE4 [62, 64]. Akt activation was
further increased by TNF treatment, resulting in eight- to ten-fold greater activity than in
1
uninfected TNF-treated cells (p < 10- °). Akt transmits many key pro-survival signals
[130], and we found that upstream PI3K inhibition with LY294002 dramatically
48
increased synergistic apoptosis mediated by Adv and TNF (p <
10 -4).
Together, these
experiments indicate that Adv infection synergizes with TNF by augmenting an existing
p38-MK2 pro-death signal and adding a new PI3K-Akt pro-survival signal to the network
(Fig. 3-1D).
3.3.Role of Akt in TNF sensitization
3.3.1. Upregulation of endogenous Akt activity in Adv-infected cells by
insulin
Despite potent Akt signaling in Adv-infected cells (Fig. 3-1C), synergistic
apoptosis still occurred after TNF-stimulation. The LY294002 inhibitor experiment
confirmed a pro-survival role of Akt by downregulation (Fig. 3-1C). However, the
contribution of some pro-survival molecules has been shown to vary depending upon
whether protein function is increased or decreased [94]. To activate endogenous Akt, we
therefore used the growth factor insulin, which has previously been shown to be
reasonably specific in its activation of Akt in HT-29 cells [47]. Additionally, insulin is
known to inhibit TNF-induced cell death [ 131] through upregulation of Akt activity in
HT-29 cells sensitized with interferon-y (IFN) [132]. To test if insulin similarly provides
protection from TNF-induced apoptosis in HT-29 cells sensitized by Adv, we added 100
nM insulin together with TNF. In contrast to IFN-sensitized cells, insulin did not
decrease TNF-induced apoptosis in Adv-infected cells (p = 0.15; Fig. 3-3A).
Surprisingly, this was not due to a lack of Akt signaling, since insulin increased Akt
activity to a similar extent in both Adv-sensitized cells and IFN-treated cells (p < 0.05;
Fig. 3-3B). There was not a significant decrease in TNF-induced MK2 activity in Advinfected cells as a result of insulin stimulation (p = 0.49; Fig. 3-3C), confirming the
selectivity of insulin as an Akt-selective agonist. To verify that IFN- and Adv-sensitized
cells behaved similarly when Akt was inhibited, we measured TNF-induced apoptosis
and Akt activity in the presence of LY294002. For both sensitizing agents, LY294002
49
Cleaved caspasecytokeratin
A
B
C
Akt activity
MK2activity
20c
._o
0 9lo
0,
-'a
C.
Im
20
I 1v
0.
ZR
.> c
10-
0
5
t-
oC
.>
0
A5A j
u
-
Insulin IFN
+
-
v
Insulin
-
+
I
+
I
I
-
+
0
Insulin-
IFN+TNF Adv+TNF
IFN+TNF Adv+TNF
Cleaved caspasecytokeratin
D
Oz
I
+
Adv + TNF
F
E
MK2 activity
Akt activity
100
I
0IE
._
Q.
O
I
8 *E
'. O 20
I
5
50
> -c
C
.> to c0
' U.a!
d)n
GJ3
Mo
'U
a
GJG>
i
0
LY -
I
i
+
-
F
+
IFN+TNF Adv+TNF
1-1
0
- 2
WW
Gi.>
0
I '"1
0
LY -
I
10
-CU
#_-_
.C
+
I
I
-
+
IFN+TNF Adv+TNF
0
LY -
+
Adv + TNF
Fig. 3-3. Insulin increases Akt activity but does not reverse TNF-induced apoptosis or reduce MK2
activity in Adv-infected cells
Adv- and IFN-sensitized HT-29 cells were treated with TNF (100 ng/ml and 50 ng/ml, respectively) in the
presence and absence of(A-C) 100 nM insulin or (D-F) 20 RM LY294002. Cells were lysed at 12 hours
and assayed for kinase activity or collected after 24 hours and assayed for apoptosis by flow cytometry for
cleaved caspase-cytokeratin. A,D, Change in TNF-induced cell death. Measurements are plotted as the
mean of six biological replicates (2 independent experiments) + s.e.m. B,E, change in TNF-induced Akt
activity. Measurements are plotted as the mean of six biological replicates (2 independent experiments) ±
s.e.m. C,F, Change in TNF-induced MK2 activity in Adv-sensitized cells. Measurements are plotted as
the mean of three biological replicates ± s.e.m. Changes are labeled as significant (*) ifp < 0.05.
inhibition significantly increased TNF-induced apoptosis (p < 10-9 ; Fig. 3-3D) and
reduced Akt activity (p < 0.001; Fig. 3-3E), as expected. Again, there was no change in
MK2 activity in Adv-sensitized cells as a result of Akt inhibition (p = 0.45; Fig. 3-3F),
indicating that MK2 pro-apoptotic signaling is separable from Akt activity changes in
Adv-infected cells. The lack of apoptotic rescue shows that the insulin-induced Akt
activation in Adv-infected cells does not result in functional anti-apoptotic signaling.
50
3.3.2. Saturation of downstream Akt effectors by Adv
The lack of pro-survival signaling by insulin-stimulated Akt could be explained
by an inability to phosphorylate downstream Akt effectors in Adv-infected cells. To test
this, we measured phosphorylation levels of multiple Akt substrates in Adv- and IFNsensitized HT-29 cells after TNF stimulation alone and in the presence of LY294002 or
insulin. We used an Akt-phosphosubstrate (Akt-pSub) antibody that binds specifically to
phosphorylated substrates of Ser/Thr kinases that recognize the RxRxxS/T motif [133].
In addition, we measured levels of phosphorylated glycogen synthase kinase-3 a and 13
(pGSK-3a/P), an established substrate of Akt [134]. Western blotting of HT-29 lysates
showed that pGSK-3a/13 and multiple Akt-pSubs were significantly upregulated in Advsensitized cells as compared to IFN-sensitized and untreated cells, even before the
addition of TNF (Fig. 3-4A). This is consistent with the high baseline Akt activity
induced by Adv (Fig. 3-3C). Two Akt-inducible proteins, pGSK-3a and Akt-pSub4 2 kDa
(Fig. 3-4A), were quantified by densitometry and normalized to the Adv- or IFNsensitized zero-minute band. For both IFN and Adv pretreatments, TNF-inducible
phosphorylation of GSK-3a and Akt-pSub4 2 kDawas reduced with LY294002 treatment
(Fig. 3-4,B-C), confirming that both proteins are reliable indicators of functional Akt
activity. In IFN-TNF-treated cells, insulin significantly increased pGSK-3a and AktpSub42kDa,but there was no significant insulin-induced phosphorylation of these proteins
in Adv-TNF-treated cells, despite clear insulin-induced activation of Akt in Adv-infected
cells (Fig. 3-3B). Thus, Akt activation in Adv-infected cells sustains phosphorylation of
downstream effectors, but further Akt activation induced by insulin fails to increase Akt
effector phosphorylation. We conclude that Adv infection in combination with TNF
saturates the ability of Akt to phosphorylate its substrates, contributing to the observed
loss of anti-apoptotic signaling.
51
A
pGSK-3a ~
pGSK-3~
__
~-
.........
....
B
_
---
42kDa
A kt-p S u b
SOkDa
-SOkDa
Adv
IFN
Controls
.,
~ ....
..
pGSK-3a
_c
~ 'E
~o
*
00
+ LY
alone
ro ~.~
"~
>-
1.0
"Vi
o ~
..r::.
s;-
VI
-Q)
o >
..r::. 0
0..
~
0.0
+ Ins
DO
+ LY
IFN + TNF
C
Akt-pSub
alone
+ Ins
Adv + TNF
42kDa
+ LY
alone
+ LY
+ Ins
IFN + TNF
alone
+ Ins
Adv+ TNF
E
Aktphospho(Serrrhr)
Substrates
..,
•
-
- .........
-
- - ..... -Controls
---
ti .....
l..
IFN-y
52
-SOkDa
42 kDa
-
V •
Adv
37 kDa
-'-25kDa
Fig. 3-4. (Previous page) Adv infection upregulates multiple Akt-pSubs before the addition of TNF and
saturates phosphorylation of downstream Akt effectors
A, Cells were treated as described in Fig. 3, lysed at 12 hours and analyzed by Western blot by probing with
anti-pGSK-3a/0 and anti-Akt-pSub. 50 [ig protein was loaded into each lane. The full blot Akt-pSub blot
is shown in E. B-C, Adv-sensitized and IFN-sensitized (B) pGSK-3a and (C) Akt-pSub4 2kDa bands were
quantified by densitometry and normalized to the sensitized (Adv or IFN) 0-min band. Measurements are
plotted as the mean of three biological replicates ± s.e.m. Changes are labeled as significant (*) ifp < 0.05.
3.4. Construction of a global Akt-survival dose-response curve
combining Adv- and IFN-mediated sensitization
Insulin caused a similar absolute increase in Akt signaling in uninfected and Adv-infected
cells, both adding roughly three Akt-activity units after stimulation (Fig. 3-3B).
However, relative to the baseline Akt signal for each treatment (Fig. 3-5A), the "fold
activation" of Akt induced by insulin was larger for IFN-sensitized cells (3.7-fold) than
for cells infected with Adv (2-fold). Recently, it has been suggested that cells become
quickly desensitized to absolute levels of signaling and are often more responsive to
gradients of signals [135]. To examine the importance of fold changes, we measured Akt
activity and TNF-induced apoptosis under a set of Adv, IFN, LY, and insulin conditions
that gave a range of TNF-induced apoptotic responses (see Fig. 3-5 legend and Methods).
Then, we preprocessed the data by normalizing IFN- and Adv-sensitized cells to their
respective zero-minute values (Fig. 3-5A). When cell viability was plotted against this
fold activation of Akt, we found that both IFN-TNF and Adv-TNF treatments collapsed
onto a common sigmoidal dose-response curve (Fig. 3-5B). Preprocessing measurements
of Akt-effector substrates in the same manner resulted in a dose-response identical to that
of Akt, with the exception of a two-fold faster transition steepness between minimumand maximum-observed viability (Fig. 3-5C). The faster viability transition for Akt
effector substrates is consistent with the known ultrasensitivity that arises from sequential
signaling cascades [136]. The existence of a single function-relating
fold activation of
the PI3K-Akt pathway to viability for a wide range of treatment conditions-suggests
that the relative change in Akt signaling is more important than its absolute level for
determining TNF-induced cell fate.
53
A
B
C
6
100
100
C
•
c --
o E
.~o
>-0
QJ
u ...
ro>
4
"Z:;
.~
~
~
ra ra
OJ
~
>-- c
.-
~ ~
"ii :v
c:: >
.s
a
Akt.p5u~'kO.
.~
60
~
~
20
(0.005. O. 18)
o
)
20
0.2
0.1
1.0
Adv
Fold activation
Fig. 3-5. Adv-infected
~
_JT'_~
60
V>
••
0
IFN
pG5K-3u
.L.:
ro>
V>
2
.0
are trapped near the plateau of a global Akt-survival
1.0
Fold phosphorylation
dose-response
curve.
A, Cells were infected with Adv or treated with IFN, lysed after 24 hours and measured for Akt activity.
Following Adv or IFN sensitization, cells were treated with TNF (100 nglml and 50 nglml,
respectively), TNF + LY294002 (20 !-lM), TNF + insulin (100 nM), or TNF + LY294002 + insulin (B
only). Conditions were chosen to generate a range of Akt activities. Measurements were taken at 12 hours
for Akt activity and effector phosphorylation and at 24 hours for apoptosis as described in Fig. 3. B, Cell
survival versus fold activation Akt activity. Akt activities were normalized to sensitized O-min values from
A. Akt measurements are the mean of six biological replicates:!: s.e.m. Survival measurements are 100% the mean of six biological replicates:!: s.e.m. The arrow indicates Adv-infected cells without L Y
inhibition. C, Plot of survival (from B) versus effector phosphorylation levels quantified in Fig. 4. Solid
red markers indicate IFN-sensitized conditions and open red markers indicate Adv-sensitized conditions
(same convention as in B). B-C, Sigmoid function and parameters were calculated as described in
Methods. Transition steepness (Ts) and the upper and lower bounds of the 90% confidence interval
(parentheses) are labeled on the graphs.
B,C,
The calculated Akt-survival
dose-response provides a quantitative explanation for
how insulin is unable to attenuate synergistic TNF-induced apoptosis in Adv-infected
cells (Fig. 3-3A). First, by dramatically increasing the baseline Akt signal (Fig. 3-5A),
Adv infection reduces insulin's ability to induce a strong fold activation of the PI3K-Akt
pathway (Fig. 3-3B). This prevents Adv-infected
cells from moving far rightward on the
dose-response curve in Fig. 3-5B. Second, our panel of experimental treatments indicates
that the role of Akt signaling in these cells is most apparent when pathway activity is
reduced to levels that are below baseline signaling. Adv-infected
positioned very near the upper viability-plateau
cells are thus
of the dose-response
(Fig. 3-5B, arrow),
suggesting a limited capacity to reduce TNF-induced apoptosis further. Together, this
indicates that Adv infection traps cell in a network state that is unable to transmit
additional anti-apoptotic information via the Akt pathway.
54
3.5. Further considerations
Our aim in this study was to explore how Adv alters human epithelial cell
signaling and apoptotic responses to the inflammatory cytokine TNF. Compared with
wild-type adenovirus, there are relatively few studies on Adv infection and its interaction
with host-cell signaling pathways. Understanding how Adv-infected cells respond in the
context of a cytokine-rich environment, such as an inflamed tissue, is critical to
improving gene therapy applications. MOIs at the injection site can be very high in vivo
[20, 110], similar to the ratios used in this study. High concentrations of Adv in vivo
activate innate immune responses independent of viral gene transcription [24, 25, 137].
Neutrophils, natural killer cells, and macrophages are recruited to the site of infection
where they secrete cytokines such as TNF. TNF and TNF-family cytokines are directly
responsible for apoptosis in the target cells [24, 138]. Adv infection, immune cell
activation, cytokine release, and host cell death are complex, overlapping events in vivo.
By studying Adv infection and apoptosis in the context of one defined cytokine in vitro,
we were able to discover that Adv and TNF cooperate to promote cell death in infected
cells. Adv-TNF synergy could be an important contributor to gene therapy side effects,
which have been reported in Adv clinical trials [139-141].
E1/E3-deleted Advs are commonly used in laboratory experiments as delivery
vectors to overexpress or inhibit pathways and elucidate function. Such studies assume
that changes caused by Adv infection and changes caused by the transgene are linearly
cumulative. If true, then infection with an Adv that lacks the transgene is an adequate
control. However, the results here show that Adv itself can significantly alter signaling
pathways in the cell, both at the basal level (e.g., Akt) and in response to TNF (e.g.,
MK2). For Adv and TNF, cell response to Adv and TNF is highly non-linear, with the
underlying impact of infection becoming apparent only after cytokine stimulation. This
reveals an important caveat for interpreting laboratory studies involving Adv. Controls
with "empty" Adv might well be complemented by delivering the transgene (and
associated control) with a different vector, such as liposomes for transient expression or
lentiviruses or adeno-associated viruses for stable expression.
55
A surprising result of our work is that regimes exist where increased Akt activity
produces limited pro-survival benefit. We showed that this is partly due to saturation of
Akt effector phosphorylation, but it remains unclear why saturation occurs. Saturation of
the PI3K-Akt pathway has also been reported in other contexts-for
to high concentrations of certain growth factors [142]-suggesting
instance, in response
that saturation is not
Adv-specific. The simplest molecular explanation for saturation in Adv-infected cells is
that Akt and its substrates are negatively regulated by different protein phosphatases. Akt
forms complexes with protein phosphatase 2A (PP2A) [143] and a protein phosphatase
2C (PP2C) family member [144], which dephosphorylate the T308 and S473 sites on
Akt. By contrast, the Akt substrate GSK-3f coassociates with protein phosphatase 1
(PPI) [145], as well as a PP1 regulatory subunit [146], which is a recognized substrate of
GSK-3 [147]. Thus, PP2A and PP2C could allow Adv- and insulin-mediated activation
of Akt, but PP1 may block additional phosphorylation of GSK-3 P. Such Aktindependent control of GSK-3 P and the closely related isoform GSK-3a could be
particularly important for Adv-TNF synergy and apoptosis, because GSK-3Pfsignaling is
important for resistance to TNF-induced cell death [148].
Our results suggest several mechanisms by which Adv and TNF may synergize to
induce apoptosis. One pathway we identified to be directly involved was via the stress
kinase MK2. Among its substrates, MK2 directly phosphorylates small heat shock
protein 27 (Hsp27), which reduces its ability to protect against oxidative stress [149].
Reactive oxygen has recently been implicated in TNF signaling [38, 83], which together
is consistent with the pro-apoptotic role we observed for MK2. At the posttranscriptional level, MK2 contributes to the regulation of several cytokines, including
TNF [80]. Adv-TNF activation of MK2 in vivo may start a positive feedback loop in
cells of the innate immune system that augments cytokine production. Adv infection may
also induce secretion of interferon-alpha (IFN-a) in host epithelial cells. Upregulation of
IFN-a could partially inhibit protein synthesis within the cell [150], limiting NF-KBmediated transcription of molecules, such as c-FLIP [151], that are activated in response
to TNF and prevent apoptosis. Any attenuation of NF-KB signaling would need to occur
downstream of IKK, because TNF-induced IKK activation was not decreased but
increased after Adv infection.
56
One possible systems-level explanation for Adv sensitization is that IKK-NF-KB
and PI3K-Akt collaborate via a common dose-response for survival. In this case,
saturation of survival signaling by Adv-mediated upregulation of Akt may proportionally
reduce the pro-survival contribution of IKK. NF-KB is a recognized determinant for
survival in the presence of TNF [75-77], but we found that small molecule inhibition of
IKK did not cause a pronounced increase in TNF-induced apoptosis in Adv-sensitized
cells (Fig. 3-1B). These data support the hypothesis that Akt activation by E4 of Adv
[62] could saturate total pro-survival signaling from both Akt and IKK-NF-KB. Akt and
IKK have been implicated in a common signaling pathway under certain circumstances
[152, 153] and apoptosis has been shown to be determined by combinations of multiple
intracellular signaling pathways [89, 97]. Cooperation between Akt and IKK may
therefore uncouple the contribution of IKK to TNF anti-apoptotic signaling.
57
CHAPTER 4
4. Testing a fundamental biological principle through data-driven
modeling of Adv-TNF synergy
4.1.Introduction: cell-specific versus common processing
The fundamental components of many intracellular signaling pathways are
common to all cells [85]. Intracellular networks arise from complex modular interactions
between the receptors, adaptors, and enzymes that together relay signals [86] (Fig. 4-1 a).
Experimental [154] and literature-based [155] reconstructions of signaling networks often
assume that there exists a common network that is qualitatively the same for all cells
[88]. However, stimulating or perturbing the molecular network as a whole with
mitogenic, pathogenic, or inflammatory stimuli often causes distinct phenotypes that are
specific to cell type [36, 156]. For example, apoptosis [157] and chemokine release [158]
are two cell-specific responses induced by the cytokine, tumor necrosis factor (TNF).
Likewise, the cellular outcomes of virus infection depend upon the particular host tissue
[159]. How are cell-specific responses achieved via a common signaling network?
Essential to answering this question are the downstream effector proteins (including
cytoplasmic substrates, transcription factors, and terminal enzymes; Fig. 4-la), which
ultimately convert (or "process") signals into cell-specific phenotypes. Cell specificity is
thus an important obstacle to understanding how signaling networks control cellular
behavior and to developing drugs that treat cellular dysfunction [160, 161].
In this Chapter, we use our Adv-TNF experimental system to explore the
fundamental biological question of signaling cell specificity by demonstrating that a
single PLSR model can predict Adv-induced apoptosis across cell types. In addition, we
use principal component analysis to develop hypotheses about Adv modification of TNFinduced signaling and propose areas for further analysis.
58
b
a
Cell-specific processing
Cytokines,
Stimuli
Substrates,
tx n. facto rs
Hela
signals
signals
~
~
HT-29
HeLa
effectors
effectors
~
~
Low cell
death
High cell
death
~
Responses
Apoptosis,
proliferation
/
HT-29
~
Effectors
Adv+ TNF
'\.
/
~
Signals
Common processing
Adv+ TNF
pathogens
Receptors,
adaptors, kinases
C
'\.
Model
HeLa
HT-29
signals
signals
'\.
Common
/
Low cell
death
/
}T
effectors
y
= f(x)
\'h"" }J Y
death
Fig. 4-1. Hypothesis description: Cell-specific processing versus common
processing.
(a) Flow of biological information from stimuli through signals and effectors to responses. (b) Diagram of
a cell-specific effector-processing mechanism
linking Adv- TNF treatment to apoptosis in HT -29 and HeLa
cells. (c) Diagram ofa common
effector-processing mechanism
linking Adv-TNF
treatment to apoptosis
in HT -29 and HeLa cells. A model of effector processing, f(x), links signals to responses.
4.2. Experimental and computational approach to testcommon
effector-processingmechanism
4.2.1. Signaling inputs from multiple cell lines
We began by addressing the problem of cell-specificityin a defined experimental
setup with clear outcomes that are cell-specific. Adenovirus (Ad) and adenoviral genetherapy vectors (Adv) have been shown to synergize with TNF in many cellularcontexts
[100-102]. Adv uptake by cells of the immune
system results in TNF release [20], and
Adv binding and entry activates many signaling pathways shared with the TNF network.
In epithelialcells,we have found that synergistic apoptosis caused by Adv- TNF
resultof Adv-induced
is a
changes in multiple key signaling pathways of the TNF network
[162]. Interestingly,the extent of Adv-TNF
synergy is specific to cell type: HeLa
cervical carcinoma cellsare much more sensitive to apoptosis than HT-29 colon
adenocarcinoma
cellswhen pre-infected with the same virus-to-cellratio and stimulated
with the same concentration ofTNF
[162] (Fig 4-2a).
59
One can envision two extreme strategies for mediating cell-specific outcomes to a
single stimulus. The first possibility is that both signal activation and effector processing
are cell specific (Fig. 4-1 b). In this scenario, the same signaling event (for instance, the
fold activation of a protein kinase) would contribute differently to a phenotype,
depending on the particular cell type in which the signal is activated. A simpler
alternative is that differences in signal activation are sufficient to confer cell-specific
variations in phenotype (Fig. 4-1 c). In this case, different cell types could use a common
effector-processing mechanism to convert cell-specific signals into cell-specific
responses. Typically, the details linking signals to effectors and responses are
incomplete. Therefore, we set out to test these two hypotheses by measuring the
intracellular network and then mathematically calculating the bulk molecular contribution
of multiple critical signals to effector processing (Fig. 4-1b,c). We show that the
resulting model strongly supports a role for common effector processing in cell
specificity (Fig. 4-1c).
Our experiments focused on an Adv-TNF treatment combination in which the
extent of HeLa apoptosis was 2.5-fold higher than that observed for HT-29 cells (Fig. 42a; see Methods for details). HeLa cell-specific sensitivity to apoptosis is a result of
Adv-TNF synergy, because TNF-induced cell death was reduced five-fold in the absence
of Adv pre-infection (Fig. 4-2b). We have previously shown that measurements of
kinase catalytic activity are particularly useful signals for monitoring information flow
a
Adv+TNF-induced
apoptosis
b
TNF-induced
HeLa
apoptosis
M
40 -
30
20
0.
o
o20
0-<
I
-
10
0
HT-29 HeLa
100 m.o.i.Adv
-
+
Fig. 4-2. Cell-specific apoptotic responses to Adv-TNF stimulation of HT-29 and HeLa cells
(a) Apoptosis in HT-29 and HeLa cells following Adv infection at 100 m.o.i. and stimulation with 100
ng/ml TNF for 48 hours. Data are presented as the mean of three biological replicates ± S.E. (b) Apoptosis
in HeLa cells uninfected or Adv-infected at 100 m.o.i. and stimulated with 100 ng/ml TNF for 24 hours.
60
HeLa cells:
a
Uninfected
versus Adv-infected
ERK
c
.2
Akt
Uninfec led
Adv-infected
10
ra
.~
u
ra
"'C
<5
u..
o
10
20
Adv-infected
IKK
MK2
,k:: ':t~ 'b '~b
o
at 100 m.o.i.:
10
20
o
10
20
o
10
20
o
10
20
Time (hr)
Time (hr)
Time (hr)
Time (hr)
Time (hr)
b
JNKl
HT-29 versus He La cells
ERK
5
.;;
ra
.~
10
I _~~~:
u
ra
"'C
<5
u...
~
o
10
20
Time (hr)
o
10
20
Time (hr)
o
10
20
o
Time (hr)
10
20
Time (hr)
o
10
20
Time (hr)
Fig. 4-3. Cell-specific signaling responses to Adv infection and TNF stimulation of HT-29 and HeLa cells
TNF-activated kinase activity time courses in (a) HeLa cells uninfected (blue) or infected with 100 m.o.i.
Adv (red) and (b) HT-29 (black) and HeLa (red) cells following Adv infection at 100 m.o.i. (HeLa data
same as in (a». Lysates were collected at 0, 5, 15, 30, and 60 minutes and 2, 4, 8, \ 2, 16, 20 and 24 hr and
kinase activities were measured by a high throughput kinase activity assay[ 111]. Results are plotted as the
mean relative activation of two biological replicates normalized to activity of the untreated, uninfected
control at zero minutes.
through a network [47, 97, 111]. Consequently,
for Adv-TNF treatment combinations,
the dynamics of five protein kinase activities (ERK, Akt, JNKl, IKK, and MK2) were
measured in both HT -29 and HeLa cells at 13 time points from 0-24 hr after TNF
addition. Kinase activities were also measured in uninfected He La cells stimulated with
TNF to confirm that Adv had perturbed these pathways en route to synergizing apoptosis.
Two-way analysis of variance (ANOVA) revealed significant alterations in ERK, JNKl,
IKK, and MK2 activity time courses between uninfected and Adv-infected HeLa cells (p
< 0.05; Fig. 4-3a), consistent with the known crosstalk between Adv and TNF signaling
pathways.
Importantly, when the network activation ofHT-29
cells and HeLa cells was
compared, we observed complex differences in the dynamics of all pathways-ERK,
IKK, JNKl, and MK2-between
cell types (p < 0.05; Fig. 4-3b). Thus, Adv infection
61
Akt,
perturbs the TNF-activated kinase network in epithelial cells [162] and the extent of this
perturbation is cell specific.
We compared the new kinase-activity
data with time courses previously measured
[69,162] in HT-29 cells stimulated with various combinations of Adv, interferon-y
(lFNy, another TNF-sensitizing
factor[163]), and TNF (Fig. 4-4a). The total dataset
included seven treatment combinations of Adv, JFNy, and TNF in HT-29 cells and two
combinations in HeLa cells. All measurements were processed via internal
normalization,
rather than control normalization
(see Section 2.4.1), consistent with the
findings in Chapter that fold activation is more important than relative activation.
One-
way hierarchical clustering of these network time courses indicated that the two treatment
combinations in HeLa cells paired more closely with HT -29 treatments than with each
other. This suggested that the HeLa activity profiles, although different from HT-29 cells
for the same stimulus, were still within the realm of network states achievable by HT-29
cells.
Because signaling dynamics between HeLa and HT -29 cells were comparable
overall (Fig. 4-4a) but the apoptotic responses were clearly different, the two effector
hypotheses could be directly tested. Jfthe common-processing
hypothesis were true, then
a processing function that is deduced from HT -29 signaling and apoptosis data (Fig. 4-1 c;
f(x)) should accurately predict the conversion of HeLa signals into HeLa-cell-specific
ERK
MK2
~ d_'
~
Adv (1000 moil
IFNy (200 u/ml)
.~t,iTl
~,LJiLInL
,
ff'
, I,
l'
Time
Omin
24 hr
Low _
.:J
I
TNF (100 ng/rnl)
Adv (100 moil + TNF (100 ng/ml)
TNF (100 ng/ml)
Adv (1000 moil + TNF (100 ng/ml)
Adv (100 moil + TNF (100 ng/ml)
IFNy (200 u/ml) + TNF (5 ng/ml)
'IFNy(200 u/ml) + TNF (100 ng/ml)
High
Fig. 4-4. Colorgram of time-dependent kinase activities in HT-29 and HeLa
HT-29 cells (black) and HeLa cells (red) treated with different combinations of Adv, lFNy, and TNF [69,
162]. Data are presented as the mean of duplicate (Adv conditions) or triplicate (IFNy conditions)
biological samples relative to the zero-minute activity for each condition.
62
apoptosis. Failure of such a model would suggest that cell-specific processing was
required (Fig. 4-lb).
4.2.2. PLS model construction and testing through prediction
To calculate the effector-processing function, we used partial least squares
regression (PLSR), which has been shown to connect cytokine-induced apoptosis to
measurements of the underlying intracellular signaling network (see Section 1.3.2) [97].
The PLSR model was constructed from the five dynamic kinase-activity profiles for each
of the seven HT-29 treatment conditions shown in Fig. 4-4, including signaling metrics
calculated from each time course (see Section 2.4.1). Using flow cytometry, we analyzed
caspase substrate cleavage in HT-29 cells at 24 and 48 hours after TNF stimulation to
provide two quantitative measures of the apoptotic phenotype that the kinase-network
dynamics must fit. The apoptotic effector-processing function of the PLSR model was
derived as a series of latent variables called principal components, which contain linear
combinations of the original kinase activity measurements [97, 118]. Principal
components are calculated iteratively according to the following general steps: 1) a
loadings vector and a scores vector are calculated that describe the line of best fit through
the data (primary variation in the data); 2) this product (the principal component
representation of the data) is subtracted from full data matrix leaving a residual (or
unexplained variation in the data); 3) the next principal component is regressed against
the residual variation in the data [118]. Therefore each successive PLS dimension
contains information not captured by the preceding component.
The root mean square error (RMSE) of apoptotic predictions, a measure of the
difference between the predicted value and the observed experimental value, decreases
monotonically for the calibrated model because all of the input treatments are included.
After several iterations, however, the residuals become very small, and new dimensions
capture only uninformative variations in the data, such as experimental noise and
measurement error. Such information is undesirable for modeling true biological
variation. It is therefore necessary to optimize the number of principal components. This
63
a
b
10
Common
Calibration
100
l
5
~
a:
c:
.!!!
HT-29
Hela
'"
0
C.
0
"Vi
UJ
Vl
0
processing model
>Oi
0.
+
III
:::f:
0
10
Prediction
\J
(l)
50
\J
t-OOt
(l)
•
~
'"
~
tQltllt
(l)
5
.D
0
0
0
~
:9
.24 h
048h
I
I
1 2
50
0
3 4 5 6
Predicted
No. of principal
+
100
% apoptosis
components
Fig. 4-5. Test of a common
effector-processing hypothesis in epithelial cells treated with Adv and TNF.
(a) RMSE
of calibration and prediction of apoptotic outputs by the PLS model as a function of increasing
number of principal components.
An optimum model with three components
was selected. (b) Correlation
between measured apoptosis and predictions of the common-processing
model. Model training with HT -29
apoptosis (black) is compared to model predictions of HeLa apoptosis (red) at 24 hr (filled) and 48 hr
(hollow). Data are presented as the mean of three biological replicateS:l: S.E., and model uncertainties were
estimated by jack-knifing[89].
•
isdone using a method known as cross-validation. In cross-validation, one observation is
leftout and the principal components
RMSE
are used to predict the excluded sample. The
of prediction isthen recalculated; however in this case itis optimized after a
smaller number of components
and increases when the "noisy" dimensions are included.
When we applied this optimization method to our model, the RMSE
three principal components
When
was minimized with
(Fig. 4-5a).
the resulting model was then used to predict TNF-induced
apoptosis in
HeLa cells with and without Adv pre-infection, we found that the model predictions
captured apoptosis measured directly in HeLa cells to within 99% accuracy (Fig. 4-5b).
These predictions of He La apoptosis were as close to experimental observation as the
calibrated fittingof the HT -29 data used to train the model. The remarkable accuracy of
the HT-29 model for predicting HeLa-specific responses suggests that these two cell
types share a common
effector-processing mechanism
apoptosis (Fig. 4-1 c).
64
that converts signal activation to
4.2.3. Biological insights through principal component analysis
The apoptotic effector-processing fimction of the PLSR model was derived as a
series of latent variables called principal components, which contain linear combinations
of the original kinase activity measurements [97, 118]. These principal components form
a set of intracellular basis axes that define the dimensions of the signaling network that
controls apoptosis. IFNy + TNF treatments projected strongly along one of the model
basis axes of the common-processing model predicting HT-29 and HeLa apoptosis (Fig.
4-4c). This axis corresponded to maximum TNF-sensitivity and included early JNK1
activity and late IKK activity (Table 4-1). The JNK1-IKK axis recapitulated the IFNy +
TNF basis axis of an earlier model of a TNF-growth factor network in HT-29 cells [97],
indicating that this combination of cytokines is a potent network activator and a strong
pro-death stimulus [163]. In addition, the common-processing model relied upon a
second axis consisting of TNF treatments without any sensitizing agents. Biochemical
Table 4-1. Top 20 kinase metrics contributing to the JNKI-IKK axis of the common-processing model.
See [97] for a complete description of the loading coefficients and signaling metrics.
Loading coefficient
0.177
0.164
0.160
0.153
0.150
0.146
Kinase
IKK
0.141
JNK1
0.139
0.139
0.137
0.136
0.136
0.135
0.135
0.135
0.132
0.130
0.130
0.130
0.122
IKK
Akt
JNKI
Signaling metric
8 hr derivative
4 hr derivative
JNK1
JNK1
30 min time point
Area under the curve, peak #1
16 hr derivative
20 hr time point
5 min derivative
24 hr time point
5 min derivative
Activation slope, peak #1
4 hr time point
0 min derivative
15 min time point
Maximum
60 min derivative
Mean
Area under the curve, peak #1
60 min derivative
Steady state
8 hr time point
IKK
IKK
IKK
JNK1
ERK
JNK
JNK1
MK2
JNKI
JNK1
JNKI
IKK
JNKI
65
Table 4-2. Top 20 kinase metrics contributing to the Akt-MK2 axis of the common-processing model.
See [97] for a complete description of the loading coefficients and signaling metrics.
Loading coefficient
0.137
0.126
0.126
0.120
0.117
0.113
0.112
0.112
0.112
0.111
0.108
0.105
0.104
0.104
0.104
0.104
0.104
0.101
0.099
0.099
Kinase
MK2
MK2
IKK
MK2
Akt
MK2
MK2
MK2
Akt
ERK
Akt
IKK
MK2
MK2
MK2
Akt
Akt
Akt
Akt
Akt
Signaling metric
24 hr time point
12 hr derivative
Activation slope, peak #2
20 hr time point
20 hr time point
0 min derivative
5 min time point
Area under the curve, peak #2
24 hr time point
2 hr derivative
16 hr derivative
Activation slope, peak #1
4 hr derivative
20 hr derivative
Steady state
60 min time point
Steady state
4 hr time point
Maximum
Activation slope, peak #3
activities overrepresented in this basis axis, which appeared to capture TNF-resistance,
including late Akt activity and late MK2 activity (Table 4-2). HT-29 projections along
the Akt-MK2 axis grouped closely with uninfected HeLa cells treated with TNF, again
illustrating a greater commonality between these two cell types than was indicated by
their raw network measurements (Fig. 4-4) or their apoptotic responses (Fig. 4-2a).
Interestingly, treatments involving Adv + TNF synergy did not require a third principal
component (data not shown), but rather fell in between the JNK1-IKK and the Akt-MK2
axes. From these reduced signaling dimensions, we conclude that cell-specificity is a
result of a stronger HeLa projection along the TNF-sensitivity axis, characterized by
increased early JNK1 activity and late IKK activity, at a lower level of infection
compared to HT-29 cells (Fig. 4-6). Given the recognized antiviral properties of certain
IFN-family members [164], it is likely that the increased projection along the JNKI-IKK
network dimension is directly related to antiviral responses that are HeLa-cell specific.
66
Model treatment projections
9.
15
HT-29
HeLa
1 Adv (1000
Vl
'x
nJ
~
~
'v,
3 TNF (100 ng/ml)
4 Adv (100 moil + TNF (100 ng/ml)
c
7.
u.
Z
I-
6.'/ synergy
~
I"
~
z
-..
2 IFN, (200 u/ml)
:~ 8.
10
moil
~
'-.Adv-TNF
5 TNF (100 ng/ml)
6 Adv (1000 moil + TNF (100 ng/ml)
7 Adv (100 moil + TNF (100 ng/ml)
5
8 IFN, (200 u/ml)
9 IFN,(200 u/ml)
+ TNF
(5 ng/ml)
+ TNF (100 ng/ml)
0
2
4
6
8
10
12
Akt-MK2 axis
Fig. 4-6. Reduced principal-component
dimensions of the common-processing
model
Adv-sensitized TNF conditions for both HeLa (red) and HT -29 (black) cells are shown.
the treatment combinations described to the right.
Numbers indicate
As demonstrated in the cytokine compendium work, interleukin-l a (lL-l a)
cooperates with TNF to activate network signals in HT -29 cells [69]. In addition, IL-l a
is released in primary hepatocytes following treatment with Adv and TNF (ll. Cosgrove,
unpublished data). While early IKK activity is generally considered to be anti-apoptotic
[38], IL-l a upregulates a late, sustained IKK activity peak that is pro-apoptotic.
time-dependent
This
nature of IKK activity in the apoptotic response was resolved in the
cytokine compendium PLS model [97]. Therefore, we examined IKK activity in our
model for evidence of the IL-I a autocrine loop.
We confirmed that the same dual role ofIKK is revealed in the common
processing model, as expected given that this is a TNF-dependent
7a).
We then compared IKK activity profiles in IFN-sensitized
29 cells with Adv-sensitized
autocrine loop (Fig. 4and Adv-sensitized
HT-
He La cells (Fig. 4- 7b). While the fold activation of IKK in
HT-29 cells infected with Adv was significantly lower than in the same cells pretreated
with IFN, the dynamics of IKK activation were nearly identical; however, temporal
activation ofIKK in HeLa cells was different.
While the second IKK activation peak
began to rise at approximately the same time as in HT -29 cells, it fell back to baseline
levels by 20 hours, suggesting that an inhibitor of IKK or IL-l a may be active in HeLa
cells. The IL-l receptor antibody (IL-lra) is another TNF-induced
67
autocrine factor
c
b
a
0.14
•
•
4
0
0
c
a
-.;:;
rc
0
N
""
-x 0_10
u
.~
~
V>
••
«
• Early IKK « 12h)
0.06
o Late IKK (> ,2h)
• Apoptosis
-0.05
•
HT-29IFN+
60
TNF
*
o HT-29Adv+TNF
'.•
•
•
He La Adv+ TNF
V>
-Vi
a
a.a
2
0<C
-c'"
40
~
eu..
20
0
0.0
0.05
Axis #1
0
10
Time (h)
20
IL-l ra
-
+ -
HT-29
IFN+TNF
+
-
+
HeLa
HT-29
Adv+ TNF Adv+ TNF
Fig. 4-7. Common-processing dimensions differentiate apoptotic role for early and late IKK activity
(a) Principal-component loading values for early IKK (filled black), late lKK (open black), and apoptosis
(blue) are plotted. (b) IKK activity in HT-29+IFN (filled black), HT-29+Adv (open black) and HeLa+Adv
(filled red) cells following 100 nglml TNF. Description of measurements same as in Fig. 4-3. (c) Potential
role of IL-I autocrine loop in Adv-infected cells. HT -29 cells and HeLa cells were treated wilth 100 ng/ml
TNF and pre-treated with either IFN or Adv (red) and HT-29 (black) in the presence ofIL-lra.
Data are
presented as the mean of 3 biological replicates :t: S.E.
operating in HT -29 cells [69]. It is possible that this factor is also induced in HeLa cells,
perhaps more strongly that in HT -29 cells.
Finally, we measured apoptosis Adv+TNF-treated
HT-29 and HeLa cells
following treatment with exogenous IL-l ra and compared the results to those in
IFN+TNF-treated HT-29 cells (Fig. 4-7c). While IL-l ra significantly reduces apoptosis
in IFN+TNF-treated HT-29 cells, it had no measurable effect in Adv+TNF treated cells.
However, this result could be consistent with a strong IL-ra loop in HeLa cells. It has
recently been shown that IL-l (a and f3) is a primary mediator of the host inflammatory
response in the liver following intravenous injection of Adv and that anti-IL-l antibodies
are an effective strategy for attenuating this response [165]. Therefore, further
investigation of this autocrine loop, including direct measurement of IL-l a and IL-l ra
with an enzyme-linked
immuno-specific
assay (ELISA), is recommended.
The overrepresentation of Akt on the axis corresponding to TNF resistance is at
first not surprising given its pro-survival role [130]. However, the only recognized
growth factor stimulus in our experimental system is Adv shedding of E40RF 1
transcripts, which we hypothesize are subsequently translated into proteins that activate
the PI3K-Akt
pathway [63]. In this case, however, Akt would correlate with an infected
68
a
b
•
0.15 -
• EarlyAkt « 12h)
o Late Akt (> 12h)
• Apoptosis
•
:>
ro
N
0.10 -
"*'
.;;:
'"
«
c:
0
.~
;;;
u
:>
"ero
2
"C
"C
(5
u.
~
2
~
•
0.05 -
~
I
-0.05
Akt acti"ityat
20h and 24h -
~
I
0.00
I
I
0.05
Axis #1
0
0
10 (h) 20
Time
0
0
0.00 -
• HT.29 high Ad"
• HeLa low Ad"
3
3
c:
.Q
C
HPV E6
= E40RFl
20
10 (h)
Time
HPV E6
= E40RFl
+
0.10
low
----
+
increasing E40RFl
TNF resistant
I
high
TNF sensitive
Fig. 4-8. HPV £6 in HeLa cells may cooperate with Ad E40RF I in mediating TNF response.
(a) Principal-component loading values for early Akt (filled black), late Akt (open black), and cleaved
caspase-cytokeratin (blue) are plotted. (b) Akt activity plots. Left: HT-29 cells infected with Adv at 100
m.oj. (black) and uninfected HeLa cells (red) cells following 100 nglml TNF. Right: HT-29 cells infected
with Adv at 1000 m.o.i. (black) and HeLa cells infected with Adv at 100 m.oj. (red) following 100 nglml
TNF. Description of measurements same as in Fig. 4-3. (c) Schematic hypothesizing a pro-survival role for
HPV E6 and Ad E40RF I at low levels and a cytotoxic role upon increasing levels of E40RF I.
cell, and therefore apoptotic, phenotype.
Therefore, we looked more closely at the
loading weights for temporal measurements
of Akt activity (Fig. 4-8a). As is evident in
the top 20 metrics corresponding to the Akt-MK2 TNR-resistance
axis, late Akt activity,
specifically at 20 and 24 hours, maps far from the cleaved caspase-cytokeratin
apoptotic
markers. We plotted the signaling time courses for uninfected HeLa cells and HT -29
cells infected with Adv at 100 m.o.i. and observed a similar temporal pattern of Akt
activity (p
= 0.44 by two-factor ANOV A) including a substantial rise in activity at 20
hours (Fig. 4-8b, left). Although Ad E40RF 1 is known to activate the PI3K-Akt
pathway [62, 63], it was surprising to observe a rise in Akt in uninfected HeLa cells.
As mentioned above, HeLa cells are transformed by the human papilloma virus
(HPV) oncoproteins E6 and E7 [123]. Recently, HPV E7 was shown to increase Akt
activity indirectly in response to serum stimulation by interacting with the catalytic
domains of PP2A and inhibiting its ability to desphosphorylate
Akt [166]. However, this
does not fully explain what gives rise to Akt activation in the first place. It has been
known for some time that both Ad E40RF 1 and HPV E6 inhibit tumor suppressor protein
69
function via interaction with their PDZ domains, which gives rise to the oncogenic
potential of these viral proteins [167]. Although to our knowledge there is no evidence
that HPV E6 has been linked directly to Akt activation, it is possible that a HPV E6 PDZ
interactions cause an increase in Akt activity in HeLa cells similar to Ad E40RF 1mediated Akt activation in HT-29 cells. This would explain the similar pattern of Akt
activation.
Interestingly, when we compared temporal Akt signaling in Adv-sensitized HeLa
cells and HT-29 cells (100 and 1000 m.o.i., respectively), Akt activity also rose and fell
in a similar pattern (p =0.90; Fig. 4-8b, right). Given our results on the saturation of Akt
anti-apoptotic signaling in HT-29 cells infected at high levels of Adv, we considered that
the combination of Ad E40RF1 and HPV E6 might have a similar effect in HeLa cells.
In fact, a recent study in epithelial tumor cell lines demonstrated that low levels of HPV
E6 were protective against TNF-induced apoptosis, while high levels of HPV E6
sensitized cells to TNF-induced apoptosis [168]. Therefore, we hypothesize that HPV E6
and Ad E4ORF1 are functionally interchangeable in this regard and, while low levels of
either protein protect against TNF-mediated apoptosis, the presence of more Ad E40RF 1
at higher levels of infection reduces the anti-apoptotic signaling of Akt and potentially
leads to a cytopathic effect (Fig. 4-8c). We recommend testing this hypothesis in followup experiments.
4.3.Common processing model as a tool for designing rational drug
therapies
4.3.1. Predictions of Akt inhibition test robustness of the common
effector-processing model
One immediate application of a common-processing model is for predicting cellspecific responses to rational drug therapies [169, 170]. In particular, drug safety in vivo
is often compromised by unintended toxicity to off-target cell types [171]. We therefore
70
examined whether the common processing model could predict apoptotic outcomes
across cell types when network signals were perturbed by inhibitors. Frequently, the
drug response for a target cell type is already known, and this information can be used to
inform predictions for another cell type. In Adv-infected HT-29 cells, for example,
inhibition of the PI3K-Akt pathway (a clinically-relevant cancer target [172])
dramatically increases TNF-induced cell death [162]. By contrast, the role of PI3K-Akt
in TNF-induced apoptosis of Adv-infected HeLa cells is unknown.
To predict the effects of PI3K inhibition on HeLa cell viability, we first included
prior information about drug sensitivity in HT-29 cells. The model was updated with
measured values [162] of Akt inhibition and apoptosis in HT-29 cells treated with
LY294002, a small molecule inhibitor of PI3K. We assumed that the potency of
LY294002 was constant throughout the 24 hours following TNF addition and that no
other kinases were affected during this time period (see Methods for details). HeLa
signaling and apoptosis data were withheld (as in Fig. 4-5b) to avoid biasing the
common-processing model toward HeLa-specific outcomes. Next, we experimentally
measured the extent to which LY294002 reduced Akt activity in HeLa cells (Fig. 4-9a).
Using the LY294002-inhibited Akt activity in combination with the corresponding HeLa
network measurements (Fig. 4-9a), we then used the updated model to predict the
response to inhibition of the PI3K--Akt pathway when HeLa cells were preinfected with
Adv and treated with TNF.
Surprisingly, the updated model predicted no significant increase in cell death in
Adv+TNF-treated HeLa cells following PI3K inhibition (Fig. 4-9b; bars). This was
despite clear inhibition of the Akt pathway by LY294002 in HeLa cells (Fig. 4-9a) and
explicit model training on LY294002 sensitivity in HT-29 cells (see above). When
apoptosis was measured directly in LY294002-treated HeLa cells, we found only a
nominal increase in TNF-induced cell death (Fig. 4-9b; circles), in close agreement with
the predictions of the common-processing model. HeLa apoptosis was also unchanged in
the presence of the mechanistically distinct PI3K inhibitor wortmannin. Thus, both
model and experiment indicate that the role of PI3K in Adv+TNF-induced apoptosis is
cell specific: HT-29 apoptosis is PI3K-sensitive [173], whereas HeLa apoptosis is PI3Kresistant (Fig. 4-9b).
71
a
b
Common
processing model
Akt inhibition
i='
.>
1.0
.+:;
'"
~
.;jj
.1::!
40
'"
.~
0
a.
«
Q)
Cell-specific
processing model
40 -
u
«
-0
C
Ci
0
8. 20
«
0.5
ra
E
0
0.0
+
LV
d
0
+
HT-29
?f.
?f.
0
0
z
~20
0
+
LV
HeLa
~~
LV
+
HeLa
Akt model
predictions
e
Akt model
HeLa
80
80
HT-29-HeLa
hybrid models
f
G
HeLa
'"
+ ERK
&
C0
o.
+IKK
~
« 40
«
+JNKl
G
'" 60
.~
.~
8.
a.
~
40
•
;:R
HeLa + MK2
0
20
~
0
0.2
1.0 2.0
Akt foldactivation
HT-29
+
LV
He La
~
~
40
50
% Apoptosis
Fig. 4-9. Common
effector processing uniquely predicts resistance of Adv-infected HeLa cells to
pharmacological
inhibition of PI3K.
(a) Akt activity measured in Adv-infected HT -29 cells and HeLa cells) 2 hr after TNF stimulation in the
presence and absence of 20 IlM L Y294002 (L Y). Data are presented as the mean of six (HT -29) or three
(BeLa) biological replicates :I: S.E. (b) Apoptosis predicted by the common-processing
model (bars)
compared with that measured experimentally (circles) in Adv-infected, TNF-treated HeLa cells in the
presence and absence of20 IlM LY294002
(LY). (c) Apoptosis predicted by a Hela cell-specific
processing model (bars) compared with that measured experimentally (circles) in Adv-infected, TNFtreated BeLa cells in the presence and absence of 20 IlM L Y294002 (L Y). (d) One-kinase relationship
linking Akt activity to apoptosis ofHT-29 cells (black) treated with combinations ofIFNy, Adv, and
TNF[162].
90% confidence interval (tan) is shown. HeLa cells (red) were treated with 100 moi Adv and
100 ng/ml TNF. (e) Apoptosis predicted by the Akt-apoptosis model (bars) and measured experimentally
(circles) in Adv-infected, TNF-treated HeLa cells in the presence and absence of 20 IlM L Y294002
(L Y).
(f) Apoptosis predicted by the common-processing
model for HT-29-BeLa
hybrid-cell profiles containing
mixtures of HT-29 signals (black) and HeLa signals (red). Data are presented as the mean of three
biological replicates:l: S.E., and model uncertainties were estimated by jack-knifing[89].
The accurate L Y294002
to compare the common
predictions in He La cellsnow provided the opportunity
model directly against competing mechanisms
of effector
processing. To mimic cell-specificprocessing (Fig. 4-lb), we trained a separate PLSR
72
processing function based only on the HeLa signaling measurements and tested if the
existing HeLa data itself was superior for relating signals to apoptosis correctly. The
HeLa-specific model was far less accurate and highly uncertain as compared to the
common-processing model (Fig. 4-9c). This emphasized that HT-29-specific
information, and thus common-effector processing, was absolutely essential for
prediction.
A reductionist alternative to common processing of the network is that there exists
one signal, which is a master regulator of TNF-Adv synergy across cell types.
Previously, we developed a mathematical relationship describing the dependence of
TNF-induced apoptosis in sensitized HT-29 cells on the relative extent of Akt activation
[162]. Using the activation of Akt at a single time point, we found that this relationship
predicted apoptosis in Adv-infected HeLa cells within a 90% confidence interval (Fig. 49d), raising the possibility that Akt signaling alone was sufficient to predict cell-specific
apoptosis. However, when the Akt model was used to test the result of LY294002
inhibition, it incorrectly predicted a four-fold increase in Adv+TNF-induced apoptosis in
HeLa cells (Fig. 4-9e), indicating the need for signaling information from multiple
pathways. We demonstrated this requirement by constructing hybrid-cell profiles, in
which HT-29 kinase measurements were individually substituted into the LY294002inhibited HeLa dataset (see Methods for details). In the common-processing model, the
HT-29-specific sensitivity to PI3K inhibition was captured by an ensemble of antiapoptotic signals from ERK and IKK and pro-apoptotic signals from JNK1 and MK2
(Fig. 4-9f), in agreement with the recognized functions of these pathways (see Table 1-1).
Among the competing mechanisms for achieving cell specificity, we therefore conclude
that common effector processing is most consistent with experimental observations.
4.3.2. Common effector-processing model predicts differential effects
of IKK inhibition in HeLa cells and HT-29 cells
The IKK-NF-KB pathway is another important therapeutic drug target for cancer
as well as inflammatory diseases [174]. However, the value of pharmacological
73
Adv+ TNF-treated HeLa cells
b
a
Predicted
?:'
.:;
Observed
3
Data
range
°e
'"
~
50
:>c:
<II
0g>
HT-29
% Apoptosis
25
0
50
~DD =DD~
~
Qj
cr:
0
0
2
0.5
8
16
24
Time (hr)
c
HT-29
HeLa
40
40
.~ 35
30
oil
HeLa
%Apoptosis
a.
25
0
a.
«
-J? 30
0
sc
AdV
+
+
+
+
+
+
+
+
25
Stan
Stop
20
~~
~
-
4
0
-
24
24
oU
10
Stan
Stop
-
4
0
24
24
Fig. 4-10. Deduction of celI-specific sensitivity to early-phase IKK inhibition.
(a) Apoptosis predicted by the common-processing
model (left) and measured experimentalIy (right) in
HT-29 (top) and HeLa (bottom) celIs treated with 100 nglml TNF in the presence or absence of20 lAM SC514 (SC) with or without Adv preinfection. Numbers indicate the treatment combinations of Adv and Sc.
(b) Range plot ofHT-29 IKK activity used for model training (tan) and IKK activity time course of Advinfected HeLa cells treated with TNF (points). Points outside the training range are highlighted in blue. (c)
TNF-induced apoptosis in Adv-infected HT-29 cells and HeLa cells following IKK inhibition with 20 lAM
SC-514 for the indicated time intervals. Data are presented as the mean of three biological replicates :f:
S.E., and model uncertainties were estimated by jack-knifing[89].
inhibitors ofIKK or NF-KB has been debated because the pathway operates very
differently across cell types [175]. We therefore used the common-processing
model to
predict how small molecule inhibition of the IKK-NF-KB pathway would affect
apoptosis in HI -29 cells and HeLa cells. For the reversible IKK inhibitor SC-514 [176],
we simulated IKK inhibition as described for Akt (see Methods for details) and then had
the model predict apoptosis given the new SC-514-inhibited
signaling inputs. No prior
information about IKK perturbation was used to update the model, because the response
of both HeLa and HT -29 cells to SC-514 was unknown.
The common-processing
model predicted cell-specific differences in response to
74
BAY
SN50
5o
50-
0
0
A
Ad
HT-29
% Apoptosis
L I
50
EDD
HeLa
% Apoptosis
n
BAYAdV-
+
-
+
+
+
I
SN5O
AdV -
I
I
I
I
+
-
+
+
+
Fig. 4-11. HeLa cells are more sensitive to IKK-NF-KB inhibition than HT-29 cells.
Comparison of apoptosis in HT-29 cells (top) and HeLa cells (bottom) following IKK inhibition with 50
[M BAY 11-7082[177] (left) or NF-KB inhibition with 40 [tM SN50[178] (right). Control cells were
treated with 0.02% DMSO for BAY or the inactive SN50M peptide for SN50.
inhibition of the IKK-NF-KB pathway. In TNF-treated HT-29 cells, apoptosis was
predicted to remain unchanged in both uninfected and Adv-infected cells (Fig. 4-1 Oa,
upper bars 1-2 and 3-4). In contrast, HeLa apoptosis was predicted to increase
substantially in uninfected cells (Fig. 4-10Oa, lower bars 1-2) and to decrease slightly in
Adv-infected cells (Fig. 4-10a, lower bars 3-4). When apoptosis was measured in
thepresence of 20 IM SC-514 experimentally, we found that IKK inhibition in HT-29
cells almost exactly mimicked the model predictions, with only a small absolute increase
in apoptosis in HT-29 cells for both conditions (Fig. 4-10a, upper bars 5-6 and 7-8). By
comparison, there was less agreement between model predictions and experiments in
HeLa cells. Although the common-processing model captured the correct relative change
in apoptosis for SC-514 inhibition of uninfected HeLa cells treated with TNF (3.0-fold
increase predicted by model versus 4.3-fold increase measured experimentally; Fig. 410
Oa, lower bars 5-6), the model predictions for Adv-infected HeLa cells were incorrect
(Fig. 4-1 Oa, lower bars 7-8). Apoptosis increased substantially after SC-514 inhibition,
rather than decreasing as predicted. These experimental results were not unique to SC75
514, because we observed the exact same pattern of cell-specific apoptosis when using
two other small-molecule inhibitors targeting different points of the IKK-NF-KB
pathway (Fig. 4-11). Thus, the common-processing model had qualitatively predicted the
existence of cell-specific responses to IKK-NF-KB pathway inhibition but did not
quantitatively predict the extent of this specificity in Adv-infected HeLa cells treated with
TNF.
The discrepancy between theory and experiment for the Adv-infected HeLa-SC514 apoptosis prediction was not necessarily due to failure of common processing as a
network mechanism for cell specificity. Rather, it was possible that the discrepancy
reflected an inadequacy of the HT-29 data used to train the PLSR model. Adv-infected
HeLa cells display a two-fold increase in IKK activity before four hours that falls outside
the range of the HT-29 IKK activity data used for model training (Fig. 4- 10
Ob). Early IKK
activity is generally considered to be anti-apoptotic [38], but because HT-29cells do not
activate IKK as strongly as HeLa cells, there may be a lack of experimental data relating
early IKK activity to an anti-apoptotic phenotype. Without such data, the commonprocessing function would not interpret this HeLa-specific signaling variation properly,
and thus would not accurately capture the change in HeLa apoptosis when IKK is
inhibited with SC-514.
An important corollary to this explanation is that early IKK activity has an antiapoptotic role in TNF-induced apoptosis in Adv-infected HeLa cells but not in Advinfected HT-29 cells. To test for different roles of early IKK activity, we performed
timed inhibition experiments with SC-514. Direct inhibition of early IKK signaling was
not possible, because removal of SC-514 caused hyperactivation of IKK (Fig. 4-12), as
reported previously [176]. To avoid possible misinterpretations stemming from IKK
hyperactivation, we measured the role of early IKK activity indirectly by selectively
inhibiting IKK from 4 to 24 hours and comparing the increase in apoptosis to that
observed when IKK was inhibited from 0 to 24 hours (Fig. 4-6c). In HT-29 cells, IKK
inhibition from 4 to 24 hours was the same as inhibiting IKK for the full 24 hours: both
resulted in only a 15-20% relative increase in apoptosis, as seen earlier (Fig. 4-6a, upper
bars 5-6). In contrast, nearly half of the SC-514-mediated increase in HeLa apoptosis
could be attributed to early-phase inhibition of IKK (Fig. 4-6c). Therefore, a lack of
76
a
SCin vitro
b
2.0
SCin vivo
2.0
>
U
Y
U
'U
1.0
Y
1.0
._
ar)
0.0
SC -
0.0
+
SC -
+
Fig. 4-12. SC-514 inhibition in vivo causes hyperactivation of IKK after removal.
(a) Immunoprecipitates from HT-29 cells treated with 100 ng/ml TNF for 30 min were incubated with 100
.tM SC-514 (SC) or 0.1% DMSO control for 15 min in vitro and then analyzed for IKK activity as
described[l 1 ]. (b) HT-29 cells were preincubated with 100 IM SC-514 or 0.1% DMSO control for
hour in vivo and then stimulated with 100 ng/ml TNF for 15 minutes. Lysates were analyzed for IKK
activity as described[ 11]. Washes during the IKK activity assay remove the SC-514 inhibitor and cause
hyperactivation of IKK as reported previously[ 176]. Data are reported as the mean normalized activity +
range of 2-3 biological replicates.
input training data, rather than a failure of the effector-processing function, accounted for
the inconsistency between model and experiment. However, this discrepancy allowed us
to deduce the existence of an early IKK activity phase that was critical for anti-apoptotic
signaling in Adv-infected HeLa cells and not present in HT-29 cells.
4.4. Further considerations
In metazoa, the need to achieve cell-specific responses with common network
components can be considered analogous to the paradox of evolution, in that the need to
change must be reconciled with the genetic conservation observed across species. A
recent evolutionary theory proposes that certain "core processes" are conserved because
they can easily be elaborated upon [ 179]. Although it is not yet clear whether common
effector processing constitutes such a core process for multicellular organisms, our results
do reveal an underlying similarity between disparate epithelial cell types. This similarity
is uncovered only when signal processing is analyzed at the network level.
77
Reconstructions of cytokine-signaling networks have frequently observed a global
hourglass topology of initiator signals (such as receptors and adaptors), transducer signals
(such as enzymatic cascades), and effectors [ 180]. Many of the kinases measured here lie
at the "waist" of these hourglasses, indicating a unique point of convergence in the
network architecture. Our results suggest that initiator signals are highly cell-specific
[86], whereas effector mechanisms are likely to be more widely conserved. Although
there probably does not exist a common-processing function that captures the responses
of all cell types, there may be common mechanisms for general cellular classes such as
epithelial cells and hematopoeitic cells.
78
CHAPTER 5
5. Conclusions and future directions
5.1. Contributions to signaling network analysis
Despite deletion of several immunogenic regions, Advs used in gene therapy still
retain some of the host cell modulators found in the original adenoviral pathogen. The
field is in need of new strategies for uncovering and addressing these problems at the
molecular level. In epithelial cells, Adv infection is revealed at the phenotypic level by
treatment with TNF. Intracellular Adv-TNF synergy was uncovered by dynamically
measuring the TNF-Adv signaling network via five key protein kinases. By perturbing
apoptotic response at high levels of adenoviral vector infection, we were able to
demonstrate the saturation of Akt effector signaling underlying Adv-TNF synergy in HT29 cells. Further systems analyses of the synergy between Adv and proinflammatory
cytokines like TNF could aid the design of next-generation viral vectors for treating
human disease [ 181].
In addition we have used our experimental system to uncover a fundamental
biological principle, which we call common effector processing. By applying a datadriven modeling approach that converts important kinase signals into phenotype, we have
identified a common effector-processing mechanism between tumor lines with cellspecific outcomes to the same stimulus. If the common-processing hypothesis had failed,
it would have meant that each cell type would require its own in-depth experiments and
model training. Our results raise the possibility that these types of models can be refined
with new data to the point that they enable accurate predictions of phenotype that are
broadly applicable. Already, the distinction between two carcinoma cell lines based on
their apoptotic response to PI3K and IKK inhibition suggests immediate applications for
combining pharmacological inhibitors [182] and cancer gene therapy [161] to treat
certain tumors.
79
5.2. Application to the development of Adv-mediated cancer gene
therapy
Adv-TNF synergy could be exploited to design Advs for cancer gene therapy, for
instance by delivering a TNF-family cytokine as a transgene [183-186]. Many tumors are
resistant to TNF-induced apoptosis [163], but presentation in the context of Adv infection
can induce apoptosis in multiple carcinomas [183, 186]. The results presented here argue
that the efficacy of these cancer gene therapy approaches is a direct product of Adv-TNF
synergy. An important consideration for cancer therapies is the role of Akt in Advinfected cells. Many tumors, especially brain, breast, and prostate cancer, have
constitutive Akt activity due to a loss of function mutation in the PTEN phosphatase gene
[187]. Our model suggests that, Adv-mediated gene transfer may serve as an even more
potent pro-death stimulus in these tumors, because the anti-apoptotic function of the Akt
pathway would already be saturated before infection. Subsequent cytokine stimulation
would therefore activate only pro-apoptotic signals, leading to rapid apoptosis of the
tumor cells. To our knowledge, there are as yet no examples of therapies that test this
prediction.
In addition to delivering a death-inducing gene to a tumor via a viral vector, some
cancer gene therapies are designed such that the virus itself is the oncolytic agent through
selective replication in tumor cells [188]. This selective replication can be achieved with
a tumor-specific promoter, as with CV706, an E3-deleted adenovirus with replication
restricted to prostate tumors by inserting a prostrate-specific antigen (PSA) promoter in
front of the Ad E1A gene [189]. In other cases, selective replication is achieved by
complementing a tumor mutation with a viral gene deletion, such that productive
infection only results in tumor cells. This is the case for ONYX-015, a mutant
adenovirus that lacks the E1B-55K protein and therefore cannot degrade p53 and carryout a productive infection unless p53 is mutated, as is the case in many cancers [98, 190].
As discussed at length in this thesis, many Adv proteins have pro- and antiapoptotic roles that could possibly be exploited to improve oncolytic therapies. However,
80
in some cases the wild type genes encoding these proteins must be removed to improve
selective replication, which can have the unwanted result of an attenuated therapeutic
response [191]. A network signaling study, comparing oncolytic adenoviruses
systematically deleted for different genes, offers one approach to studying this problem.
A cue-signal-response analysis could be pursued to identify signaling interactions that
maximize apoptosis in cancer cell lines or interactions that minimize replication in
primary cells. In addition to potentially revealing novel aspects of cancer biology, this
approach may suggest effective combination Adv cancer gene therapy strategies.
5.3. Ongoing and future directions
5.3.1. Adenoviral gene therapy in the liver
Much effort has been put into developing adenoviral-mediated gene therapy for
the liver, largely due to the natural hepatocyte-specific targeting of adenoviruses injected
directly into the bloodstream [192]. However, the Adv-induced inflammatory response,
mediated in part by TNF as discussed above, has been a major obstacle to the successful
use of liver-directed Adv therapies [21, 22, 24]. Interestingly, a healthy liver has robust
defense mechanisms with which it can exploit transient TNF-mediated stress signals to
promote proliferation; however, a liver pre-exposed to certain toxins becomes susceptible
to TNF-induced apoptosis [193]. Recently it has been demonstrated that infection with a
first-generation Adv (the same vector used in these studies) renders primary hepatocytes
sensitive to TNF-induced apoptosis (B. Cosgrove, unpublished data). With this
experimental system and the approach demonstrated in this thesis, one can ask how Adv
infection modulates cell signaling in the liver causing TNF to induce a stress response
rather than a proliferative response. This work is being pursued currently in our
laboratory.
81
5.3.2. NF-icB signaling in HIV latency
HIV latency is another viral-host interaction problem that would benefit from a
systems biology analysis of the signaling network. The majority of HIV infection events
in active T-cells result in rapid viral replication; however, a very small minority of
infections can lead to a pool of latently infected memory T cell. Because the virus cannot
directly infect memory T cells, it is unclear how this pool of cells develops. However,
this reservoir is very long lived and represents the most significant barrier to complete
elimination of virus from a patient. Infected patients must stay on costly anti-retroviral
medications (which have significant side effects) for the remainder of their lives, because
if they stop taking the medication, latent viral copies can reactivate and once again lead to
full viremia.
Due to the extreme clinical importance of latency, there has been significant
interest in understanding the molecular mechanisms that underlie its establishment [194196]. It has recently been hypothesized that stochastic effects in viral gene expression
can result in long delays before activation of viral gene expression [197], which could
maintain the virus in a transcriptionally inactive state long enough for the host T-cell to
make the slow conversion to a memory T-cell and thereby solidify the virus into a latent
state. The IKK- NF-KB signal transduction pathway is known to be a crucial regulator of
the HIV LTR promoter, and therefore it is important to understand how this signaling
pathway affects stochastic gene expression of HIV genes, potentially contributing to
latency.
One strategy for eliminating latent HIV is to purge the virus by chemically
stimulating HIV LTR transcription, thus rendering such cells susceptible to antiretroviral
agents [198]. In parallel, this approach may prevent the establishment of new latencies.
The basal nuclear environment of NF-KB, which regulates transcriptional activation, is at
least in part controlled by the upstream kinase IKK. We have demonstrated in this thesis
and elsewhere that the information contained in kinase signaling dynamics can be used to
predict cell function [97, 173]. It has also been demonstrated both computationally and
experimentally that different inflammatory stimulants induce distinct IKK temporal
profiles, which lead to different NF-KB dynamics and gene expression programs [199].
82
Thus, I have proposed a signaling network study of how chemical and pharmaceutical
activators of IKK control NF-KB-mediated transcription dynamics in HIV latency as part
of my post-doctoral work.
83
CHAPTER 6
6. Appendices
6.1.Materials
6.1.1. Adenoviral vectors
The recombinant adenovirus type 5 vectors with El and E3 regions deleted
expressing either Escherichia coli P-galactosidase (-gal) under control of the
cytomegalovirus (CMV) enhancer/promoter (Adv.P-gal) or containing the CMV
enhancer/promoter without a transgene (Adv.empty) were provided by the University of
Michigan Vector Core. Virus was provided at a concentration of 4 x 1012viral particles
(v.p.) per milliliter and reported to have an infectious plaque-forming unit (p.f.u.)
concentration of 1.7 x 101 p.f.u./ml as determined by plaque assay. Working viral stocks
were diluted in storage buffer (10 mM Tris-HCl pH 7.4, 137 mM NaC1, 5 mM KCl, 1
mM MgCl2, 10% glycerol). Storage buffer was also used for controls.
6.2.Experimental Protocols
6.2.1. Cell culture and infection
HT-29 and HeLa cells (ATCC, Manassas, VA) were grown according to the
manufacturer's recommendations. Unless otherwise specified, HT-29 cells were seeded
at 50,000 cells/cm2 , grown for 24 hours, and then infected for six hours in culture media
(50% of normal media volume) containing 1.4 x 1010v.p./ml Adv or an equivalent
84
volume of media containing storage buffer. At the end of the infection period, cells were
washed once with PBS and a full volume of fresh culture media was replaced. Using
these infection conditions, this was the lowest concentration tested that resulted in > 95%
of the cell population positive for Adv infection (see Chapter 2 for details). A similar
infection efficiency in HT-29 cells has been reported elsewhere [106]. Therefore, we
used 1.4 x 1010v.p./ml for all experiments except dose response curves (for which
concentrations are indicated in text). HeLa cells were seeded at 20,000 cells/cm2 and
infected as described for HT-29 cells at v.p. concentrations indicated in the text.
Helper-dependent adenoviral vector was provided by Anja Erdhart and Mark Kay
at Stanford. Virus was supplied at a viral titer of 9 x 109blue forming units (bfu) per ml
in storage buffer. For experiments comparing Ad.dlE3.bgal and Ad.HD, Ad.dlE3.bgal
working stock was diluted to 9 x 109 PFU/ml.
6.2.2. Beta-galactosidase Assays
Trypsinized cells were washed once in PBS and resuspended in lx Reporter Lysis
Buffer (Promega) for 15 minutes at room temperature.
-galactosidase activity was
measured in a 96-wll plate format according to manufacturers' instructions. Total protein
content of the sample was determined with a bicinchonic acid assay. Activity
measurements were reported as milliunits B-galactosidase activity per ptg protein.
6.2.3. Apoptosis assays
HT-29 or HeLa cells were seeded and infected as previously described. 24 hours
after the start of infection, TNF (Peprotech; 100 ng/ml final concentration) or an
equivalent volume of carrier (50:50 DMSO:water) was added to the culture media. For
inhibitor studies, inhibitor or DMSO carrier was added to the medium one hour before
TNF stimulation and collected 24 hours later. Inhibitors (Calbiochem) used were 20 FM
LY294002, 10 [tM U106, 10 FM SB202190, 10 [tM SP600125, and 20 FM SC-514. For
85
growth factor studies, cells were treated with 100 nM insulin (Sigma), in parallel with
TNF treatment, and collected 24 hours after stimulation. At the indicated time, cells were
rinsed with PBS and trypsinized. The supernatant and rinse were saved and combined
with the trypsinized cells to ensure capture of both floating and adherent cells. The cells
were washed with PBS, and either all or a portion of the cells were fixed in 100%
methanol and stored at -20°C. The fixed cells were stained with a fluorescein-labeled
monoclonal antibody against caspase-cleaved cytokeratin (M30; Roche) and active
cleaved caspase-3 (BD Pharmingen) with Alexa 647-conjugated anti-rabbit secondary
(Molecular Probes) according to the manufacturer's instructions. Cells were washed
once after staining and analyzed by flow cytometry (FACS-Calibur; Becton-Dickinson).
To perform parallel apoptosis assays, a portion of the original (unfixed) sample was
stained with an Alexa Fluor 488 annexin V conjugate (Molecular Probes), counterstained
with propidium iodide (PI; Molecular Probes) according to manufacturer's instructions,
and analyzed by flow cytometry. In some cases, a portion of the original infected sample
was used for [-gal activity analysis to confirm a productive infection (data not shown).
6.2.4. Kinase activity assays
For kinase activity assays, HT-29 cells were grown and infected as previously
described. 24 hours after the start of infection, TNF was added to the culture media (100
ng/ml final concentration) for the indicated times. For Akt activity studies, 20 [IM
LY294002 was added to the medium one hour before TNF stimulation and 100 nM
insulin was added in parallel with TNF stimulation, as indicated. Cells pretreated with
IFN were treated as previously described [111]. Cell lysates were prepared and analyzed
for ERK, Akt, IKK, JNK1 and MK2 activity as previously described [111]. Protein
concentrations of clarified extracts were determined with a bicinchonic acid assay
(Pierce) and all activity measurements were normalized to the total protein content of the
sample.
86
6.2.5. Quantitative polymerase chain reaction (PCR)
HT-29 cells were grown in 6-well plates and infected for the indicated times in
either full media volume at a concentration of 0.7 x 101° v.p./ml or half volume at a
concentration of 1.4 x 100 v.p./ml (all samples subject to the same overall MOI of 1000
p.f.u./cell). At the end of the infection time, cells were washed with PBS and trypsinized.
DNA was extracted using the DNeasy Tissue Kit (Qiagen, Valencia, CA) according to
the manufacturer's instructions. Genome quantification was performed using the
AbiPrism 7700 (PE Biosystems) according to the method previously reported [200] with
modifications. Taqman primers and probes were used to amplify the P-galactosidase
reporter gene, and were designed using the Primer3 software package2 acquired from
Applied Biosystems (PE Biosystems). The probe utilized a 3' TAMRA quencher
molecule attached via a linker arm and a 5' covalently linked FAM fluorescent dye. The
same Taqman primers and probe were used on known quantities of gWiz f-Gal plasmid
standards (Aldevron) to calculate a quantitative viral copy number for each sample. Data
was analyzed with ABI Sequence Detector v1.6.3 and total P3-galactosidaseDNA was
normalized to total -actin DNA for each sample using primers, probes and standards
from the Human 3-actin Gene kit (PE Biosystems). Results are presented as the average
of total P-gal DNA normalized to total
-actin DNA.
For apoptosis measurements performed in parallel with quantitative PCR analysis
infection media was aspirated at the indicated times, cells were washed once with PBS
and fresh media was replaced. Cells were then treated with TNF, collected, and analyzed
by flow cytometry as previously described.
6.2.6. RNA isolation and RT-PCR
HT-29 cells were grown in 12-well plates, infected and treated with TNF as
described previously. At the time of collection, cells were trypsinized, transferred into an
2
http://frodo.wi.mit.edu/cgibin/primer3/primer3_www.cgi
87
Eppendorf tube and washed once with PBS. 1 ml of TRIzol® (Invitrogen) was added to
the samples and vortexed until pellet was thoroughly lysed. The samples were then stored
at -80°C until further processing. RNA was extracted using the RNeasy Tissue Kit
(Qiagen) according to the manufacturer's instructions, quantified by spectrophotometer,
and stored at -80°C.
RNA samples were thawed on the day of cDNA preparation and approximately
200 ng of total RNA was used to prepare cDNA. Amplification grade DNAse I enzyme
(1 unit activity/gg of RNA) and DNAse buffer (Invitrogen) were added to the samples.
Samples were incubated at room temperature for 15 min, 1 l of EDTA (25 mM;
Invitrogen, MD) was added, and followed by denaturation of the enzyme at 65°C, for 10
min. Master mix solution was then added to the sample with final concentrations of 0.5
mM dNTPs (Qiagen, CA) mM random hexamers (Qiagen), x RT buffer (Qiagen), 2
units/pl Omniscript RT® (Qiagen), and 5 units/ml RNase (Ambion). The samples heated
for one hour in a 37°C water bath for the conversion of the RNA to cDNA and then
stored at -80°C.
Sequences for the RT-PCR E4orf3 primer were kindly provided by Matthew
Weitzman's laboratory (Salk institute). Primers were synthesized by Operon. 1 l of
cDNA was added to a mixture of 25 l of SYBR Green Master-Mix (Qiagen), 21 ,ulof
DEPC water (Qiagen), and 1.5 ,ul each of the forward and reverse primers. Amplification
was carried out in an Opticon Monitor 2 system using standard SYBR Green RT-PCR
annealing and melting protocols. Dilutions of disrupted Ad vector particles were used to
generate a standard curve to convert cycle times to E4orf3 copy number.
6.2.7. Western blots
For determination of Akt substrate activity, 50 tg of protein lysate from 12 hour
Akt activity experiment was resuspended in 40 l sample buffer (62.5 mM Tris-HCl (pH
6.8), 2% SDS, 10% glycerol, 100 mM DTT, 0.01% bromophenol blue). Samples were
boiled for 5 minutes, resolved on a 10% polyacrylamide gel and transferred to
polyvinylidene difluoride (Biorad). Membranes were blocked with 5% nonfat skim milk
88
in 20 mM Tris-HCl (pH 7.5), 137 mM NaCl, 0.1% Tween-20, and probed with antiphospho-GSK-3a/[ (1:1000; Cell Signaling) or anti-phospho-(Ser/Thr) Akt substrate
(1:1000; Cell Signaling). The membranes were then probed with horseradish peroxidase
conjugated anti-mouse or anti-rabbit secondary antibody (Amersham Pharmacia Biotech)
at 1:5000 dilution and visualized by enhanced chemiluminescence (Amersham Pharmacia
Biotech) on a Kodak Image Station (Perkin Elmer). Primary antibodies were removed
with stripping buffer (2% SDS, 62.5 mM Tris-HCl (pH 6.8), 100mM 2-mercaptoethanol).
6.2.8. PLSR model construction and refinement
Prior to all analyses, the signaling and apoptosis matrices were variance scaled to
non-dimensionalize the different measurements. The PLS model was constructed in the
SIMCA-P 10.0 (Umetrics) software suite according to the following iterative formulas:
El = X - tlplT; E2 = El - t2p2T, t2 = Elwl; Ei = Ei- - tipiT , ti = Eilwi
Fl = Y - btlqlT; F2 = F - b2t2q2T;Fi = Fi-l- bitiqiT
where Ei represents the residual of the ith principal component, with score vector ti,
weight vector wi, and loading vector pi, and T represents transpose. Fi represents the
residuals of the ith dependent principal component, with score vector ti and loading
vector qi, and bi represents the coefficient characterizing the inner relation between the
independent and dependent principal components. Model predictions were made by
leave-one-out cross-validation for the IFN-Adv-TNF treatments and by unbiased
prediction for the HeLa predictions [47]. Model uncertainties were calculated by jackknifing [89]. Treatment projections (Fig. 4-4c) were plotted using t and t2 respectively
after a 50 ° subspace rotation [201]. Centered and scaled coefficients were used as the
regression weights.
To approximate the dynamic network behavior for a given treatment in the
presence of PI3K inhibition, the level of Akt was held constant at the LY294002inhibited activity level measured at 12 hours for HT-29 (0.17) and HeLa (0.33) compared
89
to a zero-minute baseline of one. Similarly, IKK activity levels were held constant at the
SC-514-inhibited activity level measured in Supplementary Fig. S6 (0.52). With both
inhibitor predictions, time-course values of the other kinases were kept identical to the
network behavior measured in the absence of inhibitor, and metrics were extracted as
described above. For HT-29-HeLa hybrid profiles, the measured HeLa kinase activities
from Adv-TNF treatment were replaced with the corresponding kinase activities
measured in HT-29 cells, and metrics were extracted as described above.
6.2.9. Statistical analysis
For comparing two individual means, a Student's t test was used. For comparing
differences in means on a Western blot across multiple blots, a paired Student's t test was
used. A Pearson's correlation coefficient was calculated to assess linear correlation. For
comparing two time courses, a two-factor analysis of variance (ANOVA) was used. The
sigmoid function was defined as:
base +max
1+expxhf T-X
)
where y is percent viability, x is normalized Akt activity or level of phosphorylation, base
is baseline percent viability, max is maximum increase in percent viability, xhalfis the x
value at which y is at (base + max)/2 and Tsis the transition steepness. (Note: Smaller Ts
causes a faster rise). The sigmoid function was fit to data using Igor Pro Graphing
software (WaveMetrics). Base and max values were held constant at 20% and 65%,
respectively. The Levenberg-Marquardt non-linear, least-squares fitting algorithm was
used to search for values of xhalfand rate that minimize chi-square. 90% confidence
interval for sigmoid curve fit was calculated using support plane analysis [202].
90
6.3. References
1.
2.
3.
4.
Janeway, C.A., Jr. and R. Medzhitov, Innate immune recognition. Annu Rev
Immunol, 2002. 20: p. 197-216.
Shenk, T., Adenoviridae: The viruses and their replication, in Fields Virology,
D.M. Knipe and P.M. Howley, Editors. 2001, Lippincott, Williams and Wilkins:
Philadelphia, PA. p. 2265-300.
Bergelson, J.M., et al., Isolation of a common receptor for Coxsackie B viruses
and adenoviruses 2 and 5. Science, 1997. 275(5304): p. 1320-3.
Nemerow, G.R., Cell receptors involved in adenovirus entry. Virology, 2000.
274(1): p. 1-4.
5.
6.
7.
Li, E., et al.,Adenovirusendocytosisvia alpha(v)integrinsrequires
phosphoinositide-3-OH kinase. J Virol, 1998. 72(3): p. 2055-6 1.
Li, E., et al., Association ofpl3OCAS with phosphatidylinositol-3-OH kinase
mediates adenovirus cell entry. J Biol Chem, 2000. 275(19): p. 14729-35.
Suomalainen, M., et al., Adenovirus-activated PKA and p38/MAPK pathways
boost microtubule-mediated nuclear targeting of virus. Embo J, 2001. 20(6): p.
1310-9.
8.
9.
10.
Braithwaite, A.W. and I.A. Russell, Induction of cell death by adenoviruses.
Apoptosis, 2001. 6(5): p. 359-70.
Berk, A.J., Adenovirus promoters and EIA transactivation. Annu Rev Genet,
1986. 20: p. 45-79.
de Jong, R.N., P.C. van der Vliet, and A.B. Brenkman, Adenovirus DNA
replication:proteinpriming, jumping back and the role of the DNA binding
protein DBP. Curr Top Microbiol Immunol, 2003. 272: p. 187-211.
11.
12.
13.
McConnell,M.J. and M.J. Imperiale,Biology of adenovirus and its use as a
vectorfor gene therapy. Hum Gene Ther, 2004. 15(11): p. 1022-33.
Imperiale, M.J. and S. Kochanek, Adenovirus vectors: biology, design, and
production. Curr Top Microbiol Immunol, 2004. 273: p. 335-57.
Thomas, C.E., A. Ehrhardt, and M.A. Kay, Progress andproblems with the use of
viral vectors for gene therapy. Nat Rev Genet, 2003. 4(5): p. 346-58.
14.
Lusky,M., et al., In vitro and in vivo biology of recombinantadenovirusvectors
15.
Andrews, J.L., et al., Generationand characterizationof EI/E2a/E3/E4-deficient
with El, E1/E2A, or El/E4 deleted. J Virol, 1998. 72(3): p. 2022-32.
16.
17.
adenoviral vectors encoding human factor VIII. Mol Ther, 2001. 3(3): p. 329-36.
Palmer, D.J. and P. Ng, Helper-dependent adenoviral vectors for gene therapy.
Hum Gene Ther, 2005. 16(1): p. 1-16.
Ehrhardt, A. and M.A. Kay, A new adenoviral helper-dependent vector results in
long-term therapeuticlevels of human coagulationfactor IXat low doses in vivo.
18.
Blood, 2002. 99(11): p. 3923-30.
Brunetti-Pierri, N., et al., Acute toxicity after high-dose systemic injection of
helper-dependent adenoviral vectors into nonhuman primates. Hum Gene Ther,
2004. 15(1): p. 35-46.
91
19.
Yang, Y., et al., Cellular immunity to viral antigens limits El-deleted
adenovirusesfor gene therapy. Proc Natl Acad Sci U S A, 1994. 91(10): p. 440711.
20.
Muruve, D.A., The Innate Immune Response to Adenovirus Vectors. Human Gene
Therapy, 2004. 15(12): p. 1157-1166.
21.
Lieber, A., et al., The role of Kupffercell activationand viral gene expression in
early liver toxicity after infusionof recombinantadenovirusvectors.J Virol,
22.
1997. 71(11): p. 8798-807.
Lieber, A., et al., Inhibition of NF-kappaB activation in combination with bc1-2
expressionallowsfor persistence offirst-generationadenovirus vectorsin the
23.
mouse liver. J Virol, 1998. 72(11): p. 9267-77.
Otake, K., et al., Nonspecific inflammation inhibits adenovirus-mediated
pulmonarygene transfer and expressionindependentof specificacquired immune
responses. Hum Gene Ther, 1998. 9(15): p. 2207-22.
24.
25.
Muruve, D.A., et al., Adenoviralgene therapyleads to rapid inductionof multiple
chemokinesand acute neutrophil-dependenthepatic injury in vivo.Hum Gene
Ther, 1999. 10(6): p. 965-76.
Schnell, M.A., et al., Activation of innate immunity in nonhuman primates
following intraportaladministrationof adenoviralvectors.Mol Ther, 2001. 3(5
26.
Pt 1): p. 708-22.
Liu, Q. and D.A. Muruve, Molecular basis of the inflammatory response to
adenovirus vectors. Gene Ther, 2003. 10(11): p. 935-40.
27.
Tibbles, L.A., et al., Activation ofp38 and ERK signalingduring adenovirus
28.
29.
30.
vector cell entry lead to expression of the C-X-C chemokine IP-IO. J Virol, 2002.
76(4): p. 1559-68.
Bruder, J.T. and I. Kovesdi, Adenovirus infection stimulates the Raf/MAPK
signalingpathway and induces interleukin-8 expression. J Virol, 1997. 71(1): p.
398-404.
Hayden, M.S. and S. Ghosh, Signaling to NF-kappaB. Genes Dev, 2004. 18(18):
p. 2195-224.
Borgland, S.L., et al., Adenovirus vector-induced expression of the C-X-C
chemokineIP-10 is mediated throughcapsid-dependentactivationof NF-kappaB.
31.
32.
J Virol, 2000. 74(9): p. 3941-7.
Bowen, G.P., et al., Adenovirus vector-induced inflammation: capsid-dependent
induction of the C-C chemokine RANTES requires NF-kappa B. Hum Gene Ther,
2002. 13(3): p. 367-79.
Tamanini, A., et al., MAP kinases and NF-[kappa]B collaborate to induce ICAM-
I gene expressionin the earlyphase ofadenovirus infection.Virology, 2003.
307(2): p. 228-242.
33.
Elkon, K.B., et al., Tumor necrosisfactor alphaplays a central role in immunemediated clearance of adenoviral vectors. Proc Natl Acad Sci U S A, 1997.
94(18): p. 9814-9.
34.
Zhang,H.G., et al., Inhibition of tumor necrosisfactor alpha decreases
inflammationand prolongs adenovirusgene expression in lung and liver. Hum
Gene Ther, 1998. 9(13): p. 1875-84.
92
35.
36.
37.
38.
39.
40.
41.
42.
Benedict, C.A., Viruses and the TNF-related cytokines, an evolving battle.
Cytokine Growth Factor Rev, 2003. 14(3-4): p. 349-57.
Wajant, H., K. Pfizenmaier, and P. Scheurich, Tumor necrosis factor signaling.
Cell Death Differ, 2003. 10(1): p. 45-65.
Aggarwal,B.B., Signallingpathways of the TNF superfamily:a double-edged
sword. Nat Rev Immunol, 2003. 3(9): p. 745-56.
Karin, M. and Y. Ben-Neriah, Phosphorylation meets ubiquitination: the control
of NF-[kappa]B activity. Annu Rev Immunol, 2000. 18: p. 621-63.
Shaulian, E. and M. Karin, AP-I as a regulator of cell life and death. Nat Cell
Biol, 2002. 4(5): p. E131-6.
Riedl, S.J. and Y. Shi, Molecular mechanisms ofcaspase regulation during
apoptosis. Nat Rev Mol Cell Biol, 2004. 5(11): p. 897-907.
Chen, G. and D.V. Goeddel, TNF-RI signaling: a beautifulpathway. Science,
2002. 296(5573): p. 1634-5.
Yin, X.M., Signal transductionmediatedby Bid, a pro-death Bcl-2family
proteins, connectsthe death receptorand mitochondriaapoptosispathways. Cell
43.
Res, 2000. 10(3): p. 161-7.
Adams, J.M. and S. Cory, The Bcl-2 proteinfamily: arbiters of cell survival.
Science, 1998. 281(5381): p. 1322-6.
44.
Shi, Y., Apoptosome:the cellular enginefor the activationof caspase-9.
45.
46.
47.
Structure, 2002. 10(3): p. 285-8.
Barber, G.N., Host defense, viruses and apoptosis. Cell Death Differ, 2001. 8(2):
p. 113-26.
Perez, D. and E. White, EIA Sensitizes Cells to Tumor Necrosis Factor Alpha by
Downregulating c-FLIPS. J. Virol., 2003. 77(4): p. 2651-2662.
Gaudet, S., et al., A Compendium of Signals and Responses Triggered by
Prodeath and Prosurvival Cytokines. Mol Cell Proteomics, 2005. 4(10): p. 15691590.
48.
49.
Rao, L., et al., The adenovirus EIA proteins induce apoptosis, which is inhibited
by the EIB 19-kDa and Bcl-2 proteins. Proc Natl Acad Sci U S A, 1992. 89(16):
p. 7742-6.
Chiou, S.K., et al., Functional complementation of the adenovirus EIB 19-
kilodaltonprotein with Bcl-2 in the inhibitionof apoptosisin infectedcells. J
50.
51.
Virol, 1994. 68(10): p. 6553-66.
Boyd, J.M., et al., Adenovirus EIB 19 kDa and Bcl-2 proteins interact with a
common set of cellular proteins. Cell, 1994. 79(2): p. 341-51.
Perez, D. and E. White, TNF-alpha signals apoptosis through a bid-dependent
conformational change in Bax that is inhibited by EIB 19K. Mol Cell, 2000. 6(1):
p. 53-63.
52.
Berk, A.J., Recent lessonsin gene expression,cell cycle control,and cell biology
53.
from adenovirus. Oncogene, 2005. 24(52): p. 7673-85.
Friedman, J.M. and M.S. Horwitz, Inhibition of tumor necrosis factor alpha-
inducedNF-kappaB activationby the adenovirusE3-10.4/14.5Kcomplex.J
Virol, 2002. 76(11): p. 5515-21.
93
54.
Shisler, J., et al., The adenovirus E3-10.4K/14.5K complex mediates loss of cell
surface Fas (CD95)and resistanceto Fas-inducedapoptosis.J Virol, 1997.
55.
56.
71(11): p. 8299-306.
Tollefson, A.E., et al., Forced degradation of Fas inhibits apoptosis in
adenovirus-infected cells. Nature, 1998. 392(6677): p. 726-30.
Benedict, C.A., et al., Three adenovirus E3 proteins cooperate to evade apoptosis
by tumornecrosisfactor-relatedapoptosis-inducingligand receptor-i and -2. J
Biol Chem, 2001. 276(5): p. 3270-8.
57.
Ilan, Y., et al., Insertion of the adenoviralE3 region into a recombinantviral
vectorprevents antiviral humoraland cellular immune responsesandpermits
long-term gene expression. Proc Natl Acad Sci U S A, 1997. 94(6): p. 2587-92.
58.
59.
60.
61.
62.
63.
Bridge, E. and G. Ketner,Redundantcontrol of adenoviruslate gene expression
by early region 4. J Virol, 1989. 63(2): p. 631-8.
Weiden, M.D. and H.S. Ginsberg, Deletion of the E4 region of the genome
produces adenovirus DNA concatemers. Proc Natl Acad Sci U S A, 1994. 91(1):
p. 153-7.
Stracker, T.H., C.T. Carson, and M.D. Weitzman, Adenovirus oncoproteins
inactivate the Mrel l-RadSO-NBS1 DNA repair complex. Nature, 2002.
418(6895): p. 348-52.
Branton, P.E. and D.E. Roopchand, The role of adenovirus E4orf4 protein in viral
replication and cell killing. Oncogene, 2001. 20(54): p. 7855-65.
O'Shea, C., et al., Adenoviral proteins mimic nutrient/growth signals to activate
the mTOR pathwayfor viral replication. Embo J, 2005. 24(6): p. 1211-21.
Frese, K.K., et al., Selective PDZprotein-dependent stimulation of
phosphatidylinositol3-kinaseby the adenovirusE4-ORF1oncoprotein.
Oncogene, 2003. 22(5): p. 710-21.
64.
65.
Zhang, F., et al., AdenovirusE4 genepromotes selective endothelialcell survival
and angiogenesisvia activationof the vascular endothelial-cadherin/Akt
signaling pathway. J Biol Chem, 2004. 279(12): p. 11760-6.
Xia, Z., et al., Opposing effects of ERK and JNK-p38 MAP kinases on apoptosis.
Science, 1995. 270(5240): p. 1326-31.
66.
Gustin, J.A., et al., Cell type-specific expression of the IkappaB kinases
determinesthe significanceofphosphatidylinositol3-kinase/Aktsignalingto NF67.
68.
69.
kappa B activation. J Biol Chem, 2004. 279(3): p. 1615-20.
Datta, S.R., A. Brunet, and M.E. Greenberg, Cellular survival: a play in three
Akts. Genes Dev, 1999. 13(22): p. 2905-27.
Chen, W.N., et al., Induced autocrinesignalingthrough the epidermalgrowth
factor receptorcontributesto the responseof mammaryepithelialcells to tumor
necrosis factor alpha. J Biol Chem, 2004. 279(18): p. 18488-96.
Janes, K.A., et al., The response of human epithelial cells to TNF involves an
inducibleautocrine cascade.Cell (in press), 2006.
70.
71.
Allan, L.A., et al., Inhibition of caspase-9 through phosphorylation at Thr 125 by
ERK MAPK Nat Cell Biol, 2003. 5(7): p. 647-54.
Bonni, A., et al., Cell survival promoted by the Ras-MAPK signaling pathway by
transcription-dependentand -independentmechanisms.Science, 1999.
286(5443): p. 1358-62.
94
72.
73.
74.
75.
76.
77.
Reginato,M.J., et al., Integrins and EGFR coordinatelyregulate theproapoptotic protein Bim to prevent anoikis. Nat Cell Biol, 2003. 5(8): p. 733-40.
DiDonato, J.A., et al., A cytokine-responsive IkappaB kinase that activates the
transcriptionfactor NF-kappaB. Nature, 1997. 388(6642): p. 548-54.
Mercurio, F., et al., IKK-I and IKK-2: cytokine-activatedIkappaB kinases
essentialfor NF-kappaB activation. Science, 1997. 278(5339): p. 860-6.
Beg, A.A. and D. Baltimore, An essential role for NF-kappaB in preventing TNFalpha-induced cell death. Science, 1996. 274(5288): p. 782-4.
Van Antwerp, D.J., et al., Suppression of TNF-alpha-induced apoptosis by NFkappaB. Science, 1996. 274(5288): p. 787-9.
Wang, C.Y., M.W. Mayo, and A.S. Baldwin, Jr., TNF- and cancer therapy-
inducedapoptosis:potentiation by inhibitionof NF-kappaB.Science, 1996.
78.
274(5288): p. 784-7.
Raingeaud, J., et al., Pro-inflammatory cytokines and environmental stress cause
p38 mitogen-activatedprotein kinase activationby dual phosphorylationon
79.
80.
tyrosine and threonine. J Biol Chem, 1995. 270(13): p. 7420-6.
Park, J.M., et al., Macrophage apoptosis by anthrax lethalfactor through p38
MAP kinase inhibition. Science, 2002. 297(5589): p. 2048-51.
Kotlyarov, A., et al., MAPKAP kinase 2 is essentialfor LPS-induced TNF-alpha
biosynthesis. Nat Cell Biol, 1999. 1(2): p. 94-7.
81.
Minden,A., et al., Differentialactivationof ERK and JNK mitogen-activated
82.
protein kinases by Raf-l and MEKK. Science, 1994. 266(5191): p. 1719-23.
Chang, L., et al., The E3 ubiquitin ligase itch couples JNK activation to
TNFalpha-inducedcell death by inducingc-FLIP(L) turnover.Cell, 2006. 124(3):
83.
84.
p. 601-13.
Ventura, J.J., et al., JNKpotentiates TNF-stimulated necrosis by increasing the
production of cytotoxic reactive oxygen species. Genes Dev, 2004. 18(23): p.
2905-15.
Kamata, H., et al., Reactive oxygen species promote TNFalpha-induced death and
sustainedJNK activationby inhibitingMAP kinasephosphatases. Cell, 2005.
85.
86.
87.
88.
89.
120(5): p. 649-61.
Downward, J., The ins and outs of signalling. Nature, 2001. 411(6839): p. 759-62.
Pawson, T., Specificity in Signal Transduction: From Phosphotyrosine-SH2
Domain Interactions to Complex Cellular Systems. Cell, 2004. 116(2): p. 191203.
Ideker, T. and D. Lauffenburger, Building with a scaffold: emerging strategies for
high- to low-level cellular modeling. Trends Biotechnol, 2003. 21(6): p. 255-62.
Janes, K.A. and D.A. Lauffenburger, A biological approach to computational
models ofproteomic networks. Curr Opin Chem Biol, 2006. 10(1): p. 73-80.
Janes, K.A., et al., Cue-Signal-Response Analysis of TNF-Induced Apoptosis by
Partial Least SquaresRegression of Dynamic MultivariateData. Journal of
90.
91.
Computational Biology, 2004. 11(4): p. 544-561.
Perelson, A.S., Modelling viral and immune system dynamics. Nat Rev Immunol,
2002. 2(1): p. 28-36.
Jenner, R.G. and R.A. Young, Insights into host responses against pathogens
from transcriptionalprofiling. Nat Rev Microbiol, 2005. 3(4): p. 281-94.
95
92.
Nesvizhskii, A.I., et al., A statistical modelfor identifying proteins by tandem
mass spectrometrv. Anal Chem, 2003. 75(17): p. 4646-58.
93.
Sachs, K., et al., Causalprotein-signalingnetworks derivedfrom multiparameter
94.
single-cell data. Science, 2005. 308(5721): p. 523-9.
Hua, F., et al., Effects of Bcl-2 levels on Fas signaling-induced caspase-3
activation:moleculargenetic tests of computationalmodel predictions.J
95.
Immunol, 2005. 175(2): p. 985-95.
Dinh, A.T., T. Theofanous, and S. Mitragotri, A modelfor intracellular trafficking
ofadenoviral vectors. Biophys J, 2005. 89(3): p. 1574-88.
96.
Hautaniemi,S., et al., Modeling of signal-responsecascadesusing decisiontree
97.
Janes, K.A., et al.,A SystemsModel of SignalingIdentifiesa MolecularBasis Set
analysis. Bioinformatics, 2005. 21(9): p. 2027-35.
for Cytokine-InducedApoptosis. Science, 2005. 310(5754): p. 1646 - 1653.
98.
O'Shea, C.C., Viruses- seeking and destroyingthe tumorprogram. Oncogene,
99.
2005. 24(52): p. 7640-55.
Kitano, H. and K. Oda, Robustness trade-offs and host-microbial symbiosis in the
immune system. Molecular Systems Biology, 2006. doi:10.1038(msb4100039): p.
El-E10.
100.
101.
Fejer, G., et al., Adenovirus infection dramatically augments lipopolysaccharideinduced TNF production and sensitizes to lethal shock. J Immunol, 2005. 175(3):
p. 1498-506.
Kim, K.N., et al., Viral IL-10 and soluble TNF receptor act synergistically to
inhibit collagen-inducedarthritisfollowing adenovirus-mediatedgene transfer.J
Immunol, 2000. 164(3): p. 1576-81.
102.
103.
104.
105.
Philpott,N.J., et al., Adenovirus-inducedmaturation of dendritic cells through a
P13 kinase-mediated TNF-alpha induction pathway. Proc Natl Acad Sci U S A,
2004. 101(16): p. 6200-5.
Jacobson, M.D., M. Weil, and M.C. Raff, Programmed cell death in animal
development. Cell, 1997. 88(3): p. 347-54.
Lodish, H., et al., Molecular Cell Biology. Fourth ed. 2000, New York, NY: W.
H. Freeman and Company.
Leers,M.P., et al., Immunocytochemicaldetectionand mapping of a cytokeratin
18 neo-epitope exposed during early apoptosis. J Pathol, 1999. 187(5): p. 567-72.
106.
107.
108.
Saito,Y., et al., Adenovirus-mediatedtransfer of the PTEN gene inhibitshuman
colorectal cancer growth in vitro and in vivo. Gene Ther, 2003. 10(23): p. 1961-9.
Steinwaerder, D.S., C.A. Carlson, and A. Lieber, DNA replication offirstgeneration adenovirus vectors in tumor cells. Hum Gene Ther, 2000. 11(13): p.
1933-48.
Mittereder, N., K.L. March, and B.C. Trapnell, Evaluation of the concentration
and bioactivityof adenovirus vectorsfor gene therapy.J Virol, 1996.70(11): p.
109.
110.
7498-509.
Nyberg-Hoffman, C., et al., Sensitivity and reproducibility in adenoviral
infectious titer determination. Nat Med, 1997. 3(7): p. 808-11.
Mizuguchi, H. and T. Hayakawa, Targeted adenovirus vectors. Hum Gene Ther,
2004. 15(11): p. 1034-44.
96
111.
Janes, K.A., et al., A High-throughput Quantitative Multiplex Kinase Assay for
112.
Apoptosis. Mol Cell Proteomics, 2003. 2(7): p. 463-73.
Asthagiri, A.R., A.F. Horwitz, and D.A. Lauffenburger, A rapid and sensitive
MonitoringInformationFlow in SignalingNetworks: Applicationto Sepsisquantitativekinaseactivity assay using a convenient96-wellformat. Anal
113.
Biochem, 1999. 269(2): p. 342-7.
Tran, S.E., A. Meinander, and J.E. Eriksson, Instant decisions: transcription-
independentcontrol of death-receptor-mediatedapoptosis.Trends Biochem Sci,
2004. 29(11): p. 601-8.
114.
Janes, K.A., QuantitativeAnalysis of the Cytokine-MediatedApoptosis-Survival
Cell Death DecisionProcess, in BiologicalEngineering.2005, Massachusetts
Institute of Technology: Cambridge, MA.
115.
Clesham,G.J., et al., High adenoviralloads stimulateNF kappaB-dependentgene
expression in human vascular smooth muscle cells. Gene Ther, 1998. 5(2): p. 17480.
116.
Palmer, D.H., et al., Inhibition of NF-kappaB enhances the cytotoxicity of virus-
directed enzymeprodrug therapyand oncolyticadenoviruscancergene therapy.
Gene Ther, 2005. 12(15): p. 1187-97.
117.
Micheau,0. and J. Tschopp,Induction of TNF receptorI-mediated apoptosisvia
118.
two sequential signaling complexes. Cell, 2003. 114(2): p. 181-90.
Geladi, P. and B.R. Kowalski, Partial least squares regression-a tutorial.
Analytica Chimica Acta, 1986. 185: p. 1-17.
119.
Jones,N. and T. Shenk,Isolationof deletionand substitutionmutants of
120.
adenovirus type 5. Cell, 1978. 13(1): p. 181-8.
Gaynor, R.B. and A.J. Berk, Cis-acting induction of adenovirus transcription.
Cell, 1983. 33(3): p. 683-93.
121.
122.
123.
Schwarz,E., et al., Structureand transcriptionof humanpapillomavirus
sequences in cervical carcinoma cells. Nature, 1985. 314(6006): p. 111-4.
Scheffner, M., et al., Functions of human papillomavirus proteins. Curr Top
Microbiol Immunol, 1994. 186: p. 83-99.
Steinwaerder, D.S., C.A. Carlson, and A. Lieber, Human papilloma virus E6 and
E7proteins supportDNA replicationof adenovirusesdeletedfor the EIA and
124.
125.
126.
127.
128.
EIB genes. Mol Ther, 2001. 4(3): p. 211-6.
Tauber, B. and T. Dobner, Molecular regulation and biologicalfunction of
adenovirus early genes: the E4 ORFs. Gene, 2001. 278(1-2): p. 1-23.
Sarbassov, D.D., et al., Phosphorylation and regulation ofAkt/PKB by the rictormTOR complex. Science, 2005. 307(5712): p. 1098-101.
Cotten,M., et al., Psoralen treatmentof adenovirusparticles eliminates virus
replicationand transcriptionwhile maintainingthe endosomolyticactivity of the
virus capsid. Virology, 1994. 205(1): p. 254-61.
Stupack, D.G. and D.A. Cheresh, Get a ligand, get a life: integrins, signaling and
cell survival. J Cell Sci, 2002. 115(Pt 19): p. 3729-38.
Hersey, P. and X.D. Zhang, Overcoming resistance of cancer cells to apoptosis. J
Cell Physiol, 2003. 196(1): p. 9-18.
97
129.
Cowan, K.J. and K.B. Storey, Mitogen-activatedprotein kinases: new signaling
pathwaysfunctioning in cellular responses to environmentalstress. J Exp Biol,
130.
131.
2003. 206(Pt 7): p. 1107-15.
Lawlor, M.A. and D.R. Alessi, PKB/Akt: a key mediator of cellproliferation,
survival and insulin responses? J Cell Sci, 2001. 114(Pt 16): p. 2903-10.
Remacle-Bonnet, M.M., et al., Insulin-like growthfactor-Iprotects colon cancer
cellsfrom deathfactor-inducedapoptosis by potentiating tumornecrosisfactor
alpha-inducedmitogen-activatedprotein kinase and nuclearfactor kappaB
signaling pathways. Cancer Res, 2000. 60(7): p. 2007-17.
132. Abreu-Martin,M.T., et al., Divergent inductionof apoptosisand IL-8 secretion in
HT-29 cells in responseto TNF-alphaand ligation of Fas antigen.J Immunol,
1995. 155(9): p. 4147-54.
133.
Manning, B.D., et al., Identification of the Tuberous Sclerosis Complex-2 Tumor
Suppressor GeneProduct Tuberinas a Target of the Phosphoinositide3Kinase/Akt Pathway. Molecular Cell, 2002. 10(1): p. 151-162.
134.
Cross, D.A., et al., Inhibition of glycogensynthase kinase-3by insulin mediated
byprotein kinase B. Nature, 1995. 378(6559): p. 785-9.
135.
Sasagawa,S., et al., Predictionand validationof the distinct dynamics of
136.
transient and sustained ERK activation. Nat Cell Biol, 2005. 7(4): p. 365-73.
Goldbeter, A. and D.E. Koshland, Jr., Ultrasensitivity in biochemical systems
controlledby covalent modification.Interplay betweenzero-orderand multistep
137.
effects. J Biol Chem, 1984. 259(23): p. 14441-7.
Zhang, Y., et al., Acute cytokine response to systemic adenoviral vectors in mice
is mediatedby dendriticcells and macrophages.Mol Ther, 2001. 3(5 Pt 1): p.
697-707.
138. Zhang, H.G., et al., HepaticDR5 inducesapoptosis and limitsadenovirusgene
therapy product expression in the liver. J Virol, 2002. 76(11): p. 5692-700.
139.
Sangro,B., et al., Phase I Trialof IntratumoralInjection of an Adenovirus
140.
Encoding Interleukin-12 for Advanced Digestive Tumors. J Clin Oncol, 2004.
22(8): p. 1389-1397.
Wolf, J.K., et al., A phase I study ofAdp53 (INGN 201; ADVEXIN) for patients
withplatinum- andpaclitarel-resistantepithelialovarian cancer. Gynecologic
Oncology, 2004. 94(2): p. 442-448.
141. Joseph, P.M., et al., Aerosol and Lobar Administrationof a Recombinant
Adenovirusto Individualswith CysticFibrosis.I. Methods, Safety, and Clinical
142.
Implications. Human Gene Therapy, 2001. 12(11): p. 1369-1382.
Park, C.S., I.C. Schneider, and J.M. Haugh, Kinetic analysis ofplatelet-derived
growthfactor receptor/phosphoinositide3-kinase/Aktsignalinginfibroblasts. J
143.
144.
Biol Chem, 2003. 278(39): p. 37064-72.
Beaulieu, J.M., et al., An Akt/beta-arrestin 2/PP2A signaling complex mediates
dopaminergic neurotransmission and behavior. Cell, 2005. 122(2): p. 261-73.
Gao, T., F. Furnari, and A.C. Newton, PHLPP: a phosphatase that directly
dephosphorylatesAkt, promotes apoptosis,and suppresses tumorgrowth. Mol
Cell, 2005. 18(1): p. 13-24.
98
145.
Tanji, C., et al., A-kinase anchoring protein AKAP220 binds to glycogen synthase
kinase-3beta(GSK-3beta) and mediatesprotein kinaseA-dependentinhibitionof
146.
147.
GSK-3beta. J Biol Chem, 2002. 277(40): p. 36955-61.
Sakashita, G., et al., Regulation of type I protein phosphatase/inhibitor-2 complex
by glycogen synthase kinase-3beta in intact cells. J Biochem (Tokyo), 2003.
133(2): p. 165-71.
Hemmings, B.A., T.J. Resink, and P. Cohen, Reconstitution ofa Mg-A TP-
dependentprotein phosphatase and its activationthrough a phosphorylation
148.
mechanism. FEBS Lett, 1982. 150(2): p. 319-24.
Hoeflich, K.P., et al., Requirementfor glycogen synthase kinase-3beta in cell
survival and NF-kappaB activation. Nature, 2000. 406(6791): p. 86-90.
149. Rogalla, T., et al., RegulationofHsp27 oligomerization,chaperonefunction, and
protective activity against oxidativestress/tumornecrosisfactor alpha by
150.
151.
152.
153.
154.
155.
156.
157.
158.
159.
phosphorylation. J Biol Chem, 1999. 274(27): p. 18947-56.
Goldsby, R.A., T.J. Kindt, and B.A. Osborne, Kuby Immunology. Fourth ed, ed.
G.L. Hadler. 2000, New York: W.H. Freeman and Company.
Kreuz, S., et al., NF-{kappa}B Inducers Upregulate cFLIP, a CycloheximideSensitive Inhibitor ofDeath Receptor Signaling. Mol. Cell. Biol., 2001. 21(12): p.
3964-3973.
Romashkova, J.A. and S.S. Makarov, NF-kappaB is a target ofAKT in antiapoptotic PDGF signalling. Nature, 1999. 401(6748): p. 86-90.
Ozes, O.N., et al., NF-kappaB activation by tumour necrosis factor requires the
Akt serine-threonine kinase. Nature, 1999. 401(6748): p. 82-5.
Bouwmeester, T., et al., A physical andfunctional map of the human TNFalpha/NF-kappa B signal transduction pathway. Nat Cell Biol, 2004. 6(2): p. 97105.
Cho, K.H., et al., Investigationsinto the analysisand modelingof the TNF alphamediated NF-kappa B-signaling pathway. Genome Res, 2003. 13(11): p. 2413-22.
Irish, J.M., et al., Single cellprofiling ofpotentiatedphospho-protein networks in
cancer cells. Cell, 2004. 118(2): p. 217-28.
Schmitz, E.K., D.M. Kraus, and G.A. Bulla, Tissue-specificity ofapoptosis in
hepatoma-derived cell lines. Apoptosis, 2004. 9(3): p. 369-75.
Ohmori, Y. and T.A. Hamilton, Cell type and stimulus specific regulation of
chemokine gene expression. Biochem Biophys Res Commun, 1994. 198(2): p.
590-6.
Goldfeld, A.E., C. Doyle, and T. Maniatis, Human tumor necrosis factor alpha
gene regulation by virus and lipopolysaccharide. Proc Natl Acad Sci U S A,
1990. 87(24): p. 9769-73.
160. Kenakin,T., Predictingtherapeuticvalue in the lead optimizationphase of drug
161.
162.
discovery. Nat Rev Drug Discov, 2003. 2(6): p. 429-38.
McCormick, F., Cancer gene therapy: fringe or cutting edge? Nat Rev Cancer,
2001. 1(2): p. 130-41.
Miller-Jensen, K., et al., El/E3-deleted adenoviral vector saturates Akt anti-
apoptotic signalingand sensitizes human epithelialcells to tumornecrosisfactormediated apoptosis. submitted, 2006.
99
163.
Fransen, L., et al., Recombinant tumor necrosis factor: its effect and its synergism
with interferon-gammaon a variety of normaland transformedhumancell lines.
164.
165.
Eur J Cancer Clin Oncol, 1986. 22(4): p. 419-26.
Sen, G.C., Viruses and interferons. Annu Rev Microbiol, 2001. 55: p. 255-81.
Shayakhmetov, D.M., et al., Interference with the IL-I-signalingpathway
improvesthe toxicityprofile of systemicallyappliedadenovirusvectors. J
Immunol, 2005. 174(11): p. 7310-9.
166.
167.
Pim, D., et al., Activation of theprotein kinaseB pathway by the HPV-16 E7
oncoproteinoccursthrough a mechanisminvolvinginteractionwith PP2A.
Oncogene, 2005. 24(53): p. 7830-8.
Glaunsinger, B.A., et al., Interactions of the PDZ-protein MAGI-1 with
adenovirusE4-ORF1and high-riskpapillomavirusE6 oncoproteins.Oncogene,
168.
169.
170.
2000. 19(46): p. 5270-80.
Filippova, M., et al., The human papillomavirus 16 E6 protein can either protect
orfurther sensitize cells to TNF: effect of dose. Cell Death Differ, 2005.
Dancey, J. and E.A. Sausville, Issues andprogress with protein kinase inhibitors
for cancer treatment. Nat Rev Drug Discov, 2003. 2(4): p. 296-313.
Garber, K., The second wave in kinase cancer drugs. Nat Biotechnol, 2006. 24(2):
p. 127-30.
171.
Solomon,S.D., et al., Cardiovascularrisk associatedwith celecoxib in a clinical
172.
trialfor colorectal adenoma prevention. N Engl J Med, 2005. 352(11): p. 107180.
Luo, J., B.D. Manning, and L.C. Cantley, Targeting the PI3K-Akt pathway in
human cancer: rationale andpromise. Cancer Cell, 2003. 4(4): p. 257-62.
Miller-Jensen, K., K.A. Janes, and D.A. Lauffenburger, Cell-specific signaling
173.
and apoptosisare linked by a common effector-processingmechanism.submitted,
174.
175.
176.
177.
2006.
Karin, M., Y. Yamamoto, and Q.M. Wang, The IKK NF-kappa B system: a
treasure trovefor drug development. Nat Rev Drug Discov, 2004. 3(1): p. 17-26.
Karin, M. and F.R. Greten, NF-kappaB: linking inflammation and immunity to
cancer development andprogression. Nat Rev Immunol, 2005. 5(10): p. 749-59.
Kishore, N., et al., A selective IKK-2 inhibitor blocks NF-kappa B-dependent gene
expression in interleukin-1 beta-stimulated synovialfibroblasts. J Biol Chem,
2003. 278(35): p. 32861-71.
Pierce, J.W., et al., Novel inhibitors ofcytokine-induced IkappaBalpha
phosphorylationand endothelialcell adhesionmolecule expressionshow antiinflammatory effects in vivo. J Biol Chem, 1997. 272(34): p. 21096-103.
178. Lin, Y.Z., et al., Inhibition of nucleartranslocationof transcriptionfactor NFkappa B by a syntheticpeptide containing a cell membrane-permeablemotif and
179.
nuclear localization sequence. J Biol Chem, 1995. 270(24): p. 14255-8.
Kirschner, M.W. and J.C. Gerhart, The Plausibility ofLife: Resolving Darwin's
Dilemma. 2005, New Haven: Yale University Press.
180. Oda, K., et al., A comprehensivepathway map of epidermalgrowthfactor
181.
receptor signaling. Molecular Systems Biology, 2005.
Verma, I.M. and M.D. Weitzman, Gene therapy: twenty-first century medicine.
Annu Rev Biochem, 2005. 74: p. 711-38.
100
182.
Keith, C.T., A.A. Borisy, and B.R. Stockwell, Multicomponent therapeutics for
networked systems. Nat Rev Drug Discov, 2005. 4(1): p. 71-8.
183. Armeanu,S., et al., Adenoviralgene transferof tumornecrosisfactor-related
apoptosis-inducingligand overcomesan impaired responseof hepatoma cells but
causessevere apoptosis in primary human hepatocytes.Cancer Res, 2003.
184.
63(10): p. 2369-72.
Griffith, T.S., et al., Adenoviral-mediated transfer of the TNF-related apoptosis-
inducingligand/Apo-2ligandgene inducestumorcell apoptosis.J Immunol,
185.
2000. 165(5): p. 2886-94.
Rasmussen, H., et al., TNFerade Biologic: preclinical toxicology of a novel
adenovectorwith a radiation-induciblepromoter, carrying the human tumor
186.
necrosisfactor alpha gene. Cancer Gene Ther, 2002. 9(11): p. 951-7.
Voelkel-Johnson, C., D.L. King, and J.S. Norris, Resistance ofprostate cancer
cells to soluble TNF-related apoptosis-inducing ligand (TRAIL/Apo2L) can be
overcome by doxorubicinor adenoviraldeliveryoffull-length TRAIL.Cancer
Gene Ther, 2002. 9(2): p. 164-72.
187. Li, J., et al., PTEN, a putativeprotein tyrosinephosphatasegene mutatedin
188.
189.
human brain, breast, andprostate cancer. Science, 1997. 275(5308): p. 1943-7.
Everts, B. and H.G. van der Poel, Replication-selective oncolytic viruses in the
treatment of cancer. Cancer Gene Ther, 2005. 12(2): p. 141-61.
DeWeese, T.L., et al., A phase I trial of CV706, a replication-competent, PSA
selective oncolyticadenovirus,for the treatmentof locallyrecurrentprostate
190.
cancer following radiation therapy. Cancer Res, 2001. 61(20): p. 7464-72.
Chiocca, E.A., et al., A phase I open-label, dose-escalation, multi-institutional
trial of injectionwith an E1B-Attenuatedadenovirus,ONYX-015,into the
peritumoral region of recurrent malignantgliomas, in the adjuvantsetting. Mol
193.
Ther, 2004. 10(5): p. 958-66.
Liu, T.C. and D. Kim, Viruses with deletions in antiapoptotic genes as potential
oncolytic agents. Oncogene, 2005. 24(40): p. 6069-79.
Wilson, J.M., Adenovirus-mediated gene transfer to liver. Adv Drug Deliv Rev,
2001. 46(1-3): p. 205-9.
Diehl, A.M., Cytokine regulation of liver injury and repair. Immunol Rev, 2000.
194.
Jordan, A., P. Defechereux, and E. Verdin, The site of HIV-1 integration in the
191.
192.
174: p. 160-71.
humangenome determinesbasal transcriptionalactivity and responseto Tat
195.
transactivation. Embo J, 2001. 20(7): p. 1726-38.
Pierson, T., J. McArthur, and R.F. Siliciano, Reservoirs for HIV-I: mechanisms
for viralpersistence in thepresence of antiviral immuneresponses and
196.
197.
198.
antiretroviral therapy. Annu Rev Immunol, 2000. 18: p. 665-708.
Jordan, A., D. Bisgrove, and E. Verdin, HIV reproducibly establishes a latent
infection after acute infection of T cells in vitro. Embo J, 2003. 22(8): p. 1868-77.
Weinberger, L.S., et al., Stochastic gene expression in a lentiviralpositivefeedback loop: HIV-1 Tatfluctuations drive phenotypic diversity. Cell, 2005.
122(2): p. 169-82.
Korin, Y.D., et al., Effects ofprostratin on T-cell activation and human
immunodeficiency virus latency. J Virol, 2002. 76(16): p. 8118-23.
101
199.
Werner, S.L., D. Barken, and A. Hoffinann, Stimulus specificity ofgene
expressionprograms determinedby temporalcontrol of IKK activity. Science,
200.
201.
202.
2005. 309(5742): p. 1857-61.
Varga, C.M., K. Hong, and D.A. Lauffenburger, Quantitative analysis of synthetic
gene delivery vector design properties. Mol Ther, 2001. 4(5): p. 438-46.
Alter, O., P.O. Brown, and D. Botstein, Singular value decomposition for
genome-wide expression data processing and modeling. Proc Natl Acad Sci U S
A, 2000. 97(18): p. 10101-6.
Lakowicz, J.R., Principles of Fluorescence Spectroscopy (2nd ed.). 1999: p. 118124.
102