Translational Repression Mediates Activation of Nuclear Factor Kappa B by Phosphorylated Translation Initiation Factor 2
Numerous stressful conditions activate kinases that phosphorylate the alpha subunit of translation initiation factor 2 (eIF2alpha), thus attenuating mRNA translation and activating a gene expression program known as the integrated stress response. It has been noted that conditions associated with eIF2alpha phosphorylation, notably accumulation of unfolded proteins in the endoplasmic reticulum (ER), or ER stress, are also associated with activation of nuclear factor kappa B (NF-kappaB) and that eIF2alpha phosphorylation is required for NF-kappaB activation by ER stress. We have used a pharmacologically activable version of pancreatic ER kinase (PERK, an ER stress-responsive eIF2alpha kinase) to uncouple eIF2alpha phosphorylation from stress and found that phosphorylation of eIF2alpha is both necessary and sufficient to activate both NF-kappaB DNA binding and an NF-kappaB reporter gene. eIF2alpha phosphorylation-dependent NF-kappaB activation correlated with decreased levels of the inhibitor IkappaBalpha protein. Unlike canonical signaling pathways that promote IkappaBalpha phosphorylation and degradation, eIF2alpha phosphorylation did not increase phosphorylated IkappaBalpha levels or affect the stability of the protein. Pulse-chase labeling experiments indicate instead that repression of IkappaBalpha translation plays an important role in NF-kappaB activation in cells experiencing high levels of eIF2alpha phosphorylation. These studies suggest a direct role for eIF2alpha phosphorylation-dependent translational control in activating NF-kappaB during ER stress.
MOLECULAR AND CELLULAR BIOLOGY, Dec. 2004, p. 10161–10168 Vol. 24, No. 23
0270-7306/04/$08.00⫹0 DOI: 10.1128/MCB.24.23.10161–10168.2004
Copyright © 2004, American Society for Microbiology. All Rights Reserved.
Translational Repression Mediates Activation of Nuclear Factor
Kappa B by Phosphorylated Translation Initiation Factor 2
Phoebe D. Lu,
Randal J. Kaufman,
Heather P. Harding,
and David Ron
Skirball Institute of Biomolecular Medicine
and Departments of Cell Biology, Medicine, and Pharmacology,
New York University School of Medicine, New York, New York; Department of Biochemistry
Hughes Medical Institute,
University of Michigan School of Medicine, Ann Arbor, Michigan;
and Department of Biochemistry, McGill University, Montreal, Quebec, Canada
Received 11 May 2004/Returned for modiﬁcation 6 July 2004/Accepted 31 August 2004
Numerous stressful conditions activate kinases that phosphorylate the alpha subunit of translation initia-
tion factor 2 (eIF2␣), thus attenuating mRNA translation and activating a gene expression program known as
the integrated stress response. It has been noted that conditions associated with eIF2␣ phosphorylation,
notably accumulation of unfolded proteins in the endoplasmic reticulum (ER), or ER stress, are also associated
with activation of nuclear factor kappa B (NF-B) and that eIF2␣ phosphorylation is required for NF-B
activation by ER stress. We have used a pharmacologically activable version of pancreatic ER kinase (PERK,
an ER stress-responsive eIF2␣ kinase) to uncouple eIF2␣ phosphorylation from stress and found that
phosphorylation of eIF2␣ is both necessary and sufﬁcient to activate both NF-B DNA binding and an NF-B
reporter gene. eIF2␣ phosphorylation-dependent NF-B activation correlated with decreased levels of the
inhibitor IB␣ protein. Unlike canonical signaling pathways that promote IB␣ phosphorylation and degra-
dation, eIF2␣ phosphorylation did not increase phosphorylated IB␣ levels or affect the stability of the protein.
Pulse-chase labeling experiments indicate instead that repression of IB␣ translation plays an important role
in NF-B activation in cells experiencing high levels of eIF2␣ phosphorylation. These studies suggest a direct
role for eIF2␣ phosphorylation-dependent translational control in activating NF-B during ER stress.
Diverse stressful conditions lead to the phosphorylation of
translation initiation factor 2 on its alpha subunit (eIF2␣).
Phosphorylated eIF2 inhibits its guanine nucleotide exchange
factor, eIF2B, and thereby inhibits the exchange reaction re-
quired to generate active GTP-bound eIF2. As a consequence,
regulated phosphorylation of eIF2␣ serves to modulate mRNA
translation rates (18, 20). In addition to its negative impact on
global protein synthesis, eIF2 phosphorylation also promotes
gene-speciﬁc upregulation of the translation of certain
mRNAs. The two known examples of this involve the yeast
transcription factor GCN4 (19) and the mammalian transcrip-
tion factor ATF4 (12). Regulated gene expression appears to
be an important consequence of eIF2␣ phosphorylation, as
mutations that interfere with eIF2␣ phosphorylation lead to an
important defect in stress-induced gene expression (16, 28, 39).
Four known eIF2␣ kinases couple seemingly unrelated
stressful conditions to the aforementioned common transla-
tional regulatory event. PKR responds to double-stranded
RNA in virally infected cells (23), GCN2 is activated by un-
charged tRNAs in amino acid-starved cells (20), HRI is acti-
vated by heme depletion in erythroid precursor cells (3), and
PERK is activated by unfolded proteins in the endoplasmic
reticulum (ER), or ER stress (37). Mutations in each of these
four kinases have been produced, and their phenotypes reveal
the importance of eIF2␣ phosphorylation in stressed cells (6).
Nuclear factor kappa B (NF-B) encompasses a family of
stress-induced transcription factors. Like the more ancient
eIF2␣ phosphorylation-dependent signaling, NF-B signaling
is also triggered by diverse stressful conditions, and activated
NF-B has broad effects on gene expression (38). Several stud-
ies have suggested cross talk between the eIF2␣ phosphor-
ylation pathway and NF-B activation. The double-stranded-
RNA-activated eIF2␣ kinase PKR was noted to phosphorylate
the NF-B inhibitor, IB (26), and genetic and pharmacolog-
ical interventions that interfere with PKR activity attenuated
NF-B activation by cytokines (4, 27, 47) or viruses (9, 43).
There is some uncertainty regarding the role of eIF2␣ phos-
phorylation in NF-B activation by PKR, as the latter contrib-
utes to NF-B activation by both kinase-dependent (9) and
kinase-independent (8) mechanisms.
Conditions that promote accumulation of unfolded proteins
in the endoplasmic reticulum lead to high levels of eIF2␣
phosphorylation (34, 35), which is mediated by the ER-local-
ized kinase PERK (14, 15). These same conditions activate
NF-B (32). A recent study has found that ER stress-mediated
NF-B activation was attenuated both in PERK
importantly, in cells bearing two mutant alleles of EIF2A in
which serine 51 (the substrate of the stress-inducible kinases)
had been mutated to an alanine. These mutant eIF2␣
were also defective in NF-B activation by amino acid starva-
tion, as were cells lacking GCN2 (21), the kinase that phos-
phorylates eIF2␣ in amino acid-starved cells.
Together these observations point to a nonredundant role
for eIF2␣ phosphorylation in NF-B activation under various
stress conditions. But they provide little insight into the mech-
anisms involved. One of the best-characterized aspects of
* Corresponding author. Mailing address: New York University
Medical Center, SI 3-10, 540 First Ave., New York, NY 10016. Phone:
(212) 263-7786. Fax: (212) 263-8951. E-mail: email@example.com
NF-B regulation is the phosphorylation-dependent, protea-
some-mediated degradation of its inhibitor, IB. However, it is
not clear if and how eIF2␣ phosphorylation ties in to IB
levels. Because the stressful conditions used to promote eIF2␣
phosphorylation have multiple other effects (reviewed in ref-
erence 17), it is not even known whether eIF2␣ phosphoryla-
tion plays a permissive role or an instructive role in NF-B
activation, nor is it known whether the phosphorylated form of
eIF2␣ is affecting NF-B activation as a modiﬁed translation
initiation factor or by some other means. In an effort to answer
some of these questions, we have probed NF-B activation in
an experimental system that uncouples eIF2␣ phosphorylation
from stress signaling and discovered that translational repres-
sion of IB can account for activation of NF-B under condi-
tions of eIF2␣ phosphorylation.
MATERIALS AND METHODS
Cell culture, cell transfection, and treatment. The wild-type cells and
mutant cells in which the serine at position 51, the regulatory phos
phorylation site, had been replaced by an alanine have been previously described
(39). They were cultured in Dulbecco’s modiﬁed Eagle’s medium supplemented
with glutamine, nonessential amino acids, 55 M ␤-mercaptoethanol, and 10%
fetal calf serum. The establishment of stable clones of mouse ﬁbroblasts express-
ing Fv2E-PERK with deﬁned EIF2A genotypes has been previously described
wild-type mouse embryonic ﬁbroblasts described above
were transiently transfected using Fugene lipid-based gene transfer reagent (cat-
alog no. 1814443; Roche, Indianapolis, Ind.) with luciferase reporter plasmids
containing a minimal rat angiotensinogen promoter driven by four wild-type or
mutant NF-B binding sites from the rat angiotensinogen gene, as previously
described (36). One day later the cells were treated for 1 h with the indicated
concentration of AP20187 (gift of ARIAD Pharmaceuticals, Cambridge, Mass.),
washed free of the activator (to allow translation to recover), and harvested for
use in a luciferase assay 24 h later.
Cells were treated with thapsigargin (catalog no. T9033; Sigma, St. Louis, Mo.)
at a ﬁnal concentration of 400 nM or cycloheximide (catalog no. C7698; Sigma)
at 20 g/ml. Unless otherwise indicated, AP20187 was used at a concentration of
10 nM. Cells were treated with 20 ng of tumor necrosis factor alpha (TNF-␣;
catalog no. T7539; Sigma)/ml with or without the proteasome inhibitor MG132
(catalog no. 474790; Calbiochem-Novobiochem, San Diego, Calif.) at 10 M.
Immunoblotting and immunoprecipitation. Total IB␣ was detected with a
puriﬁed rabbit immune serum (catalog no. 9242; Cell Signaling, Beverly, Mass.),
and IB␣ phosphorylated on serine 32 and 36 was detected with an epitope-
speciﬁc antiserum (catalog no. 9246; Cell Signaling). GADD34 was detected with
an antiserum directed to the N terminus of the mouse protein raised in our lab
(30). PERK was detected with a 1:1 mixture of two rabbit antisera (NY97, which
detects the unphosphorylated form of the protein, and NY201, which detects
predominantly the hyperphosphorylated forms of the protein) as described pre-
viously (2). Total eIF2␣ was detected with a monoclonal antibody to human
eIF2␣, a gift of the late Edward Henshaw (40), and phosphorylated eIF2␣ was
detected with an epitope-speciﬁc antiserum (catalog no. RG0001; Research
Genetics, Huntsville, Ala.).
Pulse-chase labeling experiments were carried out in the Fv2E-PERK
type mouse embryonic ﬁbroblasts described above. Cells were switched to Dul-
becco’s modiﬁed Eagle’s medium containing 10% of the normal content of
methionine and cysteine (these levels of methionine and cysteine are sufﬁcient to
suppress activation of the eIF2␣ kinase GCN2 yet are compatible with high-level
incorporation of labeled amino acids into newly synthesized proteins) 15 min
before addition of TRANSlabel (MP Biomedical, Irvine, Calif.)
methionine-cysteine mixture at 200 Ci/ml for 10 min. The labeling pulse was
terminated by washing the unincorporated label and ﬂooding the cells with
complete medium. Following the indicated chase period, during which cells were
exposed to AP20187 and/or MG132, the cells were lysed in RIPA buffer (20 mM
Tris [pH 8.5], 100 mM NaCl, 0.2% sodium deoxycholate, 0.2% NP-40, 0.2%
Triton X-100, 0.1% sodium dodecyl sulfate, 1 mM EDTA, 1 mM dithiothreitol,
1 mM phenylmethylsulfonyl ﬂuoride, 4 g of aprotinin/ml, 2 g of pepstatin/ml),
and the lysate was clariﬁed by centrifugation at 14,000 ⫻ g for 15 min, precleared
on protein A-Sepharose beads (catalog no. 10-1042; Zymed, South San Fran-
cisco, Calif.), and subjected to immunoprecipitation with prebound anti-IB␣
rabbit immunoglobulin G (catalog no. SC-371 AC; Santa Cruz Biotech, Santa
Cruz, Calif.). Radiolabeled proteins found in the immunoprecipitate were re-
solved by reduced sodium dodecyl sulfate-polyacrylamide gel electrophoresis,
and the dried gel was exposed to autoradiography using a phosphoimaging
cassette (Molecular Dynamics, Sunnyvale, Calif.).
EMSA. NF-B DNA binding activity in nuclear extracts was detected by an
electrophoretic mobility shift assay (EMSA) performed as previously described
(21, 36). The indicated molar excess of unlabeled competitor probe or 1 lof
puriﬁed anti-p65 (catalog no. SC-7151; Santa Cruz Biotech) or anti-CHOP
antiserum (45) was added to the binding reaction together with the radiolabeled
To conﬁrm the previously reported role of eIF2␣ phosphor-
ylation in NF-B activation (21), we performed EMSA on
nuclear extracts prepared from unstressed cells and cells that
had been treated with thapsigargin (Fig. 1A). Thapsigargin-
mediated ER calcium depletion leads to rapid onset of ER
FIG. 1. NF-B activation during ER stress depends on eIF2␣ phos-
phorylation and is associated with declining levels of the NF-B inhib-
itor IB␣. (A) Autoradiogram of an NF-B EMSA performed with
nuclear extracts of thapsigargin-treated mouse ﬁbroblasts (Tg) with
) or mutant (EIF2A
) EIF2A genotypes. The free
radiolabeled probe and the labeled NF-B/DNA complex are indi-
cated. (B) Immunoblots of IB␣, GADD34, phosphorylated eIF2␣,
and total eIF2␣ from extracts of the cells shown in panel A, detected
with speciﬁc antibodies.
10162 DENG ET AL. M
stress, eIF2␣ phosphorylation (detected here by immunoblot-
ting with an antiserum speciﬁcally reactive with the phosphor-
ylated form), and subsequent ATF4-mediated activation of
downstream gene expression, measured here by accumula-
tion of the GADD34 target gene. A protein complex rapidly
formed on the NF-B binding site in nuclear extracts of treated
wild-type cells but not in extracts from cells homozygous for
mutation that substitutes the serine at position
51 of eIF2␣ with an alanine and thereby prevents regulatory
phosphorylation. Reduced levels of the NF-B inhibitory pro-
tein IB␣, detected by immunoblotting, preceded the induc-
tion of NF-B EMSA activity in thapsigargin-treated cells. The
recovery of IB␣ levels at longer treatment points correlated
with the induction of the GADD34 phosphatase and the dephos-
phorylation of eIF2␣ (Fig. 1B).
To more closely examine the role of eIF2␣ phosphorylation
in NF-B activation, we made use of an experimental system
that uncouples eIF2␣ phosphorylation from stress signaling.
PERK, the ER stress-inducible eIF2␣ kinase, is normally ac-
tivated by oligomerization in the plane of the ER membrane
(2). We fused PERK’s eIF2␣ kinase domain to a protein mod-
ule with two high-afﬁnity binding sites for the otherwise inert
bivalent compound AP20187. When expressed in cells, this
artiﬁcial kinase, Fv2E-PERK, is subordinate to AP20187
treatment (28) and is activated independently of any stress
signaling. AP20187 treatment led to high-level eIF2␣ phos-
phorylation in Fv2E-PERK
cells but had no effect on the
parental cells lacking the artiﬁcial kinase (Fig. 2A). Fv2E-
PERK was readily activated in mutant EIF2A
cells, but this
predictably failed to induce eIF2␣ phosphorylation. EMSA of
nuclear extracts showed that AP20187 induced NF-B activity
) cells but not in the mu
cells (Fig. 2B). Homologous competition bind
ing assays and antibody supershift experiments conﬁrmed the
identity of the NF-B protein-DNA complex detected in the
assay (Fig. 2C).
To gauge the functional signiﬁcance of Fv2E-PERK-medi-
ated eIF2␣ phosphorylation and activation of NF-B DNA
binding activity, we measured the activity of a transfected re-
porter gene driven by four copies of a wild-type NF-B binding
site. A brief (60-min) pulse of AP20187 induced marked acti-
vation of the wild-type reporter gene (measured 24 h later [Fig.
3]). No activation of a reporter gene driven by mutant NF-B
sites was observed. In addition, endogenous NF-B target
genes, such as those encoding the major histocompatibility
complex heavy chains (H2-Q8, H2-2KF, H2-K2, and H2-D1)
and ␤2 microglobulin (Qb-1), were induced in the Fv2E-
cells by AP20187 treatments and in wild-type mouse
ﬁbroblasts by exposure to tunicamycin (National Center for
Biotechnology Information Gene Expression Omnibue [GEO]
data set GDS405).
Fv2E-PERK-mediated eIF2␣ phosphorylation and NF-B
activation correlated with a time-dependent decrease in IB␣
levels that was not observed in the mutant EIF2A
4A). Interestingly, Fv2E-PERK activation had no measurable
effect on levels of the p65 NF-B subunit, which is consistent
with the known stability of that protein (24) and with the
induction of NF-B binding activity that we observe. eIF2␣
levels were similarly stable, attesting to the effect’s speciﬁcity to
IB␣ (Fig. 4B). Canonical activators of NF-B access signal
FIG. 2. Phosphorylation of eIF2␣ on serine 51 is sufﬁcient to acti-
vate NF-B DNA binding activity in vivo. (A) Immunoblots of ligand-
activable Fv2E-PERK (upper panel), phosphorylated eIF2␣ (P-eIF2␣;
middle panel), and total eIF2␣ (lower panel) in extracts of mouse ﬁ-
broblasts of wild-type (S/S) and EIF2A
mutant (A/A) genotypes that
do and do not stably express the chimeric eIF2␣ kinase, Fv2E-PERK.
Where indicated, the cells had been treated with the Fv2E ligand,
AP20187. Endogenous PERK is not detected at this exposure. The
asterisk marks the position of a nonspeciﬁc band reactive with the anti-
PERK sera. (B) Autoradiogram of an NF-B EMSA performed with
nuclear extracts of cells treated as described for panel A. (C) Autora-
diogram of an NF-B EMSA with nuclear extract obtained from
cells performed in the presence of the
indicated excess of an unlabeled homologous competitor oligonucleo-
tide (left panel) or in the presence of antisera to CHOP (a negative
control) or p65 (a component of the NF-B DNA binding complex)
(right panel). The positions of the free radiolabeled probe, the NF-
B/DNA complex, and antiserum supershifted complex are indicated.
OL. 24, 2004 eIF2␣ PHOSPHORYLATION AND NF-B ACTIVATION 10163
transduction pathways that promote phosphorylation of the
inhibitor IB␣ on serines 32 and 36 (38). A ubiquitin ligase
complex recognizes the phosphorylated form of IB␣, and
polyubiquitinated IB␣ is degraded by the proteasome. Fv2E-
PERK activation by AP20187 did not promote a measurable
increase in levels of phosphorylated IB␣, which remained
undetectable. However, phosphorylated IB␣ was readily de-
tectable in lysates of cells treated with the proteasome inhibi-
tor, MG132, which stabilizes the phosphorylated form of the
protein (Fig. 4B).
Because it is rapidly degraded, signal-dependent accumula-
tion of phosphorylated IB␣ is difﬁcult to detect, rendering
an Fv2E-PERK-mediated increase in IB␣ phosphorylation
potentially easy to miss. Therefore, to determine if the eIF2␣
phosphorylation-dependent decline in IB␣ levels correlated
with any increased phosphorylation on serines 32 and 36, we
exposed the AP20187-treated cells to the proteasome inhibitor
MG132. As expected, proteasome inhibition markedly in-
creased the levels of phosphorylated IB␣ in tumor necrosis
factor alpha-treated cells (Fig. 5A). Interestingly, proteasome
inhibition led to only modest stabilization of total IB␣,an
observation that is consistent with the existence of proteasome-
independent mechanisms for IB␣ degradation (5, 11).
MG132 treatment led to a progressive increase in phosphor-
ylated IB␣ levels in cells that were otherwise unperturbed
(Fig. 5A, compare lanes 1 and 3, and B, compare lane 1 with
lanes 2, 4, 6, 8, and 10). This observation is consistent with a
relatively high basal phosphorylation-dependent turnover of
IB␣ in these cells. The decline in IB␣ levels effected by
Fv2E-PERK was only slightly attenuated by proteasome inhi-
bition (compare Fig. 4, lanes 4 to 6, with 5B, lanes 7, 9, and 11).
Furthermore, proteasome inhibition promoted some eIF2␣
phosphorylation (Fig. 5, lanes 8 and 10), presumably mediated
by proteotoxic stress. Remarkably, however, Fv2E-PERK ac-
tivation and eIF2␣ phosphorylation not only failed to increase
IB␣ phosphorylation but also signiﬁcantly attenuated the ac-
cumulation of phosphorylated IB␣ in proteasome-inhibited
cells (Fig. 5B, compare odd- and even-numbered lanes). These
observations indicate that eIF2␣ phosphorylation does not
activate NF-B by accessing one of the canonical IB␣ phos-
phorylation-promoting pathways and must use a different
The original descriptions of IB emphasized the lability of
the factor, as translational inhibitors were noted to promote
NF-B DNA binding activity (1, 42). Given that eIF2␣ phos-
phorylation also inhibits protein synthesis, we decided to ex-
plore this facet of NF-B activation in more detail. NF-B
DNA binding activity was increased by cycloheximide treat-
ment of wild-type cells, as previously reported (42), and this
correlated with reduced levels of the inhibitor, IB␣ (Fig. 6A).
Cycloheximide treatment led to no measurable decrease in p65
or eIF2␣ protein levels, attesting to the stability of these pro-
teins. The effects of cycloheximide on levels of phosphorylated
IB␣ also resembled those of Fv2E-PERK activation (Fig. 4B)
in that no increase in the phosphorylated protein was observed
FIG. 3. eIF2␣ phosphorylation is sufﬁcient to activate an NF-B
reporter gene. The activity of a transiently transfected reporter gene
consisting of a minimal promoter driven by four wild-type (wt) or
mutant (mut) NF-B binding sites in mouse ﬁbroblasts stably express-
ing Fv2E-PERK is shown following treatment with the indicated con-
centration of the activating ligand AP20187. The results are expressed
as relative light units, and the activity of the reporter in untreated cells
is arbitrarily set at 1. Shown are means and standard errors of the
means of results from an experiment performed in triplicate and re-
FIG. 4. eIF2␣ phosphorylation reduces cellular levels of IB␣. (A)
Immunoblots of total IB␣ (upper panel), phosphorylated eIF2␣ (P-
eIF2␣; middle panel), and total eIF2␣ (lower panel) in extracts of
) and mutant (EIF2A
ﬁbroblasts following treatment with the activating ligand AP20187 for
the indicated periods of time. (B) Immunoblots of total IB␣, phos-
phorylated IB␣ (P-IB␣), p65 NF-B subunit, and total eIF2␣ in
extracts of wild-type (EIF2A
mouse ﬁbroblasts fol
lowing treatment with the activating ligand AP20187 or the protea-
some inhibitor (MG132) for the indicated periods of time.
10164 DENG ET AL. M
in cells treated with cycloheximide alone. Proteasome inhibi-
tor, by itself, led to a progressive increase in levels of phos-
phorylated IB␣, whereas the addition of cycloheximide
strongly attenuated this increase (Fig. 6B).
As previously noted, cycloheximide treatment induced eIF2␣
phosphorylation (21, 22) (Fig. 6A), an effect that might be
attributed to loss of the labile eIF2␣ phosphatase CReP (22).
To study the role of eIF2␣ phosphorylation in cycloheximide-
mediated activation of NF-B, we treated mutant EIF2A
cells with the protein synthesis inhibitor and studied NF-B
activation by EMSA and IB␣ levels by immunoblotting. The
genotype, which inhibits regulatory phosphorylation
of eIF2␣, had no measurable effect on NF-B activation, IB␣
phosphorylation, or total IB␣ levels in cycloheximide-treated
cells (Fig. 6C). These observations suggest that inhibition of
new protein synthesis can adequately explain the effects of
cycloheximide on NF-B activity without evoking an additional
role for eIF2␣ phosphorylation.
Induced degradation of IB␣ plays an important role in
FIG. 5. Reduction in levels of IB␣ in cells with elevated eIF2␣
phosphorylation occurs independently of IB␣ phosphorylation. (A)
Immunoblots of IB␣ phosphorylated on serines 32 and 36 (P-IB␣;
upper panel) and total IB␣ (lower panel) in extracts of mouse ﬁbro-
blasts treated with TNF-␣ and/or the proteasome inhibitor MG132.
(B) Immunoblots of phosphorylated IB␣ (P-IB␣), total IB␣, phos-
phorylated eIF2␣ (P-eIF2␣), and total eIF2␣ in extracts of Fv2E-
mouse ﬁbroblasts treated with the activating ligand AP20187
and/or the proteasome inhibitor MG132 for the indicated periods of
FIG. 6. Reduction in levels of IB␣ in cells treated with the protein
synthesis inhibitor cycloheximide occurs independently of IB␣ phos-
phorylation or eIF2␣ phosphorylation. (A) The top panel is an auto-
radiogram of an NF-B EMSA from nuclear extracts of untreated and
cycloheximide (CHX)-treated mouse ﬁbroblasts. The lower panels are
immunoblots (IB) of total IB␣, phosphorylated IB␣ (P-IB␣), phos-
phorylated eIF2␣ (P-eIF2␣), total eIF2␣, and the p65 NF-B subunit
from the same cells. (B) Immunoblots of phosphorylated IB␣ (P-
IB␣), total IB␣, and total eIF2␣ in extracts of wild-type (EIF2A
mouse ﬁbroblasts treated with cycloheximide and/or the
proteasome inhibitor MG132 for the indicated periods of time are
shown. (C) The same assays as shown in panels A and B were con-
ducted with mutant (EIF2A
OL. 24, 2004 eIF2␣ PHOSPHORYLATION AND NF-B ACTIVATION 10165
canonical activation of NF-B. To address the possibility that
eIF2␣ phosphorylation might affect this aspect of IB␣ metab-
olism (independently of IB␣ phosphorylation), we performed
pulse-chase labeling experiments, tracking the fate of newly
synthesized IB␣. The basal turnover of IB␣ in murine ﬁbro-
blasts proved very high. Less than 30% of the signal measured
at the end of the 10-min labeling pulse was present after a
20-min chase. Furthermore, activation of Fv2E-PERK dur-
ing the chase had no measurable effect on the decay of the
IB␣ signal (Fig. 7A). Addition of proteasome inhibitor during
the chase stabilized IB␣ somewhat; however, in that context,
too, activation of Fv2E-PERK during the chase did not accel-
erate IB␣ degradation and may have even contributed mod-
estly to its stability (Fig. 7B). We conclude that IB␣ turns over
rapidly in murine ﬁbroblasts and that eIF2␣ phosphorylation
does not exert its effects on the levels of the inhibitor by further
enhancing its degradation.
Next we compared the rates of synthesis of IB␣ in un-
treated cells with those in cells treated with AP20187, cyclo-
heximide, the ER stress-promoting agent thapsigargin, and
the canonical NF-B activator TNF-␣. The amount of radio-
labeled IB␣ immunoprecipitated with a speciﬁc antibody
following a short labeling pulse was markedly diminished by
activation of the eIF2␣ kinase Fv2E-PERK by AP20187, by
treatment with cycloheximide, or by exposure to conditions
that cause ER stress (thapsigargin) (Fig. 7C). The effect of
thapsigargin on IB␣ synthesis depended on eIF2␣ phosphor-
ylation, since it was abolished in the EIF2A
(Fig. 7C), and the decline in IB␣ synthesis paralleled the
global inhibition in protein synthesis in the cells exposed to
conditions promoting eIF2␣ phosphorylation (Fig. 7D). By
contrast, exposure to the canonical NF-B activator, TNF-␣,
increased IB␣ synthesis, suggesting a completely different
mechanism of action. These observations are consistent with a
role for inhibited synthesis of IB␣ in mediating the effects of
eIF2␣ phosphorylation on NF-B activation both in ER-
stressed cells and following activation of Fv2E-PERK.
Signaling through stress-induced phosphorylation of eIF2␣
is conserved among the eukaryotes and represents one of the
oldest pathways for stress-induced gene expression. Further-
more, eIF2␣ phosphorylation is concerned mostly with auton-
omous cell adaptations to stress. NF-B signaling, on the other
hand, is found in metazoans, and canonical activators of NF-B
signaling, such as cytokines, are intercellular signaling mole-
cules. However, over the years evidence that autonomous cell
phenomena, such as ER stress, are also associated with NF-B
activation has accrued, with the suggestion that ancient, au-
tonomous cell signaling pathways might be linked to NF-B
This study conﬁrms the established role of eIF2␣ phosphor-
ylation in NF-B activation by ER stress (21). Using an induc-
ible system that uncouples eIF2␣ phosphorylation from other
stress signals, we ﬁnd that eIF2␣ phosphorylation can have an
instructive role in NF-B activation. In other words, activation
of an eIF2␣ kinase provides a signal sufﬁcient for NF-B ac-
tivation in cultured mouse ﬁbroblasts. Our study also reveals
signiﬁcant differences between the mechanism used by canon-
FIG. 7. eIF2␣ phosphorylation inhibits synthesis of IB␣ but does
not destabilize the preexisting protein. (A) Autoradiogram of IB␣
immunoprecipitated from wild-type (EIF2A
ﬁbroblasts following a brief, 10-min [
S]methionine- and cysteine-
labeling pulse and cold chase of the indicated duration. The chase was
conducted in the presence or absence of the activating ligand AP20187.
The IB␣ signal intensity is expressed as a fraction of that present at
the end of the labeling pulse and is depicted beneath each lane. (B)
Same assay as shown in panel A except that the proteasome inhibitor,
MG132, was included during the chase where indicated. (C) Autora-
diogram of the radiolabeled IB␣ present at the end of the 10-min
labeling pulse in wild-type (EIF2A
) or mutant (EIF2A
mouse ﬁbroblasts treated with the indicated concentration of
AP20187 ligand (in nM), cycloheximide (in g/ml), thapsigargin (in M),
or TNF-␣ (in ng/ml) starting 30 min before and continuing throughout
the pulse. (D) Autoradiogram ([
S]methionine) of equal fractions of
the cell lysates used in panel C. The right panel is of a gel that was run
longer than the left panel, accounting for differences in appearance of
the two. (E) Coomassie stain of the gels shown in panel D.
10166 DENG ET AL. M
ical inducers of NF-B and the consequences of eIF2␣ phos-
phorylation. Unlike canonical inducers of NF-B, eIF2␣ phos-
phorylation promoted neither phosphorylation nor degradation
of IB␣. Instead, our data argue that the major impact of eIF2␣
phosphorylation on NF-B activation is inhibition of the syn-
thesis of the labile inhibitor IB␣.
The mechanism uncovered in this study suggests that the
link between eIF2␣ phosphorylation and NF-B activation de-
pends on the lability of the inhibitor, which, in turn, likely
depends on basal levels of signaling through the canonical
pathway(s) that activates NF-B. Indeed, the rapid accumula-
tion of phosphorylated IB␣ in mouse ﬁbroblasts treated with
proteasome inhibitor is consistent with high basal levels of
IB␣ kinase activity in these cells. It is worth noting that both
eIF2␣ phosphorylation and cycloheximide treatment dispro-
portionately reduced the levels of phosphorylated IB␣, com-
pared with their effect on the levels of total IB␣. Inhibited
protein synthesis may attenuate basal activity of an IB␣ ki-
nase and account for some of this effect. Alternatively, newly
synthesized IB␣ might constitute a preferred substrate for its
kinases. The plausibility of the latter explanation is supported
by evidence for the existence of multiple pools of IB␣ in cells
(25, 33, 41). The existence of more than one pool of IB␣
might also explain the discrepancy between the short half-life
of newly synthesized IB␣ (measured by the pulse-chase meth-
od [Fig. 7A and B]) and the much longer half-life inferred from
the gradual decline in total IB␣ protein levels in the cyclo-
heximide-treated and Fv2E-PERK-activated cells (Fig. 4, 5B,
and 6A and B). However, these potential complexities of IB␣
metabolism do not weaken our conclusion that attenuated
synthesis of the inhibitor plays a major role in mediating acti-
vation of NF-B by eIF2␣ phosphorylation in mouse ﬁbro-
Our ﬁndings are at odds with those reported by Jiang and
colleagues, who found no decrease in steady-state IB␣ levels
in thapsigargin-treated cells and instead uncovered evidence
for dissociation of the IB␣–NF-B complex under those con-
ditions (see Fig. 6 in reference 21). We have no explanation for
these differences; however, we do note that since the submis-
sion of the present study Wu and colleagues have reported that
induction of NF-B DNA binding activity in cells exposed to
UV light is also associated with eIF2␣ phosphorylation-depen-
dent repression of IB␣ synthesis (46).
Our study does not address the physiological signiﬁcance of
the link between eIF2␣ phosphorylation and NF-B activation.
It is worth noting that we have but an incomplete understand-
ing of the relative signiﬁcance of regulated protein synthesis
versus activation of gene expression programs as readouts of
eIF2␣ phosphorylation. In yeast it is fairly clear that mutations
in the transcription factor GCN4 phenocopy mutations in the
upstream kinase GCN2 or in the gene encoding its substrate
SUI2 (yeast eIF2␣) (6, 7). In mammalian cells too, some of the
phenotypes of loss of PERK gene function or the EIF2A
genotype are mimicked by mutations in the gene encoding the
downstream transcription factor ATF4 (16, 29, 39). Further-
more, in both yeast and mammalian cells, translation activation
of the transcription factors GCN4 and ATF4 occurs at levels of
eIF2␣ phosphorylation that have only a modest impact on
global protein synthesis (7, 44; Lu et al., unpublished observa-
tion). By contrast, our proposed mechanism of cross talk be-
tween eIF2␣ phosphorylation and NF-B signaling is propor-
tional to the repression of IB␣ translation. Such levels of
repression are easily attained in thapsigargin-treated cells (14)
or in Fv2E-PERK
cells activated by AP20187 (28) and are
clearly sufﬁcient to activate NF-B in cultured mouse ﬁbro-
blasts (Fig. 1A and 2B) (21).
The extent to which translational repression contributes to
NF-B activation in more physiological contexts in which
eIF2␣ kinases are activated is not known. However, we note
that endogenous proinﬂammatory NF-B target genes, such
as those encoding the major histocompatibility complex heavy
and light chains, the interleukin 17 receptor, and a comple-
ment receptor-related protein, were all induced in the Fv2E-
cells by AP20187 treatment and in wild-type mouse
ﬁbroblasts by exposure to tunicamycin (National Center for
Biotechnology Information GEO data set GDS405). The PERK-
dependent induction of NF-B target genes by tunicamycin is
potentially signiﬁcant, as global repression of mRNA transla-
tion is relatively modest under those conditions (14), mimick-
ing physiological stress situations. Furthermore, loss-of-func-
tion mutations in the eIF2␣ kinase PERK or HRI or the
genotype all predispose cells to programmed cell
death under physiologically stressful conditions (10, 13, 14, 39,
48); however, the role of defective activation of NF-B in this
phenotype, if any, remains to be explored.
Translational repression in response to activation of eIF2␣
kinases tends to be transient (34, 35). Translational recovery is
mediated in part by activation of GADD34, an eIF2␣-speciﬁc
regulatory subunit of a holophosphatase complex (30, 31),
which is itself a target of the eIF2␣ phosphorylation-dependent
gene expression program, the integrated stress response (16,
29, 30). GADD34-mediated translational recovery is therefore
likely to reestablish IB␣ translation and reverse the effects of
eIF2␣ phosphorylation on NF-B activity, since the stress re-
sponse is attenuated (Fig. 1B). Furthermore, while activation
of NF-B proceeds through utilization of preformed compo-
nents, the response in terms of target gene expression depends
on new protein synthesis. Thus, the biphasic nature of the
inhibition of protein synthesis, which is inherent to stressful
conditions that promote eIF2␣ phosphorylation, is also pre-
dicted to contribute to the expression of NF-B target genes.
In conclusion, our study indicates that the pathways promot-
ing eIF2␣ phosphorylation and those that activate NF-B in-
teract through translational repression of the inhibitor IB␣.
Our study also suggests that the importance of this link is likely
to be inﬂuenced by signaling through canonical NF-B activa-
tion pathways that deﬁne the turnover rate of IB␣. As such,
eIF2␣ phosphorylation and the consequent inhibition of eIF2B
might modulate NF-B signaling by parallel pathways active in
We thank Yinon Ben Neriah and Haoyuan Jiang for scientiﬁc advice
and the ARIAD Corporation for the gift of the inducible dimerization
This work was supported by NIH grants ES08681 and DK47119
(to D.R.) and DK42394 (to R.J.K.). D.R. is a Scholar of the Ellison
1. Baeuerle, P. A., and D. Baltimore. 1988. I kappa B: a speciﬁc inhibitor of the
NF-kappa B transcription factor. Science 242:540–546.
VOL. 24, 2004 eIF2␣ PHOSPHORYLATION AND NF-B ACTIVATION 10167
2. Bertolotti, A., Y. Zhang, L. Hendershot, H. Harding, and D. Ron. 2000.
Dynamic interaction of BiP and the ER stress transducers in the unfolded
protein response. Nat. Cell Biol. 2:326–332.
3. Chen, J. 2000. Heme-regulated eIF2␣ kinase, p. 529–546. In N. Sonenberg,
J. W. B. Hershey, and M. B. Mathews (ed.), Translational control of gene
expression. CSHL Press, Cold Spring Harbor, N.Y.
4. Cheshire, J. L., B. R. Williams, and A. S. Baldwin, Jr. 1999. Involvement of
double-stranded RNA-activated protein kinase in the synergistic activation
of nuclear factor-kappaB by tumor necrosis factor-alpha and gamma-inter-
feron in preneuronal cells. J. Biol. Chem. 274:4801–4806.
5. Cuervo, A. M., W. Hu, B. Lim, and J. F. Dice. 1998. IkappaB is a substrate
for a selective pathway of lysosomal proteolysis. Mol. Biol. Cell 9:1995–2010.
6. Dever, T. E. 2002. Gene-speciﬁc regulation by general translation factors.
7. Dever, T. E., L. Feng, R. C. Wek, A. M. Cigan, T. F. Donahue, and A. G.
Hinnebusch. 1992. Phosphorylation of initiation factor 2 alpha by protein
kinase GCN2 mediates gene-speciﬁc translational control of GCN4 in yeast.
8. Donze, O., J. Deng, J. Curran, R. Sladek, D. Picard, and N. Sonenberg. 2004.
The protein kinase PKR: a molecular clock that sequentially activates sur-
vival and death programs. EMBO J. 23:564–571.
9. Gil, J., J. Rullas, M. A. Garcia, J. Alcami, and M. Esteban. 2001. The
catalytic activity of dsRNA-dependent protein kinase, PKR, is required for
NF-kappaB activation. Oncogene 20:385–394.
10. Han, A. P., C. Yu, L. Lu, Y. Fujiwara, C. Browne, G. Chin, M. Fleming, P.
Leboulch, S. H. Orkin, and J. J. Chen. 2001. Heme-regulated eIF2alpha
kinase (HRI) is required for translational regulation and survival of ery-
throid precursors in iron deﬁciency. EMBO J. 20:6909–6918.
11. Han, Y., S. Weinman, I. Boldogh, R. K. Walker, and A. R. Brasier. 1999.
Tumor necrosis factor-alpha-inducible IkappaBalpha proteolysis mediated
by cytosolic m-calpain. A mechanism parallel to the ubiquitin-proteasome
pathway for nuclear factor-kappaB activation. J. Biol. Chem. 274:787–794.
12. Harding, H., I. Novoa, Y. Zhang, H. Zeng, R. C. Wek, M. Schapira, and D.
Ron. 2000. Regulated translation initiation controls stress-induced gene ex-
pression in mammalian cells. Mol. Cell 6:1099–1108.
13. Harding, H., H. Zeng, Y. Zhang, R. Jungreis, P. Chung, H. Plesken, D.
Sabatini, and D. Ron. 2001. Diabetes Mellitus and exocrine pancreatic dys-
function in Perk⫺/⫺ mice reveals a role for translational control in survival
of secretory cells. Mol. Cell 7:1153–1163.
14. Harding, H., Y. Zhang, A. Bertolotti, H. Zeng, and D. Ron. 2000. Perk is
essential for translational regulation and cell survival during the unfolded
protein response. Mol. Cell 5:897–904.
15. Harding, H., Y. Zhang, and D. Ron. 1999. Translation and protein folding
are coupled by an endoplasmic reticulum resident kinase. Nature 397:271–
16. Harding, H., Y. Zhang, H. Zeng, I. Novoa, P. Lu, M. Calfon, N. Sadri, C.
Yun, B. Popko, R. Paules, D. Stojdl, J. Bell, T. Hettmann, J. Leiden, and D.
Ron. 2003. An integrated stress response regulates amino acid metabolism
and resistance to oxidative stress. Mol. Cell 11:619–633.
17. Harding, H. P., M. Calfon, F. Urano, I. Novoa, and D. Ron. 2002. Transcrip-
tional and translational control in the mammalian unfolded protein re-
sponse. Annu. Rev. Cell Dev. Biol. 18:575–599.
18. Hershey, J. W. B., and W. C. Merrick. 2000. The pathway and mechanism of
initiation of protein synthesis, p. 33–88. In N. Sonenberg, J. W. B. Hershey,
and M. B. Mathews (ed.), Translational control of gene expression. CSHL
Press, Cold Spring Harbor, N.Y.
19. Hinnebusch, A. 1996. Translational control of GCN4: gene-speciﬁc regula-
tion by phosphorylation of eIF2, p. 199–244. In J. Hershey, M. Mathews, and
N. Sonenberg (ed.), Translational control. Cold Spring Harbor Laboratory
Press, Cold Spring Harbor, N.Y.
20. Hinnebusch, A. G. 2000. Mechanism and regulation of initiator methionyl-
tRNA binding to ribosomes, p. 185–243. In N. Sonenberg, J. W. B. Hershey,
and M. B. Mathews (ed.), Translational control of gene expression. CSHL
Press, Cold Spring Harbor, N.Y.
21. Jiang, H. Y., S. A. Wek, B. C. McGrath, D. Scheuner, R. J. Kaufman, D. R.
Cavener, and R. C. Wek. 2003. Phosphorylation of the alpha subunit of
eukaryotic initiation factor 2 is required for activation of NF-kappaB in
response to diverse cellular stresses. Mol. Cell. Biol. 23:5651–5663.
22. Jousse, C., S. Oyadomari, I. Novoa, P. D. Lu, Y. Zhang, H. P. Harding, and
D. Ron. 2003. Inhibition of a constitutive translation initiation factor 2␣
phosphatase, CReP, promotes survival of stressed cells. J. Cell Biol. 163:
23. Kaufman, R. J. 2000. The double-stranded RNA-activated protein kinase
PKR, p. 503–527. In N. Sonenberg, J. W. B. Hershey, and M. B. Mathews
(ed.), Translational control of gene expression. CSHL Press, Cold Spring
24. Krappmann, D., and C. Scheidereit. 1997. Regulation of NF-kappa B activity
by I kappa B alpha and I kappa B beta stability. Immunobiology 198:3–13.
25. Krappmann, D., F. G. Wulczyn, and C. Scheidereit. 1996. Different mech-
anisms control signal-induced degradation and basal turnover of the NF-
kappaB inhibitor IkappaB alpha in vivo. EMBO J. 15:6716–6726.
26. Kumar, A., J. Haque, J. Lacoste, J. Hiscott, and B. R. Williams. 1994.
Double-stranded RNA-dependent protein kinase activates transcription fac-
tor NF-kappa B by phosphorylating I kappa B. Proc. Natl. Acad. Sci. USA
27. Kumar, A., Y. L. Yang, V. Flati, S. Der, S. Kadereit, A. Deb, J. Haque, L.
Reis, C. Weissmann, and B. R. Williams. 1997. Deﬁcient cytokine signaling
in mouse embryo ﬁbroblasts with a targeted deletion in the PKR gene: role
of IRF-1 and NF-kappaB. EMBO J. 16:406–416.
28. Lu, P. D., C. Jousse, S. J. Marciniak, Y. Zhang, I. Novoa, D. Scheuner, R. J.
Kaufman, D. Ron, and H. P. Harding. 2004. Cytoprotection by pre-emptive
conditional phosphorylation of translation initiation factor 2. EMBO J. 23:
29. Ma, Y., and L. M. Hendershot. 2003. Delineation of a negative feedback
regulatory loop that controls protein translation during endoplasmic reticu-
lum stress. J. Biol. Chem. 278:34864–34873.
30. Novoa, I., H. Zeng, H. Harding, and D. Ron. 2001. Feedback inhibition of the
unfolded protein response by GADD34-mediated dephosphorylation of
eIF2␣. J. Cell Biol. 153:1011–1022.
31. Novoa, I., Y. Zhang, H. Zeng, R. Jungreis, H. P. Harding, and D. Ron. 2003.
Stress-induced gene expression requires programmed recovery from trans-
lational repression. EMBO J. 22:1180–1187.
32. Pahl, H., and P. Baeuerle. 1995. A novel signal transduction pathway from
the endoplasmic reticulum to the nucleus is mediated by the transcription
factor NF-B. EMBO J. 14:2580–2588.
33. Pando, M. P., and I. M. Verma. 2000. Signal-dependent and -independent
degradation of free and NF-kappa B-bound IkappaBalpha. J. Biol. Chem.
34. Prostko, C. R., M. A. Brostrom, and C. O. Brostrom. 1993. Reversible
phosphorylation of eukaryotic initiation factor 2 alpha in response to endo-
plasmic reticular signaling. Mol. Cell. Biochem. 127-128:255–265.
35. Prostko, C. R., M. A. Brostrom, E. M. Malara, and C. O. Brostrom. 1992.
Phosphorylation of eukaryotic initiation factor (eIF) 2 alpha and inhibition
of eIF-2B in GH3 pituitary cells by perturbants of early protein processing
that induce GRP78. J. Biol. Chem. 267:16751–16754.
36. Ron, D., A. R. Brasier, K. A. Wright, J. E. Tate, and J. F. Habener. 1990. An
inducible 50-kilodalton NFkB-like protein and a constitutive protein both
bind the acute-phase response element of the angiotensinogen gene. Mol.
Cell. Biol. 10:1023–1032.
37. Ron, D., and H. Harding. 2000. PERK and translational control by stress in
the endoplasmic reticulum, p. 547–560. In J. Hershey, M. Mathews, and N.
Sonenberg (ed.), Translational control. Cold Spring Harbor Laboratory
Press, Cold Spring Harbor, N.Y.
38. Rothwarf, D. M., and M. Karin. 1999. The NF-kappa B activation pathway:
a paradigm in information transfer from membrane to nucleus. Sci. STKE
39. Scheuner, D., B. Song, E. McEwen, P. Gillespie, T. Saunders, S. Bonner-
Weir, and R. J. Kaufman. 2001. Translational control is required for the
unfolded protein response and in-vivo glucose homeostasis. Mol. Cell 7:
40. Scorsone, K. A., R. Panniers, A. G. Rowlands, and E. C. Henshaw. 1987.
Phosphorylation of eukaryotic initiation factor 2 during physiological stresses
which affect protein synthesis. J. Biol. Chem. 262:14538–14543.
41. Scott, M. L., T. Fujita, H. C. Liou, G. P. Nolan, and D. Baltimore. 1993. The
p65 subunit of NF-kappa B regulates I kappa B by two distinct mechanisms.
Genes Dev. 7:1266–1276.
42. Sen, R., and D. Baltimore. 1986. Inducibility of kappa immunoglobulin
enhancer-binding protein Nf-kappa B by a posttranslational mechanism. Cell
43. Taddeo, B., T. R. Luo, W. Zhang, and B. Roizman. 2003. Activation of
NF-kappaB in cells productively infected with HSV-1 depends on activated
protein kinase R and plays no apparent role in blocking apoptosis. Proc.
Natl. Acad. Sci. USA 100:12408–12413.
44. Tzamarias, D., I. Roussou, and G. Thireos. 1989. Coupling of GCN4 mRNA
translational activation with decreased rates of polypeptide chain initiation.
45. Ubeda, M., X.-Z. Wang, H. Zinszner, I. Wu, J. Habener, and D. Ron. 1996.
Stress-induced binding of transcription factor CHOP to a novel DNA-con-
trol element. Mol. Cell. Biol. 16:1479–1489.
46. Wu, S., M. Tan, Y. Hu, J. L. Wang, D. Scheuner, and R. J. Kaufman. 2004.
Ultraviolet light activates NFB through translational inhibition of IB␣
synthesis. J. Biol. Chem. 279:34898–34902.
47. Zamanian-Daryoush, M., T. H. Mogensen, J. A. DiDonato, and B. R. G.
Williams. 2000. NF-B activation by double-stranded-RNA-activated pro-
tein kinase (PKR) is mediated through NF-B-inducing kinase and IB
kinase. Mol. Cell. Biol. 20:1278–1290.
48. Zhang, P., B. McGrath, S. Li, A. Frank, F. Zambito, J. Reinert, M. Gannon,
K. Ma, K. McNaughton, and D. R. Cavener. 2002. The PERK eukaryotic
initiation factor 2 alpha kinase is required for the development of the skel-
etal system, postnatal growth, and the function and viability of the pancreas.
Mol. Cell. Biol. 22:3864–3874.
10168 DENG ET AL. MOL.CELL.BIOL.