A Two-Dimensional ERK-AKT Signaling Code
for an NGF-Triggered Cell-Fate Decision
Jia-Yun Chen,1Jia-Ren Lin,1Karlene A. Cimprich,1and Tobias Meyer1,*
1Department of Chemical and Systems Biology, Stanford University, Stanford, CA 94305, USA
Growth factors activate Ras, PI3K, and other
signaling pathways. It is not well understood how
these signals are translated by individual cells into
a decision to proliferate or differentiate. Here, using
single-cell image analysis of nerve growth factor
(NGF)-stimulated PC12 cells, we identified a two-
(pAKT) response map with a curved boundary that
separates differentiating from proliferating cells.
The boundary position remained invariant when
different stimuli were used or upstream signaling
components perturbed. We further identified Rasa2
as a negative feedback regulator that links PI3K to
Ras, placing the stochastically distributed pERK-
pAKT signals close to the decision boundary. This
allows for uniform NGF stimuli to create a subpopu-
lation of cells that differentiates with each cycle of
proliferation. Thus, by linking a complex signaling
gain unique integration and control capabilities to
balance cell number expansion with differentiation.
Growth factor stimuli can induce different cell fates by activating
Ras, PI3K, Src, PLCg, and other signaling pathways (Lemmon
and Schlessinger, 2010). It is not well understood how cells inte-
grate such complex signaling responses to make all-or-none
cell-fate decisions. One hypothesis is that cells use multiple
pathways to better monitor the presence of neighboring cells,
growth factors, hormones, nutrient availability, and intracellular
stress. These pathways may then get integrated at specific
signaling steps that function as ‘‘bottlenecks’’ or ‘‘hubs’’ (Albert,
2005; Baraba ´si and Oltvai, 2004). In turn, multiple downstream
targets may link such an integration point to a cell fate. It is often
implicitly assumed in pharmacological or genetic studies that
signaling or transcriptional networks have such an hourglass or
hub organization with a single intermediate integration point
where a key decision is made (Friedman and Perrimon, 2007).
We investigated whether and how such signaling hubs con-
tribute to cell-fate decisions by focusing on the PI3K and Ras
pathways. These pathways are likely particularly important given
and Leo ´n, 2000; Katso et al., 2001; Okkenhaug and Vanhaese-
broeck, 2003). We chose PC12 cells as a model system since
nerve growth factor (NGF) activates both pathways and triggers
a decision between proliferation and differentiation into sympa-
thetic-like neuronal cells (Greene and Tischler, 1976). We also
of differentiation and transfectability that was difficult to match
using differentiation-proliferation models in an in vivo setting.
This offered the opportunity to ask systematic and quantitative
questions about signaling processes at the single-cell level.
We used automated imaging and single-cell image analysis to
compare the NGF-induced cell fate to the activation of the multi-
functional protein kinases ERK and AKT, important downstream
targets of Ras and PI3K signaling (Chambard et al., 2007;
Manning and Cantley, 2007). This led to the unexpected finding
that a two-dimensional (2D) pERK-pAKT response map with a
curved boundary separates regions with proliferation and differ-
entiation cell fates. The same NGF stimulus caused significant
cell-to-cell variation of pERK and pAKT signals, placing cells
onbothsidesofthe boundary,producing proliferating anddiffer-
entiating subpopulations. Furthermore, the boundary position
remained invariant when we used EGF, NGF, or serum to stimu-
late cells or when we used small molecule inhibitors or siRNA
knockdown to perturb upstream regulators. Finally, using a
targeted small interfering RNA (siRNA) screen, we identified
Rasa2 as a regulator that places the distributed pERK-pAKT
signals close to the boundary. We show that Rasa2 is a late
NGF-induced PI3K-regulated RasGAP that connects PI3K to
Ras signaling by negative feedback. Together, our study shows
that cell-fate decisions can be encoded by signaling response
maps that function as intermediate integration and decision
points. Such a response map provides mechanistic insights
how identical populations of cells are split into subpopulations
with different cell fates and how the number of differentiating
cells can be regulated within a uniform population.
A Two-Dimensional pERK-pAKT Response Map
Previous studies with PC12 cells have shown that NGF stimula-
tion of the TrkA receptor activates Ras, PI3K, and a number of
other signaling pathways to trigger neuronal differentiation
(Huang and Reichardt, 2003) (Figure 1A). The transition from
196 Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc.
Figure 1. Identification of a Two-Dimensional pERK and pAKT Signaling Response Map
(A) Schematics of growth factors (GFs) induced receptor signaling. RTKs, receptor tyrosine kinases. Inhibitors used in the study are marked in red.
(B) Automated image analysis of differentiation and proliferation after 24 hr of NGF treatment. Representative images used for the analysis are shown. Left:
Detected neurites (white) were superimposed over a merged tubulin and BrdU-stained image. Right: Overlay of BrdU and DNA-stained image. The scale bar
represents 40 mm.
(C) Time courses of differentiation and proliferation after NGF stimulation (mean ± SD of triplicate wells).
(D) Automated image analysis monitors pAKT, pERK and proliferation after 24 hr of NGF treatment.
(E) Single-cell analysis of pERK level versus the fraction of cells in S phase shows only little correlation. The percent of cells in S phase (%S) was calculated for
equally spaced bins of the ERK activity (top, mean ± 95% bootstrap confidence interval) or from the bottom and top 10 percentile of the ERK activity (bottom,
mean ± SD of five replicate wells) after NGF stimulation for 24 hr.
(F) Heat-map analysis of pERK-pAKT signaling and proliferation shows a clear boundary between proliferation and differentiation regions. Contour plots of cell
density are shown in the lower panels. The %S was calculated for equally spaced bins of the ERK and AKT activity and is marked in a color code. Cells were left
boundary (green line) was drawn across the black colored bins on the NGF heat map and overlaid on top of other plots. Each panel contains ?40,000 cells. Note
that due to day-to-day staining and imaging variations, the boundary position compares experiments done at the same time.
Cell Fate-Signaling Code
Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc. 197
a proliferative to a differentiated state occurs in most PC12 cells
in the population within the first 24 hr of NGF stimulation (Fig-
ure 1B). This switch can be tracked by the appearance of
a neuron-like morphology (quantified by average neurite length,
Figure 1C, top) that is paralleled by the reduction of cells in
S phase (monitored by BrdU incorporation, Figure 1C, bottom).
We measured the activity of the PI3K and Ras pathways by
monitoring the phosphorylation status of their downstream
targets, AKT and ERK, respectively (Figure 1A). Using immuno-
fluorescence and automated image analysis, we quantified the
levels of pERK and pAKT and the proliferation status in single
cells (Figure 1D). Given the previous evidence that sustained
pERK signals are important for the differentiation process
(Marshall, 1995), we first tested whether the level of pERK in
an individual cell correlates with the proliferation status when
both are measured 24 hr after stimulation. Figure 1E shows
that the correlation between pERK and proliferation is statisti-
cally significant but quite smalleven whenwe selected the prolif-
eration status from the bottom and top 10 percentile of pERK
intensity for comparison.
Itwas striking that the proliferative state of each cell was much
better defined by its location in a 2D pERK-pAKT plane (Fig-
ure 1F, mock and NGF stimulus). This panel shows a probability
map for proliferation of individual cells with different pERK and
pAKT levels. The axes in this plot have a log base 2, and the
proliferative state is represented by a color-coded heat map
(from 0% to 80%). The region where cells have a high probability
of proliferation is characterized by higher pAKT and lower pERK
levels, with a curved boundary separating the proliferating from
nonproliferating cells (Figure 1F).
Not only was the boundary between the two regions quite
steep, it also remained at the same location when cells were
stimulated with epidermal growth factor (EGF), NGF, low or
high serum, or a combination with a low-dose of MEK or PI3K
inhibitors (Figure 1F). The existence of an invariant boundary
between the two regions implied that each point in the pERK-
pAKT response map can be reduced to a single parameter that
reflects the distance fromthe boundary and predicts the prolifer-
NGF-Induced Signal Variation Spreads Cells across
a Sharp Boundary in the pERK-pAKT Signaling Plane
We noted that the same NGF stimulus induced a great variation
in the pERK and pAKT signals even though we activated a
homogenous cell population (Figure 2A, top and right sub-
panels). The pERK and pAKT signal distribution shows a 4.2-
and 3.1-fold difference between the bottom and top 5 percen-
tile, respectively. This wide distribution allows cells to be spread
over the narrow region boundary. We quantified the steepness
of the proliferation probability change across the boundary
(Figure 2A; green band orthogonal to the boundary) by projec-
ting the percent of cells in S phase (%S) parameter onto the
pERK and pAKT axis (Figure 2A, green curves in the top and
right subpanels). Consistent with the visual impression, the
boundary between the regions was very sharp with an approxi-
mately 30-fold increase of proliferating cells, from 2% to
60%, over only a 2-fold difference in the change in pERK or
When we extended the same analysis to cells treated with
different stimuli, we found that the transition from proliferative
to nonproliferative state was equally steep and indeed occurred
at the same site in the pERK-pAKT plane independent of the
stimulus and the position of the center of the pERK-pAKT distri-
bution (Figure 2B). In addition, when we compared the top and
bottom 10 percentile of cells farthest away from the boundary
with the cell fate was much higher (Figure 2C).
Finally, we determined whether proliferation arrest and differ-
entiation are indeed closely linked by performing a single-cell
analysis of neurite outgrowth as a marker of the differentiated
state. The plot in Figure 2D (bottom) shows that individual cells
in a region below and to the right of the proliferation region
have increased neuronal morphology. In this analysis, a third
region in the plane can also be distinguished with cells that
have low pERK and pAKT level that neither proliferate nor differ-
entiate. This region likely reflects a quiescent G0/G1-state of the
Figure 2E depicts the resulting working model that different
regions in the pERK-pAKT plane are highly predictive as to
state. The usefulness of this response map analysis is particu-
larly apparent when comparing an EGF stimulus, which primarily
promotes proliferation and has an activation vector with higher
relative pAKT, to the NGF stimulus, which promotes mostly
differentiation and has a vector with higher relative pERK (Fig-
ure 1F). As an added note, since the boundary between the
of the respective pERK and pAKT activities only partially corre-
late with the decision, implying that a 2D response map is
needed for an optimal prediction of cell-fate outcome.
An siRNA Screen Identifies Regulators of Cell-Fate
and pERK-pAKT Signals
To better understand how cells generate the pERK-pAKT
relevant molecular signaling components that we could use for
subsequent perturbation studies. We made a rat siRNA library
targeting 1308 signaling proteins using an in vitro Dicer cleavage
strategy (Myers et al., 2003) and performed a screen as outlined
in Figure 3A with automated analysis of neurite extension
and proliferation (Figure 1B). The 54 genes listed in Table S1
(available online) were hits that were validated by two sets of
siRNAs against different regions of the messenger RNA (mRNA)
with both siRNAs changing the cell-fate outcome in the same
direction (Figure S1). When surveying the identified proteins for
annotated processes, we found important regulators as well as
known NGF signaling components among the hits (Table S1).
The identified signaling regulators included for example the little
characterized proteins Arf5 and Tao1 kinase which, when
knocked down, greatly reduced or increased the differentiation
state, respectively (Figure 3B).
Shotgun Perturbation Analysis of the pERK-pAKT
Response Map and Cell Fate
Since the 54 identified genes came from a relatively unbiased
screen of signaling proteins, we used them as a signaling
Cell Fate-Signaling Code
198 Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc.
network perturbation tool. By comparing the effect of the
different siRNAs on proliferation versus neurite growth, we con-
firmed that the coupling between differentiation and suppres-
sion of proliferation is indeed very close. Most siRNAs shifted
proliferation and differentiation responses in opposite directions
Figure 2. A Sharp Boundary in the pERK-pAKT Response Map Separates Proliferating from Differentiating Cells
(A)Quantitative analysisofNGF-triggered cell-to-cellsignalvariationandproliferationprobabilitiesinthepERK-pAKTplane. Thepopulationdistributions ofpERK
and pAKT are shown in thesubpaneltop and right (gray histograms).The same histogram includesagraph (green curves) of the %S calculated from cells located
in the green band (orthogonal to the boundary shown in Figure 1F).
(B) Evidence of an invariant 2D signaling response map that determines proliferative cell fate. Proliferation changes were analyzed as shown in (A) from cells
treated with different stimuli. The analysis only included cells located within the green band. In (A) and (B), data are mean ± 95% bootstrap confidence interval.
(C) The proliferative status is better predicted by the 2D response map compared to pERK level shown in Figure 1E (bottom). The %S was compared for the
10 percentile of cells farthest above (Low) and below (High) the boundary. Inset shows the schematic diagram of the analysis region. Data are shown as the
mean ± SD of five replicate wells.
(D) Heat-map analysis showing proliferation (top) and differentiation (bottom) as a function of ERK and AKT activity at a single-cell level after 24 hr of NGF
stimulation. %S was quantified as shown in Figure 1F. Quantification of the integrated single-cell neurite parameter was achieved by measuring the presence of
neuritesproximal tothecellbodyofeachcelland calculating meanneuriteintensityfor eachcellasafunctionofpERKandpAKTlevels. Eachbincontainsatleast
(E) Different directions and amplitudes of pERK-pAKT activity vectors correlate with cell fates. The schematic also shows a quiescent state for low pERK and
pAKTlevels.EGF andNGFnotonlytriggerdifferentamplitudesofsignalactivationbutalsohavedifferentdirectionsofpERK-pAKT activityvector inthe2Dplane.
Cell Fate-Signaling Code
Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc. 199
Figure 3. siRNA Perturbation Analysis Validates the Use of the Signaling Response Map to Predict Cell Fate
(A) Protocol used to screen for siRNAs that change the fraction of proliferating and differentiating cells.
(B) Knockdown images of selected genes identified in the siRNA screen of regulators of NGF-induced differentiation. Arf5 reduces differentiation and Tao1K
increases differentiation. The scale bar represents 40 mm.
(C) Perturbation analysis with 54 siRNAs showing the correlation between proliferation and the induction of differentiation (data were from duplicate wells; robust
z score, the median absolute deviation from the control median).
(D) NGF signaling scheme and the corresponding secondary assays used to link different signaling processes to differentiation.
(E) Perturbation parameter cross-correlation analysis showing that the 24 hr pERK is the most predictive parameter for neurite extension (Nrt) and proliferation
(%S), both measured at 48 hr. All 54 siRNAs were used for the analysis. ERK50, ERK1h, and ERK24h denote measurements of pERK at 5 min, 1 hr, and 24 hr after
Color bar represents the cross-correlation values (Pearson’s correlation coefficients).
Cell Fate-Signaling Code
200 Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc.
We next determined whether siRNA-mediated signaling
can predict the changes in the differentiation or proliferation
outcomes 48 hr after NGF stimulation. The experiments shown
before in Figures 1 and 2 measured signaling and cell fate at
24 hr. We measured early pERK signaling at 5 min and at 1 hr,
and the induction of an early response gene EGR1 at 1 hr after
NGF stimulation. We also measured the late ERK and AKT
signaling at 24 hr (Figure 3D and Table S2). We then performed
a correlation analysis between the changes induced by the 54
siRNAs (Figure 3E). Interestingly, the siRNA-mediated changes
in pERK signaling at 5 min had no significant predictive value
either on the change of proliferation or differentiation at 48 hr
(Figures 3E and 3F, left). Similarly, the changes in pERK level
and EGR1 expression at 1 hr showed a higher but still small
correlation with the later cell fates. In contrast, the siRNA-medi-
ated changes of p-ERK signals 24 hr after stimulation provided
a significantly higher predictive value for the neuronal differenti-
ation and proliferation status at 48 hr (Figure 3E and 3F, right).
This is consistent with the interpretation derived from Figures 1
and 2 that sustained pERK and pAKT signals control cell fate.
This further implied that short-term signals can be altered
without changing the cell-fate outcome as long as sustained
signals are not affected.
The weaker but significant correlation of proliferation and
differentiation with changes of pAKT at 24 hr showed that most
of the siRNA impacted both pERK and pAKT signals in parallel,
keeping the center of the respective population distributions
close to the boundary. This can be seen more directly in a plot
where the centers of the relative population distributions are
shown for cells treated with different siRNAs (Figure 3G). Two
significant outliers are PTEN and Rasa2, which will be investi-
gated in more detail below.
We hypothesized that siRNA perturbations can alter the 2D
response map in at least two ways. One way is to move the
pERK-pAKT signal distribution while leaving the boundary intact
as we found before in Figures 1 and 2. This can be represented
as an orthogonal shift of the center of the population distribution
relative to the boundary of control cells (y axis in Figure 3H).
Another way would be to shift the decision boundary in the
pERK-pAKT response map without altering the population
distribution. We measured such a potential shift of the boundary
position by using a comparison of S phase probability values
between the siRNA knockdown cells and control cells (x axis in
Figure 3H; Supplemental Experimental Procedures). We antici-
pated that knockdown of signaling steps downstream of
pERK-pAKT might be of this second type. Markedly, among
different siRNAs tested, the positive cell-cycle regulators CDKs
(CDK1, CDK2, CDK4, and CDK6), MDM4, and cyclin D1/D3
(Morgan, 2007) shifted the boundary toward the proliferating
region (resulting in less proliferating cells) while the negative
away from it (Figure 3H), suggesting that they are indeed of this
second type. For comparison, the different receptor stimuli as
well as inhibition of PI3K and MEK belonged to the first type of
perturbation that shifted the population distributions along the
y axis. In this group, PTEN and Rasa2 siRNAs stood out again,
tion we examined further below.
Together, these siRNA perturbation studies showed that
proliferation and differentiation aremutuallyexclusive (Figure3E)
and that sustained pERK-pAKT signals are critical for the cell-
fate decision (Figures 1, 2, and 3F). Furthermore, downstream
cell-cycle effectors shift the decision boundary in the pERK-
pAKT response map (Figure 3H).
The Cyclin D-Mediated Shift of the Decision Boundary
Involves pERK and pAKT Regulation of Protein Stability
We were particularly interested in understanding the effect of the
combined cyclin D1/D3 knockdown since it strongly reduced
proliferation (Figure 4A) and caused the farthest shift of the
boundary among the siRNAs (Figures 3H and 4B). The shape
of the shifted boundary is quantified in Figure 4C. The knock-
down of individual isoforms had only small effects (Figure 4A),
suggesting that cyclin D1 and D3 have redundant functions.
Control experiments with the respective protein knockdowns
are shown in Figure S2A.
Cyclin D is a key cell cycle as well as differentiation regulator
(Sherr and Roberts, 2004) whose concentration is known to be
controlled by growth factor inputs acting through the Ras and
PI3K pathways (Diehl et al., 1998; Shao et al., 2000). This
made us examine how the position in the pERK- pAKT response
map is translated into a change in cyclin D1 concentration.
Inhibition of MEK increased cyclin D1 protein level whereas inhi-
bition of PI3K or AKT decreased cyclin D1 (Figures S2B and S2C
and Figures 4D and 4E), a finding that was also confirmed by
combined knockdowns of the isoforms AKT1-3 (targets of
PI3K and PIP3) or ERK1-2 (targets of MEK) (Figure 4F). When
cells were treated with proteasome inhibitor (MG132), the differ-
protein level were lost (Figure 4G and Figure S2C). Furthermore,
the opposing ERK-AKT regulation on cyclin D1 levels remained
intact when protein translation was inhibited with cycloheximide
(CHX) (Figure S2D). Together, this shows that the control of cy-
clin D1 stability is a rate-limiting regulatory step that translates
a decision made at the pERK-pAKT integration point into cell
(F) Direct correlation analysis comparing the short-term (5 min, left) and long-term ERK (24 hr, right) signaling with differentiation (neurite length) for all 54 siRNAs.
(C and F) Green lines are linear fits and R represents Pearson’s correlation coefficients.
(G) Center of population distribution for all 54 siRNAs in the pERK-pAKT plane. Each dot represents the population median of the pERK and pAKT intensity after
individual siRNA knockdown. Green line, region boundary that crosses 20% of S phase probability based on control knockdown (red circle).
(H) Separation of genes that shift the boundary (x axis) or move the population center away from the boundary of controls (y axis) upon siRNA knockdown. The
boundary shift was calculated by the sum of %S differences between specific siRNA knockdown and control per pERK-pAKT bin in the 2D plane. Movement of
population relative to theboundary wasrepresentedas theorthogonaldistance (Log2unit) fromthe centerof population distribution totheboundary (as shown in
Figure 3G). Knockdowns of cell-cycle regulators (red circle) that shifted the boundary to the negative side include cyclin D1/D3, Cdk1, Cdk2, Cdk4, Cdk6 and
Mdm4. The positive side contains the tumor suppressors Rb1, p21, and p16. SD represents the standard deviation calculated from all the siRNAs.
See also Figure S1 and Tables S2 and S3.
Cell Fate-Signaling Code
Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc. 201
Figure 4. The Cell-Fate Decision Is in Part Mediated by pERK and pAKT-Control of Cyclin D1 Protein Stability
(A) Quantitative analysis of the effect of cyclin D1/D3 single and coknockdown on proliferation. siRNA-treated cells were stimulated with NGF for 24 hr before
analysis (mean ± SD of triplicate wells).
(B) Heat-map analysis of the cyclinD1/D3knockdown effect on pERK-pAKT signaling and proliferation.The knockdown (right) shifted the boundary to thetop-left
between the differentiation and proliferation regions without significantly changing the pERK and pAKT distribution itself. Assays were performed as described in
Figures 1D and 1F.
(C) Evidence of the boundary shift with cyclin D1/D3 coknockdown. Proliferation changes were calculated from cells located in the region orthogonal to the
boundary as shown in Figure 2A (mean ± 95% bootstrap confidence interval).
(D) Time courses of the effects of PI3K (LY294002) and MEK inhibition (U0126) on cyclin D1 protein levels. LY294002 (12.5 mM) or U0126 (10 mM) was added at
24 hr after NGF stimulation for different lengths of time as indicated before immunostaining. Cyclin D1 levels were measured by automated image analysis
(mean ± SD of triplicate wells).
(E) Dose effects of U0126 and LY294002 on cyclin D1 protein level changes. Cells were treated with increasing doses of U0126 or LY294002 together with
NGF for 4 hr.
(F) Knockdown of AKT or ERK mimics the LY294002 and U0126 drug effects on cyclin D1 protein level changes. Knockdown cells were subjected to 24 hr of
NGF stimulation before analysis.
(G) The opposing regulation of cyclin D1 protein level by LY294002 and U0126 is proteasome-dependent. Cells were stimulated with NGF for 4 hr with the drug
combination as indicated. MG132 was used at 50 mM. LY294002 and U0126 were used at 12.5 mM and 10 mM, respectively.
(H) Schematics of signaling diagram showing cyclin D1 as one of the downstream mediators linking the pERK-pAKT response map to cell fates.
See also Figure S2.
Cell Fate-Signaling Code
202 Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc.
fates (Figure 4H). Thus, the observed shift of the decision
boundary in cyclin D knockdown cells (Figure 4B) reflects a
requirement for higher pAKT and lower pERK signals to increase
cyclin D protein stability and thereby make up for the reduced
cyclin D translation to restore a proliferating subpopulation.
Evidence that Rasa2 Mediates a Negative Feedback
from PI3K to Ras Signaling
When we first generated the pERK-pAKT response map in Fig-
of cells continues to proliferate upon NGF stimulation. We con-
sidered that maintaining such a subpopulation enables more
cells to differentiate over longer time periods. An effective way
to control the size of the proliferating subpopulation would be
to shift the activation vector orthogonal to the boundary, toward
the top left or bottom right. A candidate orthogonal regulator
was the lipid second messenger PIP3 that is generated by acti-
vation of PI3K. This hypothesis was based on the observation
that PTEN knockdown or PI3K inhibition both shifted the
pERK-pAKT vector along this orthogonal line but in opposite
directions (Figures 5A, 5B, 3G, 3H, and 1F). We found that cells
with knocked down PTEN, a PIP3 lipid phosphatase that lowers
PIP3 levels, not only enhances AKT signaling (Carracedo and
Pandolfi, 2008) but also lowers the average pERK signal
response (Figure 5B). Similarly, PI3K inhibition reduces PIP3
with a concomitant increase in pERK signals (Figures 1F and
5B). This explains how PIP3 changes can create an orthogonal
shift of the pERK-pAKT response compared to the shift medi-
ated by knockdown of the NGF receptor TrkA. This raised the
question of how PIP3 reduces ERK phosphorylation.
Our siRNA screen identified Rasa2 as a strong enhancer of
proliferation and suppressor of differentiation (Figures 3G, 3H,
5C, and 5D and Figure S3A). Rasa2 has a RasGAP domain
specific for Ras (Maekawa et al., 1994) that could explain the
observed effect on ERK signaling. It further provided a potential
link from PIP3 to pERK since PIP3 has been shown to bind to its
PH-domain and recruit Rasa2 to the plasma membrane (PM)
(Lockyer et al., 1999). Nevertheless, these studies did not deter-
mine whether its RasGAP activity is regulated by PIP3.
neurite growth and decreased proliferation. Furthermore, it
shifted the pERK-pAKT activity vector to the right as expected
for a knockdown of a RasGAP (Figure 5E). The specificity and
effectiveness of Rasa2 siRNA knockdowns were confirmed by
and S3C). We further confirmed that Rasa2 knockdown
enhanced the level of GTP-bound Ras (Figure S3D). Moreover,
neurite extension induced by a constitutively active (CA) Ras
(Bar-Sagi and Feramisco, 1985) could not be further enhanced
by Rasa2 knockdown while control knockdown of the down-
stream ERK could (Figure S3E). Thus, Rasa2 functions as a
RasGAP and the observed increase in pERK and differentiation
following Rasa2 knockdown is due to increased Ras activity.
We then confirmed that the PM localization of endogenous
Rasa2 and the expressed YFP-Rasa2 are both dependent on
PI3K activity (Figures 5F and 5G, top, and Figure S3F). To deter-
mine whether PIP3 regulates Rasa2 activity, we coexpressed
a CFP-tagged Ras binding RBD domain, a biosensor for active
Ras (Chiu et al., 2002), together with YFP-Rasa2. The Ras-GTP
biosensorshowedinitially onlyaweakPM localization andtrans-
located more strongly to the PM after PI3K inhibition, demon-
strating that the PIP3-mediated PM translocation of Rasa2
activates its RasGAP activity and that the dissociation of
Rasa2 from the PM rapidly increases the concentration of Ras-
GTP in the PM (Figure 5G, bottom). Furthermore, the strength
of the initial CFP-RBD translocation was markedly enhanced
when we expressed instead a mutant Rasa2 defective in PIP3
binding (R629C) (Lockyer et al., 1999), arguing that PIP3 binding
is critical for its GAP activity (Figures S3I–S3K).
The same result was obtained using a biochemical analysis of
itor led to a progressive decrease in pAKT that was paralleled by
a marked increase in Ras and ERK activity (Figures 5H and 5I).
Notably, this dose-dependent pERK increase was not observed
in Rasa2 knockdown cells, demonstrating that Rasa2 provides
the main link from PI3K to the suppression of ERK signaling (Fig-
ure 5I).Taken together, thisshowsthatPIP3 inducesPM translo-
cation and activation of Rasa2 creates a negative feedback by
which PI3K activation suppresses Ras and ERK signaling.
Rasa2 and TrkA Expression Is Upregulated by NGF
We then examined when Rasa2 acts to regulate cell fate. We
observed two waves of Ras activation following NGF stimulation
(Figure 6A). The first wave happened immediately after NGF
stimulation (2–5 min) and the second wave occurred 7–24 hr
later, coinciding with the onset of differentiation (Figure 1C).
Knockdown of Rasa2 increased the second wave of both Ras
and pERK signaling but had only a small effect on the initial
pERK activity peak (Figures 6B and 6C). We were able to explain
this delayed role of Rasa2 by the finding that Rasa2 expression
significantly increased during thesecond waveof ERK activation
(Figure 6D), arguing that Rasa2 is important during the time
window when the cell-fate decision is made but plays a minor
role in regulating short term Ras and ERK signaling. Neverthe-
less, these findings left unanswered what creates the second
peak in Ras, pERK and pAKT activities (Figures 6A–6D).
Differentiated neurons have been shown to have increased
expression of Trk family growth factor receptors (Deppmann
et al., 2008; Zhou et al., 1995). We considered that such an up-
regulation of TrkA may contribute to the delayed increase in
ERK activity. We show that TrkA expression is indeed induced
by NGF in PC12 cells with a time course that paralleled the
increase in Ras, pERK and pAKT activities as well as Rasa2
expression (Figure 6D). We would like to note that both, the
pERK and the pAKT increase have the same kinetics, consis-
tent with the interpretation that the second activation peak is
mediated by the induced expression of TrkA. Nevertheless,
since Rasa2 is upregulated in parallel with TrkA (Figure 6D),
the amplitude of the second activation peak of pERK results
from a competition between TrkA upregulation, which amplifies
NGF-triggered PI3K and Ras signaling, and Rasa2 upregula-
tion, which suppresses Ras/ERK activities. We further showed
that both Rasa2 and TrkA expression are at least partially
dependent on MEK activity since inhibition of MEK reduced
Cell Fate-Signaling Code
Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc. 203
Figure 5. Rasa2 Increases the Number of Proliferating Cells after NGFStimulation by Addinga Negative Feedback fromPI3K to Ras and ERK
(A) Heat-map analysis of PTEN and TrkA siRNA effects on pERK-pAKT signaling and proliferation. Assays were performed as described in Figure 1F.
(B) Changes in PIP3 levels cause a shift of the activation vector orthogonal to NGF activation. Data from Figures 1F and 5A were normalized to their respective
control and plotted together with robust z score units. The large ovals represent the population distributions and the small filled circles represent the centroids of
(C) Domain structure of Rasa2.
(D) Quantitative analysis of the effect of Rasa2 knockdown on reducing proliferation and increasing differentiation (mean ± SD of triplicate wells).
(E) Heat-map analysis of the Rasa2 siRNA-mediated shift of the population distribution toward higher pERK levels. Assays were performed as described in
Figure 1F. The boundary was drawn according to control cells.
(F) Membrane localization of endogenousRasa2. Cells after 24h of NGF stimulation were left untreated (left) or treated with PI3K inhibitor (LY294002 at 25 mM) for
5 min before subjected to Rasa2 antibody staining. The scale bar represents 10 mm.
Cell Fate-Signaling Code
204 Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc.
their expression (Figure 6E). Thus, Rasa2 functions as a nega-
tive feedback regulator that begins to lower Ras activity a few
hours after NGF stimulation when Rasa2 expression increases
(Figures 6F and 6G). As an added note, given the unimodal
population distributions shown in Figure 2A, the positive feed-
back resulting from NGF upregulating its own TrkA receptor
is primarily a mechanism to amplify the long term NGF signaling
response rather than creating a bistable switch for ERK and
Role of Rasa2 in Expanding the Number of Cells during
The feedback mediated by Rasa2 changes the direction of the
pERK-pAKT activity vector by reducing pERK signals as PI3K
(G) Time series of images showingYFP-Rasa2 (top) and CFP-RBD (bottom)translocation after NGF stimulation and subsequently, afterPI3K inhibitor (LY294002)
addition. Cells were cotransfected with YFP-Rasa2, CFP-RBD (Raf), and H-Ras. CFP and YFP confocal images were taken from the same representative cell.
NGF and LY (100 mM) were added as indicated. Ras activity was monitored using the relative plasma membrane translocation of CFP-RBD. The scale bar
represents 5 mm.
(H) Ras pull-down followed by western blotting showing that inhibition of PI3K is paralleled by an increase of GTP-bound Ras level. Cells were treated with
increasing dose of PI3K inhibitor (0 to 6.25 mM, 2-fold dilution from the right) for 15 min at 24 hr after NGF stimulation.
(I) Analysis of control and Rasa2 siRNA effect on ERK activity changes in response to PI3K inhibition. Cells were treated as described in (H) and assayed by
See also Figure S3.
Figure 6. NGF-Triggered Expression of Rasa2 and TrkA Directs the pERK-pAKT Activation Vector Close to the Boundary
(A) NGF stimulation triggers two waves of Ras activation.
(B) Knockdown of Rasa2 enhances Ras and pERK activities during the second wave.
(A and B) Ras pull-down assays were performed at the indicated time and assayed by western blotting.
(C) Knockdown of Rasa2 selectively enhances a second wave of pERK activation with little effect on the first peak (mean ± SD of triplicate wells).
(D) Time-course analysis of Rasa2 and TrkA expression compared to pERK and pAKT activation. Cells were assayed by western blotting. HSP90 was shown as
a protein loading control.
(E) TrkA and Rasa2 upregulation is partially dependent on MEK signaling. U0126 was used at 10 mM.
(F) Schematic representation of the feedback between PI3K, Rasa2, and Ras.
(G) Schematic model of the roles of the positive TrkA expression feedback, which increases the amplitude of the activation vector, and the negative Rasa2
expression feedback that turns the activation vector closer to the proliferation boundary.
Cell Fate-Signaling Code
Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc. 205
signaling increases. This leaves a significantly larger fraction of
cells in the proliferation region by forcing the activation vector
to turn and stay closer to the boundary as TrkA expression
increases during the 7–24 hr time window (Figure 6G). We
hypothesized that cells benefit from having the center of the
pERK-pAKT vector close to the boundary by maintaining a
balance between cell number expansion (proliferation) and
Indeed, wefound thatRasa2 has arole in maintaining apool of
proliferating cells when we monitored the fraction of proliferating
cells over a 60 hr period after NGF stimulation. Rather than
observing a near complete drop in the number of proliferating
cells by 36 hr as observed in Rasa2 knockdown cells, control
cells maintained a fraction of proliferating cells for more than
60 hr (Figure 7A). The continued proliferation comes at a small
cost since cells that upregulate Rasa2 expression take longer
to differentiate (Figure 7B). However, the continued proliferation
provides a benefit since a larger number of differentiated cells
are generated after this period in control cells compared to
Figure 7. Function of the pERK-pAKT Response
Map in Balancing Cell Number Expansion and
(A and B) Time-course analysis of proliferation (A) and
neurite extension (B) in control and Rasa2 knockdown
cells after NGF stimulation (mean ± SD of four replicate
wells). In (A), subpopulations of control cells stay prolifer-
ative over a period of 60 hr, whereas Rasa2 knockdown
cells cease to proliferate after 48 hr of NGF stimulation.
(C) Quantification of Rasa2 knockdown effect on cell
number expansion after NGF stimulation. Cells trans-
fected with control or Rasa2 siRNA were treated with
Mock or NGF for 3 days before counting of cell number
(mean ± SD of four replicate wells).
(D) Landscape scheme of the 2D pERK-pAKT response
map emphasizes the boundary between the two regions
that predict the proliferation and differentiation outcomes.
The purple circle depicts the variation of the NGF-induced
signaling response that spreads the population of cells
across the boundary. The white dashed arrow reflects the
NGF-induced shift of the activation vector and the black
solid arrow depicts the path to differentiation.
(E) Schematic showing how Rasa2 maintains a balance
between cell number expansion and differentiation.
Rasa2 knockdown cells (Figure 7C). Thus, by
positioning the population near the proliferation
competent precursor cells to create more differ-
Significance of a pERK-pAKT Signaling
When we initiated our studies, we considered
that the level of sustained ERK activation alone
might predict the decision between differentia-
tion andproliferation (Marshall, 1995).Our study
showed instead that pERK and pAKT are both critical interme-
diate signaling steps and single-cell measurements are needed
to reveal the relationship between signaling and cell fate. We
found that the position of the activation vector relative to the
boundary in the pERK-pAKT response map determines whether
a particular cell differentiates or proliferates. This boundary idea
can be applied to the probability response map because the
transition from non-proliferating to proliferating cells is steep
(Figures 2A and 2B). Our model of a response map that defines
the paths to differentiation or proliferation is schematically
shown in a landscape representation in Figure 7D. This concept
shares some similarity to the idea that complex systems use
hubs to process information (Albert, 2005; Baraba ´si and Oltvai,
Using different stimuli, small molecule inhibitors, and siRNA
knockdown of signaling proteins, we found that the curved
boundary between the regions was independent of the cellular
signaling processes that activate cells (Figures 1F and 2B),
arguing for a separation of upstream and downstream
Cell Fate-Signaling Code
206 Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc.
components from pERK and pAKT signals. We showed that the
decision of a cell in the pERK-pAKT response map is in part
translated into a cell fate by regulation of cyclin D1 stability.
Knockdown of cyclin D or other downstream cell-cycle regula-
tors shifted the decision boundary. While cyclin D1 is well known
to be induced by mitogens (Sherr and Roberts, 2004) and sepa-
rate studies have shown that its stability can be regulated by Ras
(Shao et al., 2000) or PI3K signaling (Diehl et al., 1998), our study
provides evidence that both pathways regulate cyclin D1 in
concert with opposing positive and negative regulation by
pAKT and pERK, respectively. Given that AKT and ERK also
regulate other cell cycle as well as differentiation-related
proteins (Chambard et al., 2007; Manning and Cantley, 2007;
among a broader set of effectors that cooperate to translate the
pERK-pAKT decision into a cell fate.
The existence of a pERK-pAKT response map highlights the
dual roles of PI3K and Ras signaling in regulating proliferation
and differentiation cell fates. Indeed, other than their well-known
roles in proliferation, Ras and PI3K have been implicated in
a variety of differentiation processes including those in neuronal,
myeloid, muscle, and adipocyte cells (Crespo and Leo ´n, 2000;
Katso et al., 2001). In the case of PC12 cell-fate decisions,
a number of studies have shown that the temporal differences
in pERK kinetics correlate with cell fates with EGF-mediated
transient ERK activity promoting proliferation and NGF-medi-
ated sustained ERK activity triggering differentiation outcomes
(Marshall, 1995; Santos et al., 2007; Sasagawa et al., 2005).
Our results using the response map analysis and the systemic
siRNA perturbation experiments suggest that changes of the
amplitude of the first wave of ERK signaling (2–5 min) have no
significant impact on the cell-fate decision while the sustained
pERK and pAKT activity is predictive (Figures 3E and 3F). The
relevance of long term PI3K/AKT signaling was demonstrated
using multiple perturbation experiments that shifted the
NGF-regulated proliferation and differentiation balance, further
arguing that PI3K as well as Ras signaling jointly control the
cell-fate decision (Figures 1F, 3H, and 4).
pERK-pAKT Signal Variation Enables Differentiation
and Proliferation Decisions for the Same NGF Receptor
Single-cell intrinsic noise in protein expression typically alters
protein levels by over 30% (Blake et al., 2003; Niepel et al.,
2009). Such variation can make it difficult for cells to execute
reliable signal transduction processes (Arias and Hayward,
2006). Most studies in eukaryotic cells have therefore focused
on mechanisms that make signaling less variable and outcomes
more robust (Acar et al., 2010; Colman-Lerner et al., 2005).
However,noisecan alsobeimportantin cellulardecision making
(Bala ´zsi et al., 2011; Blake et al., 2006; Spencer et al., 2009). Our
study shows that signal variation generated by stimulation of
a uniform cell population can be used to generate two subpopu-
lations with different cell fates. This required a cell-fate boundary
in the pERK-pAKT plane that is narrower (< factor of 2) than the
NGF induced signal variation (?factor of 4), allowing individual
respectively (Figure 2A). Given the multiple feedbacks and
signaling processes involved in regulating long-term AKT and
ERK activity (Campbell et al., 1998; Carracedo and Pandolfi,
2008), the relatively large variation in the pERK-pAKT response
map can plausibly be explained by the sum of smaller relative
expression differences of multiple upstream regulatory proteins.
A Rasa2-Mediated Feedback Controls Differentiation
along with Cell Number Expansion
The analysis of Rasa2 regulation on cell number and differentia-
tion in Figures 7A and 7B showed that the combination of
a response map and signal variation has a potential advantage
in the execution of an effective differentiation response. Instead
of simply differentiating all cells, the same stimulus can keep
some cells proliferating, which eventually increases the total
number of terminally differentiated cells. Figure 7E illustrates
this design, whereby each cycle of proliferation splits off
a pool of differentiating cells. This process does not require
a predefined differentiation program for each cell but rather
relies on stochastic signal variation to create the two subpopu-
lations. Thus, by combining a signaling network with (1)
a response map where the decision is made, (2) sufficient
single-cell signal variation to spread the signal over a sharp
boundary, and (3) a feedback mechanism that positions the
activity vector near the boundary, cells gain tight control over
the fraction of terminally differentiated cells. Inan in vivocontext,
we envision that this pERK-pAKT signaling code enables cells to
maintain a homeostatic balance of precursor and differentiated
cells since excessive proliferation and premature differentiation
are both harmful to multicellular organisms.
Our study introduces a 2D signaling code that describes the
NGF-induced proliferation and differentiation decision of PC12
cells. Given the broad importance of PI3K and Ras signaling,
our study likely exemplifies a general principle whereby pERK-
pAKT response maps represent key integration points where
subpopulations of cells are specified to different cell fates.
Cell Culture, Transfection, Antibodies, and Plasmids
A PC12 subline, Neuroscreen-1 (here referred to as PC12 cells) (Dijkmans
et al., 2008), was used for the study due to its reduced tendency toward cell
aggregation (Cellomics). Unless otherwise noted, cells were induced to differ-
entiate in low serum-containing media (F12K supplemented with 0.5% horse
serum [HS]) while high-serum media (Figure 1F) contained 5% HS and 0.8%
FBS.siRNA(10–20 nM)and DNAtransfection(100ngper 96-well)werecarried
out with Lipofectamine 2000 (Invitrogen) according to the manufacturer’s
protocol. For expression of multiple constructs, PC12 were electroporated
with the Amaxa system according to the manufacturer’s instructions (Lonza).
Sources for antibodies, reagents, qRT-PCR primers, and other constructs
used in this study are provided in the Supplemental Experimental Procedures.
Images for 2D response map experiments were analyzed with custom-made
MATLAB image analysis programs (Salmeen et al., 2010). In brief, nuclear
centroids were identified in images of Hoechst stain. A nucleus mask was
generated for each cell by expansion from the centroid to reach 30% of
maximum intensity. A cell mask was then generated by expansion of the
nucleus mask 5 mm to include both the nucleus and the perinuclear region.
Cell Fate-Signaling Code
Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc. 207
After local background subtraction, the pERK (cell mask), pAKT (cell mask),
and BrdU (nucleus mask) mean intensity were measured. The threshold level
used to determine BrdU-positive cells was set with a k-means clustering algo-
rithm on a well-to-well basis. Detailed analysis related to the 2D response map
was described in the Supplemental Experimental Procedures.
Error bars represent the standard deviation, standard error of the mean, or
95% bootstrap confidence interval as indicated in the legends. Statistical
comparisons (p values) were obtained from two-sided t tests. The Pearson’s
correlation coefficients (R) were calculated as indicated.
Supplemental Information includes Supplemental Experimental Procedures,
three figures, and two tables and can be found with this article online at
We would like to thank J.H. Chen for the graphic design and A. Salmeen for
help with image analysis. We are also grateful for S. Spencer and A. Winans
for comments. This work was supported by a Stanford Graduate Fellowship
to J.-Y.C., a Department of Defense (Breast Cancer Research Program)
Predoctoral Fellowship (W81XWH-09-1-0026) to J.-R.L., National Institutes
of Health (NIH) grant ES016486 to K.A.C., and NIH grant MH64801 to T.M.
Received: April 6, 2011
Revised: August 23, 2011
Accepted: November 4, 2011
Published online: December 29, 2011
Acar, M., Pando, B.F., Arnold, F.H., Elowitz, M.B., and van Oudenaarden, A.
(2010). A general mechanism for network-dosage compensation in gene
circuits. Science 329, 1656–1660.
Albert, R. (2005). Scale-free networks in cell biology. J. Cell Sci. 118, 4947–
Arias, A.M., and Hayward, P. (2006). Filtering transcriptional noise during
development: concepts and mechanisms. Nat. Rev. Genet. 7, 34–44.
Bala ´zsi, G., van Oudenaarden, A., and Collins, J.J. (2011). Cellular decision
making and biological noise: from microbes to mammals. Cell 144, 910–925.
Bar-Sagi, D., and Feramisco, J.R. (1985). Microinjection of the ras oncogene
protein into PC12 cells induces morphological differentiation. Cell 42,
Baraba ´si, A.L., and Oltvai, Z.N. (2004). Network biology: understanding the
cell’s functional organization. Nat. Rev. Genet. 5, 101–113.
Blake, W.J., KAErn, M., Cantor, C.R., and Collins, J.J. (2003). Noise in eukary-
otic gene expression. Nature 422, 633–637.
Blake, W.J., Bala ´zsi, G., Kohanski, M.A., Isaacs, F.J., Murphy, K.F., Kuang, Y.,
Cantor, C.R., Walt, D.R., and Collins, J.J. (2006). Phenotypic consequences of
promoter-mediated transcriptional noise. Mol. Cell 24, 853–865.
Campbell, S.L., Khosravi-Far, R., Rossman, K.L., Clark, G.J., and Der, C.J.
(1998). Increasing complexity of Ras signaling. Oncogene 17 (11 Reviews),
Carracedo, A., and Pandolfi, P.P. (2008). The PTEN-PI3K pathway: of feed-
backs and cross-talks. Oncogene 27, 5527–5541.
Chambard, J.C., Lefloch, R., Pouysse ´gur, J., and Lenormand, P. (2007). ERK
implication in cell cycle regulation. Biochim. Biophys. Acta 1773, 1299–1310.
Chiu, V.K., Bivona, T., Hach, A., Sajous, J.B., Silletti, J., Wiener, H., Johnson,
R.L., 2nd, Cox, A.D., and Philips, M.R. (2002). Ras signalling on the endo-
plasmic reticulum and the Golgi. Nat. Cell Biol. 4, 343–350.
Colman-Lerner, A., Gordon, A., Serra, E., Chin, T., Resnekov, O., Endy, D.,
Pesce, C.G., and Brent, R. (2005). Regulated cell-to-cell variation in a cell-
fate decision system. Nature 437, 699–706.
Crespo, P., and Leo ´n,J. (2000). Ras proteins in the control of the cell cycle and
cell differentiation. Cell. Mol. Life Sci. 57, 1613–1636.
Deppmann, C.D., Mihalas, S., Sharma, N., Lonze, B.E., Niebur, E., and Ginty,
D.D. (2008). A model for neuronal competition during development. Science
Diehl, J.A., Cheng, M., Roussel, M.F., and Sherr, C.J. (1998). Glycogen syn-
thase kinase-3beta regulates cyclin D1 proteolysis and subcellular localiza-
tion. Genes Dev. 12, 3499–3511.
Dijkmans, T.F., van Hooijdonk, L.W., Schouten, T.G., Kamphorst, J.T.,
Vellinga, A.C., Meerman, J.H., Fitzsimons, C.P., de Kloet, E.R., and
during nerve growth factor-induced differentiation. J. Neurochem. 105, 2388–
Friedman, A., and Perrimon, N. (2007). Genetic screening for signal transduc-
tion in the era of network biology. Cell 128, 225–231.
Greene, L.A., and Tischler, A.S. (1976). Establishment of a noradrenergic
clonal line of rat adrenal pheochromocytoma cells which respond to nerve
growth factor. Proc. Natl. Acad. Sci. USA 73, 2424–2428.
Huang, E.J., and Reichardt, L.F. (2003). Trk receptors: roles in neuronal signal
transduction. Annu. Rev. Biochem. 72, 609–642.
Katso, R., Okkenhaug, K., Ahmadi, K., White, S., Timms, J., and Waterfield,
M.D. (2001). Cellular function of phosphoinositide 3-kinases: implications for
development, homeostasis, and cancer. Annu. Rev. Cell Dev. Biol. 17,
Lemmon, M.A., and Schlessinger, J. (2010). Cell signaling by receptor tyrosine
kinases. Cell 141, 1117–1134.
Lockyer, P.J., Wennstro ¨m, S., Kupzig, S., Venkateswarlu, K., Downward, J.,
and Cullen, P.J. (1999). Identification of the ras GTPase-activating protein
Curr. Biol. 9, 265–268.
Maekawa, M., Li,S.,Iwamatsu, A.,Morishita, T.,Yokota,K., Imai, Y.,Kohsaka,
S.,Nakamura,S.,and Hattori, S.(1994). Anovel mammalianRasGTPase-acti-
vating protein which has phospholipid-binding and Btk homology regions.
Mol. Cell. Biol. 14, 6879–6885.
Manning, B.D.,and Cantley,L.C. (2007). AKT/PKB signaling: navigating down-
stream. Cell 129, 1261–1274.
Marshall, C.J. (1995). Specificity of receptor tyrosine kinase signaling: tran-
sient versus sustained extracellular signal-regulated kinase activation. Cell
Morgan, D.O. (2007). The Cell Cycle: Principles of Control (Sunderland, MA:
Myers, J.W., Jones, J.T., Meyer, T., and Ferrell, J.E., Jr. (2003). Recombinant
Dicerefficientlyconvertslarge dsRNAs intosiRNAssuitableforgene silencing.
Nat. Biotechnol. 21, 324–328.
Niepel, M., Spencer, S.L., and Sorger, P.K. (2009). Non-genetic cell-to-cell
variability and the consequences for pharmacology. Curr. Opin. Chem. Biol.
Okkenhaug, K., and Vanhaesebroeck, B. (2003). PI3K in lymphocyte develop-
ment, differentiation and activation. Nat. Rev. Immunol. 3, 317–330.
DUOX2 regulate cell cycle entry via a p53-dependent pathway. Oncogene 29,
Santos, S.D., Verveer, P.J., and Bastiaens, P.I. (2007). Growth factor-induced
MAPK network topology shapes Erk response determining PC-12 cell fate.
Nat. Cell Biol. 9, 324–330.
Sasagawa, S., Ozaki, Y., Fujita, K., and Kuroda, S. (2005). Prediction and vali-
dation of the distinct dynamics of transient and sustained ERK activation. Nat.
Cell Biol. 7, 365–373.
Cell Fate-Signaling Code
208 Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc.
Shao, J., Sheng, H., DuBois, R.N., and Beauchamp, R.D. (2000). Oncogenic
Ras-mediated cell growth arrest and apoptosis are associated with increased
ubiquitin-dependent cyclin D1 degradation. J. Biol. Chem. 275, 22916–22924.
Sherr, C.J., and Roberts, J.M. (2004). Living with or without cyclins and cyclin-
dependent kinases. Genes Dev. 18, 2699–2711.
Spencer, S.L., Gaudet, S., Albeck, J.G., Burke, J.M., and Sorger, P.K. (2009).
Non-genetic origins of cell-to-cell variability in TRAIL-induced apoptosis.
Nature 459, 428–432.
von Kriegsheim, A., Baiocchi, D., Birtwistle, M., Sumpton, D., Bienvenut, W.,
Morrice, N., Yamada, K., Lamond, A., Kalna, G., Orton, R., et al. (2009). Cell
fate decisions are specified by the dynamic ERK interactome. Nat. Cell Biol.
Zhou, J., Valletta, J.S., Grimes, M.L., and Mobley, W.C. (1995). Multiple levels
for regulation of TrkA in PC12 cells by nerve growth factor. J. Neurochem. 65,
Cell Fate-Signaling Code
Molecular Cell 45, 196–209, January 27, 2012 ª2012 Elsevier Inc. 209