Induced epithelial-to-mesenchymal transitions promote a similar cellular phenotype,
but act through cell-line and stimulus-specific signalling mechanisms
The stimulation of PMC42-LA and MDA-MB-468 cells with EGF, or growth of MDA-MB-468
cells under hypoxic conditions (HPX) each promoted EMT as indicated by an increased
proportion of vimentin?+?cells (red fluorescence; Figure 1b f, c g, d h). Unstimulated PMC42-ET cells express vimentin (Figure 1a), thus increases in the number of vimentin+ cells with EGF stimulation (Figure 1e) are relatively small, consistent with our previous reports on EMT within this cellular
system 40],41].
Figure 1. Stimulation of PMC42-ET and PMC42-LA cells with EGF, or stimulation of MDA-MB-468
cells with EGF or growth under hypoxic conditions (HPX) promotes a mesenchymal phenotype.
(a-h) Fluorescence images of stimulated and unstimulated cells labelled with DAPI (blue) and anti-vimentin (red). Scale bar represents 10Â ?m. Changes in mRNA transcript abundance between stimulated
and unstimulated cells for (i) EMT markers and (j) EMT-implicated transcription factors. Note the use of alternative colour-bars to
indicate statistically significant (**; q-value??0.05; red-green) and non-significant (brown/orange–teal) changes in abundance. Grey squares indicate mRNA transcripts that were not reliably
detected – normalised count data are shown in Additional file 2: Figure S1.
Examining changes in transcript abundance that occurred with the phenotypic EMT (Figure 1a-h) consistent differences were observed for several transcripts that contribute
to the canonical mesenchymal phenotype (Figure 1 j). Transcripts for VIM were significantly increased in all models of induced EMT,
including EGF-stimulated PMC42-ET cells, while several other regulatory/signalling
components implicated in EMT 13],59]-61] (further details in Additional file 1: Table S1) showed consistent changes with various degrees of significance (Figure 1i). A number of transcripts encoding cellular signalling components implicated in
EMT also showed large changes between some of the experimental models as detailed
in the sections below.
The mRNAs of transcription factors (TFs) implicated in EMT was also examined and only
FOSL1 (also known as FRA1) showed significant increases in transcript abundance across
all models of induced EMT (Figure 1j). TFs known to play a role in EMT including ETS1, SOX9 and ZEB1 showed consistent
increases in transcript abundance, while FOXO4, KLF8 and the epithelial TF GRHL2 were
consistently reduced; however, not all of these changes were statistically significant
(Figure 1j). Conversely, several well-studied TFs which drive EMT, such as SNAI1 and TWIST1,
showed vastly different expression profiles between differing cell lines and differing
stimuli, while ZEB2 and SNAI2 were not reliably detected within the MDA-MB-468 cells,
nor were FOXC2 and GSC detected across all cell lines tested (Figure 1j). Furthermore, normalised count data suggest that the mammary stem cell TF SOX9
was much more abundantly expressed in the MDA-MB-468 cells, while TWIST1 and ZEB1
had much higher transcript counts in the PMC42 sublines (Additional file 2: Figure S1).
These results indicate that a phenotypically-similar EMT process was induced across
these different cell lines and stimuli, with consistent changes in the transcripts
which mediate these canonical changes, such as VIM, CD44, CDH1 and CDH2. However,
variation in the differential abundance patterns observed for specific EMT-implicated
TFs suggests that these similar phenotypic behaviours are associated with different
regulatory mechanisms.
Pathway analysis highlights alternative signalling mechanisms which contribute to
EMT
To identify signalling pathways likely to be affected by the transcriptional changes
associated with each model of induced EMT, we first assessed the mRNA transcripts
that responded within each model and then mapped these to KEGG pathways. EGF stimulation
of PMC42-ET cells led to significant (q-value??0.05) changes in abundance for 238 transcripts (Table 1). This was the lowest number across all of our in vitro models of EMT, consistent with PMC42-ET cells being relatively mesenchymal in the
unstimulated state (Figure 1a). The EGF- and HPX-stimulated MDA-MB-468 cells had significant changes in abundance
for 3155 and 3716 transcripts, respectively, indicating a much greater response than
the EGF- stimulated PMC42-ET or PMC42-LA cells (Table 1). The number of transcripts with differential abundance for the stimulated MDA-MB-468
cells was of a similar magnitude to the inter-model comparisons between PMC42-ET and
-LA sublines in the presence or absence of EGF (3261 and 2938, respectively; Table 1). These inter-model comparisons also show that the number of transcripts with a significantly
different abundance between the PMC42-ET and -LA sublines decreased with EGF stimulation,
suggesting a potential convergence of phenotypes.
Table 1. Different signalling pathways are dysregulated between the models ofin vitroinduced EMT
Next we examined the putative signalling effects of these altered transcript abundances,
performing an over-representation analysis to identify intracellular signalling pathways
that may have been perturbed (p-value??0.05) by concerted changes to numerous signalling components during induced
EMT. Eleven signalling pathways showed some evidence of dysregulation (p??0.05) within at least one model of induced EMT (Table 1). The PI3K-Akt signalling pathway was the only pathway that showed perturbation of
components across all models of induced EMT (Table 1); however, after further correcting for multiple hypothesis testing, the EGF-stimulated
PMC42-LA cells remained as the only experimental system showing significant transcriptional
dysregulation of PI3K-Akt signalling components. The results shown in Table 1 support the observation that although a phenotypically-similar EMT is induced (Figure 1e-h 1i), as extensively characterised in previous reports by us and others 3],41]-43],46]-48], there are differences in the molecular mechanisms that drive these phenotypic changes
(Figure 1j).
Both the HIF-1 signaling pathway and Rap1 signaling pathway showed very strong transcriptional
perturbations within EGF or HPX-stimulated MDA-MB-468 cells, and there was also evidence
of HIF-1 signaling pathway dysregulation between EGF and HPX-stimulated MDA-MB-468
cells (Table 1). Strong dysregulation of Hippo, Hedgehog and TGF-beta signalling components was
observed with EGF induced EMT within the PMC42-LA cells, and in the absence of EGF,
components of the Wnt signalling pathway showed strong differences in transcript abundance
between the PMC42-ET and PMC42-LA sublines (Table 1).
To identify common signalling elements across these different pathways we examined
the frequency of components. Changes in mRNA transcript abundance of signalling proteins
which were present within at least six of the 11 KEGG maps are shown in Figure 2a. Three proteins were found across seven pathways, encoded by: MAPK1, MAPK3 and PRKCA
(Figure 2a; see membership matrix at right). Within six of the maps, the next most common proteins were encoded by: AKT1, AKT2,
AKT3, MAP2K1, MAP2K2, PIK3CA, PIK3CB, PIK3CD, PIK3CG, PIK3R1, PIK3R2, PIK3R3, PIK3R5,
PRKCB, PRKCG, and RAC1 (Figure 2a). The prevalence of MEK1/2-ERK1/2 and PI3K-Akt across these KEGG maps likely reflects
the role of these signal transducers in the integration of numerous upstream signals.
Figure 2. Numerous signalling components showed significant differences between EGF and HPX
mediated EMT. Heat maps for: (a) mRNA transcripts for signalling components which are present across at least six
perturbed signalling pathway KEGG maps (Table 1); (b, c) mRNA transcripts with significant (q-value??0.05) differences in mRNA transcript abundance within at least one PMC42
cell line condition comparison, and differences in mRNA transcript abundance going
in (b) the same, or (c) different directions for EGF or HPX-stimulated MDA-MB-468 cells compared to unstimulated,
with a significant difference in transcript abundance between the EGF- and HPX-stimulated
MDA-MB-468 cells. Membership within KEGG maps that are listed in Table 1 is shown at right (black box). Note the use of alternative colour-bars to indicate
statistically significant (**; q-value??0.05; red-green) and non-significant (brown/orange–teal) changes in abundance.
Systems-level computational analysis identified putative drug targets to alleviate
signalling pathway dysregulation that occurs with induced EMT
As described earlier, EGF stimulation and hypoxic tumour environments are both thought
to be clinically-relevant drivers of breast cancer progression in vivo. Thus, we focussed our analysis towards elucidating the convergent and divergent
alterations to intracellular signalling which may encompass therapeutic targets for
controlling EMT within MDA-MB-468 cells as a model of triple-negative breast cancer
(TNBC).
To motivate drug target selection several analyses were performed and their results
are described together below. First, transcripts showing similar or divergent patterns
of differential expression between the EGF- and HPX-stimulated MDA-MB-468 cells were
extracted (Additional file 3: Figure S2a b, respectively). Components of the dysregulated signalling pathways
(Table 1) are shown in Figure 2b c. Transcripts that changed in the same direction across all models of induced
EMT (including known EMT markers) are also shown in Additional file 4: Figure S3. Next, transcript abundance data were mapped onto an experimentally verified
protein-protein interaction network and signalling components that could be targeted
by drugs, inhibitors or antagonists were ranked by the relative level of dysregulation
to their local interactome (Table 2).
Table 2. Signalling pathway components showed variable levels of transcriptional disruption
to their local interactome
These results are discussed below with a schematic diagram showing the role of proteins
and functional relationships between signalling components, within the context of
a broader intracellular signalling network (Figure 3). These results were used to motivate pharmacological targeting of several points
within the dysregulated intracellular signalling network to examine the efficacy of
blocking EMT, as indicated within Figure 3.
Figure 3. Differences in signalling component transcript changes between EGF and hypoxia induced
EMT. Changes in transcript abundance (legend top right) for selected intracellular signalling components, within a schematic representation
of the signalling network interactions between encoded proteins. Note the use of alternative
colour-bars to indicate statistically significant (**; q-value??0.05; red-green) and non-significant (brown/orange–teal) changes in abundance. Kinase inhibitors within the families selected for screening
(described in text; shown in purple) are listed in Table 3.
Given the use of EGF within our experimental models of in vitro induced EMT (Figure 1), kinase inhibitors against EGFR/HER were included as a positive control. The local
interaction neighbourhoods of EGFR and ERBB2 were amongst the most dysregulated (Table 2); however, this may reflect the numerous feedback mechanisms which have previously
been elucidated for EGFR signalling 62]-64]. Alternative ligands for EGFR (TGFA, AREG and HBEGF) show significant changes in
transcript abundance, suggesting that autocrine/paracrine signalling mechanisms may
be activated, with HBEGF showing particularly strong differences between EGF and HPX
stimulation (Figure 3).
Activation of EGFR is known to drive signalling through both PI3K-Akt and MEK-ERK
65], and these signal transduction cascades also appear to be key integrators across
all of the dysregulated signalling pathways (Table 1; Figure 2a). Together with further details below, and the results of our pathway analysis,
this motivated our experimental screening to focus upon different classes of kinase
inhibitors targeting PI3K/mTOR, AKT and MEK1/2 as indicated (Figure 3).
Some of the strongest transcriptional changes with induced EMT were observed for integrin
subunits and corresponding ECM components (Figure 3), and these changes would be expected to influence the formation and regulation of
focal adhesion sites. Members of the Src kinase family play an important role in transducing
signals from focal adhesion sites 66] to regulate Ras signalling, and the interactomes of both LYN and FYN are relatively
enriched for disrupted binding partners, as is the homolog ABL1 (Table 2). Induction of FYN by PI3K/Akt signalling has previously been implicated as a key
mediator of cell invasiveness 40]. LYN has also been identified as an important driver of phospho-tyrosine signalling
to induce invasiveness within basal subtype breast cancers, although that study reported
a relatively low level of activated LYN within the MDA-MB-468 cell line 67].
Pharmacological modulation of PI3K/mTOR was particularly attractive for this model
of in vitro induced EMT, as MDA-MB-468 cells are known to carry an inactivating mutation in the
PIP3-phosphatase PTEN 68]. Regulatory Class IA PI3K subunits stabilise the catalytic subunits to inhibit PI3K
activity in the absence of upstream signals 69],70], and PIK3R1 (p85?) was significantly downregulated with EGF- or HPX-stimulation,
although PIK3R2 (p85?) was increased, particularly with EGF stimulation (Figure 3). Furthermore, when considering disruption to the local interactome PIK3R1 was highly
ranked, suggesting a greatly reduced threshold for signalling through PI3K, particularly
under conditions where HPX is driving EMT.
Given the evidence for signalling through PI3K as described above, it was interesting
to note changes in transcript abundance for the AKT scaffolding components Hsp90 and
Cdc37 (Figure 3) with the HSP90AA1 and HSP90AB1 local networks showing the greatest degree of disruption
(Table 2). Vivanco et al. demonstrated that GSK3B is an important downstream effector of AKT
signalling 70], which also showed a high degree of disruption to the local interactome. Furthermore,
AKT3 transcript abundance increased significantly under hypoxic conditions (Figure 3).
Increased signalling activity through MEK1/2-ERK-1/2 is the canonical downstream response
to EGFR stimulation over many cell types 64],71], and activation of EGFR signalling induces a large number of feedback mechanisms
to further modulate pathway activity 63]. This is consistent with the observation that MAPK3 (ERK1) showed some degree of
disruption to its local interactome (Table 2), and with the notion that the MEK1/2-ERK-1/2 axis is a key integrator of dysregulated
signalling pathways across the various models of induced EMT. It is possible that
under conditions where key signalling proteins have been disrupted (e.g. an inactivating mutation in PTEN), some of these feedback mechanisms may lead to
aberrant signalling. We examined differentially expressed genes with a previously
identified transcriptional signature for MEK pathway activation 72] and found many of these transcripts were significantly upregulated within the EGF
or HPX-stimulated MDA-MB-468 cells (Additional file 5: Figure S4a).
EGF- and HPX-stimulated MDA-MB-468 cells show different responses to pharmacological
inhibition of MEK-ERK and PI3K/Akt signalling
As detailed above, systems-level analysis of the mRNA transcript abundance changes
that occurred with induction of EMT identified several signalling molecules that were
likely to have dysregulated activity, and may play a role in promoting the mesenchymal
phenotype. To investigate the potential for therapeutic intervention against these
signalling components, a panel of kinase inhibitors (Table 3) was tested to determine the concentrations at which the fraction of vimentin+ cells or cell count was reduced by 50% (IC50 concentrations).
Table 3. Targeted Inhibition of signalling molecules show differential effects between EGF-
and hypoxia-induced EMT
The majority of inhibitors tested on EGF-stimulated PMC42-ET cells were efficacious
at reducing cell count; however, nearly every inhibitor tested had an IC50 for reducing the number of vimentin+ cells well above pharmacologically relevant concentrations (Table 3a–e), thus off-target effects are likely.
As expected, the panel of EGFR kinase inhibitors (Table 3a) were very effective at blocking EGF-induced EMT and cell growth in the PMC42-LA
and MDA-MB-468 cells, and with the exception of lapatinib, the IC50 values for inhibition of vimentin expression are 8–10 fold lower than the corresponding
IC50 values for reduction of cell count. Reduced levels of vimentin expression correlated
with the ability of these compounds to inhibit the phosphorylation of ERK1/2 over
a range of concentrations (Additional file 6: Figure S5), demonstrating the importance of the EGFR/MEK/ERK canonical pathway and
its associated networks in promoting EMT-associated phenotypic changes. The EGFR kinase
inhibitors also appeared to have an effect on HPX-stimulated MDA-MB-468 cells, although
IC50 values for HPX-stimulated cells were all higher than corresponding IC50 values for EGF-stimulated cells. In particular, inhibition of the HPX-induced vimentin
response in MDA-MB-468 cells occurred at drug concentrations 10-fold higher than
required for inhibition of ERK phosphorylation, indicating that the MEK/ERK pathway
may be less important for EMP and the regulation of vimentin expression under hypoxic
growth conditions. This effect may also be due to drug resistance mechanisms as discussed
below. EGFR kinase inhibitor-mediated reductions in cell count for both EGF- and HPX-stimulation
were generally observed at concentrations an order of magnitude greater than the effects
on vimentinexpression, indicating that our treatments are affecting EMT at relevant
concentrations, while reduction in cell viability at higher concentrations may be
caused by off target effects. Hypoxia-treated MDA-MB-468 cells were exposed to a small
molecule inhibitor of HIF1? accumulation and gene transcriptional activity, CAY10585,
to determine whether this could reduce the induction of EMT in these cells. At concentrations
below 1Â ?M CAY10585 did not have a significant effect on the number of vimentin+ cells; however, the number of vimentin? cells was potently reduced, suggesting this may have a deleterious effect upon the
cell population with an epithelial phenotype (Additional file 7: Figure S6).
Although EGF stimulation further increased the mRNA transcript abundance of EMT markers
(Figure 1i) the inability of EGFR inhibitors to reduce the fraction of vimentin+ PMC42-ET cells (Table 3a) suggests that the unstimulated mesenchymal phenotype of these cells is maintained
through EGFR-independent signalling mechanisms.
Inhibitors targeting the MEK1/2 (Table 3b) and Src-family kinases (Table 3c) showed a similar response profile to the EGFR inhibitors with potent blocking of
vimentin expression within the EGF-stimulated cells. For MEK inhibitors the IC50 values for inhibition of vimentin expression tended to be lower than the corresponding
IC50 values for cell count, and within MDA-MB-468 cells the IC50 values were again higher with HPX stimulation than EGF stimulation (Table 3b). A similar effect was seen for inhibition of phospho-ERK1/2 (data not shown). A
previously reported mRNA transcript signature for ‘compensatory resistance’ to AZD6244
(Additional file 5: Figure S4b) 72] shows some agreement with the observed efficacy of this MEK inhibitor (Table 3b). The EGF-stimulated PMC42-LA cells had the lowest IC50 for AZD6244 in reducing the fraction of vimentin+ cells by several orders of magnitude, and many of the AZD6244 resistance signature
genes showed a decrease in transcript abundance relative to unstimulated PMC42-LA
cells (Additional file 5: Figure S4b). Although the profile of this signature was very similar between EGF-
and HPX-stimulated MDA-MB-468 cells, several of the transcripts showed a greater degree
of upregulation with hypoxia, in agreement with the reduced efficacy of AZD6244 within
hypoxia-stimulated MDA-MB-468 cells (Table 3b). For most Src family inhibitors within EGF-stimulated MDA-MB-468 cells the values
for IC50 of vimentin+ cells were lower than the IC50 values for cell count. Conversely, within EGF-stimulated PMC42-LA cells and HPX-stimulated
MDA-MB-468 cells the IC50 for cell count is lower for most Src family inhibitors (Table 3c).
Within EGF-stimulated PMC42-LA cells GDC-0941 and GSK2126458 were the only PI3K/mTOR
inhibitors (Table 3d) with pharmacologically relevant IC50 values for reduction of vimentin+ cells, although most inhibitors were capable of reducing cell growth. The PI3K/mTOR
inhibitors were much more efficacious within the MDA-MB-468 cells, and many of the
tested compounds had a lower IC50 value for the reduction of vimentin+ cells compared to the reduction in cell count.
In contrast to PI3K/mTOR inhibitors, the majority of compounds targeting Akt kinases
(Table 3e) were only capable of reducing cell count, with A-674563 and AZD5363 the only inhibitors
with a pharmacologically relevant IC50 value for vimentin+ cells across any of the cell lines and conditions. Unexpectedly, some Akt inhibitors
and mTOR inhibitors were observed to increase the fraction of vimentin+cells and the relative cell density, particularly within HPX-stimulated MDA-MB-468
cells (Figure 4h k).
Figure 4. Hypoxia- and EGF-induced metastatic MDA-MB-468 cells show markedly different responses
to pharmacological inhibitors. Pharmacological dose–response curves showing the fraction
of vimentin-positive cells (blue; left axes) and cell-count (red; right axes) in the presence of (a-c) the MEK inhibitor AZD6244, (d-f) the PI3K inhibitor GDC-0941, (g-i) the AKT1/2/3 inhibitor AZD5363 (j-l) and the mTOR inhibitor Everolimus. (m-o) pharmacological inhibition of vimentin with a combination of MEK-1/2 (AZD6244) and
AKT1/2/3 (AZD5363) inhibitors at varying concentrations. (p) pharmacological inhibition of vimentin with a comination of MEK-1/2 (AZD6244) and
AKT1 (Akt-i-1) inhibitors at varying concentrations.
The observation that several classes of inhibitors were efficacious within EGF-stimulated
MDA-MB-468 cells, but had little effect under hypoxic growth conditions, supports
the conclusion from the transcriptome analysis that the phenotypically similar EMT
processes induced with EGF or hypoxia are driven by different signalling mechanisms.
Furthermore, given differences observed between the EGF- and HPX-induced transcriptional
profiles, particularly for signalling ligands where the receptor also has strong increases
in transcript abundance, such as HBEGF/EGFR and VEGFA/KDR, we hypothesised that pro-survival
signalling through AKT may mediate the reduced efficacy of MEK-1/2 inhibitors under
hypoxic conditions. Thus, we also applied the AKT1/2/3 inhibitors GSK690693 or AZD5363
in combination with the MEK1/2 inhibitor AZD6244. The pharmacological efficacy curves
suggest that they provide a synergistic effect to block the relative fraction of vimentin+ cells (Figure 4?m). Furthermore, this effect was not observed with the AKT1 or AKT1/2 inhibitors
tested in combination with AZD6244 (Figure 4p).