ArticlePDF Available

Neuroblastoma signalling models unveil combination therapies targeting feedback-mediated resistance

Authors:

Abstract

Very high risk neuroblastoma is characterised by increased MAPK signalling, and targeting MAPK signalling is a promising therapeutic strategy. We used a deeply characterised panel of neuroblastoma cell lines and found that the sensitivity to MEK inhibitors varied drastically between these cell lines. By generating quantitative perturbation data and mathematical modelling, we determined potential resistance mechanisms. We found that negative feedbacks within MAPK signalling and via the IGF receptor mediate re-activation of MAPK signalling upon treatment in resistant cell lines. By using cell-line specific models, we predict that combinations of MEK inhibitors with RAF or IGFR inhibitors can overcome resistance, and tested these predictions experimentally. In addition, phospho-proteomic profiling confirmed the cell-specific feedback effects and synergy of MEK and IGFR targeted treatment. Our study shows that a quantitative understanding of signalling and feedback mechanisms facilitated by models can help to develop and optimise therapeutic strategies. Our findings should be considered for the planning of future clinical trials introducing MEKi in the treatment of neuroblastoma.
RESEARCH ARTICLE
Neuroblastoma signalling models unveil
combination therapies targeting feedback-
mediated resistance
Mathurin DorelID
1,2
, Bertram KlingerID
1,2,5,6
, Tommaso MariID
8
, Joern ToedlingID
3
,
Eric BlancID
4
, Clemens MesserschmidtID
4
, Michal Nadler-Holly
8
, Matthias ZiehmID
8
,
Anja SieberID
1,2,5
, Falk HertwigID
3
, Dieter Beule
4
, Angelika Eggert
3,5,6,7
, Johannes
H. SchulteID
3,5,6,7
, Matthias SelbachID
8
, Nils Blu
¨thgenID
1,2,4,6,7
*
1Institute of Pathology, Charite
´-Universita
¨tsmedizin Berlin, Berlin, Germany, 2Integrative Research Institute
for the Life Sciences and Institute for Theoretical Biology, Humboldt-Universita¨t zu Berlin, Berlin, Germany,
3Department of Pediatric, Division of Oncology and Haematology, Charite
´-Universita
¨tsmedizin Berlin, Berlin,
Germany, 4Berlin Institute of Health, Berlin, Germany, 5German Cancer Consortium (DKTK), partner site
Berlin, Germany, 6German Cancer Research Center (DKFZ), Heidelberg, Germany, 7Berlin Institute of
Health (BIH), Berlin, Germany, 8Max Delbru¨ck Center for Molecular Medicine, Berlin, Germany
*nils.bluethgen@charite.de
Abstract
Very high risk neuroblastoma is characterised by increased MAPK signalling, and targeting
MAPK signalling is a promising therapeutic strategy. We used a deeply characterised panel
of neuroblastoma cell lines and found that the sensitivity to MEK inhibitors varied drastically
between these cell lines. By generating quantitative perturbation data and mathematical
modelling, we determined potential resistance mechanisms. We found that negative feed-
backs within MAPK signalling and via the IGF receptor mediate re-activation of MAPK sig-
nalling upon treatment in resistant cell lines. By using cell-line specific models, we predict
that combinations of MEK inhibitors with RAF or IGFR inhibitors can overcome resistance,
and tested these predictions experimentally. In addition, phospho-proteomic profiling con-
firmed the cell-specific feedback effects and synergy of MEK and IGFR targeted treatment.
Our study shows that a quantitative understanding of signalling and feedback mechanisms
facilitated by models can help to develop and optimise therapeutic strategies. Our findings
should be considered for the planning of future clinical trials introducing MEKi in the treat-
ment of neuroblastoma.
Author summary
Only few targeted therapies are currently available to treat high-risk neuroblastoma. To
address this issue we characterized the drug response of high risk neuroblastoma cell lines
and correlated it with genomic and transcriptomic data. Particularly for MEK inhibition,
we saw that our panel could be nicely separated into two groups of resistant and sensitive
cell lines. Genomic and transcriptomic markers alone did not help to discriminate
between responders and non-responders. We used signalling perturbation data to build
PLOS COMPUTATIONAL BIOLOGY
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 1 / 26
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Dorel M, Klinger B, Mari T, Toedling J,
Blanc E, Messerschmidt C, et al. (2021)
Neuroblastoma signalling models unveil
combination therapies targeting feedback-mediated
resistance. PLoS Comput Biol 17(11): e1009515.
https://doi.org/10.1371/journal.pcbi.1009515
Editor: Inna Lavrik, OvGU; Medical Faculty,
GERMANY
Received: June 7, 2021
Accepted: October 1, 2021
Published: November 4, 2021
Peer Review History: PLOS recognizes the
benefits of transparency in the peer review
process; therefore, we enable the publication of
all of the content of peer review and author
responses alongside final, published articles. The
editorial history of this article is available here:
https://doi.org/10.1371/journal.pcbi.1009515
Copyright: ©2021 Dorel et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: The data and analysis
packages are available at these repositories: RNA-
Seq data: ENA PRJEB40670 (https://www.ebi.ac.
uk/ena/browser/view/PRJEB40670) STASNet
cell line specific signalling models. Our models suggest that negative feedbacks within
MAPK signalling lead to a stronger reactivation of MEK in MEKi resistant cell lines after
MEK inhibition. Model analysis suggested that combining MEK inhibition with IGF1R or
RAF inhibition could be an effective treatment and we characterised this combination
using phosphoproteomics by mass-spectrometry and growth assays. Our study confirms
the importance of quantitative understanding of signalling and may help plan future clini-
cal trials involving MEK inhibition for the treatment of neuroblastoma.
Introduction
Neuroblastoma is the most common and devastating extracranial childhood solid tumour,
accounting for 15% of all childhood cancer deaths. The 5-year survival rate is 75% overall, but
it is below 45% for so-called high-risk neuroblastoma that represent about 40% of patients [1
3]. Telomere maintenance is a central hallmark of high-risk neuroblastoma [4], and approxi-
mately 50% of high-risk neuroblastoma harbour amplification of the MYCN oncogene [5].
Mutations activating the RAS/MAPK signalling pathway are frequent in high-risk and relapsed
neuroblastoma [6,7], with relapsed neuroblastoma being almost always fatal. Most recently,
mutations in the p53/MDM2 or RAS/MAPK pathway in the presence of telomere mainte-
nance mechanisms were shown to define a subgroup of ultra-high risk neuroblastoma with a
5-year survival below 20%. Therefore, development of novel therapies for patients with high
risk or relapsed neuroblastoma is an urgent clinical need. Mutations of anaplastic lymphoma
kinase (ALK), present in 8% of all patients at diagnosis [8,9], are the most common mutations
activating the RAS/MAPK pathway in neuroblastoma. In addition, mutations in PTPN11,
NF1, Ras and other RAS/MAPK pathway signalling elements occur in neuroblastoma [7,10].
This makes RAS/MAPK pathway inhibition a promising treatment option for neuroblas-
toma, and ALK and MEK inhibitors are already being tested in early clinical trials [11]. How-
ever, tumour responses to targeted inhibitors were inconsistent, and early progression pointed
towards development of resistance, giving a strong incentive to understand mechanisms of pri-
mary and secondary resistance and how to overcome these mechanisms.
Resistance to targeted therapies of signalling pathways are often mediated by feedbacks that
re-wire or re-activate signalling. For example, resistance to PI3K/mTOR inhibition in breast
cancer is often mediated by feedbacks that lead to activation of JAK/STAT signalling [12]. Sim-
ilarly, in colon cancer, MAPK-directed therapy is counteracted by a negative feedback that
leads to hyper-sensitisation of the EGF receptor and ultimately reactivation of MAPK and
AKT signalling [13,14]. Additionally, a very strong feedback from ERK to RAF leads to re-
activation of MAPK signalling upon MEK inhibition in many cancer types [1517]. One
approach to overcome feedback-mediated resistance is by combinatorial therapy that co-tar-
gets the feedback [18].
We report here how a more quantitative understanding of feedback mechanisms might
help to optimise combinatorial treatment. We used a neuroblastoma cell line panel represent-
ing the class of very high-risk neuroblastoma, which we profiled for drug sensitivity, genomic
and transcriptomic alterations. We observed strong differences in the sensitivity to MEK inhi-
bition. To arrive at a mechanistic understanding of resistance to MEK inhibition, we generated
systematic perturbation data and quantified signalling using data-driven models. By this we
described qualitative and quantitative differences in feedback structures that might confer the
observed robustness to MEK inhibition. We then identified potential combinations capable of
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 2 / 26
package: GitHub (https://github.com/molsysbio/
STASNet/releases/tag/Dorel2020)
Phosphoproteomics: https://itbgit.biologie.hu-
berlin.de/dorel/phosphoproteomics_tnb_
perturbations.
Funding: We acknowledge funding from the Berlin
Institute of Health (CRG1 Terminate NB) and from
the Federal Ministry of Education and Research
/BMBF/ (grant MSTARS, to NB and MS). The
funders had no role in study design, data collection
and analysis, decision to publish, or preparation of
the manuscript.
Competing interests: The authors have declared
that no competing interests exist.
sensitising highly resistant cell lines to MEK inhibition, and tested these combinations
systematically.
Results
Drug sensitivity in a panel of very-high-risk neuroblastoma cell lines
We collected a panel of 9 neuroblastoma cell lines (CHP212, LAN6, NBEBC1, SKNAS, NGP,
SKNSH, N206, KELLY and IMR32) and performed molecular profiling of these cells (RNA-
sequencing and exome sequencing, see Fig 1A and S1 File). We noticed that all cell lines har-
bour a mutation in at least one of the RAS pathway genes with all cell lines having a mutation
in either KRAS, NRAS, NF1, BRAF or ALK. One cell line (IMR32) had two mutations in the
pathway: a mutation in KRAS and an atypical BRAF mutation. Most cell lines also have a
mutation in one of the p53 pathway genes: ATRX, ATM, ATR, PRKDC, CDKN2A and TP53.
Additionally, all express telomerase as seen by TERT expression, except for LAN6 which is
known to have an alternative mechanism to lengthen the telomeres (ALT) [4]. We saw strong
variability in the expression of MYCN, with 4 cell lines expressing low levels of MYCN, and 5
cell lines displaying high levels of MYCN. When considering mutations of individual genes,
we found a strong heterogeneity within our panel, but overall the frequency of mutations in
individual genes reflects that of high risk tumours [6]. Taken together, those data indicate that
the chosen cell line panel can be seen as representative for the group of very-high risk
neuroblastoma.
To further characterise the cell line panel, we measured drug sensitivity for 6 inhibitors that
target components of the pathways shown to be affected by mutations (MAPK/PI3K/mTOR),
using live cell imaging and computing growth rates from confluency measurements (Fig 1B).
In this panel of cell lines, there was no notable difference in the sensitivity to the AKT inhibitor
MK2206 or to the RAF/pan-tyrosine kinase-inhibitor Sorafenib. In contrast, pronounced vari-
ation in IC50 across the panel can be seen for mTORC1 inhibitor Rapamycin and MEK inhibi-
tor AZD6244. When comparing to published drug sensitivity data, the IC50 for AZD6244
largely correlate with those derived for a different MEK inhibitor (binimetinib) [19]. All 6
NRAS wild type cell lines showed similar sensitivity to Rapamycin while the 3 NRAS mutant
cell lines exhibited either strong resistance (SKNSH and SKNAS) or sensitivity (CHP212).
This is only partly in agreement with previous literature that described CHP212 but also
SKNAS as sensitive to sub-nanomolar concentrations of Everolimus, a Rapamycin analog
[20]. AZD6244 is the drug with the most variable drug response, with a subset of 6 cell lines
cell lines being very resistant to AZD6244 (IC50 >10μM, Fig 1C and S1 Fig) and another sub-
set of 3 cell lines showing extreme sensitivity (IC50 10–100 nM). When correlating inhibitor
sensitivity with mutations, we found no notable correlation for AZD6244 and Rapamycin (S2
Fig). Drug sensitivities also did not correlate significantly with selected expression data
(adjusted p>0.93 for the 1000 most variable genes and adjusted p>0.94 for GO signal trans-
duction genes, S3 Fig). Also a PCA analysis could not separate cells according to MEKi sensi-
tivity for those two expression groups (S4 and S5 Figs). For instance, previous reports showed
that NF1 expression is linked to sensitivity to MEK inhibitors [19], however we only found a
weak and non-significant correlation with AZD6244 sensitivity (R
2
= 0.34, p= 0.10, S6 Fig).
Taken together, this data establishes that this cell line panel represents a heterogeneous group
of very high risk neuroblastoma that differ in drug sensitivity, most prominently against MEK
inhibitors. Furthermore, it suggests that the difference cannot be explained by single mutations
or expression of marker genes alone.
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 3 / 26
Fig 1. Mutations are insufficient to explain sensitivity variations to RAS/PI3K drugs in neuroblastoma cell line
panel. A: Oncoprint of 9 neuroblastoma cell lines for RAS/p53/PI3K related genes along with MYCN and TERT
mRNA expression. Bold font indicates MYCN-amplified cell lines. B: Relative IC50 of the same 9 neuroblastoma cell
lines as in A for drugs targeting the PI3K and MAPK pathways (n = 2). C: Viability concentration curves for the MEK
inhibitor AZD6244 on the neuroblastoma cell line panel along with the calculated IC50 (intersection with dotted line).
Points represent measurements (n = 2).
https://doi.org/10.1371/journal.pcbi.1009515.g001
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 4 / 26
Perturbation-response data unveils heterogeneity in signalling
To get insights into the underlying mechanisms of resistance to the MEK inhibitor AZD6244,
we selected 6 neuroblastoma cells lines that represented the spectrum of sensitivity to MEK
inhibition (sensitive: CHP212, LAN6; resistant: SKNAS, SKNSH, KELLY and IMR32). Using
these cell lines, we performed perturbation experiments, in which we stimulated the cells by
growth factors for 30 minutes, and additionally inhibited specific pathways for 90 minutes
(Fig 2A). After perturbation, we then monitored pathway activity by measuring phospho-
proteins.
We designed the experiments such that they probe the AKT/mTOR and MAPK signalling
pathways (Fig 2B). Specifically, we selected ligands that might activate those pathways based
on the expression of growth factor receptors in the cell lines. As expression of receptors was
heterogeneous (S7 and S8 Figs), we chose a set of growth factors such that each cell line had
robust expression of receptors for at least two provided ligands. Inhibitors were chosen such
that they block key steps of the pathway. The position of perturbations and readouts in the sig-
nalling network is shown in Fig 2B. We perturbed the 6 cell lines with 4 ligands (PDGF, EGF,
IGF1 and NGF, shown in blue) and 7 inhibitors (GS4997 (ASK1i), MK2206 (AKTi), Rapamy-
cin (mTORC1i), AZD6244/Selumetinib (MEKi), Sorafenib (RAFi), TAE684 (ALKi) and
GDC0941 (PI3Ki), shown in red) alone or in combinations. Subsequently, we measured 6
phosphoproteins (MEK, ERK, AKT, S6K, p38 and cJUN, yellow background) for each pertur-
bation using a sandwich ELISA where a first bead-bound antibody captures the protein and a
second recognises the phosphosite of interest. All experiments were performed in two biologi-
cal replicates.
Overall, the perturbation experiments yielded 240 data points per cell line, which are visual-
ised in a heatmap in Fig 2C. Inspection of the heatmap shows that the perturbation-response
data has similar patterns in different cell lines, but there are also clear differences. For instance,
inhibition of mTOR leads to down-regulation of phospho-S6K across all cell lines, but inhibi-
tion of AKT and PI3K has diverging effects on S6K. Similarly, application of MEKi leads to an
increase of phospho-MEK across all cell lines, but ALK inhibition had varying effects in differ-
ent cell lines.
To get further insights into this high-dimensional data set, we performed principal compo-
nent analysis (PCA) on the perturbation data (Fig 2D top and S9 Fig). The PCA highlights 3
groups of cell lines. The first component (42% of variance) separates the cell lines according to
the effect of Sorafenib and TAE684 on AKT and S6K. The second component (26%) separates
IMR32 and KELLY based mainly on the MEK response to MEK inhibition. The third compo-
nent (18%) contains the effects of IGF1, GS4997 and Rapamycin on AKT and S6K and mainly
separates KELLY and IMR32 (S10 Fig).
When we applied hierarchical clustering on the cell line panel, SKNSH was clustered sepa-
rately, suggesting that it has a very atypical response to the perturbations, with a generally very
high response to all ligands, and an especially strong response to PDGF (Fig 2D bottom). This
atypical status of SKNSH is also present in the mRNA expression, with a PCA on the most var-
iables genes or on the genes in the GO term “signal transduction” separating it from the other
cell lines. Interestingly, CHP212 also separated from the other cell line in a PCA based on gene
expression data, but not when considering the response to the perturbations. When grouping
cells by MEK inhibitor sensitivity, we noticed that simple multivariate analysis by PCA does
not separate cells into groups that correspond to sensitive or resistant cells (Fig 2D top and S9
Fig), and also hierarchical clustering does not separate sensitive from resistance cell lines (Fig
2D bottom).
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 5 / 26
Signalling models highlight differential feedback regulation of MEK
To get further, more mechanistic, insights into potential resistance mechanisms, we used the
perturbation data to parameterise signalling models. We applied our previously developed
method that has been derived from Modular Response Analysis (MRA, implemented as R
Fig 2. Neuroblastoma cell lines show heterogeneous responses to signalling perturbations. A: Outline of the perturbation experiments. A panel of cell lines was
treated with growth factors and small molecule inhibitors, and the resulting effect on selected phosphoproteins was measured using multiplexed bead-based ELISAs. B:
Graphical representation of the perturbation scheme on a literature signalling network. Blue and red contour highlights ligand stimulation and kinase inhibition,
respectively; yellow filling shows measured phosphoproteins. C: Perturbation data obtained from applying all combinations of 4 ligands or BSA control and 7
inhibitors or DMSO control to 6 neuroblastoma cell lines. Each measurement is normalised by the BSA+DMSO control of the corresponding cell line and represents at
least 2 biological replicates. Readouts are phospho-proteins p-MEK1
S217/S221
, p-p38
T180/Y182
, p-ERK1
T202/Y204
, p-cJUN
S63
, p-AKT
S473
and p-S6K
T389
. D: Global non-
mechanistic analysis of the perturbation data presented in C: TOP first two components of a principal component analysis and BOTTOM hierarchical clustering. Colour
scale corresponds to the IC50 for AZD6244 treatment (see also Fig 1C).
https://doi.org/10.1371/journal.pcbi.1009515.g002
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 6 / 26
package STASNet, [21]) to fit signalling network models to each cell line. This modelling pro-
cedure requires a literature network and the perturbation data as input, and then estimates
response coefficients corresponding to link strengths using a maximum likelihood estimate
(see Fig 3A, first step). By using the statistical framework of the likelihood ratio test, the model-
ling procedure then allows to test if any extension of the literature network is required to
describe the data (see Fig 3A, second step). To compare parameters between cell lines, it is
essential to harmonise parameters between all cells that can practically not be identified alone,
i.e. parameters for inhibitors (see Fig 3A, third step). This finally yields a parameter map that
allows to compare signalling strength between cell lines (see Fig 3A, final step).
When starting with a canonical literature network (see Materials and methods), we
obtained reasonable fits for 4 of the 6 cell lines, as judged by the sum of weighted squared
residuals that is in the order of number of data points (Fig 3B, red bars), and the normal distri-
bution of residuals (S11 Fig). When we systematically tested if extensions of the network
improve the fit using a likelihood ratio test, we found that significant improvements were still
possible for most cell lines. We therefore performed successive rounds of extensions for each
cell line independently (Fig 3A and S3 File). While SKNSH required no extension of the litera-
ture network, CHP212, LAN6, SKNAS required two or three extensions. KELLY and IMR32,
the two cell lines that initially had the poorest fit, required four extensions (Fig 3C). After the
extension the sum of weighted squared residuals was in the order of the number of data points
for all cell lines except KELLY (Fig 3B green bar). The high residuals still exhibited by KELLY
could be narrowed down to uncertainties in individual data points (see S3 File). Two network
extensions (ASK1!MEK and p38!S6K) were significant in at least 3 cell lines and corre-
spond to an effect of the ASK1 inhibitor GS4997 on the MEK/ERK MAPK pathway and S6K.
Both links are negative which suggests an antagonism between the p38 MAPK and the MEK/
ERK MAPK pathways in neuroblastoma cell lines. This negative crosstalk from p38 to MEK/
ERK has also been described in other cell systems, e.g. after p38 knockdown in HeLa cells [22].
All extended models had similar, but different, parameters for the inhibitor strength. How-
ever, there is a strong interdependence of the inhibitor strength and link strength downstream
of the inhibitor which render comparison between those link strengths in different cells diffi-
cult (see S3 File). As all cell lines received the same inhibitor concentration we therefore har-
monised the inhibitor parameters by fixing them to the mean value between all models (Fig
3A, fixed inhibitor parameters). The resulting harmonised models maintained a good agree-
ment with the data (Fig 3B, blue bars) and were used for inter-model comparisons (Fig 3D and
3E).
When inspecting the parameters for ligand-induced pathway activation, we noticed that
they reflected a strong heterogeneity in ligand response between the cell lines. Reassuringly,
they matched the expression of the corresponding receptors in many cases (Fig 3D and S12
Fig). The parameters for pathways downstream of NGF correlated mostly with NTRK1 expres-
sion and not with NGFR expression, which might indicate that NGF signalling is mediated
mostly via NTRK1 in those cell lines. The parameters for IGF-induced signals correlated with
IGF1R or IGF2R for MEK and AKT, respectively, indicating that both receptors mediate IGF1
signalling independently. Interestingly, the parameters for the pathway from EGF to MEK did
not correlate with EGFR expression, but they do for EGF to AKT, which might suggest that
differences in adaptor protein expression shape routing into downstream signalling in the vari-
ous cell lines. Indeed, the expressions of GAB2 and SRC are very different between the cell
lines and could explain that IMR32 and LAN6 are activated by EGF as strongly as SKNAS and
SKNSH despite their lower EGFR expression (Fig 2C and S6 Fig). Another potential cause for
the attenuated activation of MEK/ERK is that in NRAS mutant cell lines (CHP212, SKNAS
and SKNSH), MEK/ERK activity is less inducible by receptors, as also parameter values of the
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 7 / 26
Fig 3. Receptor expression and topology variations explain the heterogeneity in perturbation response. A: Starting from a
literature-derived network, a model was fitted for each cell line (Initial model fit) and extended following suggestions from the
model (Model extensions and refit). Those models with different network structures were then harmonised by fixing the inhibition
parameters to a consensus value (Fixed inhibitor parameters) to make the parameters directly comparable (Parameter comparison).
B: Model residuals before and after model extension and harmonisation. The black line represents the number of data points, which
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 8 / 26
routes from PDGF, EGF, NGF and IGF into MAPK signalling are lower in those cell lines.
Conversely, these cell line models display a slightly more inducible PI3K pathway. This obser-
vation is in agreement with a recent comparative study of G12V-mutated RAS isoforms in
colorectal SW48 cells, where the NRAS mutated cell line showed a weaker coupling of recep-
tors to MEK and a stronger coupling to PI3K than in the parental cell line [23]. This would
suggest that an activation of the MEK/ERK pathway is relayed predominantly by NRAS while
the PI3K pathway activation is mediated by other proteins [24]. Taken together, this shows
that the wiring and routing of ligand induced signalling in these cell lines is varying and is
mostly explainable by the expression of the corresponding receptor and RAS mutation status.
In contrast to the receptor-associated parameters, the strength of intra-cellular kinase paths
are less variable, and most paths are comparable between cell lines (Fig 3E). The most promi-
nent exception is the negative feedback in MAPK signalling from ERK to RAF. When com-
pared to the other cell lines, this feedback appears to be 3 to 4 times stronger in KELLY and
IMR32, which are two cell lines that are highly resistant to AZD6244. A strong RAF-mediated
feedback is a known resistance mechanism against MEK inhibitors [1517], where relieve of
inhibition of upstream components post inhibition can partially reactivate signalling. This sug-
gests that AZD6244 resistance could be mediated by a differential regulation of this feedback.
Apart from the RAF-mediated feedback, MAPK signalling is also controlled by receptor-
mediated feedbacks. In the KELLY cell line, our modelling procedure extended the model by a
negative feedback from S6K to IGFR that could then explain the strong accumulation of
pMEK by IGF following AZD6244 treatment (Fig 3C and S3 File). Receptor-mediated feed-
backs are also known to mediate resistance, notably to MAPK inhibitions [13,18,2527], by
reactivating the pathway and other parallel pathways.
In summary, the signalling parameters derived from the perturbation data by our models
show that cell lines diverge in receptor expression and feedback regulation, with strong multi-
layered feedbacks for some of the resistant cell lines.
Differential quantitative wiring of resistant cell lines
A hallmark of negative feedbacks is that they lead to re-activation of the pathway after pathway
inhibition. In agreement with this, we observe an increase of phosphorylated MEK upon
MEKi treatment (AZD6244) that is more pronounced in the cell lines IMR32 and KELLY
compared to the other cell lines modelled, including the most sensitive cell lines CHP212 and
LAN6 (Fig 4A and S13 Fig). We also tested the most resistant cell line in our panel, N206,
which also showed a strong feedback response (Fig 4A). To more precisely dissect the feedback
wiring, we generated additional focused perturbation data for those cells with high feedback
(KELLY, IMR32 and N206) to MEK inhibition. We stimulated cells with different growth fac-
tors (IGF and NGF or EGF), and blocked MAPK signalling with MEK and RAF inhibitors,
and subsequently monitored six phosphoproteins (Fig 4B). Subsequently, we used this data to
parameterise a focused MRA model that additionally either contained or did not contain the
only receptor-mediated feedback found in the first modelling round from S6K!IGF1 (Figs
is equal to the expected mean of the error if the model explains all the data. C: Cell-line-specific network extensions (dashed arrows)
relative to the literature network. Colour of the extended link was matched to cell line colour if required in only one cell line model
and black otherwise. D: Model paths from the receptors to the first measured downstream node and correlation with the
corresponding receptor expression. The colours correspond to the value of the path scaled by the maximum absolute value of that
path between all cell lines. E: Model paths between non-receptor perturbed nodes and measured nodes for routes present in at least
2 cell lines. Colour scale is the same as in D. Cells are ordered from left to right from most sensitive to most resistant to the MEK
inhibitor AZD6244. Due to the absence of ASK1 basal activity in IMR32 ASK1->p38 and ASK1->MEK represent in this cell line
NGF->ASK1->p38 and NGF->ASK1->MEK respectively.
https://doi.org/10.1371/journal.pcbi.1009515.g003
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 9 / 26
3C and 4A). Inclusion of the IGF receptor-mediated feedback led to a significantly better fit of
the data for N206 and KELLY (χ
2
p<0.05), but did not improve the IMR32 model (Fig 4C and
4D). Interestingly, the S6K!IGF1!RAF!MEK feedback is stronger in the N206 models, but
the pathway-intrinsic feedback (ERK!RAF!MEK) is stronger in KELLY (Fig 4D). This
highlights that all these cells display negative feedback regulation, but the strengths of the two
layers of feedbacks are different between cell lines.
Parallel inhibition of MEK and IGFR leads to synergistic effects on the
phosphoproteome
To gain a more systematic understanding of the effect of MEK and IGFR inhibition on the sig-
nalling states of the cells, we generated deep (phospho-)proteomics profiles using tandem
mass-tag (TMT) based mass spectrometry [28,29]. We reasoned that inhibition of IGFR in
Fig 4. AZD6244 resistant cell lines have strong feedback control of MAPK signalling. A: Mean pMEK log2-fold change relative to control after AZD6244 treatment
in 7 neuroblastoma cell lines measured with bead-based ELISAS. Error bars represent 95% confidence interval. B: Measurement of 6 phosphoproteins (columns) after
perturbation of N206, IMR32 and KELLY by either EGF (KELLY, N206) or NGF (IMR32) (together referred to as GF), IGF1, or control BSA in combination with
Sorafenib (RAFi), AZD6244 (MEKi) or control DMSO. Values are expressed in log2-fold change to BSA+DMSO control. C: Starting model and S6K!IGF1 receptor
extension for the high pMEK responder cell lines. D: (top panel) Model residuals for N206, IMR32 and KELLY models with (black) or without (blue) an S6K!IGF1
receptor feedback link and corresponding p-value(χ
2
test with df = 1). (bottom panel) Parameter values of the high pMEK responder models including the S6K!IGF1
receptor link.
https://doi.org/10.1371/journal.pcbi.1009515.g004
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 10 / 26
combination with MEK should have a synergistic effect in N206 compared to IMR32. We mea-
sured the phospho- and total protein levels in IMR32 and N206 cells after 4h treatment with
MEK and/or IGFR inhibitors and control cells. Although a similar number of phosphosites
were dis-regulated in both cell lines (448 in IMR32, 615 in N206, FDR <0.05), there was little
overlap in the phospho-peptides differentially regulated between the two cell lines (Fig 5A),
and this overlap was mostly limited to phospho-peptides affected by MEK inhibition (S14 Fig).
In IMR32, IGFR inhibition had little effect, while the presence of MEK inhibition strongly
affected the phosphoproteome (Fig 5B left). Moreover the effect of the combination of MEK
and IGFR inhibitors was dominated by the effect of the MEK inhibition, with about two thirds
of the differential phosphopeptides (96/149) being also regulated by MEK inhibitor alone.
Accordingly, differentially phosphorylated peptides in IMR32 are enriched in MAPK targets
(S15 Fig). In contrast, both MEK as well as IGFR inhibition induce strong alterations in the
phosphoproteome in N206 (S14 Fig), affecting both mTOR and MAPK signalling targets (S15
Fig), and the combination exhibits a synergistic effect (Fig 5B right). Overall, 24 differentially
phosphorylated sites in N206 show synergistic regulation, as defined by a significant deviation
of the combination from the sum of the individual treatment effects. Of these, 18 phosphosites
were synergistically down-regulated, and 7 sites showed up-regulation. Moreover, 11 of those
synergistically downregulated phosphosites are known or putative targets of the PI3K/AKT
signalling axis. This suggests that MEK/ERK signalling influences AKT signalling in a IGFR
dependend way. In contrast, only two sites showed synergy in IMR32 (Fig 5C). Among the
synergistically downregulated phospho-sites in N206 was S425 of the Eukaryotic translation
initiation factor 4B (EIF4B), a protein involved in regulation of translation and a known nexus
between AKT and MAPK signalling [30]. We performed a kinase substrate enrichment analy-
sis [31] to explore how the signalling networks were affected by the inhibitions (Fig 5D). For
IMR32 cells, this analysis showed a decreased phosphorylation of MEK and JAK targets and
an increased phosphorylation of ARAF and BRAF targets in response to MEK inhibition.
Interestingly, in combination with IGFR inhibition the RAF activation is partially reversed
whereas other kinase targets seem rather unaffected. Overall this indicates a feedback activa-
tion of RAF that does not totally compensate the loss of MEK activity. In N206 cells, the
response to MEK inhibition and the attenuation of the activation of RAF targets following
double inhibitor treatment is similar to the response in IMR32. However, in IMR32 cells IGFR
inhibitor treatment had little impact on the kinome whereas a massive down-regulation of tar-
gets of a range of kinases occurred in N206 cells, covering the PI3K/AKT/mTOR pathway
(SGK1–3,AKT1,p70S6K), MAPK pathway (p90RSK) and many members of the Protein
Kinase C Family. This suggests a central role of IGFR signalling on central growth and survival
pathways.
When we investigated the phosphorylation of components of the MAPK pathway more
closely, we found many RAF negative feedback/crosstalk sites to be down-regulated after MEK
inhibition (BRAF: T401, S750, T753; RAF1: S29, S642, S259) in both cell lines (Fig 5E). MEK1
S222/S226 phosphorylation is increased and pERK S204 decreased in both cell lines after MEK
inhibition, in line with corresponding measurements using bead-based ELISAs. Among those
down-regulated phosphosites that were only significant in the combination in N206 we
detected many MYCN-phosphosites, notably MYCN S62, which is regulated by MAPK via
CDK1 [32]. Interestingly, this loss of S62 phosphorylated MYCN is associated with reduced
MYCN levels (Fig 5F) despite the association of MYCN S62 with increased MYCN degrada-
tion [33]. The decreased detection of MYCN S62 might be a consequence of the loss of total
MYCN protein but is likely not causing this loss itself. This downregulation was observed in
IMR32 and N206 cells upon single inhibition (IGFRi for N206 and MEKi for both cell lines),
but only in N206 cells an even stronger downregulation could be observed upon double
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 11 / 26
Fig 5. Phosphoproteomics analysis reveals important variations in the response to combination treatment. A-B. Venn diagrams
showing the overlap in differentially regulated phosphosites Abetween IMR32 and N206 or Bbetween treatments for each cell line. C:
Phosphopeptides synergistically altered by MEKi+IGFRi combination (black outline) when compared to the sum of individual inhibitor
treatments. AKT, mTOR or P70S6K bona fide targets (bold font) and putative targets (italic font; top 5 predicted kinases by PhosphoNET
Kinase Predictor www.phosphonet.ca) are indicated. D: Kinase substrate enrichment score using PhosphoSitePlus annotations. E: Log-
fold change to DMSO for RAF/MAPK and MYCN phosphopeptides. C-E Black outline highlights significant changes in activity (limma
moderated t-test, FDR<5%) F-H: Relative levels compared to control of the total proteins levels for MYCN (F) and CCND1 (H)
measured with mass spectrometry and MYCN measured with Western blot (G).
https://doi.org/10.1371/journal.pcbi.1009515.g005
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 12 / 26
inhibition (Fig 5F). We confirmed these effects in Western blots for IMR32 and N206 cells
(Fig 5G), and also found downregulation of MYCN upon IGFRi as well as MEKi treatment
but no synergetic decrease after the combination treatment (Fig 5G). Another interesting pro-
tein that is regulated synergistically in N206 is Cyclin D1 (Fig 5H), a protein that is involved in
cell cycle progression and whose loss likely mediates MYCN loss. It should be noted that only
5 proteins (PHGDH, DERL1, AMPD3, ARHGEF16 and CCND1) were found differentially
affected with an FDR <10%, highlighting that on this time scale phospho-protein changes
dominated.
Taken together, the proteomics data is coherent with the model that MAPK signalling in
N206 is controlled by a dual feedback structure involving RAF and IGFR, whereas it is mainly
controlled by a RAF-mediated feedback in IMR32. It furthermore supports the notion that
treatment with MEK and IGFR inhibitors would show synergy in N206.
Vertical inhibition can break feedback-mediated resistance
Feedback regulation is often a central aspect for drug resistance that could be overcome by a
vertical inhibition strategy, where an inhibition of an upstream node prevents pathway reacti-
vation. Based on our models, we tested if the additional application of an inhibitor targeting
the feedback nodes (RAF and IGFR) would sensitise resistant cells toward MEK inhibition
(Fig 6A). We quantified growth reduction after inhibiting IMR32, KELLY and N206 with dif-
ferent dose combinations of inhibitors against MEK (AZD6244), IGFR (AEW541) and RAF
(LY3009120) (Fig 6B). In agreement with our model predictions of strong IGFR-mediated
feedback in N206 (Fig 4D), there was a strong synergistic effect as evaluated by the Bliss score
[34] of the combination of MEK and IGFR inhibitions on growth in N206 but little in KELLY
or IMR32 (Fig 6C and see S16 Fig for Loewe score).
When trying to overcome the model-derived strong ERK-RAF feedback found in all three
cell lines with a combination of MEK and RAF inhibition we only found a synergistic effect for
two of the three cell lines (N206 and KELLY), whereas IMR32 remained resistant and no syn-
ergy could be detected. We hypothesised that this observed resistance in IMR32 might be
either because the vertical inhibition by MEKi and RAFi was molecularly not effective or that
IMR32 might no longer depend on ERK signalling for survival and growth. To distinguish the
former from the latter we decided to compare model simulation and measurements for pertur-
bation effects of selected inhibitor combinations on pMEK and pERK in IMR32 and KELLY
cells.
Based on the model simulations, in both cell lines the vertical inhibition of MEK + RAF
inhibitor was predicted to suppress MAPK signalling much stronger than MEK inhibitor
alone or in combination with an ERK inhibitor. Moreover, the suppressive effect was predicted
to be even more profound in IMR32 than in KELLY (Fig 6D left). We then measured the effect
on pMEK and pERK of MEK inhibitor alone and in combination with the RAF inhibitor
LY3009120 or ERK inhibitor SCH772984 (Fig 6D right). The measurements qualitatively sup-
ported the model simulations showing that RAF inhibitor suppressed MEK feedback activa-
tion by AZD6244, and that this suppression is stronger in IMR32. Addition of the ERK
inhibitor neither suppressed this feedback activation nor could it decrease ERK phosphoryla-
tion more than RAF inhibition, as also predicted by the model. This suggests that in agreement
with the model simulations the combination of RAFi and MEKi is most effective in IMR32 to
effectively suppress ERK activation and feedback-mediated re-activation. However, since the
growth is least affected by this combination IMR32 seems not to depend on ERK activity.
As both KELLY and N206 have strong multi-layered feedbacks (Fig 4D), we also tried triple
combinations of IGFRi, RAFi and MEKi. We observed that only in KELLY, triple inhibitor
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 13 / 26
treatment seems to have an additional benefit compared to the best combination of two inhibi-
tors. (S17 Fig and S2 File).
Discussion
Neuroblastoma is a complex disease with distinct subtypes that display radically different out-
comes, ranging from spontaneous regression in low-risk groups to only 50% survival of
patients in the high risk neuroblastoma group. Mutations in RAS/MAPK signalling are a hall-
mark of high risk neuroblastoma, and also define a subgroup of patients with ultra-high-risk
neuroblastoma and an even worse survival. Therefore targeted treatment might be a valid
strategy to treat those patients. However, response to MEK inhibitors are very variable, and it
is thus important to understand mechanisms of resistance and how to circumvent these.
Fig 6. AZD6244 resistant cell lines can be sensitised with combined inhibition with the IGFR inhibitor AEW541 or the RAF inhibitor LY3009120. A: Model-
inferred targeting strategy of dual inhibition. B: Growth inhibition measurements for various combinations of the MEK inhibitor AZD6244 with the RAF inhibitor
LY3009120 or the IGFR inhibito AEW541. Values over 100 indicate cell death. n = 2. C: Bliss synergy corresponding to the measurements in B. D: LEFT: Model
predictions of pERK and pMEK activity for MEK inhibition alone and in combination with inhibition of upstream kinase RAF or downstream kinase ERK for KELLY
and IMR32. Values are log-fold changes to IGF1 condition with inhibitor strength set to -1. D: RIGHT: pERK and pMEK plex measurements in KELLY and IMR32 after
90min treatment of the MEK inhibitor AZD6244 in combination with either DMSO, SCH772984 (ERKi, 10μM) or LY3009120 (RAFi, 5μM) in cells grown with 10%
FCS. Values are log-fold change to FCS medium condition.
https://doi.org/10.1371/journal.pcbi.1009515.g006
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 14 / 26
In this work, we explored how a more quantitative understanding of signalling can be used
to design combinatorial treatments to counteract drug resistance. We used a panel of deeply
profiled cell lines representing high risk neuroblastoma and showed that the response to MEK
inhibitors is variable, with some cell lines responding at low doses in the nM range, whereas
others are highly resistant. By using signalling perturbation-response data, we characterised
the signalling network surrounding MAPK. Analysis of that perturbation data with the model-
ling framework of modular response analysis unveiled that MAPK signalling is controlled by a
multi-layered feedback with variable strength. A central finding was that MEK-inhibitor sensi-
tive cells are controlled by low feedbacks within the MAPK cascade, whereas a subset of resis-
tant cell lines shows strong multi-layered feedbacks that may be causal for resistance.
Simulation of cell-line specific models suggested that different combinations of inhibitors can
be used to overcome resistance, and experiments could confirm these predictions in two out of
three cell lines.
Our work highlights that systematic perturbation data are a powerful source to probe intra-
cellular signalling pathways. The connectivity of signalling pathways implies that minor quan-
titative alterations of the network can lead to many changes in response, not all of which alter
the phenotype. In this work, we saw that multivariate analysis of the perturbation data alone
was not fruitful to separate cell lines with respect to their drug sensitivity. In contrast, integra-
tion of data by models highlighted that variations of only a few links is enough to explain the
differences between those cell lines. Modelling was therefore key to integrate the data and to
unveil feedback loops as potential sources of resistance.
In our work we used a maximum likelihood version of MRA, but there are multiple other
methods that might be suited to reconstruct semi-quantitative signalling networks from per-
turbation data. [35] proposed a bayesian variant which overcomes the linearity assumption of
MRA using chemical kinetics to guide the inference and fuzzy-logic models such as used by
[36] also show good performance to reconstruct network topology from signalling data. How-
ever getting quantitative values for the interactions between components of a signalling net-
works from a small set of perturbations requires MRA variants [21,37] or necessitates time-
resolved perturbation data which limits the number of perturbations that can be studied simul-
taneously [38]. While boolean models are very good strategies to model large signalling net-
works and complex synergies [39], they would be unable to capture quantitative differences in
feedback regulation, which are the key resistance mechanisms uncovered in this work.
Drug resistance to targeted therapies have been attributed to negative feedback loops in
multiple tumours. Most importantly, sensitivity to MEK inhibitors is strongly influenced by a
pathway-intrinsic feedback, where ERK phosphorylates RAF at multiple sites [1517]. This
feedback has been shown to be very strong in epithelial cells leading to pathway robustness
[16], which can be overcome by vertical inhibition of RAF [17]. Another mode of feedback
regulation is the inhibition of receptors by pathways. An example is the inhibitory regulation
of EGFR by the MAPK pathway [13,14]. When inhibiting MAPK signalling by MEK or RAF
inhibitors, this feedback leads to hyper-sensitisation of EGFR, which in turn reactivates MAPK
signalling and additionally activates other downstream pathways such as PI3K/AKT signalling.
Also in this case vertical inhibition can help to overcome this mode of resistance, by co-target-
ing the MAPK pathway and the upstream receptor.
Our modelling analysis suggested that some neuroblastoma cell lines possess two major lay-
ers of feedback in MAPK signalling. One of these feedbacks is pathway-intrinsic (from ERK to
RAF) and one is a feedback to the IGF receptor. Interestingly, different cell lines show different
relative strength of feedbacks from ERK to RAF and IGFR, and simulations show that those
require different strategies for vertical inhibition. For the cell line KELLY, modelling unveiled
an extremely strong negative feedback from ERK to RAF. This suggests that a combination of
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 15 / 26
MEK and RAF inhibitor will be more potent than a combination of MEK and IGFR inhibitor.
In contrast, in the cell line N206, both feedbacks have similar strength, suggesting that both
combinations might be potent. In line with these predictions, experiments showed that in
KELLY indeed the combination of MEK and RAF inhibitors is much more potent to reduce
growth compared to the combination of MEK and IGFR. In contrast, in N206 both combina-
tions reduce growth.
Our phospho-proteomics analysis shows that the combination of MEK and IFGR also has dif-
ferent effects in the two cell lines: Whereas it shows clearly synergistic effects of the combination
in N206, there is no sign of synergy in IMR32. By aggregating the phosphoproteome to kinase
activities using kinase enrichment scores, one can also get insights into the re-wiring of signal-
ling after perturbation. In our case, it clearly shows how the re-activation of RAF after MEK
inhibition is inhibited by the treatment with IGFR inhibitors, and IGFR and MEK inhibitors
synergize in reducing AKT activity in N206. The phosphoproteome also showed that the dual
treatment of IGFR and MEK manifests itself in synergistic downregulation of important pro-
teins that are regulated by convergent signalling of MEK and AKT, such as MYCN and EIF4B.
Interestingly, a third resistant cell line, IMR32, showed no response in growth to MEK
inhibitor in vertical combination with either RAF and/or IGFR inhibitor on growth, even
though it’s cellular ERK signalling was strongly responsive. This highlights that cancer cells
might lose ERK-mediated cell cycle control, suggesting that coupling of cellular phenotype to
signalling pathways is not necessarily strict [40,41]. To more directly model changes on cellu-
lar phenotypes such as growth or viability, models of signalling would need to be connected to
phenotypic readouts [42]. In addition, it might be beneficial to include downstream readouts
such as cyclin levels or CDK activation that are more directly involved in cell cycle progression
and can be deregulated in cancer [43,44]. Our model attributes signalling differences between
cell lines to an apparent feedback from MAPK signalling to IGFR and/or RAF. However, our
model is too coarse-grained to distinguish feedback regulation from other, potentially non-lin-
ear mechanisms of cross-talk. Ultimately, only mechanistic studies that e.g. include the use of
cell lines that have mutant feedback will unveil if the feedbacks are responsible for the observed
signalling phenotypes and inhibitor synergies. It should be also pointed out that our measure-
ments only encompass one time point and that later dynamics of the MAPK pathway, such as
transcriptional feedbacks, could also explain IMR32 resistance to vertical inhibition.
In summary, our results show that a quantitative understanding of differences in signalling
networks can be very helpful to rationalize resistance, and to derive effective treatments.
Future work should investigate if those feedback mechanisms exist in tumours in vivo and
whether they could explain relapses. Our description of the wiring of the RAS/MAPK pathway
in neuroblastoma will support the design of clinical trials using combinatorial treatments to
prevent or overcome therapy resistance. In addition, the framework described here could be
used to analyse signalling in tumours of individual patients While it will be technically chal-
lenging to assess signalling network responses in tumour patients, ex vivo cultures—so-called
avatars—could be an option [45,46]. We envision that learning features of robustness and vul-
nerability of tumours from signalling models on cell line panels might greatly reduce the
required set of perturbations in those avatars that are sufficient to inform a model, and allow
reliable stratification and prediction of treatment options.
Materials and methods
Cell lines
The neuroblastoma cell lines were obtained by courtesy of the Deubzer lab (Charite
´, Berlin) as
part of the Terminate-NB consortium. The identity of the cell lines was confirmed with STR
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 16 / 26
profiling (see S2 Table), which were generated by Eurofins Cell Line Authentification Test and
matched with the Cellosaurus STR similarity research tool [47]. All cell lines were grown in
DMEM (Gibco, Life Technologies) with 3.5 g/L glucose (Sigma), 5 mM glutamine (Gibco, Life
Technologies) and 10% FCS (Pan Biotech).
Whole exome sequencing
DNA was extracted from the human neuroblastoma cell lines (see above), using the Nucleos-
pin Tissue kit (Macherey-Nagel) according to the manufacturer’s protocol. From the DNA,
libraries for whole-exome sequencing were prepared using the SureSelect Human All Exon V7
kit (Agilent) and the Illumina TruSeq Exome kit. The libraries were sequenced on Illumina
HiSeq 4000 and Illumina NovaSeq 6000 sequencers. The read sequences and base quality
scores were demultiplexed and stored in Fastq format using the Illumina bcl2fastq software
v2.20. Adapter remnants and low-quality read ends were trimmed off using custom scripts.
The quality of the sequence reads was assessed using the FastQC software. Reads were aligned
to the human genome, assembly GRCh38, using the bwa mem software version 0.7.10 [48],
and duplicate read alignments were removed using samblaster version 0.1.24 [49]. Copy-num-
ber alterations were determined using cnvkit version 0.1.24 [50]. Single-nucleotide variants
(SNVs) were identified using strelka version 2.9.10 [51]. Afterwards, potential germline vari-
ants were filtered out by excluding all SNVs that had also been observed in at least 1% of sam-
ples in cohorts of healthy individuals, namely the 1000 Genomes Project [52] and the NHLBI
GO Exome Sequencing Project [53] cohorts. The raw data are available on ENA under the
accession number PRJEB40670.
RNA sequencing
The cell lines were sequenced in 3 separate batches. The IMR32, KELLY, SKNAS, LAN6,
NBEBC1 cell lines were prepared in triplicate, using a paired-end stranded protocol with 2x75
cycles per fragment and 2 more cell lines (NGP, SKNSH) were prepared in duplicate, using a
paired-end stranded protocol with 2x150 cycles. Two more libraries (CHP212 and N206) were
prepared using a paired-end stranded protocol with 2x75 cycles per fragment.
Raw sequencing data were rigorously checked for quality using FastQC. The reads were
aligned to the human genome GRCh38 (without patches or haplotypes) and the GENCODE
transcript annotation set using the STAR aligner software [54]. The read counts per gene were
obtained using the featurecounts [55] method from the subread software package. The raw
data are available on ENA under the accession number PRJEB40670.
Drug sensitivity assay
Cells grown for 1 day in full medium were treated with the indicated drugs in 4 different con-
centrations (0.1, 1, 10 and 100 μMFig 1B) along with the corresponding DMSO controls on
the same plate. The growth of the cells was tracked by phase contrast imaging for 72h with 4
images per well taken every 2h using the Incucyte Zoom instrument (Essen BioScience) and
the confluency estimated using the Incucyte Zoom Analysis software (Essen BioScience). The
growth rate was estimated with a linear fit on the log-transformed confluency, and the IC50
was determined by fitting a sigmoid of the form:
V¼1
1þexpð logðCÞ þ IC50Þ  S
to normalised growth rates (implemented in https://github.com/MathurinD/drugResistance).
Vis the growth rate relative to DMSO control, Cis the concentration and the parameters IC50
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 17 / 26
and slope Sare fitted. See S1 Table for the fitted parameters and S2 File for the raw data and
analysis scripts, as well as S18 and S19 Figs for example images.
Synergy estimation
For the synergy assay, cells seeded the day before were treated with different concentrations of
AZD6244 (0.1, 1, 10, 30 and 50 μM, Selleck Chemicals) in combination with NVP-AEW541
(0.1, 0.3, 1, 3 and 10 μM, Cayman Chemical) or LY3009120 (0.1, 0.3, 1, 3 and 15 μM, Selleck
Chemicals). The synergy scores were determined using the R package synergyfinder [56] with
the relative growth rates thresholded between 0 and 1 as input (0 meaning no growth or cell
death and 1 meaning growth as fast as the DMSO control).
Perturbation assay
Cells were seeded in 24 well plates and grown for 2 days in full medium followed by 24h in
FCS-free medium before treatment with the same concentrations of ligands and inhibitors.
All inhibitors were dissolved in DMSO and cells were treated for 90 minutes at the follow-
ing concentrations: GDC0941 (1 μM, Selleck Chemicals), AZD6244/Selumetinib (10 μM, Sell-
eck Chemicals), MK2206 2HCl (10 μM, Selleck Chemicals), Rapamycin (10 μM, Selleck
Chemicals), Sorafenib (10 μM, Selleck Chemicals), GS-4997 (10 μM, Selleck Chemicals) and
TAE684 (10 μM, Selleck Chemicals).
The cells were treated for 30 minutes (60 minutes after inhibitor treatment) with ligands in
a 0,1% PBS/BSA carrier solution at the following concentrations: EGF (25 ng/mL, Peprotech),
PDGF (10 ng/mL, Peprotech), NGF (50 ng/mL, Peprotech) and IGF1 (100 ng/mL, Peprotech).
The cells were then lysed using BioRad Bio-Plex Cell Lysis Kit and measured using the Bio-
Plex MAGPIX Multiplex Reader with a custom kit from ProtAtOnce with analytes p-cJUN
(S63), p-p38 (T180/Y182), p-AKT (S473), p-ERK1/2 (T202/Y204,T185/Y187), p-MEK1 (S217/
S221), p-S6K (T389) and p-RSK1 (S380). The p-RSK1 (S380) readout was discarded because of
a low dynamic range.
The same procedure and analytes were used for the other perturbation assays in this paper.
Refer to the main text for the exact inhibitors and concentrations used for each experiment.
Signalling models
The model for each cell line was fitted separately from the corresponding perturbation data
with the createModel function from the R package STASNet [21]. STASNet implements the
variation of Modular Response Analysis (MRA) described in [13] and [21] that implements a
dual effect of inhibitors as both a negative stimulus and a disruption of signal propagation.
Under the hypothesis of pseudo-steady-state and locally linear dependencies between nodes,
MRA models the response to a perturbation as
R¼  ~
rkSð1Þ
where R
ij
is the global response of node jafter perturbation of node i,~
rk
ij is the local response of
node jafter perturbation of node itaking into account the effect of inhibition of node k, and
S
ik
is the sensitivity of node ito perturbation k. The pAKT readout was systematically removed
if AKT inhibition was present because the AKT inhibitor MK2206 blocks AKT autophosphor-
ylation [57], i.e acts upstream of the AKT node, while STASNet expect inhibitors to act down-
stream of their annotated target.
We designed a literature network consisting of the MAPK and PI3K/AKT signalling path-
way as annotated in KEGG (https://www.genome.jp/kegg/pathway/hsa/hsa04010.html and
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 18 / 26
https://www.genome.jp/kegg-bin/show_pathway?hsa04151) with intermediate nodes sup-
pressed, the addition of the well documented ERK->RAF feedback and all receptors corre-
sponding to RTK. Each cell line was fitted first on the literature network then we extended the
networks independently using a greedy hill climbing approach, until no significant link could
be added. We then performed successive rounds of reduction to identify the non-significant
links. Most removed links relate to receptor connections. Only three connections not related
to receptors were removed during this procedure, each for one cell line only. To facilitate
model comparison, these links were ultimately retained in the model, as otherwise the model
parameters would not be comparable. Those models with final topology yielded similar values
for the inhibition parameters so we generated new models with those parameters fixed to the
mean value across all 6 models and re-fitted each cell line with inhibitor values fixed. With this
fitting strategy the links between models became directly comparable as the non identifiability
induced by the inhibitor parameters was removed (Fig 3A). The high pMEK responder cell
line models were fitted using the same procedure.
Western blot
Cells were grown to confluency for 3 days in full medium and treated with AEW541 10μM
and/or AZD6244 10μM or control DMSO for 4h then lysed using BioRad Bio-Plex Cell Lysis
Kit. The lysates were run for 3h at a constant 45 mA in 10% acrylamid gels and blotted for 45
minutes at 400 mA on nitrocellulose. The membrane were stained for total protein using
Pierce Reversible Protein Stain (Thermofischer 24580) and blocked for 30 minutes in 1:1 PBS:
Odyssey blocking buffer. The primary antibodies were incubated overnight at 4C one at a time
and the corresponding secondary during the following day for 2h at room temperature in 1:1
PBST/Odyssey. We used the following primary antibodies: pIGF1R beta
Y1135/Y1136
1:1000
(CST 3024), pAKT
S473
1:2000 (CST 4060), total MYCN 1:200 (Santa Cruz sc-53993) and
pMEK
S217/S221
1:1000 (CST 9154).
TMT (phospho-)proteomics
For the proteomics and phosphoproteomics cells were grown to confluency for 3 days in full
medium and treated with AEW541 10μM and/or AZD6244 10μM or control DMSO for 4h.
We used an adapted version of the TMT workflow [28]: samples were reduced, alkylated
and digested with a combination of LysC (Wako) and Trypsin (Promega) using the the single-
pot, solid-phase-enhanced sample preparation [58]. For each sample, an equal amount of pep-
tide was then chemically labelled with TMTpro reagents [29]. Samples were randomly assigned
to one of the first 15 TMT channels, while the 16th channel was composed of a superset of all
the samples to allow multi-plex normalisation. Equal amounts of the labelling reactions were
combined in two TMT16 plexes, desalted via SepPak columns (Waters) and fractionated via
high-pH fractionation [59] on a 96 minutes gradient from 3 to 55% acetonitrile in 5 mM
ammonium formate, each fraction collected for 1 minute then combined into 24 fractions.
From each fraction, an aliquot was used to measure the total proteome while the remaining
peptides were combined into 12 fractions and used as input for an immobilised metal affinity
chromatography using an Agilent Bravo system. For the total proteome analysis, peptides were
on-line fractionated on a multi-step gradient from 0 to 55% acetonitrile in 0.1% formic acid
prior injection in a QExactive HF-x mass spectrometer. Samples were acquired using a data
dependent acquisition strategy with MS1 scans from 350 to 1500 m/z at a resolution of 60 000
(measured at 200 m/z), maximum injection time (IT) of 10 ms and an automatic gain control
(AGC) target value of 3 ×10
6
. The top 20 most intense precursor ions with charges from +2 to
+6 were selected for fragmentation with an isolation window of 0.7 m/z. Fragmentation was
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 19 / 26
done in an HCD cell with a normalised collision energy of 30% and analysed in the detector
with a resolution of 45 000 (200 m/z), AGC target value of 10
5
, maximum IT of 86 ms. We
used the same parameters for phosphoproteome analysis with the exception of MS2 maximum
IT that was set to 240 ms.
The acquired raw files were analysed using MaxQuant v1.6.10.43 [60], with TMTpro tags
manually added as fixed modifications and used for quantitation The correction factors for
purity of isotopic labels was set according to vendor specification and minimum reporter pre-
cursor intensity fraction was set to 0.5. The resulting protein groups were filtered for potential
protein contaminants, protein groups only identified via peptides decorated with modification
or hits in the pseudo-reverse database used for FDR control. The resulting intensities of each
sample channel were normalised to the intensity of the 16th reference channel, then median-
centered and normalised according to the median-absolute deviation. Identified phosphopep-
tides were similarly filtered, with the exception of filtering based on modified sites, and nor-
malised using the same strategy.
Differentially expressed phosphopeptides were called using the limma package [61] with a
false discovery rate of 0.05 on treatment minus control constrasts. Synergies were computed
using a contrast fit of the combination minus the sum of single treatments. Kinase substrate
activity was implemented in R using the ratio of the mean z-score as described in [31] and
computed for kinase-substrate sets from PhosphoSitePlus [62]. The normalised intensities and
scripts used for the analysis can be found at https://itbgit.biologie.hu-berlin.de/dorel/
phosphoproteomics_tnb_perturbations.
Supporting information
S1 Fig. Annotated IC50 of all measured drugs and cell lines with MYCN and TERT expres-
sion information.
(PDF)
S2 Fig. IC50 and mutations. t-test comparison of the IC50 in mutant (Mut) versus wild type
(WT) for RAS/P53 and associated genes with mutation frequency between 30% and 70% in
our panel.
(PDF)
S3 Fig. IC50 and gene expression. Top correlation between IC50 and mRNA transcript per
million for the 1000 most variable genes (top, adjusted p>0.93) and GO signal transduction
genes (bottom, adjusted p>0.94).
(PDF)
S4 Fig. PCA on the 1000 most variable genes. Principal component analysis of the 1000 most
variable genes. All components up to the first one explaining less than 10% of the variance are
shown.
(PDF)
S5 Fig. PCA on the signal transduction genes. Principal component analysis of the 5262 sig-
nal transduction genes. All components up to the first one explaining less than 10% of the vari-
ance are shown.
(PDF)
S6 Fig. Selected gene-drug correlations. Correlation of NF1 expression with AZD6244 IC50
and ALK expression with TAE684 IC50.
(PDF)
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 20 / 26
S7 Fig. Receptors RNA expression (TPM).
(PDF)
S8 Fig. Adaptros and ERBB receptor family RNA expression (TPM).
(PDF)
S9 Fig. Perturbation data PCA. Pair-plot of the principal components from the perturbation
in 2. All components up to the first one explaining less than 10% of the variance are shown.
(PDF)
S10 Fig. Perturbation PCA loadings. Main loadings in the first 3 principal components of the
perturbation data PCA. Colors correspond to the component for which the condition has the
highest absolute weight. Table indicates the weight for the top 10 conditions of the first 3 prin-
cipal components.
(PDF)
S11 Fig. Models qq-plots. Quantile-quantile plots of the initial models using the (A) literature
topology and (B) the final model after extension.
(PDF)
S12 Fig. Correlation of signaling and receptor expression. Correlation between the fitted
path value from ligands to readouts and the expression of the matching receptor or receptor
family. IGFRsum and PDGFRsum are the sum of the isoforms expression for IGFR and
PDGFR respectively.
(PDF)
S13 Fig. AZ6244 perturbation versus IC50. Linear model fit of AZD6244 IC50 response to
perturbations including AZD6244. Points are indenpendent replicates, n = 2.
(PDF)
S14 Fig. Differential phosphopetides. Differentially measured phosphopeptides in IMR32
and N206 after 4h inhibition (FDR <0.05, n = 3) classified by treatment(s) where the phos-
phosite is differentially expressed.
(PDF)
S15 Fig. KEGG enrichment of phosphoproteomics. KEGG enrichment of unique genes cor-
responding to phosphopeptides differentially expressed after MEKi, IGFRi or MEKi+IGFRi
treatment in (A) IMR32, (B) N206 or (C) both strictly. Enrichment was computed using the R
package enrichKEGG.
(PDF)
S16 Fig. Loewe synergy. Loewe synergy for the combinations of AD6244 with (A) AEW541 or
(B) LY3009120 shown in 6B. Synergy scores were computed with the R package synergyfinder.
Positive scores indicate synergy, negative scores indicate antagonism.
(PDF)
S17 Fig. Viability to combination treatments. Relative viability of IMR32, KELLY and N206
after treatment with AZD6244, AEW541 and RO5126766 alone or in combination. Con-
fluency was tracked for 72h use the Incucyte Zoom. Growth rate was fitted to the confluency
curve and normalised to the average growth rate of the corresponding DMSO controls. black
crosses indicate the mean value for each cell line for the corresponding treatment.
(PDF)
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 21 / 26
S18 Fig. Incucyte CHP212. Incucyte image of the AZD6244-sensitive cell line CHP212 imme-
diately after and 72h after DMSO or AZD6244 treatment.
(PDF)
S19 Fig. Incucyte IMR32. Incucyte image of the AZD6244-resistant cell line IMR32 immedi-
ately after and 72h after DMSO or AZD6244 treatment.
(PDF)
S1 Table. IC50 data.
(CSV)
S2 Table. STR profiling results.
(XLSX)
S1 File. Sequencing data and related analysis scripts.
(ZIP)
S2 File. Dose response data and related analysis scripts.
(ZIP)
S3 File. Model data and fitting summary.
(ZIP)
Acknowledgments
We thank Aleixandria McGearey for technical assistance with preparing the whole-exome
sequencing libraries, Martha Hergesekke for help in cell culture, as well as Jasmin Wu¨nschel
for providing the cell lines.
Author Contributions
Conceptualization: Bertram Klinger, Nils Blu¨thgen.
Data curation: Mathurin Dorel, Tommaso Mari, Joern Toedling, Eric Blanc, Clemens Mes-
serschmidt, Michal Nadler-Holly, Matthias Ziehm, Falk Hertwig.
Funding acquisition: Angelika Eggert, Matthias Selbach, Nils Blu¨thgen.
Investigation: Mathurin Dorel, Tommaso Mari.
Methodology: Bertram Klinger, Anja Sieber.
Project administration: Nils Blu¨thgen.
Resources: Matthias Ziehm, Falk Hertwig, Johannes H. Schulte.
Software: Mathurin Dorel, Bertram Klinger, Nils Blu¨thgen.
Supervision: Bertram Klinger, Dieter Beule, Angelika Eggert, Johannes H. Schulte, Matthias
Selbach, Nils Blu¨thgen.
Visualization: Mathurin Dorel.
Writing – original draft: Mathurin Dorel.
Writing – review & editing: Bertram Klinger, Nils Blu¨thgen.
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 22 / 26
References
1. De Bernardi B, Nicolas B, Boni L, Indolfi P, Carli M, Di Montezemolo LC, et al. Disseminated neuroblas-
toma in children older than one year at diagnosis: Comparable results with three consecutive high-dose
protocols adopted by the Italian Co-Operatiye Group for Neuroblastoma. Journal of Clinical Oncology.
2003; 21(8):1592–1601. https://doi.org/10.1200/JCO.2003.05.191 PMID: 12697885
2. Maris JM, Hogarty MD, Bagatell R, Cohn SL. Neuroblastoma. Lancet. 2007; 369(9579):2106–2120.
https://doi.org/10.1016/S0140-6736(07)60983-0 PMID: 17586306
3. Kyo Y, Tanaka T, Hayashi K, Iehara T, Kaneko M, Hosoi H, et al. Identification of therapy-sensitive and
therapy-resistant neuroblastoma subtypes in stages III, IVs and IV. Cancer Letters. 2011; 306(1):27–
33. https://doi.org/10.1016/j.canlet.2011.02.016 PMID: 21486686
4. Peifer M, Hertwig F, Roels F, Dreidax D, Gartlgruber M, Menon R, et al. Telomerase activation by geno-
mic rearrangements in high-risk neuroblastoma. Nature. 2015; 526(7575):700–704. https://doi.org/10.
1038/nature14980 PMID: 26466568
5. Barone G, Anderson J, Pearson ADJ, Petrie K, Chesler L. New strategies in neuroblastoma: Therapeu-
tic targeting of MYCN and ALK. Clinical cancer research: an official journal of the American Association
for Cancer Research. 2013; 19(21):5814–21. https://doi.org/10.1158/1078-0432.CCR-13-0680 PMID:
23965898
6. Ackermann S, Cartolano M, Hero B, Welte A, Kahlert Y, Roderwieser A, et al. A mechanistic classifica-
tion of clinical phenotypes in neuroblastoma. Science. 2018; 362(6419):1165–1170. https://doi.org/10.
1126/science.aat6768 PMID: 30523111
7. Eleveld TF, Oldridge DA, Bernard V, Koster J, Daage LC, Diskin SJ, et al. Relapsed neuroblastomas
show frequent RAS-MAPK pathway mutations. Nature genetics. 2015; 47(8):864–71. https://doi.org/10.
1038/ng.3333 PMID: 26121087
8. Bresler SC, Weiser DA, Huwe PJ, Park JH, Krytska K, Ryles H, et al. ALK Mutations Confer Differential
Oncogenic Activation and Sensitivity to ALK Inhibition Therapy in Neuroblastoma. Cancer Cell. 2014;
26(5):682–694. https://doi.org/10.1016/j.ccell.2014.09.019 PMID: 25517749
9. Hallberg B, Palmer RH. The role of the ALK receptor in cancer biology. Annals of Oncology. 2016; 27
(suppl3):iii4–iii15. https://doi.org/10.1093/annonc/mdw301 PMID: 27573755
10. Pugh TJ, Morozova O, Attiyeh EF, Asgharzadeh S, Wei JS, Auclair D, et al. The genetic landscape of
high-risk neuroblastoma. Nature Genetics. 2013; 45(3):279–284. https://doi.org/10.1038/ng.2529
PMID: 23334666
11. Johnsen JI, Dyberg C, Fransson S, Wickstro
¨m M. Molecular mechanisms and therapeutic targets in
neuroblastoma. Pharmacological Research. 2018; 131:164–176. https://doi.org/10.1016/j.phrs.2018.
02.023 PMID: 29466695
12. Britschgi A, Andraos R, Brinkhaus H, Klebba I, Romanet V, Mu¨ller U, et al. JAK2/STAT5 Inhibition Cir-
cumvents Resistance to PI3K/mTOR Blockade: A Rationale for Cotargeting These Pathways in Meta-
static Breast Cancer. Cancer Cell. 2012; 22(6):796–811. https://doi.org/10.1016/j.ccr.2012.10.023
PMID: 23238015
13. Klinger B, Sieber A, Fritsche-Guenther R, Witzel F, Berry L, Schumacher D, et al. Network quantifica-
tion of EGFR signaling unveils potential for targeted combination therapy. Molecular Systems Biology.
2013; 9(1):673. https://doi.org/10.1038/msb.2013.29 PMID: 23752269
14. Prahallad A, Sun C, Huang S, Di Nicolantonio F, Salazar R, Zecchin D, et al. Unresponsiveness of
colon cancer to BRAF(V600E) inhibition through feedback activation of EGFR. Nature. 2012; 483
(7387):100–104. https://doi.org/10.1038/nature10868 PMID: 22281684
15. Friday BB, Yu C, Dy GK, Smith PD, Wang L, Thibodeau SN, et al. BRAF V600E disrupts AZD6244-
induced abrogation of negative feedback pathways between extracellular signal-regulated kinase and
Raf proteins. Cancer Research. 2008; 68(15):6145–6153. https://doi.org/10.1158/0008-5472.CAN-08-
1430 PMID: 18676837
16. Fritsche-Guenther R, Witzel F, Sieber A, Herr R, Schmidt N, Braun S, et al. Strong negative feedback
from Erk to Raf confers robustness to MAPK signalling. Molecular Systems Biology. 2011; 7(1):489–
489. https://doi.org/10.1038/msb.2011.27 PMID: 21613978
17. Sturm OE, Orton R, Grindlay J, Birtwistle M, Vyshemirsky V, Gilbert D, et al. The mammalian MAPK/
ERK pathway exhibits properties of a negative feedback amplifier. Science Signaling. 2010; 3(153):
ra90–undefined. https://doi.org/10.1126/scisignal.2001212 PMID: 21177493
18. Klinger B, Blu¨thgen N. Consequences of feedback in signal transduction for targeted therapies. Bio-
chemical Society Transactions. 2014; 42(4):770–775. https://doi.org/10.1042/BST20140130 PMID:
25109956
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 23 / 26
19. Woodfield SE, Zhang L, Scorsone KA, Liu Y, Zage PE. Binimetinib inhibits MEK and is effective against
neuroblastoma tumor cells with low NF1 expression. BMC cancer. 2016; 16:172. https://doi.org/10.
1186/s12885-016-2199-z PMID: 26925841
20. Kiessling MK, Curioni-Fontecedro A, Samaras P, Lang S, Scharl M, Aguzzi A, et al. Targeting the
mTOR complex by everolimus in NRAS mutant neuroblastoma. PLoS ONE. 2016; 11(1):e0170851–
undefined. https://doi.org/10.1371/journal.pone.0147682 PMID: 26821351
21. Dorel M, Klinger B, Gross T, Sieber A, Prahallad A, Bosdriesz E, et al. Modelling signalling networks
from perturbation data. Bioinformatics. 2018; 34(23):4079–4086. https://doi.org/10.1093/bioinformatics/
bty473
22. Finch AR, Caunt CJ, Perrett RM, Tsaneva-Atanasova K, McArdle CA. Dual specificity phosphatases 10
and 16 are positive regulators of EGF-stimulated ERK activity: Indirect regulation of ERK signals by
JNK/p38 selective MAPK phosphatases. Cellular Signalling. 2012; 24(5):1002–1011. https://doi.org/10.
1016/j.cellsig.2011.12.021 PMID: 22245064
23. Hood FE, Klinger B, Newlaczyl AU, Sieber A, Dorel M, Oliver SP, et al. Isoform-specific Ras signaling is
growth factor dependent. Molecular Biology of the Cell. 2019; 30(9):1108–1117. https://doi.org/10.
1091/mbc.E18-10-0676 PMID: 30785867
24. Yang HW, Shin MG, Lee S, Kim JR, Park WS, Cho KH, et al. Cooperative Activation of PI3K by Ras
and Rho Family Small GTPases. Molecular Cell. 2012; 47(2):281–290. https://doi.org/10.1016/j.molcel.
2012.05.007 PMID: 22683270
25. Corcoran RB, Ebi H, Turke AB, Coffee EM, Nishino M, Cogdill AP, et al. EGFR-mediated reactivation of
MAPK signaling contributes to insensitivity of BRAF-mutant colorectal cancers to RAF inhibition with
vemurafenib. Cancer Discovery. 2012; 2(3):227–235. https://doi.org/10.1158/2159-8290.CD-11-0341
PMID: 22448344
26. Rozengurt E, Soares HP, Sinnet-Smith J. Suppression of feedback loops mediated by pi3k/mtor
induces multiple overactivation of compensatory pathways: An unintended consequence leading to
drug resistance. Molecular Cancer Therapeutics. 2014; 13(11):2477–2488. https://doi.org/10.1158/
1535-7163.MCT-14-0330 PMID: 25323681
27. Lake D, Corrêa SAL, Mu¨ller J. Negative feedback regulation of the ERK1/2 MAPK pathway; 2016. Avail-
able from: http://link.springer.com/10.1007/s00018-016-2297-8.
28. Mertins P, Tang LC, Krug K, Clark DJ, Gritsenko MA, Chen L, et al. Reproducible workflow for multi-
plexed deep-scale proteome and phosphoproteome analysis of tumor tissues by liquid chromatogra-
phy-mass spectrometry. Nature Protocols. 2018; 13(7):1632–1661. https://doi.org/10.1038/s41596-
018-0006-9 PMID: 29988108
29. Li J, Van Vranken JG, Pontano Vaites L, Schweppe DK, Huttlin EL, Etienne C, et al. TMTpro reagents:
a set of isobaric labeling mass tags enables simultaneous proteome-wide measurements across 16
samples. Nature Methods. 2020; 17(4):399–404. https://doi.org/10.1038/s41592-020-0781-4 PMID:
32203386
30. Shahbazian D, Roux PP, Mieulet V, Cohen MS, Raught B, Taunton J, et al. The mTOR/PI3K and
MAPK pathways converge on eIF4B to control its phosphorylation and activity. TheEMBO Journal.
2006; 25(12):2781–2791. https://doi.org/10.1038/sj.emboj.7601166 PMID: 16763566
31. Casado P, Rodriguez-Prados JC, Cosulich SC, Guichard S, Vanhaesebroeck B, Joel S, et al. Kinase-
substrate enrichment analysis provides insights into the heterogeneity of signaling pathway activation in
leukemia cells; 2013. Available from: http://www.ncbi.nlm.nih.gov/pubmed/23532336.
32. Sjostrom SK, Finn G, Hahn WC, Rowitch DH, Kenney AM. The Cdk1 complex plays a prime role in reg-
ulating N-myc phosphorylation and turnover in neural precursors. Developmental Cell. 2005; 9(3):327–
338. https://doi.org/10.1016/j.devcel.2005.07.014 PMID: 16139224
33. Gustafson WC, Meyerowitz JG, Nekritz EA, Chen J, Benes C, Charron E, et al. Drugging MYCN
through an Allosteric Transition in Aurora Kinase A. Cancer Cell. 2014; 26(3):414–427. https://doi.org/
10.1016/j.ccr.2014.07.015 PMID: 25175806
34. Tang J, Wennerberg K, Aittokallio T. What is synergy? The Saariselka
¨agreement revisited; 2015. Avail-
able from: http://journal.frontiersin.org/article/10.3389/fphar.2015.00181.
35. Oates CJ, Hennessy BT, Lu Y, Mills GB, Mukherjee S. Network inference using steady-state data and
goldbeter-koshland kinetics. Bioinformatics. 2012; 28(18):2342–2348. https://doi.org/10.1093/
bioinformatics/bts459 PMID: 22815361
36. Terfve CDA, Wilkes EH, Casado P, Cutillas PR, Saez-Rodriguez J. Large-scale modelsof signal propa-
gation in human cells derived from discovery phosphoproteomic data. Nature Communications. 2015; 6
(1):1–11. https://doi.org/10.1038/ncomms9033 PMID: 26354681
37. Santra T, Kolch W, Kholodenko BN. Integrating Bayesian variable selection with Modular Response
Analysis to infer biochemical network topology. BMC systems biology. 2013; 7:57. https://doi.org/10.
1186/1752-0509-7-57 PMID: 23829771
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 24 / 26
38. Invergo BM, Beltrao P. Reconstructing phosphorylation signalling networks from quantitative phospho-
proteomic data. Essays in Biochemistry. 2018; 62(4):525–534. https://doi.org/10.1042/EBC20180019
PMID: 30072490
39. Niederdorfer B, Toure
´V, Vazquez M, Thommesen L, Kuiper M, Lægreid A, et al. Strategies to Enhance
Logic Modeling-Based Cell Line-Specific Drug Synergy Prediction. Frontiers in Physiology. 2020;
11:862. https://doi.org/10.3389/fphys.2020.00862 PMID: 32848834
40. Cerezo A, Guadamillas MC, Goetz JG, Sa
´nchez-Perales S, Klein E, Assoian RK, et al. The Absence of
Caveolin-1 Increases Proliferation and Anchorage- Independent Growth by a Rac-Dependent, Erk-
Independent Mechanism. Molecular and Cellular Biology. 2009; 29(18):5046–5059. https://doi.org/10.
1128/MCB.00315-09 PMID: 19620284
41. Castro AF, Campos T, Babcock JT, Armijo ME, Martı
´nez-Conde A, Pincheira R, et al. M-Ras induces
Ral and JNK activation to regulate MEK/ERK-independent gene expression in MCF-7 breast cancer
cells. Journal of Cellular Biochemistry. 2012; 113(4):1253–1264. https://doi.org/10.1002/jcb.23458
PMID: 22121046
42. Korkut A, Wang W, Demir E, Aksoy BA, Jing X, Molinelli EJ, et al. Perturbation biology nominates
upstream-downstream drug combinations in RAF inhibitor resistant melanoma cells. eLife. 2015; 4:
e04640. https://doi.org/10.7554/eLife.04640 PMID: 26284497
43. Keyomarsi K, Pardee AB. Redundant cyclin overexpression and gene amplification in breast cancer
cells. Proceedings of the National Academy of Sciences of the United States of America. 1993; 90
(3):1112–1116. https://doi.org/10.1073/pnas.90.3.1112 PMID: 8430082
44. Sung WW, Lin YM, Wu PR, Yen HH, Lai HW, Su TC, et al. High nuclear/cytoplasmic ratio of cdk1
expression predicts poor prognosis in colorectal cancer patients. BMC Cancer. 2014; 14(1):951. https://
doi.org/10.1186/1471-2407-14-951 PMID: 25511643
45. Brandt R, Sell T, Lu¨then M, Uhlitz F, Klinger B, Riemer P, et al. Cell type-dependent differential activa-
tion of ERK by oncogenic KRAS in colon cancer and intestinal epithelium. Nature Communications.
2019; 10(1):2919. https://doi.org/10.1038/s41467-019-10954-y PMID: 31266962
46. Saez-Rodriguez J, Blu¨thgen N. Personalized signaling models for personalized treatments. Molecular
Systems Biology. 2020; 16(1):e9042–undefined. https://doi.org/10.15252/msb.20199042 PMID:
32129942
47. Robin T, Capes-Davis A, Bairoch A. CLASTR: The Cellosaurus STR similarity search tool—A precious
help for cell line authentication. International Journal of Cancer. 2019; 146(5):1299–1306. https://doi.
org/10.1002/ijc.32639 PMID: 31444973
48. Li H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. arXivorg.
2013;.
49. Faust GG, Hall IM. SAMBLASTER: fast duplicate marking and structural variant read extraction. Bioin-
formatics. 2014; 30(17):2503–2505. https://doi.org/10.1093/bioinformatics/btu314 PMID: 24812344
50. Talevich E, Shain AH, Botton T, Bastian BC. CNVkit: Genome-Wide Copy Number Detection and Visu-
alization from Targeted DNA Sequencing. PLOS Computational Biology. 2016; 12(4):e1004873.
https://doi.org/10.1371/journal.pcbi.1004873 PMID: 27100738
51. Kim S, Scheffler K, Halpern AL, Bekritsky MA, Noh E, Ka
¨llberg M, et al. Strelka2: fast and accurate call-
ing of germline and somatic variants. Nature Methods. 2018; 15(8):591–594. https://doi.org/10.1038/
s41592-018-0051-x PMID: 30013048
52. Auton A, Abecasis GR, Altshuler DM, Durbin RM, Bentley DR, Chakravarti A, et al. A global reference
for human genetic variation; 2015. Available from: http://www.nature.com/articles/nature15393.
53. Fu W, O’Connor TD, Jun G, Kang HM, Abecasis G, Leal SM, et al. Analysis of 6,515 exomes reveals
the recent origin of most human protein-coding variants. Nature. 2013; 493(7431):216–220. https://doi.
org/10.1038/nature11690 PMID: 23201682
54. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: Ultrafast universal RNA-
seq aligner. Bioinformatics. 2013; 29(1):15–21. https://doi.org/10.1093/bioinformatics/bts635 PMID:
23104886
55. Liao Y, Smyth GK, Shi W. FeatureCounts: An efficient general purpose program for assigning sequence
reads to genomic features. Bioinformatics. 2014; 30(7):923–930. https://doi.org/10.1093/
bioinformatics/btt656 PMID: 24227677
56. Ianevski A, He L, Aittokallio T, Tang J. SynergyFinder: A web application for analyzing drug combination
dose-response matrix data. Bioinformatics. 2017; 33(15):2413–2415. https://doi.org/10.1093/
bioinformatics/btx162 PMID: 28379339
57. Yan L. MK-2206: A potent oral allosteric AKT inhibitor. Cancer Research. 2009; 69(9 Supplement):
DDT01–1.
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 25 / 26
58. Hughes CS, Moggridge S, Mu¨ller T, Sorensen PH, Morin GB, Krijgsveld J. Single-pot, solid-phase-
enhanced sample preparation for proteomics experiments. Nature Protocols. 2019; 14(1):68–85.
https://doi.org/10.1038/s41596-018-0082-x PMID: 30464214
59. Batth TS, Francavilla C, Olsen JV. Off-line high-pH reversed-phase fractionation for in-depth phospho-
proteomics. Journal of Proteome Research. 2014; 13(12):6176–6186. https://doi.org/10.1021/
pr500893m PMID: 25338131
60. Tyanova S, Temu T, Cox J. The MaxQuant computational platform for mass spectrometry-based shot-
gun proteomics. Nature Protocols. 2016; 11(12):2301–2319. https://doi.org/10.1038/nprot.2016.136
PMID: 27809316
61. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. Limma powers differential expression analy-
ses for RNA-sequencing and microarray studies. Nucleic Acids Research. 2015; 43(7):e47. https://doi.
org/10.1093/nar/gkv007 PMID: 25605792
62. Hornbeck PV, Zhang B, Murray B, Kornhauser JM, Latham V, Skrzypek E. PhosphoSitePlus, 2014:
Mutations, PTMs and recalibrations. Nucleic Acids Research. 2015; 43(D1):D512–D520. https://doi.
org/10.1093/nar/gku1267 PMID: 25514926
PLOS COMPUTATIONAL BIOLOGY
Signalling models unveil combination therapies targeting feedback-mediated resistance
PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1009515 November 4, 2021 26 / 26
... According to emerging evidence from molecular studies, the tumor microenvironment uses key paracrine factors such as insulin-like growth factor 1 (IGF-1) to fuel tumorigenesis (1)(2)(3). Certainly, in the current global obesity epidemic, IGFs have attracted attention for their potential role as a link between cancer and insulin resistance syndrome (4-6), a metabolic syndrome of increasing incidence that encompasses hyper/ hypoinsulinemia, hyper/hypoglycemia, dyslipidemia and hypertension, with a marked increase in risk to develop diabetes mellitus, cardiovascular diseases, and cancer (7)(8)(9)(10). Notably, recent reports show that antidiabetic drugs, such as metformin, can mediate anticancer effects partly by silencing IGF-1 expression (9,10). ...
... Insulin-like peptides (ILPs), which include IGF-1, IGF-2, insulin, and seven relaxin-related peptides sharing the same basal fold in humans, are evolutionary conserved factors that play a central role in the regulation of energy metabolism, cell growth and proliferation, and neurotransmission (7,8,17). ...
... Further supporting the complexity of crosstalk between IGF-1R and other tyrosine kinase receptor pathways, unclear synergistic actions of EGF, PDGF, TGF-b, and IGF induce stemness, cancer progression and metastasis, drug resistance, and tumor relapse in various cancer types (33,39,53,77). A study using a panel of cell lines of high-risk neuroblastoma, a cancer characterized by increased MAPK signaling, found drastic variations in sensitivity to serine/tyrosine/threonine kinase inhibitors between these cell lines (7). Surprisingly, mathematical modeling of these variations revealed that MAPK signaling negative feedback via IGF-1R reactivated MAPK signaling to mediate cancer cell drug resistance, highlighting the concomitant targeting of MEK and IGF-1R/MAPK signaling pathways as a potential therapeutic strategy in high-risk neuroblastoma. ...
Article
Full-text available
The tumor microenvironment fuels tumorigenesis and induces the development of resistance to anticancer drugs. A growing number of reports support that the tumor microenvironment mediates these deleterious effects partly by overexpressing insulin-like growth factor 1 (IGF-1). IGF-1 is known for its role to support cancer progression and metastasis through the promotion of neovascularization in transforming tissues, and the promotion of the proliferation, maintenance and migration of malignant cells. Anti-IGF therapies showed potent anticancer effects and the ability to suppress cancer resistance to various chemotherapy drugs in in vivo and in vitro preclinical studies. However, high toxicity and resistance to these agents are increasingly being reported in clinical trials. We review data supporting the notion that tumor microenvironment mediates tumorigenesis partly through IGF-1 signaling pathway. We also discuss the therapeutic potential of IGF-1 receptor targeting, with special emphasis on the ability of IGF-R silencing to overcome chemotherapy drug resistance, as well as the challenges for clinical use of anti-IGF-1R therapies.
... Most importantly, modeling indicated that ERK-RAF feedback was weakened, which led us to predict MEK inhibitor sensitivity as a new collateral sensitivity that was confirmed in our small inhibitor screen in the NF1 knockout cell line model. Most treatment-naïve neuroblastoma cells are only intermediately sensitive to MEK inhibitors [65], in line with the failure of MEK inhibitors in clinical trials to treat neuroblastoma [66]. Our NF1 knockout models were, contrastingly, hypersensitive to MEK inhibitor treatment. ...
... Cell lines sensitive to MEK inhibitors, however, show only a weakened or absent negative ERK to RAF feedback, which we propose as the mechanistic reason for MEK inhibitor sensitivity in neuroblastoma cell lines lacking NF1. Accordingly, we recently demonstrated in work available on the bioRxiv preprint server that the strength of ERK to RAF feedback correlates with the level of MEK inhibitor sensitivity in neuroblastoma cell lines [65]. Surprisingly, even though NF1 and NRAS Q61K mutations both increase RAS/MAPK signaling, thereby conferring ALK inhibitor resistance, cells harboring these alterations respond differently to MEK inhibitor treatment. ...
Article
Full-text available
Background Development of resistance to targeted therapies has tempered initial optimism that precision oncology would improve poor outcomes for cancer patients. Resistance mechanisms, however, can also confer new resistance-specific vulnerabilities, termed collateral sensitivities. Here we investigated anaplastic lymphoma kinase (ALK) inhibitor resistance in neuroblastoma, a childhood cancer frequently affected by activating ALK alterations. Methods Genome-wide forward genetic CRISPR-Cas9 based screens were performed to identify genes associated with ALK inhibitor resistance in neuroblastoma cell lines. Furthermore, the neuroblastoma cell line NBLW-R was rendered resistant by continuous exposure to ALK inhibitors. Genes identified to be associated with ALK inhibitor resistance were further investigated by generating suitable cell line models. In addition, tumor and liquid biopsy samples of four patients with ALK -mutated neuroblastomas before ALK inhibitor treatment and during tumor progression under treatment were genomically profiled. Results Both genome-wide CRISPR-Cas9-based screens and preclinical spontaneous ALKi resistance models identified NF1 loss and activating NRAS Q61K mutations to confer resistance to chemically diverse ALKi. Moreover, human neuroblastomas recurrently developed de novo loss of NF1 and activating RAS mutations after ALKi treatment, leading to therapy resistance. Pathway-specific perturbations confirmed that NF1 loss and activating RAS mutations lead to RAS-MAPK signaling even in the presence of ALKi. Intriguingly, NF1 loss rendered neuroblastoma cells hypersensitive to MEK inhibition. Conclusions Our results provide a clinically relevant mechanistic model of ALKi resistance in neuroblastoma and highlight new clinically actionable collateral sensitivities in resistant cells.
Article
Full-text available
Discrete dynamical modeling shows promise in prioritizing drug combinations for screening efforts by reducing the experimental workload inherent to the vast numbers of possible drug combinations. We have investigated approaches to predict combination responses across different cancer cell lines using logic models generated from one generic prior-knowledge network representing 144 nodes covering major cancer signaling pathways. Cell-line specific models were configured to agree with baseline activity data from each unperturbed cell line. Testing against experimental data demonstrated a high number of true positive and true negative predictions, including also cell-specific responses. We demonstrate the possible enhancement of predictive capability of models by curation of literature knowledge further detailing subtle biologically founded signaling mechanisms in the model topology. In silico model analysis pinpointed a subset of network nodes highly influencing model predictions. Our results indicate that the performance of logic models can be improved by focusing on high-influence node protein activity data for model configuration and that these nodes accommodate high information flow in the regulatory network.
Article
Full-text available
Isobaric labeling empowers proteome-wide expression measurements simultaneously across multiple samples. Here an expanded set of 16 isobaric reagents based on an isobutyl-proline immonium ion reporter structure (TMTpro) is presented. These reagents have similar characteristics to existing tandem mass tag reagents but with increased fragmentation efficiency and signal. In a proteome-scale example dataset, we compared eight common cell lines with and without Torin1 treatment with three replicates, quantifying more than 8,800 proteins (mean of 7.5 peptides per protein) per replicate with an analysis time of only 1.1 h per proteome. Finally, we modified the thermal stability assay to examine proteome-wide melting shifts after treatment with DMSO, 1 or 20 µM staurosporine with five replicates. This assay identified and dose-stratified staurosporine binding to 228 cellular kinases in just one, 18-h experiment. TMTpro reagents allow complex experimental designs—all with essentially no missing values across the 16 samples and no loss in quantitative integrity.
Article
Full-text available
Dynamic mechanistic models, that is, those that can simulate behavior over time courses, are a cornerstone of molecular systems biology. They are being used to model molecular mechanisms with varying degrees of granularity—from elementary reactions to causal links—and to describe these systems by various dynamic mathematical frameworks, such as Boolean networks or systems of differential equations. The models can be based exclusively on experimental data, or on prior knowledge of the underlying biological processes. The latter are typically generic, but can be adapted to a certain context, such as a particular cell type, after training with context‐specific data. Dynamic mechanistic models that are based on biological knowledge have great potential for modeling specific systems, because they require less data for training to provide biological insight in particular into causal mechanisms, and to extrapolate to scenarios that are outside the conditions they have been trained on.
Article
Full-text available
Despite an increased awareness of the problematic of cell line cross‐contamination and misidentification, it remains nowadays a major source of erroneous experimental results in biomedical research. To prevent it, researchers are expected to frequently test the authenticity of the cell lines they are working on. STR profiling was selected as the international reference method to perform cell line authentication. While the experimental protocols and manipulations for generating a STR profile are well described, the available tools and workflows to analyze such data are lacking. The Cellosaurus knowledge resource aimed to improve the situation by compiling all the publicly available STR profiles from the literature and other databases. As a result, it grew to become the largest database in terms of human STR profiles, with 6,474 distinct cell lines having an associated STR profile (release 31, July 2019). Here we present CLASTR, the Cellosaurus STR similarity search tool enabling users to compare one or more STR profiles with those available in the Cellosaurus cell line knowledge resource. It aims to help researchers in the process of cell line authentication by providing numerous functionalities. The tool is publicly accessible on the SIB ExPASy server (https://web.expasy.org/cellosaurus‐str‐search) and its source code is available on GitHub under the GPL‐3.0 license. This article is protected by copyright. All rights reserved.
Article
Full-text available
Oncogenic mutations in KRAS or BRAF are frequent in colorectal cancer and activate the ERK kinase. Here, we find graded ERK phosphorylation correlating with cell differentiation in patient-derived colorectal cancer organoids with and without KRAS mutations. Using reporters, single cell transcriptomics and mass cytometry, we observe cell type-specific phosphorylation of ERK in response to transgenic KRASG12V in mouse intestinal organoids, while transgenic BRAFV600E activates ERK in all cells. Quantitative network modelling from perturbation data reveals that activation of ERK is shaped by cell type-specific MEK to ERK feed forward and negative feedback signalling. We identify dual-specificity phosphatases as candidate modulators of ERK in the intestine. Furthermore, we find that oncogenic KRAS, together with β-Catenin, favours expansion of crypt cells with high ERK activity. Our experiments highlight key differences between oncogenic BRAF and KRAS in colorectal cancer and find unexpected heterogeneity in a signalling pathway with fundamental relevance for cancer therapy.
Article
Full-text available
HRAS, NRAS, and KRAS isoforms are almost identical proteins that are ubiquitously expressed and activate a common set of effectors. In vivo studies have revealed that they are not biologically redundant; however, the isoform specificity of Ras signaling remains poorly understood. Using a novel panel of isogenic SW48 cell lines endogenously expressing wild-type or G12V-mutated activated Ras isoforms, we have performed a detailed characterization of endogenous isoform-specific mutant Ras signaling. We find that despite displaying significant Ras activation, the downstream outputs of oncogenic Ras mutants are minimal in the absence of growth factor inputs. The lack of mutant KRAS-induced effector activation observed in SW48 cells appears to be representative of a broad panel of colon cancer cell lines harboring mutant KRAS. For MAP kinase pathway activation in KRAS-mutant cells, the requirement for coincident growth factor stimulation occurs at an early point in the Raf activation cycle. Finally, we find that Ras isoform-specific signaling was highly context dependent and did not conform to the dogma derived from ectopic expression studies.
Article
Full-text available
Cascades of phosphorylation between protein kinases comprise a core mechanism in the integration and propagation of intracellular signals. Although we have accumulated a wealth of knowledge around some such pathways, this is subject to study biases and much remains to be uncovered. Phosphoproteomics, the identification and quantification of phosphorylated proteins on a proteomic scale, provides a high-throughput means of interrogating the state of intracellular phosphorylation, both at the pathway level and at the whole-cell level. In this review, we discuss methods for using human quantitative phosphoproteomic data to reconstruct the underlying signalling networks that generated it. We address several challenges imposed by the data on such analyses and we consider promising advances towards reconstructing unbiased, kinome-scale signalling networks.
Article
Full-text available
We describe Strelka2 ( https://github.com/Illumina/strelka ), an open-source small-variant-calling method for research and clinical germline and somatic sequencing applications. Strelka2 introduces a novel mixture-model-based estimation of insertion/deletion error parameters from each sample, an efficient tiered haplotype-modeling strategy, and a normal sample contamination model to improve liquid tumor analysis. For both germline and somatic calling, Strelka2 substantially outperformed the current leading tools in terms of both variant-calling accuracy and computing cost.
Article
Neuroblastoma is a pediatric tumor of the sympathetic nervous system. Its clinical course ranges from spontaneous tumor regression to fatal progression. To investigate the molecular features of the divergent tumor subtypes, we performed genome sequencing on 416 pretreatment neuroblastomas and assessed telomere maintenance mechanisms in 208 of these tumors. We found that patients whose tumors lacked telomere maintenance mechanisms had an excellent prognosis, whereas the prognosis of patients whose tumors harbored telomere maintenance mechanisms was substantially worse. Survival rates were lowest for neuroblastoma patients whose tumors harbored telomere maintenance mechanisms in combination with RAS and/or p53 pathway mutations. Spontaneous tumor regression occurred both in the presence and absence of these mutations in patients with telomere maintenance–negative tumors. On the basis of these data, we propose a mechanistic classification of neuroblastoma that may benefit the clinical management of patients.