Article

CRISPR screens identify tumor-promoting genes

conferring melanoma cell plasticity and resistance
Arthur Gautron1, Laura Bachelot1, Marc Aubry1,2,† , Delphine Leclerc3,†, Ana€ıs M Quemener1,†,
Sebastien Corre1,† , Florian Rambow4,5, Ana€ıs Paris1, Nina Tardif1, Helo€ıse M Leclair1,

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

Oskar Marin-Bejar4,5, Cedric Coulouarn3, Jean-Christophe Marine4,5, Marie-Dominique Galibert1,6,* &
David Gilot1,‡,**

Abstract Introduction
Most genetic alterations that drive melanoma development and Identifying molecular cancer drivers is critical for precision oncology.
resistance to targeted therapy have been uncovered. In contrast, Last year, the cancer genome atlas (TCGA) identified 299 driver genes
and despite their increasingly recognized contribution, little is by focusing on point mutations and small indels across 33 cancer
known about the non-genetic mechanisms that drive these types (Bailey et al, 2018). It represents the most comprehensive effort
processes. Here, we performed in vivo gain-of-function CRISPR thus far to identify cancer driver mutations. Complementary studies
screens and identified SMAD3, BIRC3, and SLC9A5 as key actors of are required to elucidate the role of copy-number variations, genomic
BRAFi resistance. We show that their expression levels increase fusions, and methylation events in the 33 TCGA projects.
during acquisition of BRAFi resistance and remain high in persister Moreover, there is increasing evidence that non-genetic reprogram-
cells and during relapse. The upregulation of the SMAD3 transcrip- ming leading to cancer cell dedifferentiation, stemness, invasiveness
tional activity (SMAD3-signature) promotes a mesenchymal-like also contribute to tumor growth and therapy resistance (Puisieux
phenotype and BRAFi resistance by acting as an upstream tran- et al, 2014; Bai et al, 2019). Thus, deciphering the signaling pathways
scriptional regulator of potent BRAFi-resistance genes such as that drive such processes may also lead to innovative cancer thera-
EGFR and AXL. This SMAD3-signature predicts resistance to both pies. Recent gene expression quantifications performed at single-cell
current melanoma therapies in different cohorts. Critically, chemi- level by single-cell RNA sequencing (scRNA-Seq) demonstrated that
cal inhibition of SMAD3 may constitute amenable target for cancer cells operate a dedifferentiation process, for instance in
melanoma since it efficiently abrogates persister cells survival. glioblastoma and melanoma (Patel et al, 2014; Tirosh et al, 2016;
Interestingly, decrease of SMAD3 activity can also be reached by Rambow et al, 2018), promoting tumor growth, stemness, and therapy
inhibiting the Aryl hydrocarbon Receptor (AhR), another druggable resistance. This “onco-dedifferentiation” seems to be independent of
transcription factor governing SMAD3 expression level. Our work de novo mutations and could offer new targets/strategies to cure
highlights novel drug vulnerabilities that can be exploited to cancer. However, these scRNA-Seq studies are mainly descriptive; the
develop long-lasting antimelanoma therapies. tumor growth capability of each gene/RNA is not yet investigated at
the genome-scale. Such functional analyses are nowadays feasible
Keywords Aryl hydrocarbon Receptor; CRISPR-SAM; melanoma; SMAD3; using clustered regularly interspaced short palindromic repeats
targeted therapy resistance (CRISPR)-Cas9 screens (Shalem et al, 2015). The majority of the
Subject Categories Cancer; Skin CRISPR-Cas9 screens is based on the invalidation of coding genes but
DOI 10.15252/emmm.202013466 | Received 16 September 2020 | Revised 9 modulation of gene expression is reachable with the CRISPR-Cas9
February 2021 | Accepted 11 February 2021 | Published online 16 March 2021 synergistic activation mediator (SAM) approach (Konermann et al,
EMBO Mol Med (2021) 13: e13466 2015). It corresponds to an engineered protein complex for the tran-
scriptional activation of endogenous genes. Importantly, SAM can
further be combined with a human genome-wide library to activate all

1 CNRS, IGDR (Institut de genetique et developpement de Rennes)-UMR 6290, Univ Rennes, Rennes, France
2 Plateforme GEH, CNRS, Inserm, BIOSIT - UMS 3480, US_S 018, Univ Rennes, Rennes, France
3 INSERM U1242, Centre Eugene Marquis, Rennes, France
4 Department of Oncology, KU Leuven, Leuven, Belgium
5 VIB Center for Cancer Biology, VIB, Leuven, Belgium
6 Service de Genetique Moleculaire et Genomique, CHU Rennes, Rennes, France
*Corresponding author. Tel: +33 0 223234705; E-mail: [email protected]
**Corresponding author (Lead Contact). Tel: +33 0 223234441; E-mail: [email protected]
†
These authors contributed equally to this work
‡
Present address: INSERM U1242, Centre Eugene Marquis, Rennes, France

ª 2021 The Authors. Published under the terms of the CC BY 4.0 license EMBO Molecular Medicine 13: e13466 | 2021 1 of 22
EMBO Molecular Medicine Arthur Gautron et al

known coding isoforms from the RefSeq database (23,430 isoforms) Results
for gain-of-function screening without a priori. To date, CRISPR
screens are mainly performed in vitro using cell lines or primary Identification of tumor-promoting genes by in vivo
cultures (Meyers et al, 2017). A pan-cancer CRISPR-Cas9 knockout gain-of-function CRISPR screen
screen was performed in vitro (324 human cancer cell lines from 30
cancer types) to identify essential genes for cancer cell fitness (defined Since the tumor environment influences, at least in part, the tumor
as genes required for cell growth or viability) and to prioritize candi- growth capability of cancer cells, we performed an in vivo genome-
dates for cancer therapeutics (Behan et al, 2019). However, because wide CRISPR-Cas9 SAM screen to identify in vivo tumor-promoting
the contribution of the tumor environment in tumor growth is increas- genes, defined as genes whose expression support tumor growth
ingly recognized, it seems important to perform such screens in the (in contrast to driver genes bearing a driver mutation such as
relevant patho-physiological context and, for instance, take advantage BRAF (V600E)). To select the most appropriate cellular model, we

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

of animal models. classified melanoma biopsies from The Cancer Genome Atlas
We selected cutaneous melanoma as a paradigm since novel ther- (TCGA) cohort (n = 458) in function of differentiation states
apeutic strategies are critically needed (Bai et al, 2019). Targeted according to the most recent melanoma profiling data (Fig 1A; Tsoi
therapies such as BRAF inhibitors (BRAFi) initially showed great et al, 2018). As anticipated, the vast majority of these tumors,
promise in patients with BRAF(V600)-mutated metastatic melanoma. which are almost all drug naive, exhibited a differentiated profile
Unfortunately, the vast majority of patients that initially respond to (89%; melanocytic and transitory). We selected the 501Mel cell
these drugs, almost inevitably develop resistance. Although combi- line since (i) these cells display a melanocytic differentiation state
nation therapies (BRAF and MEK inhibitors) enhance the response as the majority of diagnosticated melanoma, (ii) they harbor the
and delay relapse, the overall survival remains unsatisfactory high- BRAF(V600E) mutation as ~50% of cutaneous melanoma, (iii) they
lighting the need of new therapeutic targets (Ascierto et al, 2016). are highly sensitive to BRAFi with an IC50 value of 0.45 µM to
The mechanisms underlying resistance are numerous and proba- vemurafenib [PLX4032] (Halaban et al, 2010; Corre et al, 2018),
bly not mutually exclusive (Sullivan & Flaherty, 2013; Welsh et al, and importantly (iv) they are unable to generate tumor in nude
2016; Song et al, 2017). Resistance can be driven by a small pre- mice (Ohanna et al, 2011). This latter characteristic may allow to
existing subpopulation, harboring-specific genetic alterations that identify tumor-promoting genes.
confer them with resistance to the inhibitors (Wagle et al, 2011). To generate the CRISPR-SAM cell library, we modified the
Such alterations may also occur de novo, during treatment (Welsh 501Mel cells, to express constitutively defective-Cas9 and the
et al, 2016). In addition, there is increased evidence that non-genetic required cofactors for CRISPR-SAM technology (Konermann et al,
reprogramming may confer drug-tolerant and/or resistant pheno- 2015). These engineered cells were infected with the single-guide
types to melanoma cells (Rambow et al, 2018; Corre et al, 2018; RNA (sgRNA) lentivirus library that contained at least three dif-
Tsoi et al, 2018; Hugo et al, 2015a; Talebi et al, 2018; Rapino et al, ferent guides per coding gene (Konermann et al, 2015; Fig 1B). The
2018; preprint: Marin-Bejar et al, 2020). Earlier works demonstrated infection was performed at a multiplicity of infection (MOI) of 0.2
that phenotype switching from a proliferative to an invasive/mes- ensuring that only one guide is expressed per infected cell. Infected
enchymal-like state is also likely to contribute to therapy resistance cells were positively selected using antibiotic selection during
(Hoek & Goding, 2010; Konieczkowski et al, 2014; M€ uller et al, 7 days. By DNA sequencing, we observed a normal distribution
2014; Verfaillie et al, 2015; Boshuizen et al, 2018). Paradoxically, of the sgRNAs in two cell library replicates (Fig 1C). Only 78
MITF-induced differentiation into a slow cycling, pigment-producing sgRNAs were not detected in our cell library, which validate our
state was also reported to confer tolerance to BRAFi (M€ uller et al, protocol and the cell library (>70,100 sgRNAs were detected;
2014; Smith et al, 2016). It therefore seems that various drug-toler- Table EV1). Thus, the controls were proper to identify in vivo
ant subpopulations can emerge under therapeutic pressure and that tumor-promoting genes.
these cells can provide a pool from which resistance develops. The cell library (30 × 106 cells) was fractionated and subcuta-
Targeting these populations of persister cells is therefore crucial to neously xenografted in 10 nude mice (3 × 106 cells/mouse) and
achieve effective personalized therapies (Nassar & Blanpain, 2016). tumor growth was monitored using caliper over a 5-month period
Here, we performed unbiased screens to identify genes promoting (Fig 1D and Table EV2). As previously demonstrated (Ohanna et al,
tumor growth from persister cells and conferring resistance to BRAFi 2011), we confirmed that parental 501Mel are unable to form
using CRISPR-Cas9 SAM methodology. We demonstrate that, in addition tumors in nude mice (n = 6). In contrast, seven tumors were
to promote melanoma development, Mothers against decapentaplegic obtained from the CRISPR-engineered cells xenografted in 10 mice
homolog 3 (SMAD3), Baculoviral IAP repeat-containing protein 3 (Fig 1D). The nature of the sgRNAs, their abundance and occur-
(BIRC3), and Sodium/hydrogen exchanger 5 (SLC9A5) also support rence across these 7 tumors were determined by DNA-Seq (Fig 1E
relapse since they promote both BRAFi-resistance and tumor growth and F, and Table EV3).
capability of persister cells. Their expression levels correlated with BRAFi By comparing the most represented genes (sgRNAs) in each
resistance and relapse. Consequently, their inhibition strongly reduced tumor (Tum), we identified 3 common genes (Fig 1F). An enrich-
the number of persister cells. Moreover, we demonstrate that the tran- ment of SMAD3, BIRC3, and SLC9A5 sgRNAs was found in tumors
scription factor AhR governs SMAD3 expression levels and in turn when compared to their starting abundance (cell library; orange
SMAD3 drives the expression of a set of genes associated with BRAFi points; Fig 1G), in contrast to EGFR sgRNAs. Thirty-six other genes
resistance and mesenchymal phenotype. These experiments identify inte- were recurrently retrieved in the tumors but not in all (Table EV3).
grated AhR-SMAD3 signaling as a key driver of melanoma growth and Interestingly, YAP1 which has already been identified as melanoma
relapse, pointing to a new therapeutic vulnerability in melanoma. growth-promoting gene was also found (Table EV3). This supports

2 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors

Arthur Gautron et al EMBO Molecular Medicine

A B C

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

D E F

G H I

J K

Figure 1.

ª 2021 The Authors EMBO Molecular Medicine 13: e13466 | 2021 3 of 22

 EMBO Molecular Medicine Arthur Gautron et al

◀ Figure 1. Identification of tumor-promoting genes by in vivo gain-of-function CRISPR screen.

A Determination of differentiation states of skin cutaneous melanoma (SKCM) biopsies from the TCGA cohort (n = 458) according to (Tsoi et al, 2018). The vast majority
of these tumors exhibited a differentiated profile (89%; melanocytic and transitory states). The others display a dedifferentiated profile (11%; neural crest-like cells
and undifferentiated states). Human melanoma 501Mel cell line is classified as differentiated melanoma cells (melanocytic cells), according to the gene expression
profile (four categories were defined by (Tsoi et al, 2018); melanocytic, transitory, neural crest-like, and undifferentiated cells). 501mel cell line was selected for the
CRISPR screens.
B Workflow depicting the in vivo CRISPR-SAM screen to identify tumor-promoting genes. Parental cells and CRISPR-SAM cell library were xenografted on nude mice
(3 × 106 cells/mouse, n = 6 and n = 10, respectively) and tumor growth was monitored during 5 months. Seven tumors were collected and analyzed by DNA-Seq to
identify the sgRNAs (Tables EV1 and EV2).
C sgRNAs distribution in the cell library. sgRNA mean reaches 550  2.
D Tumor growth curves for the 7 tumors arising from the CRISPR-SAM-engineered cells xenografted in nude mice as detailed in panel B. (No tumor for the parental cell
line (501Mel cells)) (Table EV2).
E Distribution of sgRNAs in cell library and in the 7 tumors (log10(sgRNAs counts)). Blue and red lines indicated the enrichment ≥ 10 fold or ≥ 100 fold (tumors vs cell

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

library (in vitro)) (Table EV1).
F From the seven tumors, the top hundred genes (enriched) have been selected and common genes are SMAD3, BIRC3, and SLC9A5 (Table EV3).
G sgRNA counts in tumors versus in CRISPR-SAM cell library (respectively, black and orange points). Each black point corresponds to one tumor. Two replicates have
been shown for CRISPR-SAM cell library (orange points).
H Workflow depicting the validation step: 501Mel cells overexpressing SMAD3, BIRC3, or SLC9A5 (obtained by CRISPR-SAM) were xenografted on nude mice and tumor
volume was monitored using caliper. 3 × 106 cells/mouse. n = 7, 6, and 6 mice, respectively.
I SMAD3 expression levels in 501Mel cells overexpressing the cofactors for CRISPR-SAM approach and the control sgRNA (501Mel 3 + backbone) or the SMAD3 sgRNA.
SLC9A5 and BIRC3 sgRNAs are used as controls to show the specificity of the SMAD3 overexpression. HSC70 serve as loading control.
J SMAD3, BIRC3, and SLC9A5 mRNA expression levels in melanoma cell lines described in H and I. n = 7 independent biological experiments. Dashed lines for medians.
K Tumor growth curves from 501Mel cells overexpressing SMAD3, BIRC3, or SLC9A5.
Data information: Western blot results are representative of at least two independent experiments. Source data and unprocessed original blots are available in
Appendix Fig S1 source data. See also Fig EV1.
Source data are available online for this figure.

the robustness of the screen (Lamar et al, 2012; Verfaillie et al, almost inevitably occurs (Bai et al, 2019). To examine the genes
2015; Hugo et al, 2015b). The majority of the tumor-promoting promoting BRAFi resistance and relapse, we performed an in cellulo
genes (Table EV4) identified here are not considered as genes screen using the same cell library in the presence of BRAFi (Fig 2).
required for cell growth or viability (except the essential genes Briefly, the CRISPR-SAM 501Mel cell library (40 × 106 cells) was
YAP1, SLC25A41, and TGIF1; Behan et al, 2019) and are not treated for 14 days with BRAFi (2 µM), using either the BRAFi used
frequently altered in melanoma (Fig EV1A). Moreover, the trans- in clinical practice (vemurafenib), the next-generation inhibitor that
forming growth factor (TGF)-b pathway seemed well-represented is still under investigation in clinical trials (PLX8394), or the solvent
among the tumor-promoting genes (Table EV4). Since our results (dimethyl sulfoxide (DMSO)) as control (Fig 2A). This procedure
suggest that a high expression level of these tumor-promoting genes allows for the enrichment of sgRNAs (genes) conferring resistance.
is sufficient to promote melanoma tumor growth, we evaluated the The nature of the sgRNA present in the resistant population and
biological consequences of the inhibition of one tumor-promoting their abundance was determined by DNA-Seq (Fig 2B). The best hit
gene, BIRC3. The chemical inhibitor Birinapant reduced the was the Epidermal growth factor receptor gene (EGFR), a well-
SKMel28R and Me1402 cell density (Fig EV1B and C), suggesting known BRAFi-resistance gene (Sun et al, 2014; Shaffer et al, 2017).
that the BIRC3 protein is involved in cell proliferation of these mela- By examining the enrichment of sgRNAs targeting EGFR promoter
noma cells as previously demonstrated (Krepler et al, 2013; Vetma (Fig 2B), we decided to retain genes with at least two sgRNAs
et al, 2017). among the enriched sgRNAs present in BRAFi-exposed cells (with a
Next, we examined in vivo the ability of these tumor-promoting false discovery rate, FDR < 0.05) (Tables EV5 and EV6,
genes to promote tumor growth by xenografting cell populations Appendix Fig S1) since sometimes one of the three sgRNAs designed
overexpressing the sgRNAs individually. We focused on the top 3 per gene is not detected or not enriched as observed for EGFR and
tumor-promoting genes: SMAD3, BIRC3, and SCL9A5 (Fig 1H–K). BIRC3 (Fig 2C). A recent publication confirmed that sgRNAs are not
We generated three new CRISPR-engineered cell lines, and we eval- all functional in CRISPRa libraries and it could be interesting to
uated the overexpression levels of these genes by Western blot increase the number of sgRNAs per target and to cover more TSS
experiments (SMAD3) and RT–qPCR (SMAD3, BIRC3, and SLC9A5) per gene (Sanson et al, 2018).
(Fig 1I and J). Finally, we confirmed that they independently foster Apart sgRNAs targeting the EGFR promoter, the sgRNAs enrich-
tumor development (Fig 1K). Altogether, our results demonstrated ment in BRAFi-exposed cells was unexpectedly weak (Tables EV5
that in vivo CRISPR-SAM screen identifies new tumor-promoting and EV6). Thus, to select the best BRAFi-resistance genes, we exam-
genes, which may constitute amenable target for melanoma. ined the gene expression levels of these potential BRAFi-resistance
genes identified by CRISPR screen in 12 melanoma cell lines (Fig 2D
Genome-wide CRISPR activation screen identifies BRAFi- and E, Appendix Fig S1). We postulated that BRAFi-resistance genes
resistance genes in cutaneous melanoma are highly expressed in BRAFi-resistant cells (n = 6) as already
demonstrated by other approaches for NRP1, AXL, and NGFR. To
BRAFi provokes tumor shrinkage in the vast majority of patients this end, we confronted CRISPR-SAM candidates (identified in
with BRAF(V600)-mutated metastatic melanoma but resistance 501Mel cells) to gene expression data from six melanoma cell lines

4 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors

Arthur Gautron et al EMBO Molecular Medicine

A B

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

E F G

Figure 2. Genome-wide CRISPR activation screen identifies BRAFi-resistance genes in cutaneous melanoma.
A CRISPR-SAM workflow. Cell library was exposed to DMSO (solvent of BRAFi) or BRAFi 2 µM (vemurafenib (Vem) or PLX8394 (PB, Paradox Breaker)) during 14 days.
40 × 106 cells per arm. Experiments were done in duplicate.
B Plot showing sgRNAs detected in BRAFi-resistant cells versus in control cells (cell library exposed to DMSO) with a fold change > 1.5. Raw data are available in
Table EV5 (BRAFi for PBV). Appendix Fig S1 depicted the comparison Vem vs. PB.
C sgRNAs counts in BRAFi-resistant cells versus in DMSO-exposed cells for EGFR and BIRC3 (BRAFi for PBV, Appendix Fig S1)
D Determination of differentiation states of 12 human BRAF(V600) melanoma cell lines from CCLE according to (Tsoi et al, 2018).
E BRAF(V600) melanoma cell lines from CCLE were distributed in two groups (Sensitive and “intrinsically” Resistant, n = 6 cell lines per group) according to their BRAFi
IC50 (µM, half maximal inhibitory concentrations) (Barretina et al, 2012). Error bars reflect mean  s.d.
F Volcano plot showing the expression levels of BRAFi-resistant genes (identified in 501Mel by CRISPR-SAM screen) in 12 human melanoma cell lines. The fold change
corresponds to the ratio of expression levels found in resistant and sensitive cell lines. In red: selected genes. EGFR is considered as positive control and SMAD3, BIRC3,
and SLC9A5; selected as favorite genes for the next steps. SMAD3, BIRC3, SLC9A5, and AFAP1 are BRAFi-resistance genes and potent tumor-promoting genes (Fig 1).
Raw data are available in Table EV5 (BRAFi for PBV, Appendix Fig S1)
G Heat map recapitulating the expression levels of the selected hits (red dots in Fig 2F) in BRAFi-resistant and BRAFi-sensitive cell lines. Markers of differentiation (MITF,
MLANA, OCA2, TYRP1, TYR). In red: resistance genes already published (NRP1, AXL, ZEB1, LPAR1). Scale corresponds to Z scores.
Source data are available online for this figure.

that were highly resistant to BRAFi according to the Cancer Cell our candidate genes identified using this workflow, we added well-
Lines Encyclopedia (CCLE) versus six sensitive cell lines (Barretina established and validated BRAFi-resistant genes (NRP1, AXL, ZEB1,
et al, 2012; Fig 2E). It is important to note that the CCLE BRAFi- and LPAR1) (M€ uller et al, 2014; Konermann et al, 2015; Rizzolio
resistant cell lines are therapy-na€ıve and intrinsically resistant. et al, 2018) and five genes associated with melanoma cell differenti-
We next focused on candidate genes which were both enriched ation (MITF, OCA2, MLANA, TYR, and TYRP1) (Levy et al, 2006).
in CRISPR screen and highly expressed in the majority of BRAFi- All BRAFi-resistant cell lines presented a dedifferentiated profile as
resistance cell lines (Fig 2F and G). To better evaluate the validity of anticipated. Importantly, our candidate genes including SMAD3,

ª 2021 The Authors EMBO Molecular Medicine 13: e13466 | 2021 5 of 22

EMBO Molecular Medicine Arthur Gautron et al

SLC9A5, and BIRC3 (Fig 2G, black color) displayed a similar expres- selected genes were highly and selectively expressed in both of
sion profile than observed for the well-established and validated these cell populations at mRNA level (Fig 3D and E). To reinforce
BRAFi-resistant genes (Fig 2G, red color), strongly suggesting that these observations, we performed SMAD3 immunostainings in four
these genes may also confer BRAFi resistance. EGFR and platelet- BRAF-mutant PDXs exposed to BRAF/MEK inhibitors until resis-
derived growth factor receptor (PDGFR)-b showed high expression tance (recently characterized in (preprint: Marin-Bejar et al, 2020),
only in a few BRAFi-resistant cell lines, as previously observed in Appendix Fig S2A and B). Immunostainings showed the emergence
patients (Sun et al, 2014). of SMAD3+ cells in Dabrafenib + Trametinib resistant lesions from
Together, our results confirmed the robustness of the functional the MEL003 and MEL006 PDXs in contrast to PDXs characterized by
in cellulo CRISPR-SAM screen. an intrinsic resistance mechanism (MEL007 and MEL037).
Moreover, we found that EGFR-expressing cells sorted from mela-
Validation of BRAFi SAM-selected resistance genes noma tumors displayed comparatively high expression levels of the

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

BRAFi SAM-selected genes (Fig EV2B and C). Importantly, these
To evaluate the contribution of the CRISPR-SAM-selected genes in EGFR-positive cells are able to proliferate in the presence of BRAFi
BRAFi resistance, we examined the transcriptome of the differenti- and generate BRAFi-resistant colonies (Shaffer et al, 2017). In addi-
ated cell line M229 (transitory cell state) at different stages during tion, high expression levels of the BRAFi SAM-selected genes have
acquisition of resistance (Fig 3A; Song et al, 2017). As described by been found in invasive cells when compared to proliferative mela-
Song et al (2017), cell lines were exposed to chronic exposure to noma cell lines (Verfaillie et al, 2015; Fig EV2D). Together, these data
BRAFi and analyzed at different days of treatment (P: parental cells, confirm the upregulated expression of BRAFi SAM-selected genes in
2D: two days of treatment, DTP: drug-tolerant persister cells, DTPP: cells shown to contribute to relapse, suggesting their involvement in
drug-tolerant proliferating persister cells, SDR: single-drug-resistant establishing drug-tolerant and/or resistant phenotypes in vivo.
cell). We observed a sequential upregulation in the expression of To evaluate the clinical relevance of the SAM-selected BRAFi-
BRAFi SAM-selected genes: a group of genes (MMP2, SEMA3B, resistance genes, we compared their expression levels (median) in
BIRC3, TIPARP, etc) being expressed earlier than a 2nd group (IL6, two independent BRAFi drug naive/drug-resistant patient cohorts
EGFR, AFAP1, etc). The majority of the candidates were overex- (Fig 3F and G, n = 21 (Hugo et al, 2015b) and n = 16 (Rizos et al,
pressed while the cells exhibited resistance to a single BRAFi agent 2014) patients, respectively). The expression levels of the selected
(single-drug resistance, SDR). Comparable results were obtained resistant candidate genes increased during relapse in drug-resistant
with the M238 melanoma cell line (Figs 3B and EV2A). Importantly, patients (48 and 31%). Notably, none of these genes have been
we found that combining the BRAFi with MEKi (DDR) led to compa- implicated in recurrent gene-amplification events that are some-
rable upregulation of the BRAFi SAM-selected genes than observed times identified in drug-naive lesions (cBioPortal, TCGA; Gao et al,
in cells exposed to BRAFi alone (SDR) (Fig 3C). These results indi- 2013) (Fig EV2E–H). These data indicate that the increase in expres-
cate that a common gene expression program can confer resistance sion of the SAM-selected genes in BRAFi-resistant cells is likely
to inhibitors of MAPK pathway as previously reported (Moriceau associated with a (non-genetic) dedifferentiation process of mela-
et al, 2015; Corre et al, 2018; Lee et al, 2020). noma cells induced by the therapy. Together, these in vitro and
Having shown that combination of BRAFi and MEKi promotes in vivo gene expression analyses strongly support a BRAFi-resis-
sequential upregulation of BRAFi SAM-selected genes, we moni- tance function for the SAM-selected genes.
tored their expression levels in an in vivo preclinical patient-derived
xenograft (PDX) model (Rambow et al, 2018). Using scRNA-Seq, BRAFi-resistance genes promote tumor growth
our collaborators reported the presence of dedifferentiated drug-
tolerant cells exhibiting a neural crest stem cell (NCSC) and invasive Long-term effect of BRAFi is reduced by the ability of persister cells
profiles at minimal residual disease isolated from the MEL006 PDX to resist to BRAFi but also to promote the tumor growth (relapse).
model (Rambow et al, 2018). Here, we showed that the BRAFi SAM- Thus, we investigated the capability of the BRAFi-persister cells

Figure 3. Validation of BRAFi SAM-selected resistance genes.

A Expression levels of our hits in M229 melanoma cells during the BRAFi-resistance acquisition. P: parental cells, 2D: two days of treatment, DTP: drug-tolerant persister
cells, DTPP: drug-tolerant proliferating persister cells, SDR: single-drug-resistant cells (BRAFi) (Song et al, 2017). Scale corresponds to Z scores.
▸
B Expression levels of our hits in M238 melanoma cells (Song et al, 2017) during the BRAFi-resistance acquisition. Scale corresponds to Z scores.
C Expression levels of our hits in parental, single-drug-resistant (SDR, BRAFi), or dual drug-resistant (DDR, BRAFi + MEKi) SKMel28 melanoma cell lines (Song et al,
2017). Scale corresponds to Z scores.
D T-distributed Stochastic Neighbor Embedding (t-SNE) plot showing the 4 drug-tolerant states (NCSC (neural crest stem cells), invasive, SMC (starved-like melanoma
cells) and pigmented cells) according to single-cell RNA-seq (scRNA-Seq) performed in PDX MEL006 model exposed to BRAFi + MEKi (Rambow et al, 2018). Our BRAFi-
resistance genes are mainly expressed in NCSC and invasive cells.
E AUCell score for each drug-tolerant states. ****P < 0.0001, Mann–Whitney test.
F Expression level of our genes (median) in cutaneous melanoma biopsies before the BRAFi treatment (baseline) and during the relapse. Cohort from (Hugo et al,
2015b). The expression level of our BRAFi-resistance genes is increased in relapse samples (48%, n = 21).
G Expression level of our genes (median) in cutaneous melanoma biopsies before the BRAFi treatment and during the relapse. Cohort from (Rizos et al, 2014). The
expression level of our BRAFi-resistance genes is increased in relapse samples (31%, n = 16).
Data information: See also Fig EV2.
Source data are available online for this figure.

6 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors

Arthur Gautron et al EMBO Molecular Medicine

A B C

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

D E

F G

Figure 3.

(obtained from the in vitro CRISPR screen, Fig 2) to promote tumor (Fig 4C and Table EV7), and we detected 97 genes (sgRNAs)
growth into nude mice. The subset of BRAFi-resistant/persister cells detected in all tumors. Interestingly, we recovered the tumor-
(Fig 4A) was engrafted (36 × 106 cells, 3 × 106/mouse), and tumor promoting genes SMAD3, BIRC3, and SLC9A5 identified above
growth was monitored (Fig 4B). (Fig 4D). We looked for the enrichment of these sgRNAs in each
These BRAFi-persister cells formed tumors, in contrast to tumor developed from persister cells (Fig 4E). EGFR sgRNA was not
parental 501Mel cells (Ohanna et al, 2011). We determined the frequently enriched in tumors (as previously described for human
nature and abundance of sgRNAs present in each emerging tumor melanoma tumors (Prahallad et al, 2012; Sun et al, 2014; Shaffer

ª 2021 The Authors EMBO Molecular Medicine 13: e13466 | 2021 7 of 22

EMBO Molecular Medicine Arthur Gautron et al

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

B C

D E

Figure 4.

8 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors

 Arthur Gautron et al EMBO Molecular Medicine

◀ Figure 4. BRAFi-resistance genes promote tumor growth.

A Workflow to identify genes involved in tumor growth from BRAFi-persister cells. Cell populations were subcutaneously grafted on nude mice flanks (3 × 106 cells per
mouse); Vem-persister cells (n = 11 mice) and PLX8394 (PB)-persister cells (n = 12 mice) (Table EV2). (BRAFi for PBV, Appendix Fig S1)
B Tumor growth curves from BRAFi-persister cells (monitored during 5 months; Table EV2). V for Vem-resistant cells and PB for PLX8394-resistant cells.
C Distribution of sgRNAs in BRAFi-resistant cells (before xenograft) and in the 7 tumors emerging from the BRAFi-resistant cells (log10(sgRNAs counts)). Blue and red
lines indicated the enrichment ≥ 10 fold or ≥ 100 fold (tumors versus BRAFi-resistant cells (in vitro)). Raw data are available in Table EV7.
D From the seven tumors arising from BRAFi-resistant cells, the common genes (enriched) have been extracted. Ninety-seven genes including SMAD3, BIRC3, and
SLC9A5 are detected in all these 7 tumors. Raw data are available in Table EV7.
E sgRNAs counts in tumors versus sgRNA detected in BRAFi-resistant cell library (in vitro) (respectively, black and orange points) for selected candidates. Each black
point corresponds to one tumor. EGFR was the most potent BRAFi-resistant gene (Fig 2). Two replicates have been shown for CRISPR-SAM cell library (orange points).
SMAD3, BIRC3, and SLC9A5 were identified as hits in the three screens (Figs 1, 2, and 4).
Source data are available online for this figure.

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

et al, 2017)) in contrast to SMAD3, BIRC3, and SLC9A5 (Fig 4E and To transfer this strategy (SMAD3 inhibition + BRAFi) into clinic,
Appendix Fig S3). Together this suggests that these 3 genes are we looked for an efficient inhibitor of SMAD3. We identified the
potential interesting targets for antimelanoma therapy. chemical inhibitor SIS3 (SMAD3 inhibitor, SMAD3i; Jinnin et al,
2006; Wu et al, 2020). Firstly, we validated the inhibitory efficiency
Functional validation of genes involved in BRAFi resistance of SMAD3i in melanoma cells since it decreased the levels of phos-
and relapse pho-SMAD3 Ser423/425 induced by TGFb (Fig 5J) in accordance
with previous studies (Jinnin et al, 2006; Chihara et al, 2017). Since
We focused on the transcription factor SMAD3 as a model gene for the SMAD3 regulation by phosphorylation is not fully understood
monitoring BRAFi resistance and relapse due to its critical function and the phospho-SMAD3 Ser423/425 status is not strictly correlated
downstream of the TGFb pathway. Although this pathway is known with SMAD3 transcriptional activity (Ooshima et al, 2019), we eval-
to promote melanoma phenotype switching/dedifferentiation (Sun uated the effect of SMAD3i using a reporter assay. We demonstrated
et al, 2014), to support melanoma growth (Berking et al, 2001) and that SMAD3i reduced the transcriptional activity of SMAD3 in
metastasis, little is known about the role of SMAD3 in melanoma response to TGFb exposure in 4 melanoma cell lines (Fig 5K). To
biology and as a modulator of resistance to targeted therapy. know if the combination (SMAD3i + BRAFi) could be broadly used
We confirmed that SMAD3 mRNA is highly expressed in dedif- to eradicate persister cells emerging in response to BRAFi treatment,
ferentiated cells (Fig 5A–C) and in BRAFi-resistant cells (Fig 2G). we selected three melanoma cell lines in function of SMAD3 expres-
To reinforce the role of SMAD3 in BRAFi resistance, we showed sion levels (Fig 5L). The differentiated cells (SMAD3low) are highly
that gain-of-function of SMAD3 significantly increases the BRAFi sensitive to BRAFi (decrease of cell density: > 80% at 5 µM BRAFi,
resistance of melanoma cells when compared to different control Fig 5M)) in contrast to dedifferentiated cells (SMAD3high), which
cells (parental 501Mel cells and the CRISPR-engineered cells: are highly resistant to BRAFi (decrease of cell density: ~50% at
501Mel cells expressing dCas9 and HSF1-p65-MS2 (named here 5 µM BRAFi, Fig 5M).
501Mel 2+) and the 501Mel 2+ cells expressing a control guide We showed that SMAD3i (SIS3) reduced the cell density of all
(named 3+ backbone; Fig 5D). In addition, we showed that gain-of- melanoma cell lines (Fig 5N). These results are in agreement with
function of SMAD3 also promotes the three-dimensional (3D) tumor the SMAD3 knock-down results (Fig 5G and H). Our experiments
spheroid invasion capability of melanoma cells. Similar results were also indicated that melanocyte survival is weakly affected by
obtained for SLC9A5 (Figs 5E and EV3A). These results strongly SMAD3 inhibition (up to 20 µM), in agreement with the non-toxicity
suggest that a high expression level of SMAD3 confers BRAFi resis- of this inhibitor observed in vivo (Tang et al, 2017; Wu et al, 2020).
tance and invasion capability in melanoma cells. The inhibitory effect of SMAD3i on melanoma cells seems to be
Thus, we investigated if SMAD3 impairment re-sensitizes cells to associated with the total SMAD3 expression levels (Fig 5L). Our
BRAF inhibitor, using small-interfering RNA (siRNA) (Fig 5F–I). As results might suggest that these melanoma cells could be “addicted”
the dedifferentiation status correlated with BRAFi resistance, we to SMAD3 activity.
selected two BRAFi-resistant cell lines, SKMel28 BRAFi-resistant We finally investigated the interest to combine BRAFi (Vem, alone
cells (SKMel28R, Fig 5F and G; Hugo et al, 2015b) and Me1402 or in combination with MEKi (Cobi)) and SMAD3i (SIS3) to eradicate
melanoma cells (Fig 5H). In contrast to the SKMel28R, the resis- the BRAFi-resistant cell lines (SKMel28R, M229R, and M238R)
tance of which was created by chronic exposure to non-lethal doses (Figs 5O and EV3K). We showed that SMAD3 inhibitor alone or in
of BRAFi, Me1402 cells are intrinsically resistant. combination with BRAFi (Vem 5 µM) or BRAFi + MEKi (Cobi 1 µM)
Surprisingly, the single SMAD3 depletion decreased the cell might be a promising treatment to reduce the amount of persister cells
density in a similar magnitude than BRAFi treatment in these (melanoma). Together, these results identified SMAD3 as an amenable
BRAFi-resistant cells (Figs 5G and H, and EV3B). The SMAD3 deple- target to limit resistance to BRAFi and tumor growth.
tion did not modify the ERK pathway (Fig 5I) in contrast to the
BRAFi. The combo (SMAD3 depletion and BRAFi (5 µM)) efficiently The transcription factor AhR drives SMAD3 expression
reduced the number of resistant/persister cells in these two cell
lines, suggesting that SMAD3 is an interesting target to limit resis- Having shown that SMAD3 expression mediates BRAFi-resistance
tance to BRAFi. Similar results were obtained by targeting BIRC3, and tumor growth, we explored the transcriptional program promot-
EGFR, IL6, or AQP1 (Fig EV3C–J). ing its expression in BRAFi-resistant melanoma cells. We recently

ª 2021 The Authors EMBO Molecular Medicine 13: e13466 | 2021 9 of 22

EMBO Molecular Medicine Arthur Gautron et al

reported that the Aryl hydrocarbon Receptor (AhR), a ligand-depen- We postulated that AhR may govern SMAD3 expression in
dent transcription factor is an upstream central node regulating the BRAFi-resistant cells. We identified three putative canonical binding
expression of BRAFi-resistance genes and melanoma dedifferentia- sites for AhR (XRE for xenobiotic responsive element) in the proxi-
tion (Corre et al, 2018; Leclerc et al, 2021; Paris et al, 2021). mal promoter of SMAD3 (Fig 6A) in accordance with chromatin

A B C

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

D E F

G H I

J K L

M N O

Figure 5.

10 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors

 Arthur Gautron et al EMBO Molecular Medicine

◀ Figure 5. Functional validation of genes involved in BRAFi resistance and relapse.

A PCA analysis of melanoma cell lines in function of their dedifferentiation states (generated by the webtool http://systems.crump.ucla.edu/dediff/index.php).
B SMAD3 expression increases with melanoma cell dedifferentiation. PCA analysis of SMAD3 expression in melanoma cell lines in function of their dedifferentiation
states. Scale: red color corresponds to a high SMAD3 expression level (unit of the scale: Log2 FPKM).
C SMAD3 expression in these four subtypes of melanoma cells (U, undifferentiated; NC, neural crest-like; T, transitory; M, melanocytic). Number in each group: U = 10,
NC = 14, T = 12, M = 17. Error bars reflect mean  s.d. Multiple comparisons have been done using ordinary one-way ANOVA **P < 0.01, ***P < 0.001.
D SMAD3 gain-of-function increases BRAFi resistance. Determination of BRAFi half maximal inhibitory concentrations (IC50 values) for 501Mel cell lines (Log[Vem] (nM)).
Parental 501Mel cells (n = 5), 501Mel cells expressing dCas9 and HSF1-p65-MS2 (named here 501Mel 2+, n = 5), the 501Mel 2 + cells expressing a control guide
(backbone, n = 5) and the 501Mel 2 + cells overexpressing SMAD3 (n = 3 biologically independent experiments). A representative experiment has been chosen.
Dotted line is used to determine the IC50 values.
E SMAD3 and SLC9A5 gain-of-function increases invasion capability of melanoma cells. Invasion assays for engineered cell lines (melanoma spheroids): CTR; 501Mel 2+,
SMAD3; 501Mel 2 + cells overexpressing SMAD3. Two other cell lines overexpressing SLC9A5 or BIRC3 have been tested. Explanation for ratio calculation is detailed in
Fig EV3A. Results obtained from two biologically independent experiments. (n = 5, 4, 6, and 3 spheroids, respectively). Error bars reflect mean  s.d. Multiple

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

comparisons have been done using ordinary one-way ANOVA, *P < 0.05, ***P < 0.001.
F Characterization of SKMel28 sensitive and resistant cell lines (SKMel28S and SKMel28R). Determination of BRAFi half maximal inhibitory concentrations (IC50 values)
(Log[Vem] (nM)). A representative experiment has been chosen among two experiments. Dotted line is used to determine the IC50 values.
G SMAD3 depletion (siRNA#1 & #2) decreased cell density and increased BRAFi effect (vemurafenib) on BRAFi-resistant cells (SKMel28R). CTR for non-targeting siRNA.
DMSO for dimethylsulfoxide (solvent of vemurafenib; BRAFi). n = 3 biologically independent experiments. Each histogram represents the mean  s.d.; Multiple
comparisons have been done using ordinary one-way ANOVA, **P < 0.01.
H SMAD3 depletion (siRNA#1 & #2) decreased cell density and increased BRAFi effect (vemurafenib, BRAFi) on BRAFi-resistant cells (Me1402). CTR for non-targeting
siRNA. DMSO for dimethylsulfoxide (solvent of vemurafenib). n = 3 biologically independent experiments. Each histogram represents the mean  s.d.; multiple
comparisons have been done using ordinary one-way ANOVA, *P < 0.05, ** P < 0.01.
I Validation of SMAD3 knock-down by Western blot experiments in Me1402 cells exposed or not to vemurafenib (BRAFi (Vem) 5 µM, 2 days). Cells were exposed to
BRAFi 24 h after siRNA transfection. Vemurafenib inhibitory effect on mutated BRAF was evaluated by analyzing the phospho-ERK1/2 levels. ERK and HSC70 serves
as loading control.
J Validation of SMAD3 inhibitor (SIS3, SMAD3i). Effect of SMAD3i (10 µM) on the level of phospho-SMAD3 Ser423/425 in response to TGFb (2 ng/ml, 1 h) (or
solvent: HCl 4 mM + Bovine serum albumin 1 mg/ml) in SKMel28R cells (expressing a high endogenous level of SMAD3 mRNA). Serum starved cells (500,000
per well) were pretreated with SMAD3i 10 µM (or control solvent) during 2 h before TGFb addition. SMAD3 and HSC70 serve as loading control for Western
blot experiments.
K Inhibitory effect of SMAD3i (SIS3) on the SMAD-responsive luciferase activity. Vector encodes the Firefly luciferase reporter gene under the control of a minimal (m)
CMV promoter and tandem repeats of the SMAD Binding Element (SBE). Cells (10,000 per well) were pretreated with SMAD3i 10 µM or control solvent during 1.5 h,
and next cells were exposed to TGFb 10 ng/ml (or solvent: HCl 4 mM + Bovine serum albumin 1 mg/ml) for 6 h. n = 3 biologically independent experiments. Each
histogram represents the mean  s.d.; Bilateral Student test (with non-equivalent variances); TGFb vs TGFb+SMAD3i, *P < 0.05, **P < 0.01.
L SMAD3, Phospho-SMAD3 Ser423/425, and BIRC3 expression levels in melanoma cell lines. Four subtypes of melanoma cells (U, undifferentiated; NC, neural crest-like;
T, transitory; M, melanocytic) have been compared. 501Mel and M249 cells are melanocytic cells in contrast to dedifferentiated BRAFi-resistant cells (R). Three
couples of melanoma cell lines (Sensitive (S) and (R)) have been used to illustrate the SMAD3 and BIRC3 increases in BRAFi-resistant cell lines. HEK293 cells are used
as control (kidney). HSC70 serves as loading control for Western blot experiments. Each antibody has been evaluated on individual membrane (explaining the three
HSC70, loading controls).
M The vemurafenib decreased melanoma cell density. Cell lines have been exposed to Vem (1 or 5 µM, 84 h) to define two groups of cell lines (sensitive (blue) and
“resistant” (red) cell lines). Normal human melanocytes (NEHM) have been used to evaluate the effect of Vem on normal cells (mean of 3 independent donors).
Representative values (mean of biological triplicates) of n = 3 biologically independent experiments.
N The SMAD3 inhibitor decreased melanoma cell density. Cell lines have been exposed to SMAD3i (0, 3, 10, 15, or 20 µM, 84 h). The cell lines defined as S and R to
BRAFi in the item 5 M are indicated in blue and red. Normal human melanocytes (NEHM) have been used to evaluate the effect of SMAD3i on normal cells (n = 3
independent donors). The SMAD3i effect on NHEMs is weak and manageable for these normal cells (NHEM). Representative values (mean of biological triplicates) of
n = 2 biologically independent experiments.
O The chemical inhibition of SMAD3 by SIS3 (SMAD3i) improved current therapy effect (BRAFi + MEKi; Vem 5 µM and Cobi 1 µM) on BRAFi-resistant cells (SKMel28R,
M229R & M238R). Cells have been treated as detailed for panel M. Representative values (mean of biological triplicates) of n = 2 biologically independent
experiments.
Data information: Western blot results are representative of at least two independent experiments. Source data and unprocessed original blots are available in
Appendix Fig S2 source data. See also Fig EV3.
Source data are available online for this figure.

immunoprecipitation coupled to massively parallel DNA sequencing (Fig 6D). Long-term chemical inhibition of AhR activity reduced
data (AhR ChIP-Seq) showing AhR binding on SMAD3 promoter (Lo SMAD3 and TIPARP expression levels. In accordance with these
& Matthews, 2012; Yang et al, 2018). To demonstrate the SMAD3 results, SMAD3 expression levels decreased in AhR KO SKMel28
induction by AhR, 501Mel cells were exposed to the most potent cells (Fig 6E).
and well-known AhR ligands (TCDD for 2,3,5,7-tetrachlorodibenzo- Together, our results support the hypothesis that AhR
dioxin; ITE for 2-(1H-Indol-3-ylcarbonyl)-4-thiazolecarboxylic acid activity drives SMAD3 expression along with the acquisition of
methyl ester). These AhR ligands increased SMAD3 expression, in BRAFi resistance.
an AhR-dependent manner (Fig 6B). Comparable results were
obtained with a canonical target gene of AhR; the TCCD-induced SMAD3 drives phenotype switching and resistance to
poly(ADP ribose) polymerase gene (TIPARP), supporting the role of melanoma therapies
AhR in regulating SMAD3 expression (Fig 6C). The need of an acti-
vated AhR-promoting SMAD3 expression was further confirmed by To explore the mechanism underlying therapy sensitization upon
the use of an AhR antagonist (CH-223191) in SKMel28 cells SMAD3 inhibition, we examined the transcriptional program

ª 2021 The Authors EMBO Molecular Medicine 13: e13466 | 2021 11 of 22

EMBO Molecular Medicine Arthur Gautron et al

regulated by SMAD3 (Fig 7). We hypothesized that the transcription and EGFR (SMAD3-signature). We next showed that basal SMAD3-
factor SMAD3 may regulate the expression levels of several resistant signature is higher in the three BRAFi-resistant cell lines when
genes and thereby induce a multifactorial effect, in which multiple compared to parental cell lines (SKMel28S, M229S, and M238S)
drug resistance pathways are activated. We compared the SMAD3 (Fig 7B). The stimulation of the TGFb-SMAD3 pathway by the
ChIP-Seq (Ramachandran et al, 2018) data to BRAFi-resistance gene recombinant TGFb promoted the SMAD3 signature in these mela-
sets established from three different sources, namely the enclosed noma cell lines (Fig 7C). The inducibility is higher in parental cell
screen, the screen performed in A375 (Konermann et al, 2015), and lines (S) since the basal expression level of genes forming the
other established BRAFi-resistance genes such as AXL or NRP1 SMAD3 signature is weak in parental cells (Fig 7B). Next, we exam-
(Fig 7A). We established a list of SMAD3-regulated genes, which ined the SMAD3 signature in a large panel of melanoma cell lines
includes SLIT2, RUNX2, NRP1, MMP2, JUNB, ITGB5, AXL, AFAP1, representative of the distinct differentiation states (U - NC -T - M)

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

A

B C

D E

Figure 6.

12 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors

 Arthur Gautron et al EMBO Molecular Medicine

◀ Figure 6. The Transcription Factor AhR Drives SMAD3 Expression.

A AhR binding sites (xenobiotic responsive element (XRE); GCGTG) in human SMAD3 proximal promoter.
B AhR activation by exogenous and endogenous ligands promotes SMAD3 induction. 501Mel cells AhR wild-type or knockout have been exposed to exogenous and
endogenous AhR ligands; TCDD (5 nM) or ITE (10 µM) or the solvent (DMSO) during 10 days. n = 5 biologically independent experiments for AhR WT cells and n = 3
for AhR KO cells. Each histogram represents the mean  s.d.; multiple comparisons have been done using ordinary one-way ANOVA, ***P < 0.001.
C AhR activation by exogenous and endogenous AhR ligands promotes TIPARP induction. 501Mel cells have been treated as described in B. n = 5 biologically
independent experiments for AhR WT cells and n = 3 for AhR KO cells. Each histogram represents the mean  s.d.; multiple comparisons have been done using
ordinary one-way ANOVA, **P < 0.01, ***P < 0.001.
D AhR antagonist (CH-223191) reduces SMAD3 and TIPARP expression levels. SKMel28 cells (AhR wild type) have been exposed to CH-223191 (5 µM) or the solvent
(DMSO) during 7 days. n = 6 biologically independent experiments. Each histogram represents the mean  s.d.; Bilateral Student test (with non-equivalent
variances): ***P < 0.001.
E Loss of AhR reduces SMAD3 expression levels. SMAD3 expression has been investigated in SKMel28 cells AhR wild-type (WT) or knockout (KO). SKMel28R has been
obtained from SKMel28S by chronic exposure to non-lethal doses of BRAFi (Hugo et al, 2015b). R for BRAFi-resistant SKMel28 cells and S for sensitive. n = 2

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

biologically independent experiments. Each histogram represents the mean  s.d.
Source data are available online for this figure.

(n = 53) (Fig 7D) and cutaneous melanoma (n = 118, TCGA cohort, (proneural, classical, and mesenchymal GBM) (Jin et al, 2017). We
tumors not exposed to targeted therapy; Fig 7E). The SMAD3 signa- found that the SMAD3 signature overlaps with the mesenchymal
ture correlated with a dedifferentiation status as suggested above GBM signature (Fig 7J). Mesenchymal GBM is the most aggressive
(Figs 2G and 5B, and EV4A). Importantly, a subset of BRAF(V600E) GBM usually associated with poor overall survival (Patel et al, 2014;
melanoma patients (~20%) expressing the SMAD3-signature was Jin et al, 2017). The SMAD3 signature was also associated with
identified (Fig 7E), suggesting that these tumors contained dedif- epithelial–mesenchymal transition (EMT) in hepatoma (Fig EV4B
ferentiated melanoma cells with potential intrinsic resistance to and C).
BRAFi. In addition, SMAD3 signature could be useful to clinicians to Altogether, these results indicate that the transcription factor
propose immune checkpoint immunotherapy to their patients since SMAD3 and its downstream target genes confer resistance to
SMAD3 signature identified anti-PD1 non-responders (prior the targeted therapies by promoting a mesenchymal-like phenotype.
selection of the treatment) (Fig 7F). Therefore, these results strongly Our work identifies AhR-SMAD3 axis as a target to overcome ther-
suggest that the SMAD3-signature may be useful to identify a popu- apy resistance of melanoma.
lation of pre-existing drug-resistant cells within drug-naive lesions.
To further illustrate the clinical relevance of our results, we assessed
the expression levels of the SMAD3 signature in drug-naive and Discussion
BRAFi-resistant patients using publicly available dataset. SMAD3
signature increased in 14/16 patients exposed to BRAFi (baseline vs. Targeted therapy and immunotherapy have greatly improved the
relapse, Fig 7G). Altogether, these results indicated that the SMAD3 prognosis of patients with cancer, but resistance to these treatments
signature is associated with resistance to both current melanoma restricts the overall survival of patients. Increasing evidence indi-
therapies. cates that transcriptomic reprogramming is associated with persister
The analyses of the SMAD3 signature in tumor samples highlighted cell emergence (Puisieux et al, 2014; Bai et al, 2019) but the mecha-
the tumor heterogeneity (mRNA expression level of genes forming the nism underlying resistance from this pool of cells remains elusive.
SMAD3 signature is variable between patients, and all genes forming Targeting tumor-promoting genes leading reprogramming could
the SMAD3 signature are not high in each tumor.). As expected, this therefore constitute an attractive approach to prevent relapse, at
heterogeneity has been retrieved in melanoma cell lines (Fig 7H). In least in some specific contexts (Bailey et al, 2018). Here, using a
response to SMAD3 depletion, AXL and EGFR decreased in the two whole genome approach, we searched for pathways that trigger the
models in contrast to other genes such as MMP2, which decreased in transcriptional reprogramming of persister cells into drug-resistant
only one model. Altogether, our results indicate that the SMAD3 cells. Based on CRISPR screen, data mining, and in vivo experi-
signature assessment is probably more appropriate than quantification ments, we identified and validated three genes (SMAD3, BIRC3, and
of specific mRNAs such as EGFR or SLIT2 to track pre-existing drug- SLC9A5) able to promote both BRAFi-resistance and tumor growth.
resistant cells within drug-naive lesions. Our work expands our understanding of the biology of persister cells
As the shift toward the mesenchymal-like state confers broad and highlights novel drug vulnerabilities that can be exploited to
resistance to therapies (Redfern et al, 2018), we postulated that the develop long-lasting antimelanoma therapies.
SMAD3 signature may be associated with this particular dedifferen- Even if CRISPR-SAM screen is a leading-edge genetic tool, several
tiated phenotype. Comparing the SMAD3 signature with the classi- concerns must be considered. As observed for all screening
cal mesenchymal-like signature (Mak et al, 2016) of melanoma approaches, false positives and false negatives are engendered
(TCGA cohort) highlighted a significant correlation (Figs 7I and rendering the validation experiments a crucial step. In this study,
EV4A). To further confirm the link between the SMAD3 signature we clearly showed that the number of sgRNAs per target is an
and a mesenchymal state, we searched for similarities with other important parameter. For our best hit, EGFR, only two sgRNAs were
mesenchymal states identified in two other cancers (glioblastoma enriched in BRAFi-treated cells. Thus, it is highly likely that we
[GBM] and hepatoma). As described for the cutaneous melanoma, missed interesting BRAFi-resistance genes (false negatives) due to
different differentiation states have been characterized for GBM the number of sgRNA/gene (at least 3 sgRNAs/gene in this library).

ª 2021 The Authors EMBO Molecular Medicine 13: e13466 | 2021 13 of 22

EMBO Molecular Medicine Arthur Gautron et al

A recent publication confirmed that sgRNAs are not all functional in (Fig EV5). These two sgRNAs promote the expression of the longest
CRISPRa libraries and it could be interesting to increase the number SMAD3 isoform; the SMAD3 mRNA expressed in our model (501Mel
of sgRNAs per target and to cover more TSS per gene (Sanson et al, cells) (Fig EV5A–E). It is important to note that the nine other
2018). Interestingly, the sgRNA library used in our study displays sgRNAs targeting SMAD3 are not able to confer the BRAFi resistance
for several genes up to 27 sgRNAs. These sgRNAs target different since they target other SMAD3 isoforms. Based on these observa-
isoforms (or TSS) of these genes. By examining the sgRNAs target- tions, we believe that mRNA isoforms identified by RNA sequencing
ing SMAD3, we found that 2 sgRNA are enriched (BRAFi resistance) should be considered during the sgRNAs selection for each model.

A B C D

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

E F G

I J

Figure 7.

14 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors

 Arthur Gautron et al EMBO Molecular Medicine

◀ Figure 7. SMAD3 drives phenotype switching and resistance to melanoma therapies.

A SMAD3 signature has been established by comparing BRAFi-resistance genes and SMAD3-regulated genes identified by chromatin immunoprecipitation followed by
DNA sequencing (ChIP-Seq) (Ramachandran et al, 2018).
B Basal SMAD3 signature in six melanoma cell lines (S for sensitive to BRAFi and R for Resistant). n = 2 biologically independent experiments. Each histogram
represents the median with interquartile range.
C Inducibility of SMAD3-signature in six melanoma cell lines exposed to TGF-b (10 ng/ml, 48 h). Data were normalized to cell lines exposed to solvent (4 mM
HCl + 1 mg/ml human BSA). Values obtained with the TGF-b stimulated M238R cell line have been set to 1. n = 2 biologically independent experiments. Each
histogram represents the median with interquartile range.
D SMAD3 signature discriminates differentiation states of 53 melanoma cell lines (Tsoi et al, 2018). The SMAD3 signature in four subgroups of melanoma cell lines. Each
point represents a cell line (n = 17, 12, 14, and 10, respectively, for melanocytic, transitory, neural crest-like, and undifferentiated cell lines); each histogram
represents the median with interquartile range; multiple comparisons have been done using ordinary one-way ANOVA; **P < 0.01, ***P < 0.001.
E Heat map depicting mRNA levels of SMAD3 signature in BRAF (V600E) non-treated melanoma patients (dataset from TCGA; SKCM, BRAF(V600E) mutated: n = 118).
Three pigmentation genes (MITF, MLANA, and TYR) have been added to highlight the differentiation states of tumors. ~20% of tumors are considered as

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

dedifferentiated tumors (red box) with a high SMAD3 signature. Scale corresponds to Z scores.
F SMAD3 signature in pre-treatment biopsies from responders and non-responders to anti-PD-1 treatment (from Hugo’s cutaneous melanoma cohort, n = 15 and 13,
respectively) (Hugo et al, 2016). Each histogram represents the median with interquartile range, one-tailed Mann–Whitney test; *P = 0.0324.
G SMAD3 signature in two groups (before BRAFi treatment or during relapse) of V600E patients from Rizos’s cohort (Rizos et al, 2014) (14 on 16 patients displayed an
upregulation of SMAD3 signature during relapse).
H SMAD3 depletion decreases expression of SMAD3-regulated genes. SMAD3 knock-down by siRNA decreased mRNA expression of BRAFi-resistance genes in SKMel28R
and Me1402. NRP1 mRNA was not detected in our experimental conditions. n = 3 biologically independent experiments. Each histogram represents the mean  s.d.;
Bilateral Student test (with non-equivalent variances) *P < 0.05, **P < 0.01, *** P < 0.001. Dotted line highlights the value of 1.
I SMAD3 signature overlapped with mesenchymal signature in melanoma tumors (TCGA, n = 459) (Akbani et al, 2015). Melanocytic signature highlights the
differentiation states of tumors (Corre et al, 2018). SMAD3 signature and mesenchymal signature correlate in melanoma tumors (TCGA, n = 459) (Mak et al, 2016).
Scale corresponds to Z scores.
J The SMAD3 signature overlaps with the glioblastoma mesenchymal subtype (Jin et al, 2017). Scale corresponds to Z scores.
Source data are available online for this figure.

By this way, it would be easy to disqualify a part of the sgRNAs of which promotes the expression of various genes including potent
the library targeting weakly expressed isoforms. This attitude could BRAFi-resistance genes (AXL and EGFR). The robust transactivation
increase the score per isoform. In other words, the use of the z-score of genes encoding a transcription factor, a transporter, or an enzyme
calculated per gene may discard interesting candidates. So, an anal- is more inclined to be enriched with our protocol, especially if the
ysis based on RNA isoforms could reduce the false negatives. basal gene expression is low. Here, we identified and validated
The second lesson of this CRISPR screen is the weak sgRNAs in vitro and in vivo the transcription factor SMAD3 and the trans-
enrichment in BRAFi-exposed cells. Except for EGFR, the sgRNAs porter SLC9A5, validated in vitro and in vivo.
enrichment is about 2. Despite these values, we showed that these BRAFi resistance relies, at least in part, on the phenotypic plastic-
candidates are robust as for SMAD3 (in vitro, in vivo, and in ity of melanoma cells (Tsoi et al, 2018; Rambow et al, 2018; Corre
patients). It is tempting to explain this fact by the duration of treat- et al, 2018). These cells may escape the deleterious effect of drug
ment (BRAFi exposure) and the dose (2 µM). By increasing the dose combinations such as BRAFi + MEKi. Among these cells, those
(i.e., 5 µM), we showed that BRAFi killed more than 90% of harboring a mesenchymal-like phenotype (usually named invasive
501Mel cells in four days (Fig 5M). This protocol is not achievable cells) display high intrinsic resistance to MAPK therapeutics
because it would induce too many false negatives. The other option (Konieczkowski et al, 2014; M€ uller et al, 2014; Verfaillie et al, 2015;
consists to increase the duration of BRAFi treatment (and keep a Shaffer et al, 2017). Enrichment in AXLhigh subpopulation (consid-
low BRAFi concentration, i.e., 2 µM). However, a recent publication ered as invasive and mesenchymal-like cells) is a common feature
demonstrated that a long-term BRAFi exposure promotes a dedif- of drug-resistant melanomas (M€ uller et al, 2014; Konieczkowski
ferentiation process conferring BRAFi resistance (Tsoi et al, 2018; et al, 2014). Targeting mesenchymal-like cells using an antibody-
Bai et al, 2019). This alternative protocol could be perilous by drug conjugate, AXL-107-MMAE, showed promising effects in a
inducing cell resistance to BRAFi independently of the sgRNA preclinical model of melanoma (Boshuizen et al, 2018). The emer-
expression. Here, we selected a melanoma cell model exhibiting a gence of AXLhigh cells is currently explained by the decrease in MITF
differentiated profile as the vast majority of metastatic melanoma activity, but the mechanism of resistance to MAPK therapeutics
tumors (89% in the TCGA cohort), a short period of treatment remains unclear. Here, we demonstrate that the AhR-SMAD3 axis
(14 days) and an intermediate dose of BRAFi (2 µM). In fact, we governs the expression levels of potent BRAFi-resistance genes,
followed, except the cell line, the protocol established by Feng including AXL, EGFR, and MMP2.
Zhang’s laboratory, who developed the CRISPR-SAM library (Koner- The dedifferentiation process, conferring BRAFi resistance,
mann et al, 2015). The differentiation status of the cell line could be requires transcriptomic reprogramming by transcription factors.
important since the transactivation mediated by CRISPR-SAM is Retinoic acid receptor gamma (RXRc) was identified as a crucial
possible only for active promoter in basal condition and the magni- transcription factor that promotes the emergence of the drug-toler-
tude of transactivation relies on the basal expression level. Here, we ant subpopulation of NCSCs (Rambow et al, 2018). An increase in
showed that 501Mel cells express low level of SMAD3 in basal AXLhigh-positive cell population was reported following MAPK inhi-
condition and the transactivation obtained by CRISPR-SAM is bition in the presence of an RXRc antagonist. This increase may
substantial (Fig 1I). Moreover, SMAD3 is a transcription factor explain why this co-treatment only delays but does not completely

ª 2021 The Authors EMBO Molecular Medicine 13: e13466 | 2021 15 of 22

EMBO Molecular Medicine Arthur Gautron et al

prevent relapse in PDXs (preprint: Marin-Bejar et al, 2020). This et al, 2015; Hugo et al, 2015b), our results suggest that regulation of
confirms the need to develop strategies that prevent melanoma BRAFi-resistance genes expression is multiparametric and probably
dedifferentiation during BRAFi treatment. Thus, our data strongly more sophisticated than initially though. Nonetheless, the elucida-
suggest that SMAD3 is a key transcriptional factor involved in the tion of these transcriptional programs and networks governing
emergence of drug-resistant mesenchymal-like cells in response to BRAFi-resistance genes and relapse is important for optimal
MAPK and identify a clinically compatible approach (SMAD3i) that target selection and the development of rationale and effective
might abrogate such a trajectory. Other transcription factors have combination strategies.
been associated to BRAFi resistance such as JUN (Titz et al, 2016) BRAFi resistance may be achieved through the exposure of
and AhR (Liu et al, 2017; Corre et al, 2018). We recently showed melanoma cells to TGF-b, demonstrating that transcriptome repro-
that a sustained AhR activation promotes the dedifferentiation of gramming may confer resistance without the need for pre-existing
melanoma cells and the expression of BRAFi-resistance genes (Corre or de novo mutations (Viswanathan et al, 2017). The TGF-b path-

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

et al, 2018). As proof-of concept, we demonstrated that differenti- way promotes a shift toward the mesenchymal state (Antony et al,
ated and BRAFi-sensitive cells can be directed toward an AhR- 2019). The resulting dedifferentiation modifies the expression of
dependent resistant program using AhR agonists. To abrogate the the adhesion molecules in the cell, supporting a migratory and
deleterious AhR sustained-activation, we identified Resveratrol, a invasive behavior. Together, our results strongly indicate that the
clinically compatible AhR antagonist. Combined with BRAFi, SMAD3-regulated genes are critical players in melanoma resistance
Resveratrol reduces the number of BRAFi-resistant cells and delays to therapies by promoting an EMT-like process. EMT reversal
relapse. Recently, an independent team confirmed that AhR inhibi- represents a powerful approach, as it may reduce the invasive
tion is reachable in vivo using other AhR antagonists (Kyn 101, behavior of cancer cells and favor re-differentiation, synonymous
Ikena Oncology, or the CH-223191; Campesato et al, 2020) to of a decrease in BRAFi-resistance gene expression (Giannelli et al,
improve the melanoma therapy. AhR antagonists validated in mice 2014; Rod on et al, 2015). By combining anti-EMT drug and
are currently evaluated in clinical trials (NCT04200963, Ikena targeted therapy such as SMAD3i and BRAFi, we should efficiently
Oncology) and (NCT04069026, Bayer). In addition, a water-soluble reduce amount of persister cells. We anticipate that SMAD3 inhibi-
SMAD3 inhibitor has been recently published for the in vivo treat- tion should limit the risk of resistance to therapies, since a
ment (Wu et al, 2020), suggesting that clinical trials should start decrease of expression levels of several BRAFi-resistance genes is
soon. Another option overcoming the potential pharmacological obtained with SMAD3i (SIS3). SMAD3 inhibition is expected to be
caveat would be to use antisense oligonucleotides (ASO) targeting more efficient than inhibitors targeting single downstream targets
AhR or SMAD3 (Leclerc et al, 2021). We and others demonstrated in such as AXL or EGFR. In conclusion, our work highlights novel
melanoma that ASO strategy is feasible in vivo by targeting drug vulnerabilities that can be exploited to develop long-lasting
SAMMSON mRNA and the lncRNA TYRP1 (Leucci et al, 2016; Gilot antimelanoma therapies.
et al, 2017). Given the plasticity of melanoma cells and the capability of
In this study, we propose an AhR-SMAD3 impairment as a strat- tumor microenvironment to produce TGF-b (Chakravarthy et al,
egy to overcome melanoma resistance. Recently, conditional dele- 2018), our work also warrants further investigation of the source of
tion of Smad7, a negative regulator of TGF-b/SMAD pathway, led to TGF-b as another approach to prevent acquisition of the therapy-
sustained melanoma growth and at the same time promoted metas- resistant mesenchymal phenotype.
tasis formation (Tuncer et al, 2019), confirming that TGF-b/SMAD
pathway is a promising target for melanoma (Javelaud et al, 2007).
In addition, Rizos’s team further illustrated the link between the Materials and Methods
TGF-b and melanoma therapy resistance. They showed that TGF-b
promotes a dedifferentiation phenotype, which is a common mecha- Reagents
nism of resistance to PD-1 inhibitors (Lee et al, 2020).
Several questions remain unsolved. We previously reported that • DMSO—Sigma-Aldrich (D8418)
activated AhR reprograms the transcriptome of melanoma cells • BRAF inhibitors: Vemurafenib (PLX4032)—Selleckchem (RG7204);
mediating BRAFi resistance. In this study, we demonstrate that a Paradox Breaker (PLX8394)—MedChemExpress (HY-18972)
SMAD3-regulated gene expression program promotes therapy resis- • SMAD3 inhibitor: SIS3—Santa Cruz Biotechnology (sc-222318)
tance in cutaneous melanoma and EMT. Importantly, SMAD3 • MEK inhibitor: Cobimetinib—Selleckchem (GDC-0973)
expression levels during resistance acquisition are dependent, at • 2,3,7,8-TCDD (TCDD)—Sigma-Aldrich (48599)
least in part, on AhR. Thus, it would be interesting to precisely • 2-(1H-Indol-3-ylcarbonyl)-4-thiazolecarboxylic acid methyl ester
define the role of AhR and SMAD3 in the induction of each BRAFi- (ITE)—MedChemExpress (HY-19317)
resistance gene. To date, no physical interaction between AhR and • CH-223191—Selleckchem (S7711)
SMAD3 proteins has been reported, suggesting that AhR acts as an • TGF-b recombinant—Santa Cruz Biotechnology (240-B-010)
upstream regulator of SMAD3 axis. It is noteworthy that the • BIRC inhibitor: Birinapant—Selleckchem (S7015)
increased expression levels of SMAD3 by AhR expands the possibil-
ity of fine tuning gene expression since SMAD3 interacts with Cell lines and culture conditions
SMAD2 but also with JUN, TEADs, and YAP1 (Zhang et al, 1998;
Fujii et al, 2012; Piersma et al, 2015). Because these three transcrip- 501Mel, Me1402, and HEK293T cell lines were obtained from ATCC.
tion factors have also been associated with therapies resistance in SKMel28 S & R cell lines were obtained from J.C Marine’s laboratory
melanoma (Nallet-Staub et al, 2014; Ramsdale et al, 2015; Verfaillie at VIB Center for Cancer Biology, VIB, Leuven, Belgium. 501Mel

16 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors

Arthur Gautron et al EMBO Molecular Medicine

and SKMel28 AhR knockout cell lines have been established as To identify tumor-promoting genes in vivo (Fig 1D), two cell
previously described (Corre et al, 2018). M229S, M229R, M238S, populations were subcutaneously xenografted on female NMRI nude
M238R, and M249 were obtained from Thomas Graeber’s laboratory mice flanks (3 × 106 cells per mouse); 501Mel (6 mice) and 501Mel
at department of Molecular and Medical Pharmacology, University CRISPR-SAM cell library (10 mice, Table EV2) for Fig 1. For xeno-
of California, Los Angeles, USA. All melanoma cell lines were grown graft of other libraries (Fig 4): Vem-persister cells (n = 11 mice) and
in humidified air (37°C, 5% CO2) in RPMI-1640 medium (Gibco (PLX8394)-persister cells (n = 12 mice) (Table EV7). Tumor volume
BRL, Invitrogen, Paisley, UK) supplemented with 10% fetal bovine was assessed according to the formula V = (W(2) × L)/2, during
serum (FBS) (PAA cell culture company) and 1% penicillin–strepto- 20 weeks. After mice sacrifice, tumors were sampled and conserved
mycin (PS) antibiotics (Gibco, Invitrogen). HEK293T was grown in at 80°C for further sgRNA enrichment analysis.
DMEM (Gibco BRL, Invitrogen, Paisley, UK) supplemented as mela- For individual validation of tumor-promoting genes (Fig 1H–K):
noma cell lines media. SKMel28R, M229R, and M238R are cultivated 3 × 106 of 501Mel cells overexpressing either SMAD3, BIRC3, or

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

in presence of 0.1 µM vemurafenib. All cell lines have been routi- SLC9A5 were subcutaneously xenografted on female NMRI nude
nely tested for mycoplasma contamination (Mycoplasma contamina- mice flanks (6 mice per group). In each group, mice were injected
tion detection kit; rep-pt1; InvivoGen—San Diego—CA). on both flanks: sgRNA 1 on right flank and sgRNA 2 on left flank
(see Table EV8 for sgRNA sequences). Tumor growth was assessed
CRISPR-SAM screens as previously described (Gilot et al, 2017) until the endpoint
(600 mm3).
A detailed protocol is available as supplementary Methods. Briefly, In order to identify BRAFi-resistance and tumor growth genes
lentiviral productions have been performed as recommended in vivo (Fig 4), three cell populations were subcutaneously xeno-
(http://tronolab.epfl.ch), using HEK293T cells, psPAX2, pVSVG, grafted on female NMRI nude mice flanks (3 × 106 cells per mice);
and vectors required for CRISPR-SAM according to Zhang lab 501Mel (6 mice), 501Mel CRISPR-SAM vemurafenib resistant (10
(Joung et al, 2017). Infections were performed overnight in presence mice), and 501Mel CRISPR-SAM Paradox Breaker resistant (12 mice).
of 4 or 8 lg of polybrene per ml. All vectors have been provided by Tumor growth was assessed as previously described (Gilot et al,
Addgene. SgRNA Library was amplified and prepared as described 2017), during 20 weeks. After mice sacrifice, tumors were sampled
by Zhang Lab (Joung et al, 2017). 501Mel cells were transduced to and conserved at 80°C for further sgRNA enrichment analysis.
stably express dCAS-VP64 (cat. no. 61425) and MS2-P65-HSF1 (cat. These experiments are compliant with all relevant ethical regula-
no. 61426), before to transduce them with SAM sgRNA library tions regarding animal research.
(lentiSAMv2, 3-plasmid system, cat. no. 1000000057) at a MOI of
0.2. Infected cells have been selected using antibiotics: Blasticidin sgRNA enrichment analysis
(2 µg/ml, 5 days), Hygromycin B (200 µg/ml, 5 days), and Zeocine
(600 µg/ml, 5 days). In vitro CRISPR-SAM screens (Fig 2) were Genomic DNAs from cell pellets (>36.106 cells) and tumors
conducted as described by Zhang laboratory (Joung et al, 2017). (> 400 mg) were extracted using the Zymo Research Quick-gDNA
The 501Mel cells expressing the sgRNA library were split into three MidiPrep according to the manufacturer’s protocol. PCR amplifi-
groups: DMSO (solvent for BRAFi), vemurafenib (PLX4032, 2 µM, cations and quality controls have been done as described by Zhang
Selleckchem), and Paradox Breaker (PLX8394, 2 µM, MedChemEx- laboratory (Joung et al, 2017).
press). After 14 days, resistant cell populations have been amplified.
A minimum of 36 × 106 cells has been pellet and further sgRNA sgRNA sequencing
enrichment analysis. Cell libraries have been cryopreserved at
80°C for further in vivo experiments. The results presented are Sequencing was performed by the Human & Environmental Geno-
pooled data from two independent screens. To generate 501Mel cells mics core facility of Rennes on a HiSeq 1500 (Rapid SBS kit v2
overexpressing individually SMAD3, BIRC3, or SLC9A5, 501Mel 1x100 cycles, Illumina). Base Calling was performed with Illumina’s
expressing dCAS-VP64 and MS2-P65-HSF1 were transduced to CASAVA pipeline (Version 1.8).
stably express specific sgRNAs (Table EV8). Infected cells have been
selected using zeocin (600 µg/ml, 5 days). Manipulations of lenti- Bioinformatic analysis of sgRNA and gene hits
virus were performed in the biosafety level 3 containment labora-
tory core facility of the Biology and Health Federative Research Data processing was conducted using the MAGeCK v0.5.6 soft-
Structure of Rennes (Biosit). ware (Li et al, 2014). Briefly, read counts from different samples
are first median-normalized to adjust for the effect of library sizes
Xenograft and read count distributions (mageck count with option: norm-
method median). Then, in an approach similar to those used for
Nude mice (4 weeks) were purchased from Janvier Labs and main- differential RNA-Seq analysis, the variance of read counts is esti-
tained under specific pathogen-free conditions in our accredited mated by sharing information across features and a negative
animal house (A 35_238_40). The animal study follows the 3R (re- binomial model is used to test whether sgRNA abundance differs
place_reduce_refine) framework and has been filed with and significantly between the treatment conditions and the DMSO
approved by the French Government Board (No. 04386.03). Animal control. Positively or negatively selected sgRNA are ranked
welfare is a constant priority: animals were thus euthanized under according to adjusted P-values (false discovery rate) and gene log
anesthesia. No blinding was done for animal studies. The animals fold changes computed with the modified robust ranking aggrega-
were randomly allocated into different groups. tion algorithm implemented in MAGeCK (mageck test with

ª 2021 The Authors EMBO Molecular Medicine 13: e13466 | 2021 17 of 22

EMBO Molecular Medicine Arthur Gautron et al

options: norm-method median, gene-lfc-method alphamedian, Western blot experiments

andadjust method fdr).
Experiments were performed as previously described (Gilot et al,
RNA interference 2017). Membranes were probed with suitable antibodies and signals
were detected using the LAS-3000 Imager (Fuji Photo Film). The
All siRNAs were transfected at 66 nM using Lipofectamine primary and secondary antibodies are described in Table EV8.
RNAiMAX (Invitrogen). For survival assay, 4,000 cells were seeded
in 96-well plates, in quadruplicates. The following day, cells have SMAD-luciferase assay
been FBS-starved (1% FBS overnight). Thirty-six hours after trans-
fection, cells were exposed to DMSO or BRAFi (vemurafenib Experiments are based on the SMAD-responsive element: Cignal
(PLX4032); Selleckchem; 1 µM). After 84 h of treatment, cell Lenti SMAD Reporter (luc) (CLS-017L from Qiagen) and as control

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

density was measured by methylene blue assay, as previously the Cignal Lenti Negative Control (luc). The SMAD-responsive luci-
described (Gilot et al, 2011). For RNA analysis, 50,000 cells were ferase vector encodes the Firefly luciferase reporter gene under the
seeded in 12-well plates. Cells were harvested 48 h after transfec- control of a minimal (m)CMV promoter and tandem repeats of the
tion. All siRNAs were purchased from IDT DNA (Table EV8). SMAD Binding Element (SBE). Melanoma cells line were infected as
previously described (Gilot et al, 2017), and infected cells were
Treatment experiments selected using antibiotic selection (puromycin 2 µg/ml, 7 days).
Cells were exposed to TGFb  SMAD3i as detailed in Fig 5K legend.
8,000 (SKMel28R) or 12,000 (Me1402) cells were seeded in 96-well Luciferase assays were then performed with a Promega kit accord-
plates, in quadruplicates. For SMAD3i + vemurafenib combination ing to the manufacturer’s instructions. Data were expressed in arbi-
treatment: 5,000 (501Mel) or 8,000 (M249, SKMel28R) cells were trary units, relative to the value of luciferase activity levels found in
seeded in 96-well plates, in triplicates. Cells were exposed, 6 h after TGFb-exposed cells, arbitrarily set at 100%. Firefly luciferase activ-
seeding, to either DMSO, BRAFi (vemurafenib; 5 µM), or MEKi (Cobi, ity was normalized to protein content using Bicinchoninic Acid Kit
1 µM) or combination BRAFi (5 µM) + SMAD3i (0, 3, 10, 15, or from Sigma-Aldrich and measured with using a luminometer (Cen-
20 µM or solvent)  MEKi (1 µM) for 4 days. 8,000 (SKMel28R) or tro XS3 LB960, Berthold Technologies).
12,000 (Me1402) cells were seeded in 96-well plates, in quadruplicates.
For Birinapant treatment (96 h): cells were exposed, 6 h after seeding, Immunohistochemistry
to either DMSO or Birinapant (100 nM). Cell density was evaluated
using methylene blue assay, as previously described (Gilot et al, 2011). Tissue samples from representative lesions (preprint: Marin-Bejar
et al, 2020) were collected and fixed in 4% paraformaldehyde for
Half maximal inhibitory concentrations (IC50 values) 24 h and then processed for paraffin embedding (HistoStarTM
Embedding Workstation). Sections of 5 µm of thickness obtained
Cell sensitivity to vemurafenib or SMAD3i (SIS3) has been estab- from the paraffin-embedded tissues (Thermo Scientific Microm
lished by cell density measurement and calculation of the IC50 using HM355S microtome) were mounted on SuperFrostTM Plus Adhesion
GraphPad (PRISM8.0) as previously described (Corre et al, 2018). slides (Thermo Scientific).
5,000 (501Mel) or 8,000 cells (other cell lines) were plated and
exposed to inhibitor(s) at indicated concentrations for 4 days. Cells Immunofluorescence with and without tyramide
have been exposed to inhibitors 6 h after plating. signal amplification

Melanoma spheroids Antibody targeted SMAD3 was used for detecting the following
protein: (rabbit, 1:100, Cell Signaling Technologies, #9513, incuba-
Spheroids were prepared using the liquid overlay method. Briefly, tion for 30 min at RT).
500 ll of melanoma cells (10,000 cell/ml) were added to a 24-well Furthermore, the Akoya Opal Polaris 7 Color Automation IHC
plate coated with 1.5% agar (Invitrogen). Plates were left to incubate Detection Kit (Akoya, NEL871001KT) was used for the tyramide
for 72 h; by this time, cells had organized into 3-dimensional (3D) signal amplification according to the manufacturer’s protocol. All
spheroids. Spheroids were then harvested and added into 1 ml of a dewaxing and staining steps were performed using the Leica Bond
solution of collagen I (2 mg/ml – Corning) with MEM 1X (Gibco), RX slide stainer with the standard settings of the Opal 7-color (v5.2
acetic acid 0.02N and neutralization buffer (HEPES 200 mM pH 7.4; plus) IHC protocol provided by the manufacturer. The steps specific
sodium bicarbonate 2.2%; NaOH 0.2 N). The suspension was then for the staining is as follows: Antigen retrieval was performed for
added to a 12-well plate coated with 1.5% agar. Normal medium 20 min at 100°C with Bond TM Epitope Retrieval 1 solution (Leica,
was overlaid on top of the solidified collagen after 2 h of incubation. AR9961). Tissue was blocked for 15 min using a blocking buffer
After 48 h, medium was renewed. Pictures of the invading spheroids (TBS with 1% BSA (VWR, 22013, bovine serum albumin (BSA),
were monitored at different times using a Zeiss microscope. fraction V, biotium (50 g))) with 10% normal goat serum
(Invitrogen, 10000C). For introduction of the secondary-HRP, the
RNA extraction and RT–qPCR expression Envision+/HRP goat anti-Rabbit (Dako Envision + Single Reagents,
HRP, Rabbit, Code K4003) was used for the antibody raised in
Experiments have been done as previously described (Gilot et al, 2017). rabbit (SMAD3). The protein SMAD3 was detected using the OPAL
Primers used for RT–qPCR experiments are available in Table EV8. 690 reagent.

18 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors

Arthur Gautron et al EMBO Molecular Medicine

After the staining in the Leica Bond RX, slides were washed
The paper explained
twice for 2 min in TBS-Tween-20 (0.5%) and were then mounted
using ProLong Diamond Antifade Mountant (Thermo Fisher, Problem
P36961). Numbers of count cells are indicated in Appendix Fig S2 Precision medicine has greatly improved the survival of patients with
source data file. cutaneous melanoma, but relapse limits the long-lasting responses.
Relapse is explained by the capability of persister cells to survive
despite targeted therapies; however, the molecular mechanism confer-
In silico analyses ring resistance is not fully understood

Heat maps were generated with R-packages heatmap3 (Zhao et al, Results
2014). We performed a genetic screen (CRISPR activation) to identify genes
Gene Set Enrichment Analysis was performed using the Broad conferring resistance to BRAF inhibitors, promoting survival of

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

Institute software. persister cells and relapse. The current study finds that genes that
enable tumor growth of therapy-na€ıve melanoma cells are also
SKCM TCGA expression data were obtained using OncoLNC
enriched during acquisition of resistance to BRAFi. SMAD3 targeted
portal (http://www.oncolnc.org; (Anaya, 2016)). Cell state catego- genes (SMAD3-signature) could be useful to find subpopulations of
rization into four differentiation states (Undifferentiated, Neural pre-existing BRAFi-resistant cells within therapy-na€ıve lesions.
crest-like, Transitory, Melanocytic) of SKCM TCGA tumors and were Moreover, chemical inhibition of SMAD3 decreases persister cells
performed using expression data of previously established gene sets population.
(Tsoi et al, 2018).
Impact
Genomic alterations of SKCM TCGA tumors were analyzed using
We established in a clinical context that a high expression level of
cBioPortal (http://www.cbioportal.org). several tumor-promoting genes fuels cancer cell plasticity, therapy
CCLE cell lines expression data and IC50 were obtained from resistance, and relapse. By highlighting the role of tumor-promoting
GSE36139 and from the original publication (Barretina et al, 2012), genes, this work expands our understanding of the biology underlying
respectively. tumor growth. Moreover, we identify novel drug vulnerabilities that
can be exploited to develop long-lasting anticancer therapies.
Expression data of 501Mel were obtained from our previously
published RNA-seq (GSE95589) (Gilot et al, 2017).
Expression data of M229, M238, and SKMel28 at different resis-
tance steps were obtained from GSE75313 (Song et al, 2017).
Patient’s median expression of our 18 hits in baseline vs relapse t-distributed stochastic neighbor embedding (tSNE) and colored
(BRAFi) is based on expression data from GSE65186 (Hugo et al, according to their drug-tolerant state (DTC) identity (Rambow et al,
2015b) and GSE50509 (Rizos et al, 2014). 2018). In a second step, the activity of the resistance gene expres-
Expression data of invasive vs proliferative cell lines were sion signature (18 genes) was quantified per cell using the AUCell
obtained from GSE60666 (Verfaillie et al, 2015). algorithm (Aibar et al, 2017) resulting into an AUCell score
Analysis of our 18 hits expression in EGFR-positive sorted cells (0 < range<0.4), which was used to gradient-color the tSNE plot.
was realized based on GSE97682 (Shaffer et al, 2017). Finally, the activity of the resistance gene signature was quantified
Analysis of SMAD3, BIRC3, and EGFR expression among 52 cell per DTC population using GraphPad (****P < 0.0001, Mann–
lines previously categorized as Undifferentiated, Neural crest-like, Whitney test).
Transitory, or Melanocytic were performed using http://systems.c
rump.ucla.edu/dediff and GSE80829. Statistical analyses
SMAD3 ChIP-Seq data have been obtained from GSE92443
(Ramachandran et al, 2018). Data are presented as mean  s.d. unless otherwise specified,
The comparison of the median SMAD3 signature in anti-PD-1 and differences were considered significant at a P-value of less
responsive vs resistant patients has been performed via expression than 0.05. In the box plots, the line within the box is the
data from GSE78220 (Hugo et al, 2016). median, the bottom and top of the box are, respectively, the
SMAD3 signature median expression was compared to melano- first and the third quartiles, and the whiskers represent the mini-
cytic one (MITF, OCA2, MLANA, TYR, DCT) and mesenchymal one mum and maximum of all the data points. Comparisons between
(52 genes; (Mak et al, 2016)) via expression data from SKCM TCGA groups normalized to a control were carried out using bilateral
obtained from OncoLNC. Student’s test (with non-equivalent variances). When at least
The comparison between median expression of SMAD3 signature two factors were compared between two groups, a two-way
with classical, proneural, and mesenchymal ones in a cohort of ANOVA (without adjustment) was used as specified in figure
glioblastoma patients was based on expression data from legends. The statistical significance between two independent
GSE103366 (Jin et al, 2017). groups of patient samples was examined using the Mann–Whit-
Gene Set Enrichment Analysis on Huh28  TGF-b was realized ney test. All statistical analyses were performed using Prism 8
based on GSE102109 (Merdrignac et al, 2018). Gene Set Enrichment software (GraphPad, La Jolla, CA, USA) or Microsoft Excel
Analysis on 3sp cells (mesenchymal) versus 3p cells (epithelial) was software. All experiments were performed three or more times
realized based on GSE26391 (van Zijl et al, 2011). independently under similar conditions, unless otherwise speci-
Single-cell RNA-seq data of a BRAF-mutant PDX model undergo- fied in the figure legends and statistics table containing the raw
ing BRAF and MEK inhibition were retrieved (GSE116237). The 674 data (Appendix Fig S2–S3 source data) and Appendix Table S1
single cells were projected into a two-dimensional space using (P-values).

ª 2021 The Authors EMBO Molecular Medicine 13: e13466 | 2021 19 of 22

EMBO Molecular Medicine Arthur Gautron et al

Data availability Ascierto PA, McArthur GA, Dreno B, Atkinson V, Liszkay G, Di Giacomo AM,
Mandala  M, Demidov L, Stroyakovskiy D, Thomas L et al (2016)
The datasets generated during current study are available: https:// Cobimetinib combined with vemurafenib in advanced BRAFV600-mutant
www.ebi.ac.uk/arrayexpress/experiments/E-MTAB-8595/. melanoma (coBRIM): updated efficacy results from a randomised, double-
blind, phase 3 trial. Lancet Oncol 17: 1248 – 1260
Expanded View for this article is available online. Bai X, Fisher DE, Flaherty KT (2019) Cell-state dynamics and therapeutic
resistance in melanoma from the perspective of MITF and IFNc pathways.
Acknowledgements Nat Rev Clin Oncol 16: 549 – 562
The authors thank the Gene Expression and Oncogenesis team from the CNRS Bailey MH, Tokheim C, Porta-Pardo E, Sengupta S, Bertrand D, Weerasinghe
UMR6290 especially M. Migault, A. Forestier, and L. Boussemart; the Rennes A, Colaprico A, Wendl MC, Kim J, Reardon B et al (2018) Comprehensive
FHU CAMIn team; and E. Watrin and C. Hitte for providing scientific expertise. Characterization of Cancer Driver Genes and Mutations. Cell 174:

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

The authors acknowledge the SFR Biosit core facilities of Rennes 1 University 1034 – 1035
with the ARCHE animal housing facility, the cell culture L3 facility, and the Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S,
Human and Environmental Genomics platform for their help and support. This r J, Kryukov GV, Sonkin D et al (2012) The Cancer Cell Line
Wilson CJ, Leha
study received financial support from the following: Fondation ARC pour la Encyclopedia enables predictive modeling of anticancer drug sensitivity.
Recherche (ARC labellized team); Ligue Nationale Contre le Cancer (LNCC); Nature 483: 603 – 607
Departements du Grand-Ouest; Region Bretagne; University of Rennes 1; CNRS; Behan FM, Iorio F, Picco G, Gonçalves E, Beaver CM, Migliardi G, Santos R,
SFR Biosit, Association Vaincre le Cancer. Further support was provided by a Rao Y, Sassi F, Pinnelli M et al (2019) Prioritization of cancer therapeutic
“Ligue Nationale Contre le Cancer” (LNCC) Grand Ouest fellowship (AG) and targets using CRISPR-Cas9 screens. Nature 568: 511 – 516
from the Region Bretagne (AG), and Fondation ARC pour la Recherche (AG) and Berking C, Takemoto R, Schaider H, Showe L, Satyamoorthy K, Robbins P
the LNCC (AQ) and from French Ministry of Research (NT and DL) and from (2001) Transforming growth Factor-b1 increases survival of human
Institut National contre le Cancer (INCa) (AP). The authors are grateful to Feng melanoma through stroma remodeling. Cancer Res 61: 8306 – 8316
Zhang for providing the Human CRISPR 3-plasmid activation pooled library Boshuizen J, Koopman LA, Krijgsman O, Shahrabi A, Van Den Heuvel EG,
(SAM) (Addgene). Ligtenberg MA, Vredevoogd DW, Kemper K, Kuilman T, Song JY et al (2018)
Cooperative targeting of melanoma heterogeneity with an AXL antibody-
Author contributions drug conjugate and BRAF/MEK inhibitors. Nat Med 24: 203 – 212
Conceptualization: AG and DG. Methodology: AG, LB, AMQ, MA, and DG. Soft- Campesato LF, Budhu S, Tchaicha J, Weng C-H, Gigoux M, Cohen IJ, Redmond
ware: AG, MA, SC, CC, and FR. Formal analysis: AG, LB, AMQ, MA, DL, SC, OM-B, D, Mangarin L, Pourpe S, Liu C et al (2020) Blockade of the AHR restricts a
CC, and FR. Investigation: AG, LB, AMQ, AP, MA, DL, SC, AP, HML, OM-B, NT, and Treg-macrophage suppressive axis induced by L-Kynurenine. Nat Commun
DG. Writing-original draft: AG and DG. Writing-review and editing: AG, DG, LB, 11: 4011
AMQ, SC, M-DG, FR, MA, CC, and J-CM. Visualization: AG, LB, AMQ, MA, SC, FR, Chakravarthy A, Khan L, Bensler NP, Bose P, De Carvalho DD (2018) TGF-b-
CC, and DG. Supervision: DG. Project Administration: DG and M-DG. Funding: associated extracellular matrix genes link cancer-associated fibroblasts to
DG, SC, and M-DG. immune evasion and immunotherapy failure. Nat Commun 9: 4692
Chihara Y, Shimoda M, Hori A, Ohara A, Naoi Y, Ikeda J, Kagara N, Tanei T,
Conflict of interest Shimomura A, Shimazu K et al (2017) A small-molecule inhibitor of
The authors declare that they have no conflict of interest. SMAD3 attenuates resistance to anti-HER2 drugs in HER2-positive breast
cancer cells. Breast Cancer Res Treat 166: 55 – 68
For more information Corre S, Tardif N, Mouchet N, Leclair HM, Boussemart L, Gautron A, Bachelot
• The Graeber laboratory provides a webtool to investigate the expression L, Perrot A, Soshilov A, Rogiers A et al (2018) Sustained activation of the
profile of your favorite RNAs in 53 melanoma cell lines, classified in four Aryl hydrocarbon Receptor transcription factor promotes resistance to
categories based on the differentiation status. BRAF-inhibitors in melanoma. Nat Commun 9: 4775
• https://systems.crump.ucla.edu/dediff/ Fujii M, Toyoda T, Nakanishi H, Yatabe Y, Sato A, Matsudaira Y, Ito H,
Murakami H, Kondo Y, Kondo E et al (2012) TGF-b synergizes with defects
in the Hippo pathway to stimulate human malignant mesothelioma
References growth. J Exp Med 209: 479 – 494
Gao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO, Sun Y, Jacobsen
Aibar S, Gonz
alez-Blas CB, Moerman T, Huynh-Thu VA, Imrichova H, A, Sinha R, Larsson E et al (2013) Integrative analysis of complex cancer
Hulselmans G, Rambow F, Marine J-C, Geurts P, Aerts J et al (2017) genomics and clinical profiles using the cBioPortal. Sci Signal 6: pl1
SCENIC: single-cell regulatory network inference and clustering. Nat Giannelli G, Villa E, Lahn M (2014) Transforming growth factor- as a
Methods 14: 1083 – 1086 therapeutic target in hepatocellular carcinoma. Cancer Res 74: 1890 – 1894
Akbani R, Akdemir KC, Aksoy BA, Albert M, Ally A, Amin SB, Arachchi H, Arora Gilot D, Le Meur N, Giudicelli F, Le Vee M, Lagadic-Gossmann D, Theret N,
A, Auman JT, Ayala B et al (2015) Genomic classification of cutaneous Fardel O (2011) RNAi-based screening identifies kinases interfering with
melanoma. Cell 161: 1681 – 1696 dioxin-mediated up-regulation of CYP1A1 activity. PLoS One 6: e18261
Anaya J (2016) OncoLnc: linking TCGA survival data to mRNAs, miRNAs, and Gilot D, Migault M, Bachelot L, Journe F, Rogiers A, Donnou-Fournet E, Mogha
lncRNAs. PeerJ Comput Sci 2: e67 A, Mouchet N, Pinel-Marie ML, Mari B et al (2017) A non-coding function
Antony J, Thiery JP, Huang RY-J (2019) Epithelial-to-mesenchymal transition: of TYRP1 mRNA promotes melanoma growth. Nat Cell Biol 19: 1348 – 1357
Lessons from development, insights into cancer and the potential of Halaban R, Zhang W, Bacchiocchi A, Cheng E, Parisi F, Ariyan S,
EMT-subtype based therapeutic intervention. Phys Biol 16(4): 041004 Krauthammer M, McCusker JP, Kluger Y, Sznol M (2010) PLX4032, a

20 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors

Arthur Gautron et al EMBO Molecular Medicine

selective BRAFV600Ekinase inhibitor, activates the ERK pathway and Liu Y, Liang X, Yin X, Lv J, Tang K, Ma J, Ji T, Zhang H, Dong W, Jin X et al
enhances cell migration and proliferation of BRAFWTmelanoma cells. (2017) Blockade of IDO-kynurenine-AhR metabolic circuitry abrogates IFN-
Pigment Cell Melanoma Res 23: 190 – 200 c-induced immunologic dormancy of tumor-repopulating cells. Nat
Hoek KS, Goding CR (2010) Cancer stem cells versus phenotype-switching in Commun 8: 15207
melanoma. Pigment Cell Melanoma Res 23: 746 – 759 Lo R, Matthews J (2012) High-resolution genome-wide mapping of AHR and
Hugo W, Shi H, Sun L, Piva M, Song C, Kong X, Moriceau G, Hong A, Dahlman KB, ARNT binding sites by ChIP-Seq. Toxicol Sci 130: 349 – 361
Johnson DB et al (2015a) Non-genomic and immune evolution of melanoma Mak MP, Tong P, Diao L, Cardnell RJ, Gibbons DL, William WN, Skoulidis F,
acquiring MAPKi resistance HHS public access. Cell 162: 1271 – 1285 Parra ER, Rodriguez-Canales J, Wistuba II et al (2016) A patient-derived,
Hugo W, Shi H, Sun L, Piva M, Song C, Kong X, Moriceau G, Hong A, Dahlman pan-cancer EMT signature identifies global molecular alterations and
KB, Johnson DB et al (2015b) Non-genomic and immune evolution of immune target enrichment following epithelial-to-mesenchymal
melanoma acquiring MAPKi resistance. Cell 162: 1271 – 1285 transition. Clin Cancer Res 22: 609 – 620

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

Hugo W, Zaretsky JM, Sun L, Song C, Moreno BH, Hu-Lieskovan S, Berent- Marin-Bejar O, Rogiers A, Dewaele M, Femel J, Karras P, Pozniak J, Bervoets G,
Maoz B, Pang J, Chmielowski B, Cherry G et al (2016) Genomic and Van Raemdonck N, Pedri D, Swings T et al (2020) A neural crest stem cell-
transcriptomic features of response to anti-PD-1 therapy in metastatic like state drives nongenetic resistance to targeted therapy in melanoma.
melanoma. Cell 165: 35 – 44 bioRxiv https://doi.org/10.1101/2020.12.15.422929 [PREPRINT]
Javelaud D, Mohammad KS, McKenna CR, Fournier P, Luciani F, Niewolna M, Merdrignac A, Angenard G, Allain C, Petitjean K, Bergeat D, Bellaud P, Fautrel
Andre J, Delmas V, Larue L, Guise TA et al (2007) Stable overexpression of A, Turlin B, Clement B, Dooley S et al (2018) A novel transforming growth
Smad7 in human melanoma cells impairs bone metastasis. Cancer Res 67: factor beta-induced long noncoding RNA promotes an inflammatory
2317 – 2324 microenvironment in human intrahepatic cholangiocarcinoma. Hepatol
Jin X, Kim LJY, Wu Q, Wallace LC, Prager BC, Sanvoranart T, Gimple RC, Wang Commun 2: 254 – 269
X, Mack SC, Miller TE et al (2017) Targeting glioma stem cells through Meyers RM, Bryan JG, McFarland JM, Weir BA, Sizemore AE, Xu H, Dharia NV,
combined BMI1 and EZH2 inhibition. Nat Med 23: 1352 – 1361 Montgomery PG, Cowley GS, Pantel S et al (2017) Computational
Jinnin M, Ihn H, Tamaki K (2006) Characterization of SIS3, a novel specific correction of copy number effect improves specificity of CRISPR-Cas9
inhibitor of Smad3, and its effect on transforming growth factor-beta1- essentiality screens in cancer cells. Nat Genet 49: 1779 – 1784
induced extracellular matrix expression. Mol Pharmacol 69: 597 – 607 Moriceau G, Hugo W, Hong A, Shi H, Kong X, Yu CC, Koya RC, Samatar AA,
Joung J, Konermann S, Gootenberg JS, Abudayyeh OO, Platt RJ, Brigham MD, Khanlou N, Braun J et al (2015) Tunable-combinatorial mechanisms of
Sanjana NE, Zhang F (2017) Genome-scale CRISPR-Cas9 knockout and acquired resistance limit the efficacy of BRAF/MEK cotargeting but result
transcriptional activation screening. Nat Protoc 12: 828 – 863 in melanoma drug addiction. Cancer Cell 27: 240 – 256
Konermann S, Brigham MD, Trevino AE, Joung J, Abudayyeh OO, Barcena C, €ller J, Krijgsman O, Tsoi J, Robert L, Hugo W, Song C, Kong X, Possik PA,
Mu
Hsu PD, Habib N, Gootenberg JS, Nishimasu H et al (2015) Genome-scale Cornelissen-Steijger PDM, Foppen MHG et al (2014) Low MITF/AXL ratio
transcriptional activation by an engineered CRISPR-Cas9 complex. Nature predicts early resistance to multiple targeted drugs in melanoma. Nat
517: 583 – 588 Commun 5: 5712
Konieczkowski DJ, Johannessen CM, Abudayyeh O, Kim JW, Cooper ZA, Piris A, Nallet-Staub F, Marsaud V, Li L, Gilbert C, Dodier S, Bataille V, Sudol M, Herlyn
Frederick DT, Barzily-Rokni M, Straussman R, Haq R et al (2014) A M, Mauviel A (2014) Pro-invasive activity of the hippo pathway effectors
melanoma cell state distinction influences sensitivity to MAPK pathway YAP and TAZ in cutaneous melanoma. J Invest Dermatol 134: 123 – 132
inhibitors. Cancer Discov 4: 816 – 827 Nassar D, Blanpain C (2016) Cancer stem cells: basic concepts and
Krepler C, Chunduru SK, Halloran MB, He X, Xiao M, Vultur A, Villanueva J, therapeutic implications. Annu Rev Pathol Mech Dis 11: 47 – 76
Mitsuuchi Y, Neiman EM, Benetatos C et al (2013) The novel SMAC Ohanna M, Giuliano S, Bonet C, Imbert V, Hofman V, Zangari J, Bille K, Robert
mimetic birinapant exhibits potent activity against human melanoma C, Bressac-de Paillerets B, Hofman P et al (2011) Senescent cells develop a
cells. Clin Cancer Res 19: 1784 – 1794 PARP-1 and nuclear factor- B-associated secretome (PNAS). Genes Dev 25:
Lamar JM, Stern P, Liu H, Schindler JW, Jiang Z-G, Hynes RO (2012) The Hippo 1245 – 1261
pathway target, YAP, promotes metastasis through its TEAD-interaction Ooshima A, Park J, Kim SJ (2019) Phosphorylation status at Smad3 linker
domain. Proc Natl Acad Sci 109: E2441 – E2450 region modulates transforming growth factor-b-induced epithelial-
Leclerc D, Staats Pires AC, Guillemin GJ, Gilot D (2021) Detrimental activation mesenchymal transition and cancer progression. Cancer Sci 110: 481 – 488.
of AhR pathway in cancer: an overview of therapeutic strategies. Curr Paris A, Tardif N, Galibert MD, Corre S (2021) AhR and cancer: from gene
Opin Immunol 70: 15 – 26 profiling to targeted therapy. Int J Mol Sci 22: 752
Lee JH, Shklovskaya E, Lim SY, Carlino MS, Menzies AM, Stewart A, Pedersen Patel AP, Tirosh I, Trombetta JJ, Shalek AK, Gillespie SM, Wakimoto H, Cahill
B, Irvine M, Alavi S, Yang JYH et al (2020) Transcriptional downregulation DP, Nahed BV, Curry WT, Martuza RL et al (2014) Single-cell RNA-seq
of MHC class I and melanoma de- differentiation in resistance to PD-1 highlights intratumoral heterogeneity in primary glioblastoma. Science
inhibition. Nat Commun 11: 1897 344: 1396 – 1401
Leucci E, Vendramin R, Spinazzi M, Laurette P, Fiers M, Wouters J, Radaelli E, Piersma B, Bank RA, Boersema M (2015) Signaling in fibrosis: TGF-b, WNT,
Eyckerman S, Leonelli C, Vanderheyden K et al (2016) Melanoma addiction and YAP/TAZ converge. Front Med 2: 59
to the long non-coding RNA SAMMSON. Nature 531: 518 – 522 Prahallad A, Sun C, Huang S, Di Nicolantonio F, Salazar R, Zecchin D,
Levy C, Khaled M, Fisher DE (2006) MITF: master regulator of melanocyte Beijersbergen RL, Bardelli A, Bernards R (2012) Unresponsiveness of colon
development and melanoma oncogene. Trends Mol Med 12: 406 – 414 cancer to BRAF(V600E) inhibition through feedback activation of EGFR.
Li W, Xu H, Xiao T, Cong L, Love MI, Zhang F, Irizarry RA, Liu JS, Brown M, Liu Nature 483: 100 – 103
XS (2014) MAGeCK enables robust identification of essential genes from Puisieux A, Brabletz T, Caramel J (2014) Oncogenic roles of EMT-inducing
genome-scale CRISPR/Cas9 knockout screens. Genome Biol 15: 554 transcription factors. Nat Cell Biol 16: 488 – 494

ª 2021 The Authors EMBO Molecular Medicine 13: e13466 | 2021 21 of 22

EMBO Molecular Medicine Arthur Gautron et al

Ramachandran A, Viz €ller P,

an P, Das D, Chakravarty P, Vogt J, Rogers KW, Mu Tang PM-K, Zhou S, Meng X-M, Wang Q-M, Li C-J, Lian G-Y, Huang X-R, Tang Y-
Hinck AP, Sapkota GP, Hill CS (2018) TGF-b uses a novel mode of receptor J, Guan X-Y, Yan BP-Y et al (2017) Smad3 promotes cancer progression by
activation to phosphorylate SMAD1/5 and induce epithelial-to- inhibiting E4BP4-mediated NK cell development. Nat Commun 8: 14677
mesenchymal transition. Elife 7: 1 – 29 Tirosh I, Izar B, Prakadan SM, Wadsworth MH, Treacy D, Trombetta JJ, Rotem
Rambow F, Rogiers A, Marin-Bejar O, Aibar S, Femel J, Dewaele M, Karras P, A, Rodman C, Lian C, Murphy G, et al (2016) Dissecting the multicellular
Brown D, Chang YH, Debiec-Rychter M et al (2018) Toward minimal ecosystem of metastatic melanoma by single-cell RNA-seq. Science 352:
residual disease-directed therapy in melanoma. Cell 174: 843 – 855 189 – 196
Ramsdale R, Jorissen RN, Li FZ, Al-Obaidi S, Ward T, Sheppard KE, Bukczynska Titz B, Lomova A, Le A, Hugo W, Kong X, ten Hoeve J, Friedman M, Shi H,
PE, Young RJ, Boyle SE, Shackleton M et al (2015) The transcription Moriceau G, Song C et al (2016) JUN dependency in distinct early and late
cofactor c-JUN mediates phenotype switching and BRAF inhibitor BRAF inhibition adaptation states of melanoma. Cell Discov 2: 16028
resistance in melanoma. Sci Signal 8: ra82 Tsoi J, Robert L, Paraiso K, Galvan C, Sheu KM, Lay J, Wong DJL, Atefi M,

Downloaded from https://www.embopress.org on August 11, 2024 from IP 2400:79e0:8040:9ff0:856b:52c9:dd97:af07.

Rapino F, Delaunay S, Rambow F, Zhou Z, Tharun L, De Tullio P, Sin O, Shirazi R, Wang X et al (2018) Multi-stage differentiation defines
Shostak K, Schmitz S, Piepers J et al (2018) Codon-specific translation melanoma subtypes with differential vulnerability to drug-induced iron-
reprogramming promotes resistance to targeted therapy. Nature 558: dependent oxidative stress. Cancer Cell 33: 890 – 904.e5
605 – 609 Tuncer E, Calçada RR, Zingg D, Varum S, Cheng P, Freiberger SN, Deng C-X,
Redfern AD, Spalding LJ, Thompson EW (2018) The Kraken Wakes: induced Kleiter I, Levesque MP, Dummer R et al (2019) SMAD signaling promotes
EMT as a driver of tumour aggression and poor outcome. Clin Exp melanoma metastasis independently of phenotype switching. J Clin Invest
Metastasis 35: 285 – 308 129: 2702 – 2716
Rizos H, Menzies AM, Pupo GM, Carlino MS, Fung C, Hyman J, Haydu LE, Verfaillie A, Imrichova H, Atak ZK, Dewaele M, Rambow F, Hulselmans G,
Mijatov B, Becker TM, Boyd SC et al (2014) BRAF inhibitor resistance Christiaens V, Svetlichnyy D, Luciani F, Van den Mooter L et al (2015)
mechanisms in metastatic melanoma: spectrum and clinical impact. Clin Decoding the regulatory landscape of melanoma reveals TEADS as
Cancer Res 20: 1965 – 1977 regulators of the invasive cell state. Nat Commun 6: 6683
Rizzolio S, Cagnoni G, Battistini C, Bonelli S, Isella C, Van Ginderachter JA, Vetma V, Rozanc J, Charles EM, Hellwig CT, Alexopoulos LG, Rehm M (2017)
Bernards R, Di Nicolantonio F, Giordano S, Tamagnone L (2018) Examining the in vitro efficacy of the IAP antagonist birinapant as a
Neuropilin-1 upregulation elicits adaptive resistance to oncogene-targeted single agent or in combination with dacarbazine to induce melanoma cell
therapies. J Clin Invest 128: 3976 – 3990 death. Oncol Res Featur Preclin Clin Cancer Ther 25: 1489 – 1494
n J, Carducci M, Sepulveda-Sanchez JM, Azaro A, Calvo E, Seoane J,
Rodo Viswanathan VS, Ryan MJ, Dhruv HD, Gill S, Eichhoff OM, Seashore-Ludlow B,
~a I, Sicart E, Gueorguieva I, Cleverly A et al (2015) Pharmacokinetic,
Bran Kaffenberger SD, Eaton JK, Shimada K, Aguirre AJ et al (2017) Dependency
pharmacodynamic and biomarker evaluation of transforming growth of a therapy-resistant state of cancer cells on a lipid peroxidase pathway.
factor-b receptor I kinase inhibitor, galunisertib, in phase 1 study in Nature 547: 453 – 457
patients with advanced cancer. Invest New Drugs 33: 357 – 370 Wagle N, Emery C, Berger MF, Davis MJ, Sawyer A, Pochanard P, Kehoe SM,
Sanson KR, Hanna RE, Hegde M, Donovan KF, Strand C, Sullender ME, Johannessen CM, Macconaill LE, Hahn WC et al (2011) Dissecting
Vaimberg EW, Goodale A, Root DE, Piccioni F et al (2018) Optimized therapeutic resistance to RAF inhibition in melanoma by tumor genomic
libraries for CRISPR-Cas9 genetic screens with multiple modalities. Nat profiling. J Clin Oncol 29: 3085 – 3096
Commun 9: 5416 Welsh SJ, Rizos H, Scolyer RA, Long GV (2016) Resistance to combination
Shaffer SM, Dunagin MC, Torborg SR, Torre EA, Emert B, Krepler C, Beqiri M, BRAF and MEK inhibition in metastatic melanoma: where to next? Eur J
Sproesser K, Brafford PA, Xiao M et al (2017) Rare cell variability and drug- Cancer 62: 76 – 85
induced reprogramming as a mode of cancer drug resistance. Nature 546: Wu N, Lian G, Sheng J, Wu D, Yu X, Lan H, Hu W, Yang Z (2020) Discovery of
431 – 435 a novel selective water-soluble SMAD3 inhibitor as an antitumor agent.
Shalem O, Sanjana NE, Zhang F (2015) High-throughput functional genomics Bioorg Med Chem Lett 30: 127396
using CRISPR–Cas9. Nat Rev Genet 16: 299 – 311 Yang SY, Ahmed S, Satheesh SV, Matthews J (2018) Genome-wide mapping
Smith MP, Brunton H, Rowling EJ, Ferguson J, Arozarena I, Miskolczi Z, Lee JL, and analysis of aryl hydrocarbon receptor (AHR)- and aryl hydrocarbon
Girotti MR, Marais R, Levesque MP et al (2016) Inhibiting drivers of non- receptor repressor (AHRR)-binding sites in human breast cancer cells. Arch
mutational drug tolerance is a salvage strategy for targeted melanoma Toxicol 92: 225 – 240
therapy. Cancer Cell 29: 270 – 284 Zhang Y, Feng XH, Derynck R (1998) Smad3 and Smad4 cooperate with c-Jun/
Song C, Piva M, Sun L, Hong A, Moriceau G, Kong X, Zhang H, Lomeli S, Qian c-Fos to mediate TGF-beta-induced transcription. Nature 394: 909 – 913
J, Yu CC et al (2017) Recurrent tumor cell-intrinsic and -extrinsic Zhao S, Guo Y, Sheng Q, Shyr Y (2014) Advanced heat map and clustering
alterations during MAPKi-induced melanoma regression and early analysis using heatmap3. Biomed Res Int 2014: 986048
adaptation. Cancer Discov 7: 1248 – 1265 van Zijl F, Mall S, Machat G, Pirker C, Zeillinger R, Weinhaeusel A, Bilban M,
Sullivan RJ, Flaherty KT (2013) Resistance to BRAF-targeted therapy in Berger W, Mikulits W (2011) A human model of epithelial to mesenchymal
melanoma. Eur J Cancer 49: 1297 – 1304 transition to monitor drug efficacy in hepatocellular carcinoma
Sun C, Wang L, Huang S, Heynen GJJE, Prahallad A, Robert C, Haanen J, Blank progression. Mol Cancer Ther 10: 850 – 860
C, Wesseling J, Willems SM et al (2014) Reversible and adaptive resistance
to BRAF(V600E) inhibition in melanoma. Nature 508: 118 – 122 License: This is an open access article under the
Talebi A, Dehairs J, Rambow F, Rogiers A, Nittner D, Derua R, Vanderhoydonc terms of the Creative Commons Attribution License,
F, Duarte JAG, Bosisio F, Van den Eynde K et al (2018) Sustained SREBP-1- which permits use, distribution and reproduction in
dependent lipogenesis as a key mediator of resistance to BRAF-targeted any medium, provided the original work is properly
therapy. Nat Commun 9: 2500 cited.

22 of 22 EMBO Molecular Medicine 13: e13466 | 2021 ª 2021 The Authors