The role of the cancer stem cell marker CD271 in DNA damage response and drug resistance of melanoma cells

Several lines of evidence have suggested that stemness and acquired resistance to targeted inhibitors or chemotherapeutics are mechanistically linked. Here we observed high cell surface and total levels of nerve growth factor receptor/CD271, a marker of melanoma-initiating cells, in sub-populations of chemoresistant cell lines. CD271 expression was increased in drug-sensitive cells but not resistant cells in response to DNA-damaging chemotherapeutics etoposide, fotemustine and cisplatin. Comparative analysis of melanoma cells engineered to stably express CD271 or a targeting short hairpin RNA by expression profiling provided numerous genes regulated in a CD271-dependent manner. In-depth analysis of CD271-responsive genes uncovered the association of CD271 with regulation of DNA repair components. In addition, gene set enrichment analysis revealed enrichment of CD271-responsive genes in drug-resistant cells, among them DNA repair components. Moreover, our comparative screen identified the fibroblast growth factor 13 (FGF13) as a target of CD271, highly expressed in chemoresistant cells. Further we show that levels of CD271 determine drug response. Knock-down of CD271 in fotemustine-resistant cells decreased expression of FGF13 and at least partly restored sensitivity to fotemustine. Together, we demonstrate that expression of CD271 is responsible for genes associated with DNA repair and drug response. Further, we identified 110 CD271-responsive genes predominantly expressed in melanoma metastases, among them were NEK2, TOP2A and RAD51AP1 as potential drivers of melanoma metastasis. In addition, we provide mechanistic insight in the regulation of CD271 in response to drugs. We found that CD271 is potentially regulated by p53 and in turn is needed for a proper p53-dependent response to DNA-damaging drugs. In summary, we provide for the first time insight in a CD271-associated signaling network connecting CD271 with DNA repair, drug response and metastasis.


INTRODUCTION
Despite recent progress in treatment options, malignant melanoma metastasized to liver, lung or brain remains to be a non-curable disease. Overall, the therapy of stage IV melanoma by chemotherapeutics and targeted therapies results in median progression-free survival of approximately 1.5-7 months and a 5-year survival period is observed for 10% of patients only. The major obstacle to long-term patient survival is resistance acquired under therapy. Although BRAF V600E mutated melanomas, which represent 40-60% of this tumor entity, are effectively targetable with the BRAF inhibitor vemurafenib, 1 relapse occurs as early as within approximately 5 months. 2 Resistant tumors exhibit upregulation of receptor tyrosine kinase receptors PDGFRB 3 or EGFR, 4 signaling mediators such as CRAF or NRAS, as well as mutations in MEK1, MEK2 and NRAS resulting in the stimulation of the RAS/RAF/mitogen-activated protein kinase pathway (reviewed in Spagnolo et al. 2 ). A recent publication suggested that resistance to vemurafenib was also dependent on the levels of nerve growth factor receptor (NGFR, CD271) and of drug-induced nuclear factor κB (NFκB) signaling. 5,6 For therapy of melanomas lacking BRAF or other targetable mutations, the chemotherapeutics paclitaxel, cisplatin and dacarbacine are in clinical use, either administered in combination or as monotherapy. 7 Dacarbazine, its prodrug temozolomide, as well as fotemustine are favorably used to treat brain metastases because these inhibitors are capable of passing the blood-brain barrier. 8,9 Clinical trials combining these drugs with immunological or targeted therapies are ongoing. 10,11 It is well known that intrinsic resistance to chemotherapeutics involves the expression of active membrane transporters, [12][13][14] anti-apoptotic genes, 15,16 genes associated with stemness 17,18 and epigenetic modulators. 17 The increased DNA repair capacity of melanoma cells 18 provides an additional mechanism conferring intrinsic resistance to targeted drugs and chemotherapeutics. Although chemotherapy of melanoma cells leads to acquired resistance, DNA-damaging agents induce upregulation of DNA damage-binding protein 2 and xeroderma pigmentosum genes. 19 It has been shown that acquired resistance of melanoma and glioblastoma to fotemustine is directly linked to high expression of the de-alkylating enzymes. 20 Thus, acquired increase of the DNA repair capacity is likely to be a major mechanism of melanoma cells in evading chemotherapeutic interventions (reviewed in Soengas and Lowe 21 ).
In view of the complexities of resistance mechanisms in chemoresistant melanoma, it is desirable to better understand the interplay of the determining factors and potential ways of their regulation. Therefore, we set out to explore new factors and regulatory mechanisms of therapy resistance by profiling chemoresistant BRAF wild-type melanoma cells. In addition, 1 we analyzed the consequence of drug-induced changes in gene expression in non-resistant cells.

Steady-state and induced levels of CD271 protein in resistant and sensitive melanoma cells
We performed microarray-based genome-wide expression profiling of the drug-sensitive melanoma cell line MeWo (MeWo Par ) and of derivatives selected for resistance to etoposide (MeWo Eto ), fotemustine (MeWo Fote ), vindesine (MeWo Vind ) or cisplatin (MeWo Cis ). 22 By comparative analysis of profiling data, we identified subsets of genes whose elevated expression was either common or unique to the chemoresistant cells (Figure 1a and Supplementary Tables S1A and B). MeWo Cis cells were exceptional in that there were no marked expression alterations compared with MeWo Par cells. Among the strongly (fold change, FC ⩾ 2, P ⩽ 0.05) upregulated factors within the subset of 39 genes common to MeWo Fote and MeWo Vind cells, CD271 (NGFR) attracted our attention because of its propensity to mediate vemurafenib resistance 7 and stemness in melanoma 23 (Figure 1a). We verified the elevated transcription of CD271 by quantitative PCR (qPCR) and upregulation at the protein level by immunoblotting ( Figure 1b). In addition, we quantified cell surface localization of CD271 by flow cytometry (Figure 1c). On average, we found enrichment of CD271-expressing cells by a factor of 2.23 ± 0.38 in MeWo Fote , MeWo Vind and MeWo Eto cells, equivalent to a median proportion of CD271-expressing cells of 62.0 ± 11.4% compared with MeWo Par cells (27.9 ± 2.6%).
Next, we asked whether increased CD271 expression occurred as a direct response to DNA-interfering/DNA-damaging therapeutics or as a secondary effect, nonspecifically induced and maintained during the process of selection for resistant cells. We treated MeWo Par cells with the potent topoisomerase II inhibitor etoposide or the mediator of DNA crosslinks cisplatin in a dosage and time-dependent manner. Etoposide and cisplatin treatment augmented CD271 expression in MeWo Par cells at low and high doses, whereas MeWo Eto or MeWo Cis cells exhibited a constant expression over all concentrations or showed increased CD271 only at highest dose, respectively (Figures 1d and e). Enhanced CD271 protein levels remained elevated for 18 h after etoposide withdrawal ( Figure 1e, upper panels). This effect was more evident in the cisplatin time course (Figure 1e, lower panels). We also confirmed upregulation of CD271 cell surface expression following short-term treatment of MeWo Par cells with etoposide and fotemustine but not vindesine by flow cytometry. Treatment with cisplatin showed a very similar effect on CD271 expression (Figure 1f), suggesting transient upregulation at the protein level. Fotemustine treatment of MeWo Fote cells showed no further increase of the CD271 level (Supplementary Figure S1A). The dose-dependent treatment of MeWo Par or MeWo Vind cells showed only marginal or no increase of CD271, respectively (Supplementary Figure S1B). This may suggest a direct response of CD271 to DNA-interfering drugs, acting differently than microtubule-interfering drugs like vindesine. Quantification of Ki67-positive MeWo Par cells revealed no change in response to DNA-interfering drugs (Figure 1f, right panel).
A CD271-associated gene signature uncovers a regulatory linkage with DNA repair Drug resistance is associated with numerous alterations of gene expression, 24 hence we asked if CD271 can modulate drug response by impacting on the transcriptome. To identify CD271responsive transcriptional targets, we performed expression profiling of patient-derived melanoma cells T20/02 and the A375 cell line, both engineered to stably express CD271. Cell lines transfected with the CD271/green fluorescent protein (GFP) transgene showed increased endogenous and ectopic CD271 protein levels and increased cell surface localization (Supplementary Figure S2A). Genome-wide expression profiling revealed 5233 and 736 differentially regulated genes (P ⩽ 0.05) in A375 NGFR/CD271 cells and T20/02 NGFR/CD271 cells, respectively (Supplementary Figures S2B and C). Of those, 235 genes were upregulated in the two cell lines by CD271 (Figure 2a,  Supplementary Table S3). In addition, we contrasted the expression data set of T20/02 NGFR/CD271 cells with those obtained after silencing endogenous CD271 in the same cell strain. 23 This comparison identified 577 genes modulated in conjunction with the deregulation of CD271 (Figure 2a). Among those were 340 genes transcriptionally stimulated (⩾1.5-fold) following overexpression of CD271 (P ⩽ 0.05) and downregulated (⩽0.5-fold) by CD271 knock-down (P ⩽ 0.05), (Supplementary Table S4). In all, 237 genes followed a converse regulation, that is, were repressed by forced CD271 expression (Supplementary Figure S2D, Supplementary Table S5). For a deeper exploration of profiling data, we performed DAVID analysis 25,26 of gene subsets and gene set enrichment analysis (GSEA) 27 Figure S3A). To further investigate the association between CD271-responsive genes and DNA repair, we performed GSEA, particularly considering a DNA repair gene signature deposited in Broad Institute's molecular signature database (MSigDB, http://software.broadinstitute.org/ gsea/msigdb/index.jsp). Indeed, GSEA revealed enrichment of genes involved in DNA repair (nominal enrichment score = 1.76) in CD271-overexpressing cells (T20/02 NGFR/CD271 ) and conversely regulated in CD271 knock-down cells (T20/02 k.d. ) (Figure 2b). We confirmed the differential, CD271-dependent expression of RAD21, RAD51 (homologs of S. pombe and S. cervisiae, respectively), RAD51-associated protein 1 (RAD51AP1) genes, responsible for double-strand break repair, and of DNA damage-binding protein 2 and xeroderma pigmentosum genes involved in nucleotide excision repair (Wiese et al. 28 Figure S3C). To further prove the relationship between CD271 and increased expression of repair genes, we sorted three different patient-derived melanoma cell strains for high endogenous and low CD271 expression by fluorescence-assisted cell sorting. The expression pattern of the repair genes RAD21, RAD51, RAD51AP1 and MSH6 was correlated with that of CD271 in positively and negatively sorted cells (Figure 2e and Supplementary Figure S2E). CD271 regulates expression of FGF13, a mediator of chemoresistance To further exploit the relationship between CD271 expression and chemoresistance, we performed a supervised analysis of the expression profiles of MeWo Fote , MeWo Vind , MeWo Eto and MeWo Par cells by GSEA based on a consensus set of 516 upregulated CD271-responsive genes identified in T20/02 NGFR/CD271 and A375 NGFR/CD271 cells (Supplementary Table S6). Compared with MeWo Par cells, we observed enrichment of CD271-responsive genes most evident in MeWo Fote and MeWo Vind , but not MeWo Eto cells (Figure 3a). We further asked for specific enrichment of genes involved in DNA repair and cell cycle in chemoresistant cells by using of GSEA and a subset of genes of the CD271 consensus signature, constituting both processes. We found highest enrichment of CD271-responsive genes associated with DNA damage and cell cycle processes in MeWo Fote and MeWo Vind but not CD271 in DNA-repair and drug response T Redmer et al Figure S4A). Genes of this subset followed a CD271-dependent regulation as validated in engineered cells (Supplementary Figure S4A, lower panels). In addition, GSEA led to identification of a CD271-responsive gene signature in MeWo Fote and MeWo Vind cells (Figure 3b), comprising important components of the DNA damage repair machinery such as PRKDC (DNA-dependent protein kinase catalytic subunit) and RAD51 and the cisplatin resistance associated gene encoding fibroblast growth factor 13 (FGF13). 30 We confirmed high expression of FGF13 in MeWo Fote cells by qPCR   In (b, d, e), whole-cell lysates were analyzed, β-tubulin served as loading control; in (c, f) 10 000 cells were recorded following incubation with either CD271-PE (top panels) or mouse IgG1 isotype control (mIgG1, lower panels). Shown are mean values ± s.d. as indicated.  . Integrative analysis of melanoma cells with overexpression and/or knock-down of CD271 links CD271 with DNA repair. (a) Venn diagram depicting commonly upregulated genes induced in A375 and T20/02 cells engineered to stably express CD271 (A375 NGFR/CD271 , T20/02 NGFR/CD271 ) or stably transfected with a CD271-targeting shRNA (shCD271, T20/02 k.d. ). Numbers of identified genes found commonly upregulated in A375 NGFR/CD271 and T20/02 NGFR/CD271 (235 genes) or differentially regulated in T20/02 NGFR/CD271 and T20/02 k.d. cells (577 genes, comprising up and downregulated genes) are shown. Right panel: DNA repair cluster from DAVID analysis of a set of 340 CD271-responsive genes upregulated in T20/02 cells, red or blue indicates positive or negative GO-contribution, respectively. (b) GSEA of profiling data of T20/02 NGFR/CD271 and T20/02 k.d. cells using a DNA repair signature (230 genes) as provided by MSigDB. 18 Figure 4a). Next, we assessed the susceptibility of CD271 knock-down cells to fotemustine. In MeWo Fote-shCD271 cells (clones shCD271_5-2 and shCD271_5-1), the susceptibility to fotemustine was partially restored as indicated by reduced viability of clone 5-2 at 100 μg/ml (76.9 ± 2.6% vs 109.9 ± 10.1%) and clone 5-1 at 300 μg/ml (47.5 ± 11.5% vs 78.6 ± 7.1%), as well as by enhanced apoptosis following 48-h treatment ( Figure 4b). In addition, we explored the response of clone 5-1 to cisplatin, vindesine and etoposide and observed that none of these showed a comparable response as compared with fotemustine (Supplementary Figure S5B). Moreover, we performed a stable knock-down of CD271 in MeWo Vind cells, by two independent short hairpin RNAs (shRNAs; sh#3, sh#4) yielding in a robust downregulation of CD271 ( CD271-responsive genes confer a metastatic phenotype on melanoma cells As selective pressure occurring upon drug-based therapies can lead to changes in expression levels of genes favoring metastasis and relapse, 31,32 we asked whether expression of CD271 and its responsive genes may be involved in mediating a migratory/ metastatic phenotype. We took advantage of a life cell imaging system based wound-healing assay, which allowed us to monitor migration of A375 NGFR/CD271 and MeWo Par-NGFR/CD271 and control cells (A375 GFP , MeWo Par-GFP ) within 2 days. We observed that overexpression of CD271 in A375 and MeWo cells specifically increased the migratory phenotype and accelerated wound closure in both cellular systems ( Figure S7A). The CD271-responsive genes RAD51AP1 and the known resistance mediators NIMA-related kinase 2 (NEK2) 33 and Hyaluronan-mediated motility receptor (HMMR), 34 Figure 2d). To further assess whether expression of CD271-responsive genes may not only promote chemoresistance but also metastasis formation, we took advantage of publicly available expression profiling data of primary (n = 31) and    Table S7). Among them, we found NEK2, RAD51AP1 and LYPD1 (LY6/PLAUR domain containing 1), a neuronal transmembrane protein potentially associated with brain metastases. 36 Statistical analysis revealed that expression of the CD271-responsive genes NEK2 (P = 1.56E-06), RAD51AP1 (P = 2.67E-04), HMMR (P = 5.47E-07) and TOP2A (P = 1.56E-06) but not FGF13 (P = 0.3122) significantly discriminated primary tumors and melanoma metastases (Figure 5f and Supplementary Figures S7B (d) Immunoblot analysis of T20/02 cells stably transfected with shCD271 or control, following dose-dependent treatment with etoposide for levels of CD271, p-p53 S15 , p21 CIP and γH2AX. (e) Analysis of cells described in d but stably transfected with different CD271-targeting shRNAs (sh#2, sh#3, sh#4) or control (shCtl.) for levels of p53, p21 CIP , CD95, NOXA and BAX. Scale indicates relative expression levels ± s.d. of independent triplicates, *P ⩽ 0.05; **P ⩽ 0.01; ***P ⩽ 0.001. In (a, b, d), whole-cell lysates were analyzed, β-tubulin served as loading control. (f) Panel-based next-generation sequencing (NGS) revealed detection of a TP53 mutation (p.E258K, 5′-GAA-4AAA-3'; antisense (3′-CTT → TTT-5′)) in MeWo Par cells, not present in wild-type T20/02 cells. Shown are sequences, as well as read depths. (g) Left panel: immunoblot analysis of cells described in d for levels of CD271 and p21 CIP following a dose-dependent treatment with MDM2 inhibitor nutlin-3a. Right panel: nutlin-3a treatment of MeWo Par cells.  Figure S7D). 37 Keratin 17 was found inversely correlated with expression of CD271-responsive genes in melanoma metastases (Figure 5f), upregulated upon CD271 knock-down (data not shown). The role of CD271 as mediator of melanoma metastasis is further underpinned by the inverse regulation of the known suppressor of melanoma metastasis KISS1 38 and DMBT1 (deleted in brain tumors 1). DMBT1 expression is frequently lost in lung cancer 39 and glioblastoma. 40 We observed a strong upregulation of both genes upon knock-down of CD271 in MeWo derivatives and T20/02 cells ( Supplementary Figures S6D and E), suggesting that CD271 expression leads to repression of both metastasis inhibitors, facilitating melanoma metastasis.

MeWo Eto cells (Supplementary
As we identified mediators of melanoma metastasis among CD271-responsive genes, we next asked whether these genes can be induced in MeWo Par cells by fotemustine treatment. We observed increased expression of NEK2 and RAD51AP1 (Supplementary Figure S7E, left panel) as well as strong increased levels of the ligands of CD271, nerve growth factor (NGF, 13.5-fold) and neurotrophin 4 (~3.5-fold) (Figure 5g) following fotemustine treatment for 24 h. Brain-derived neurotrophic factor was only marginally increased, melanoma antigen recognized by T cells 1 showed inverse correlation with CD271 and neurotrophins. In conclusion, the enhanced expression of neurotrophins like NGF upon drug treatment and in drug-resistant cells (MeWo Fote ) (Supplementary Figure S7E, right panel) suggests that the genetic programs governing drug resistance and metastasis may overlap at least partially.

CD271 is induced in a p53-dependent manner
To assess the mechanisms regulating the drug-mediated induction of CD271, we explored signaling processes potentially induced on drug treatment of MeWo Par cells. Pathway components included stabilized p53 (p-p53 S15 ), p21 CIP (CDKN1A), mitogen-activated protein kinase (extracellular signal-regulated kinases 1 and 2 (ERK1/2), p38 kinase), AKT signaling as well as checkpoint-kinase 1 (CHK1), a direct target of ataxia telangiectasia mutated and ataxia telangiectasia and RAD3-related signaling. The latter is the DNA damage sensor pathway and marks sites of DNA damage via formation of γH2AX a crucial step initiating DNA repair. 29 We also determined phosphorylation of IκBα and the nuclear localization of NFκB/p65, as activation of NFκB signaling was earlier recognized as a proposed mechanism of CD271 regulation. 7 We observed a strong activation of mitogen-activated protein kinase signaling (p-ERK1/2) and the p53-signaling pathway following treatment with cisplatin (Cis), etoposide (Eto) or temozolomide, as well as a moderate activation of p38 kinase and CHK1 (Figure 6a), however, we did not detect activation of canonical NFκB signaling (Figure 6a and Supplementary Figure S8A). In contrast to MeWo Par cells, p53 was not stabilized in chemoresistant MeWo Eto and MeWo Cis cells upon drug treatment (Figure 6b). Moreover, known p53 targets 41 and CD271 were co-regulated in etoposide-treated MeWo Par but not in MeWo Eto cells (Figure 6c), supporting the conclusion that the p53 pathway is impaired in resistant cells. We observed a lower induction of p53 targets upon treatment with cisplatin in MeWo Par cells as compared with etoposide and also observed a weak regulation of targets in MeWo Cis cells (Supplementary Figure S8B). To further assess the connection between p53 and CD271, we investigated the response of the p53 pathway in a patient-derived cell strain following stable knock-down of CD271 and dosedependent treatment with etoposide. Etoposide treatment increased levels of γH2AX, p-p53 S15 and p21 CIP in control cells but not in CD271 knock-down cells (Figure 6d). The levels of p53 itself, of p53 targets, as well as of the DNA damage sensors PRKDC, Nijmegen breakage syndrome 1 and ataxia telangiectasia mutated were downregulated in CD271 knock-down cells. Expression of ataxia telangiectasia and RAD3-related was not affected (Figure 6e and Supplementary Figure S9A). In line with downregulation of DNA repair genes, knock-down cells showed increased DNA damage (Supplementary Figure S9A, center and right panels). These data suggest that endogenous levels of CD271 are required for proper drug response. We observed comparable CD271dependent regulations of p53 targets and sensors of DNA damage repair in MeWo Fote and MeWo Vind cells upon knock-down of CD271 albeit with lower significance (Supplementary Figures S9A  and B).
The co-regulation of p53 and CD271 in drug-exposed MeWo Par cells may suggest that CD271 upregulation is controlled by p53. The CD271 promoter harbors three potential binding sites matching the consensus sequence of p53 as predicted by p53FamTaG algorithm (http://p53famtag.ba.itb.cnr.it/) (Supplementary Figure S10A). To elucidate, if p53 indeed controls CD271 expression, we treated T20/02 and MeWo Par cells with the MDM2 inhibitor nutlin-3a, which stabilizes p53 leading to activation of downstream signaling, in p53 wild-type cells. Panel-based next-generation sequencing of MeWo Par , resistant derivatives and T20/02 cells revealed the presence of a TP53 p.E258K mutation in MeWo Par and resistant cell lines and evidenced the wild-type status of T20/02 cells (Figure 6e, Supplementary Figure S10B and Supplementary Table S2A). In line with this observation, we observed a strong increase in p21 CIP but not CD271 in T20/02 (shCtl.) but not CD271 knock-down (sh#3) or MeWo Par cells (Figure 6f). These data suggest that knockdown of CD271 impairs activation of p53-dependent signaling processes by inhibition of MDM2 as well as DNA-damaging drugs. In addition, qPCR and immunofluorescence (IF) microscopy showed comparably low levels of mutant p53 in all MeWo derivatives ( Supplementary Figures S10C and D).

DISCUSSION
Stem cell-like properties are considered as a major cause for rendering cancer cells intrinsically resistant to therapeutics. 42,43 NGFR (tumor necrosis factor superfamily member 16 or CD271) has been used as a marker to enrich for tumor-initiating cells in melanoma. 44 Our own previous findings had indicated that CD271 not only regulates stem cell functions, for example, tumorigenicity and plasticity, but was also highly expressed in patient-derived cell cultures from melanoma metastases. 23 These results prompted the current investigation as to whether CD271 was involved in chemoresistance, particularly in wild-type BRAF melanomas not eligible for anti-BRAF kinase therapy. We show here that selection of MeWo cells for acquired resistance to chemotherapeutics triggering DNA damage via alkylation, cross-linking of DNA or topoisomerase inhibition was accompanied by high CD271 expression. High expression of CD271 was maintained in resistant cells, with exception of MeWo Cis cells, which may have developed a different mechanism to overcome cisplatin-induced DNA damage without the need for CD271. However, short-term exposure of drug-sensitive cells to chemotherapeutics induced CD271 expression already after 4 h, suggesting that CD271 upregulation and drug exposure were directly linked rather than due to selection of highly expressing cells present in a CD271 in DNA-repair and drug response T Redmer et al heterogeneous cell population. Moreover, silencing of CD271 partially re-sensitized resistant cells toward drug treatment and identified a link between enhanced CD271 expression and the regulation of FGF13, a known mediator of cisplatin resistance. 30 Upregulation of CD271 in response to drugs occurred in a p53-dependent manner. However, the relationship between CD271 and p53 appeared to be more complex and additional general mechanisms involved in stress response 45 may trigger CD271 expression. In drug-sensitive melanoma cells, in which intrinsically moderate CD271 expression was silenced by introduction of shRNA, the p53-dependent DNA repair system was impaired because of low expression or activation of p53 and of crucial proteins acting as DNA damage sensors PRKDC, Nijmegen breakage syndrome 1 and ataxia telangiectasia mutated. This suggested that a sufficient level of CD271 in turn is required for the proper response of cells to chemotherapeutic drugs. This conclusion is based on the results of gene expression profiling of cells in which CD271 was either silenced or enhanced by ectopic expression of a recombinant CD271 expression vector. CD271related gene signatures comprised predominantly DNA repair genes, cell cycle progression genes and genes involved in metastasis. This may explain why DNA repair genes are expressed at a higher level in CD271-expressing chemoresistant melanoma cells in the absence of p53 expression levels exceeding those in drug-sensitive cells. However, a CD271-independent upregulation of the DNA repair machinery cannot be excluded at this stage. Moreover, although our survey points out a role of CD271 in drug resistance, we only observed a partial sensitization of MeWo Fote or MeWo Vind cells to certain drugs. In addition, overexpression of CD271 was unable to fully mimic the resistant phenotype of observed in any of the resistant cells lines. This suggests that additional mechanisms are present in chemoresistant melanoma cells and clonal heterogeneity supposedly lead to variabilities in drug response.
Besides DNA repair genes, ectopic CD271 expression impinged on target genes, which have previously been associated with melanoma relapse and metastasis. 18,46 Our survey of potential CD271-associated genes in publicly available melanoma expression profiles recovered as many as 110 CD271-responsive genes predominantly expressed in metastases, among them the NEK2 gene (Never in mitosis gene A-related kinase 2), previously found associated with drug resistance in myeloma 33 and RAD51AP1 previously associated with metastasis of melanoma and breast cancer 18,47 as well as TOP2A (topoisomerase II). Expression of TOP2A was associated with worse prognosis in melanoma. 48 NEK2 was also identified as a regulator of aldehyde dehydrogenase 1, a marker of cancer-initiating cells, 49  Metastasis to the brain is frequently observed in melanoma patients, therapeutically treated with drugs able to pass the blood-brain barrier, such as fotemustine, temozolomide or dacarbacine. Recent publications demonstrated that expression of CD271 is frequently observed in melanoma brain metastases. 50 These findings are in line with our observation defining a crucial role of CD271 in regulating metastasis gene expression and conferring chemoresistance. Moreover, CD271 expression was found in brain metastatic melanoma cells resistant to vemurafenib. 8 These observations underpin our finding that CD271-responsive genes were also predominantly expressed in a small set of matched melanoma brain metastases. Notably, NGF mediates melanoma cell migration and metastasis as reported previously. 51 High expression of metastasis-associated CD271-responsive genes NEK2, RAD51AP1, TOP2A and NGF in drug-resistant MeWo cells and their drug-dependent regulation, may have consequences for the priming of melanoma cells for metastasis. Although FGF13 shows no statistical significant discrimination between melanoma primary tumors and metastases, we observed that subsets of melanoma metastases show high expression of either FGF13 or LYPD1. The role of both genes are poorly understood in melanoma metastases, however, FGF13, which is a special type of FGFs, is not only associated with signaling processes but also has a role in microtubule stabilization and the regulation of neuronal polarization and migration. 52 Hence, FGF13 may present a new driver of melanoma metastasis.
In general, these findings lead to the provocative hypothesis that sustained expression of CD271 may activate a genetic program promoting simultaneously drug resistance and metastasis. CD271-expressing cells potentially start to metastasize earlier than CD271-negative cells. This is further underpinned by a very recent work 53 and our observation that melanoma cells with a high endogenous level or overexpression of CD271 showed increased migration into a scratch wound, suggesting CD271 as a determining factor in melanoma metastasis and a candidate factor determining drug response.

Cell culture
All primary low passage melanoma cell strains, as well as their derivatives were cultured in Quantum 263 tumor growth medium (Q236, Capricorn Scientific, Marburg, Germany) supplemented with 1% penicillin/streptomycin (Invitrogen, Carlsbad, CA, USA). The patient tumor-derived cell strain T20/02 has been established from a lymph node metastasis. A375 cells (ATCC, Manassas, VA, USA) and MeWo cell lines were cultured in Leibovitz's L-15 medium (Invitrogen) supplemented with 10% fetal bovine serum, 1%, penicillin/streptomycin (Invitrogen), transferrin/fetuin, minimum essential medium-vitamins, insulin, glucose and sodium bicarbonate. Nextgeneration sequencing of T20/02 and MeWo cells revealed absence of mutations in BRAF or RAS (K/H/N), Supplementary Table S2A. Cells were cultured in 5% CO 2 , 37°C and routinely passaged at 80% confluence. Medium was changed every 3 days. Stably shRNA or control transfected fotemustine-resistant cells were selected for puromycine resistance (10 μg/ ml) and sub-cloned by limited dilution assays, which yielded in clones 5-1 and 5-2. Vindesine resistant cells were transfected with shRNA or control plasmids, selected for generation of cell pools.

Drug treatment assays
Drug treatments were performed for at least 48 h in six-well plates (2.5 × 10 5 cells seeded) for RNA/protein extraction or 96-well plates (2500 cells per well seeded) for viability assays. Stock solutions of etoposide (10 mM, dimethylsulfoxide), cisplatin (1670 mM, sodium chloride, 0.9%), temozolomide (10 mM, dimethylsulfoxide) and nutlin-3a (10 mM, dimethylsulfoxide), (all from Sigma-Aldrich, Munich, Germany) and fotemustine (50 mg/ml, ethanol, 80%) were used. Fotemustine was received either directly from the hospital and administered to cells at day of preparation, or received from company (Sigma-Aldrich). Viability was determined by MTT assay using 50 μl of MTT solution (5 mg/ml) added 48 h after drug treatment. Concentration of dimethylsulfoxide-solubilized formazan was detected at 570 nm after 1 h of incubation.
Flow cytometry/fluorescence-assisted cell sorting Fluorescence-assisted cell sorting and flow cytometry were performed as reported previously. 23 Data analysis was done with FlowJo software (http://www.flowjo.com/).

Gene expression profiling
Whole-genome expression profiling of chemoresistant and sensitive melanoma cells (MeWo, GSE78898) or cells of a patient-derived cell strain (T20/02) stably transfected either with a CD271-targeting (T20/02 k.d. ) or a control shRNA (GSE52456) plasmid, as well as T20/02 and A375 cells stably expressing either NGFR/CD271 or GFP (GSE78155) was performed with three biological replicates. Gene expression of parental (n = 1) and chemoresistant derivatives (Eto, Fote n = 3; Vind, Cis n = 2) was measured on two Illumina BeadChips (HumanHT-12V4, Illumina Inc., San Diego, CA, USA). After summarization via BeadStudio, data on each chip was quantile normalized (R package preprocessCore) and filtered (detection Po0.01 for at least one sample). After combination of chips by quantile normalization and batch correction (R package sva), significance thresholds for differential regulation were set to mean FC ⩾ 2 and adjusted P-value o0.05 (R package limma). Plots and heat maps were generated using R packages ggplot2 and heatmap.3 with spearman's rank correlation as similarity measure and complete linkage.
Gene set enrichment analysis GSEA was performed with appropriate software (GSEA version 2.2.2, Cambridge, MA, USA), as well as signatures specific for DNA repair, melanoma relapse and melanoma metastasis were received from the latest version of the Molecular Signatures Database v5.1 (MSigDB, GSEA) of the Broad Institute (http://software.broadinstitute.org/gsea/ downloads.jsp). The CD271-responsive gene signature represents a consensus (n = 516 genes) of genes upregulated in A375 NGFR/CD271 and/or T20/02 NGFR/CD271 cells. DAVID analysis was used to derive a sub-signature (n = 187 genes) from the consensus signature comprising genes associated with DNA repair and cell cycle processes.

Meta-analysis
Exploration for presence of CD271-responsive genes, publicly available data were used comprising primary melanoma with melanoma metastases of different organs (GSE8401) or extracranial with brain metastases (GSE50496). From the latter study, only matched pairs of extracranial and brain metastases were taken into consideration for analysis. Data were downloaded from GEO (http://www.ncbi.nlm.nih.gov/geo). Statistical analysis of CD271-responsive genes differentially expressed among melanoma metastases and primary melanoma was done in R using a modified version of the linear model function (lm).
Panel-based sequencing and copy number variation (CNV) Panel-based next-generation sequencing was performed following the manufacturer's instructions. Briefly, 10 ng of genomic DNA were used for preparation of barcoded PCR-libraries using the comprehensive cancer panel. Libraries (100 pM of each primer pool per cell line) were mixed, amplified and bead-coupled by emulsion PCR using the OneTouch 2 system. Sequencing of duplexed samples was performed on the Ion Personal Genome Machine (PGM) System and a 318v2 chip (all equipment from Thermo Fisher Scientific, Waltham, MA, USA). Variant call files were analyzed by our in-house pipeline and mutations were selected regarding prediction of potential impact on protein function by algorithms SIFT/ PolyPhen 54,55 and MutationTaster. 56 Analysis of CNV was performed with R package CNVPanelizer using BAM/BAI sequencing output files. Determination of CNVs of resistant MeWo cells was referenced to sensitive (MeWo Par ) cells.

Statistics
All quantitative real-time reverse transcriptase-PCR data are presented as mean ± s.d. values. Where indicated, Student's t-test (two-tailed) was used to compare two groups. A P-value ⩽ 0.05 is considered statistically significant. Center values shown represent median values. Data sets subjected to pairwise comparisons performed using Student's t-test were checked for normality and variance among groups via calculating Pearson's coefficients fell within a range of ± 0.8. For statistics of microarray data, see section Gene expression profiling.