Spitz nevi (SN) are clinically and histologically distinct pigment cell lesions that commonly arise in children and adolescents, but may also occur in older individuals 1. Spitzoid neoplasms were first described by the American pathologist Sophie Spitz in 1948 as ‘juvenile melanomas’ or as ‘melanomas of childhood’ 2. SN are characterized by distinct histological properties. They are composed of large epitheloid and/or spindle-shaped melanocytes with large nuclei that contain vesicular chromatin and often prominent nucleoli (Fig. 1c and d3). Spitzoid neoplasms with marked atypia and an aggressive clinical behavior are termed spitzoid malignant melanoma (SMM).
The distinction of SN from intermediate lesions such as atypical spitzoid tumors (AST) or spitzoid tumors of unknown malignant potential (STUMP) that may recur locally or metastasize to regional lymph nodes, but not toward distant sites, may be challenging 4–6. Molecular alterations commonly observed in melanomas are typically absent from SN and translocations that occur in SN are usually not found in melanomas 7–9.
A major obstacle in the molecular understanding of spitzoid neoplasms is the lack of basic knowledge of the molecular–genetic events that distinguish SN from common nevocellular nevi (NCN). We hypothesized that the unique features of SN, such as their cellular and nuclear characteristics as well as their stromal features, may reflect a distinct molecular differentiation profile, and that this profile may be altered in AST/STUMP and SMM and, hence, may help in the differential diagnosis in daily surgical pathology practice.
In the current study, we used the RNA NanoString nCounter Gene Expression Platform (number of genes=770) to explore the molecular differences in 15 NCN compared with 25 SN. To minimize interindividual differences, we first studied gene expression in a training group consisting of seven patients in whom both a NCN and a SN were removed. Subsequently, the resulting differential gene profile was validated on a series of eight NCN and 18 SN.
Patients and methods
Patients and tissues
Formalin-fixed and paraffin-embedded (FFPE) excision specimens of patients harboring a SN, a NCN, or both were retrieved from the archives of the Department of Pathology at the University of Leuven, KUL Belgium, and the Maastricht Pathology Tissue Collection from the Department of Pathology, Maastricht University Medical Center, MUMC+, the Netherlands. All respective samples had been excised for diagnostic and therapeutic reasons. All use of tissue and patient data was in agreement with the Dutch Code of Conduct for Observational Research with Personal Data (2004) and Tissue (2001, www.fmwv.nl) with written informed consent from all participants in accordance with the Declaration of Helsinki. Diagnoses were defined previously by histology in routine diagnostics and were confirmed by three experienced dermatopathologists (V.W., J.VdO., L.M.H.). A total of 40 cases were included in the analysis. Of these, 15 were NCN (Fig. 1a and b) and 25 were histologically diagnosed as SN (Fig. 1c and d). In seven patients, both an NCN and an SN were available for study and represented the training set. SN were defined as benign melanocytic nevi composed of large epithelioid, oval, or spindled melanocytes arranged in nests and/or fascicles without significant cytonuclear atypia (Fig. 1c and d). The study was approved by the Maastricht Ethic Committee of the University of Maastricht, the Netherlands and by the Institutional Review Board of the University Hospitals of Leuven, Belgium (project number S 59659, 14 Februrary 2017).
The training group consisted of seven patients with five women and two men ranging in age between 18 and 46 years (mean: 29.14 years, median: 26 years, Table 1). The lesions were localized on the extremities or on the corpus. No lesions were derived from the head and neck region. Histologically, one SN was diagnosed as dermal, two SN as junctional, and five as compound lesions. One compound SN (SN3) showed discrete dysplastic features with mild cytonuclear atypia and variation in cellular morphology, but did not fulfill the criteria for the diagnosis of an AST or STUMP. The same accounted for the patient’s NCN (NCN3) with some degree of histological atypia, but not fulfilling the criteria for a dysplastic nevus. From all patients, the NCN and SN were removed at the same age, with the exception of patient number 7, with excision of his SN at the age of 18, followed by an excision of his NCN at the age of 24 years.
The validation group consisted of 26 patients, with 15 female and 10 male patients showing a similar age distribution ranging from 13 to 62 years (mean: 34.08 years, median: 33 years, Table 1). From one patient, clinical data are not available (SN16). The NCN group consisted of 14 dermal, four junctional, and eight compound histological subtypes. Like the SN group, the NCN group showed localization on the extremities and the corpus, with the exception of NCN9, which had been excised from the temporal zone with a histological dermal papillomatous morphology, and SN12, which had been excised from the cervical area.
Tissue preparation and RNA isolation
Serial sections, 5 µm thick, were cut from FFPE tissue blocks using a microtome. One section per sample was mounted onto a glass slide and stained with Hematoxylin and Eosin, which served as the corresponding reference section to obtain at least 70% tumor purity. For RNA and DNA isolation, FFPE sections were collected in Eppendorf tubes, deparaffinized with xylol washing steps (1–4 times), and rehydrated in a graded alcohol series, starting at 100%, proceeding with 96%, and completing with 70% ethanol. Each time, tubes were washed with xylol/ethanol for 5 min in a centrifugation step for 5 min at 13000g. Subsequently, the pellets were dried for 1 h at 56°C. RNA and DNA were isolated using the AllPrep DNA/RNA FFPE kit (Qiagen, Hildesheim, Germany) according to the manufacturer’s instructions. Purified RNA was measured in a spectrophotometer (Nanodrop, 2000; Thermo Scientific, Landsmeer, The Netherlands). Finally, another section was cut from the rest of the FFPE block and stained with Hematoxylin and Eosin, and the inclusion criteria were evaluated again on this section.
RNA expression analysis
mRNA expression was analyzed using the NanoString nCounter Gene Expression Platform (NanoString Technologies, Center of Medical Biotechnology, University Duisburg, Essen, Germany). The methodology detects the expression of up to 770 genes (including 40 housekeeping genes) from 13 canonical pathways including hallmark cancer genes (PanCancer Pathways Panel: https://www.nanostring.com/products/gene-expression-panels/hallmarks-cancer-gene-expression-panel-collection/pancancer-pathways-panel). The mRNA array data are deposited in the NCBI’s Gene Expression Omnibus database (accession number: GSE110589).
Bioinformatic and statistical analysis
The samples were dichotomized into a training set and a validation set. The training set included the seven NCN and the seven SN samples from the same seven patients to minimize interindividual genomic background. The second group of samples included eight NCN and 18 SN from different patients. All analyses were carried out using R-/Bioconductor (https://www.bioconductor.org/).
Positive and negative controls
Each assay included six positive and eight negative controls. A positive control normalization factor was calculated (with the geometric mean method). The negative controls were used to estimate the background (mean+2SD). The average positive control normalization factor was low across assays, with 1.04 (range: 0.71–1.64). NanoString recommends a value between 0.3 and 3. The negative controls consistently produced a low value for the background (mean=23.4; range: 6.68–40.97).
Marker discovery and validation
Testing for differential expression of the nCounter data in SN versus NCN was performed using the limma and the voom method in R 10. The limma procedure has the advantage that it borrows information across all genes, which makes the analysis stable even when the sample size is small 11. The voom procedure was used to nonparametrically estimate the mean–variance relationship of the log-transformed gene counts 12. This procedure takes as input the raw counts and total library size (total number of counts) per sample is used as a scaling factor to normalize the counts. The voom procedure generates precision weights, which are incorporated into the limma linear modelling procedure and this removes the mean–variance relationship in the log-counts. The normalized log-counts were used in subsequent analyses.
The training set was first used to identify differentially expressed genes (n=14; paired). All 770 genes were used as input for the analysis and adjusted for the paired study design, similar to a paired t-test. All significant genes (P<0.05) were then further tested in the validation set (n=26, red bars, Fig. 2c). Validated gene expression markers were those that were associated with SN versus NCN in the same direction in both datasets using a P value threshold of 0.05. False discovery rate (FDR) q values, adjusted for the tests that were performed in the validation dataset, were also calculated. A heatmap of the expression data was generated using pheatmap in R.
Building gene signatures
Gene signatures for SN versus NCN were generated using two statistical learning procedures: random forests (RF, Fig. 3a and b) and least absolute shrinkage and selection operator (LASSO, Fig. 3c and d) in R. To increase study power, all tissue samples were combined for these analyses (15 SN vs. 25 NCN). A gene classifier was generated using RF in R. The optimal value for the number of candidate gene expression variables that should be tried at every split was tuned (n=27). To prevent overfitting, all the genes were used as input for the analyses. Five-hundred bootstrap samples were used. The out-of-bag (OOB) error was estimated, which is the average performance of the RF predictor using the samples not selected during bootstrap. The OOB error is therefore a valid estimate for the test error.
A LASSO signature was generated using glmnet in R. Five-fold cross-validation and the binomial deviance criterion were used to find the optimal tuning parameter for classification (lowest binomial deviance in cross-validation). Using this method, a model of gene expression markers was identified, where every marker had a corresponding LASSO coefficient, for distinguishing SN and NCN.
Gene set analysis
A pathway analysis was carried out using the camera method (limma in R) to identify sets of genes that are significantly upregulated or downregulated in SN compared with NCN 13. A matrix of gene expression levels (number of genes=770) in the different samples was used as input for the analysis. To increase power, the training and validation samples were combined similarly as described before for the RF and LASSO analyses. The following predefined Molecular Signatures Database (MSigDB) gene set categories were considered: Hallmark, Kegg, and Reactome. These gene sets were downloaded from http://bioinf.wehi.edu.au/software/MSigDB/.
Differential gene expression in Spitz nevi versus nevocellular nevi
In the training set, 197 out of the 770 genes showed differential expression in the SN versus the NCN group (P<0.05; Fig. 2a and Supplementary Fig. 1S, Supplemental digital content 1, http://links.lww.com/MR/A62). Seventy-five transcripts in the SN group had a higher mean expression level with a positive log-fold change (log FC) and 122 transcripts showed a lower mean expression level (negative log FC) compared with the NCN group. Of the 197 gene transcripts identified, 74 genes were validated in the second dataset (P≤0.05, FDR q≤0.13, Fig. 2b). Out of this set, 41 transcripts showed a positive log FC and 33 showed a negative log FC. Figure 2c shows a heatmap of the 74 identified transcripts in all samples (n=40). The gene transcript with the lowest log FC was COL24A1 (log FC=−3.12). PLA2G2A showed the highest log FC (log FC=2.80). The gene transcripts COL24A1, FGF14, GZMB, and PLA2G2A showed an absolute log FC of at least 2. Table 2 presents an overview of the discovered gene transcripts with log FC and P value for the discovery, that is, training set, and for the validation set as well as the FDR q value for each identified gene.
A Gene Set Enrichment Analysis was carried out to identify upregulated and downregulated pathways in SN compared with NCN using the camera method. Out of the 677 investigated MSigDB gene sets from Hallmark, KEGG, and Reactome, a top of nine gene pathways was identified. The MSigDB gene sets are included in the following broad categories: (i) cell cycle/proliferation; (ii) cytokine/immune/inflammatory; (iii) matrix/adhesion; (iv) hormone/receptor/signal transduction; and (v) transport and other. Figure 2d shows each of the top-three identified Hallmark, KEGG, and Reactome gene sets. All gene pathways were upregulated in SN compared with NCN, with the exception of the Reactome metabolism of the mRNA pathway. We found that four out of the top-ranked nine gene sets (all the Hallmark gene sets and the KEGG hematopoietic cell lineage gene pathway) were related to immunomodulatory and inflammatory processes. Three out of the nine gene sets showed involvement of ECM processes and interactions (two Reactome pathways and one Kegg pathway; see also Supplemental Table S1, Supplemental digital content 1, http://links.lww.com/MR/A62).
Identification of a molecular gene expression signature
Two different statistical learning methods (RF and LASSO) were used to create molecular profiles of the SN versus NCN group. For this analysis, the training and testing samples were combined. Using RF, a multigene classifier was created with an OOF error rate of 15% (Fig. 3a). The optimal value for the number of candidate gene expression variables that should be tried at every split was tuned (n=27). The Gini impurity criterion was used to measure the importance of the individual genes obtained from the RF analysis (Fig. 3b). As such, the two top-ranked genes are ITGA3 and IGFR1.
Then, the LASSO method was used. Five-fold cross-validation identified the value for the tuning parameter that resulted in the lowest binomial deviance for classification (Fig. 3c and d). From the analysis, a molecular signature consisting of 15 top-ranked differentially expressed gene transcripts was derived. Figure 3e shows a supervised heatmap of the identified genes. CAPN2, CDC25B, ITGA3, LIG4, HSPA6, DUSP10, and JAK3 were upregulated in the SN compared with the NCN group. IGFR1, IDH1, HES1, BAIAP3, SUV39H2, SHC2, FGF2, and MSH6 transcripts showed lower expression in the SN group. Table 3 provides detailed information about the top-ranked differentially expressed gene transcripts. The identified gene transcripts showed genome-wide localization (including chromosomes 1, 2, 3, 4, 10, 13, 15, 16, 17, 19, 20). The distribution of the differentially expressed gene transcripts across the genome is highlighted in a Manhattan plot (Supplementary Fig. 2S, Supplemental digital content 1, http://links.lww.com/MR/A62).
It is noteworthy that four of the top-ranked signature genes identified with the LASSO and RF procedure (JAK3, HES1, LIG1, and IDH1, Table 3 and Fig. 3a–e) are not within the group of the 74 genes differentially expressed genes in SN versus NCN (Table 2 and Fig. 2–c). The reason for this different statistical outcome is that the LASSO and RF analysis uses all samples in a single cohort. In contrast, the total group of samples is divided into a paired training set and validated in a validation set in the differential expression analysis using limma. In the differential expression analysis, the P value for those four genes is greater than 0.05 either in the training set or in the validation set. The P value for HES1, LIG4, and IDH1 exceeds 0.05 in the training group (0.07, 0.08, and 0.41, respectively). For JAK3, the P value is 0.09 in the validatiom group. Increasing the number of samples in the combined analysis yields statistically significant results.
In the present study, we have identified and validated which genes are differentially expressed in SN compared with NCN using NanoString nCounter expression profiling with a probe set of 770 genes related to cancer, growth, and invasion. To minimize interindividual differences, we studied expression levels of both types of nevi within the same individuals and validated the results in a larger group of lesions. Following this approach, we could identify a molecular signature that can distinguish SN from NCN. Our findings strongly indicate that spitzoid melanocytic lesions represent a distinct subgroup of melanocytic neoplasms 9,14. Many molecular studies have been carried out on the various types of spitzoid lesions, but the basic molecular characteristics of benign spitzoid lesions and their differences from common NCN are known to a lesser degree 15. Therefore, the aim of this study was to identify a discriminating expression profile of SN in relation to NCN and define a list of genes that may be able to provide new insights into this gap in knowledge from a molecular standpoint.
The expression profile of SN in comparison with NCN was investigated using NanoString nCounter expression profiling, which is based on the detection of RNA expression differences. This platform has been used to identify RNA expression patterns to accurately diagnose and classify different tumor types and grades including pancreatic, breast, and colorectal cancer 16–18. We applied customized NanoString expression analysis using the PanCancer pathway panel, which does not require reverse transcription or amplification. The measurement of gene expression at the RNA level yields information on the actual functional state of a cell. However, as mRNA levels do not necessarily reflect protein expression, we are currently investigating the identified signature genes on the protein level using immunohistochemistry staining.
The limma software package was used to test for differential expression in the SN group versus the NCN group 10. An important feature of limma is that it borrows information across genes, which makes it powerful even when the total sample size is small 11. Our results showed distinct expression profiles for SN and NCN that resulted in the separation of cases into two clusters. We could clearly differentiate NCN from SN in the validation cohort when the discriminatory criteria obtained from the training cohort were used. Two different learning methods (RF and LASSO) were used to identify the molecular signature of SN in relation to NCN. Different from the RF method, the LASSO method performs model fitting and variable selection. The method was used to generate a model of important genes that optimally distinguishes SN from NCN samples.
In the training set, 122 genes showed a negative log FC and 75 genes showed a positive log FC. Out of this set, 41 transcripts validated with a positive log FC and 33 showed a negative log FC (Fig. 2a and b). The reason why many downregulated genes did not validate in the final analysis might be skipping of gene filtering before the analysis. Gene filtering was not applied to prevent any form of selection bias to the subsequent analytical steps. Furthermore, many NCN in the training set showed a positive normalization factor and a relatively small library size with a lower RNA content. Although the samples were adjusted for this, it cannot be ruled out that very low expression levels might actually represent missing information. However, there should not be too much concern about this finding because gene discovery analysis with high-dimensional data usually leads to the discovery of many false positives. It should also be noted that we did not adjust for multiple testing in the training set, but used a P value less than 0.05 as criterion. Most importantly, we validated the data on a larger validation set and could remove many false positives immediately.
Gene set enrichment analysis showed that the molecular profile of SN was related to inflammatory/immunomodulatory processes and ECM interactions as well as angiogenesis-related processes. This is in line with histological observations; many SN are characterized by infiltration of lymphocytes and their stroma may show pronounced desmoplasia and increased numbers of capillaries and postcapillary venules 3,19,20. In the group of the 74 differentially expressed genes (Table 2), there was upregulation of GZMB (Granzyme B) in the SN group versus the NCN group, with a positive log FC of 2.2 (validation group) and log FC of 1.7 (training group). As assumed earlier in the literature, the reason might be an increased T-cell-bound CD8+ cytotoxic immune response in SN compared with NCN 21.
The identified gene transcripts of the SN signature comprise transcription factors (e.g. HES1) regulating cell division and growth, DNA repair molecules (e.g. MSH6, LIG4, SUV39H2), mRNA splicing proteins (e.g. HSPA6), tyrosine kinase receptores (e.g. IGFR1), kinases (e.g. JAK3), phosphatases (e.g. SHC2, DUSP10, CDC25B), proteases (e.g. CAPN2), cell adhesion, and ECM-associated molecules (e.g. ITGA3, FGF2) as well as transcripts associated with angiogenesis-associated processes (e.g. BAIAP3). Furthermore, downregulation of isocitrate dehydrogenase 1 (IDH1) was identified in the SN signature (log FC: −0.133). Previously, mutations in IDH1 have been detected in an integrated genomic analysis of human glioblastoma multiforme 22. IDH1 alterations often represent the first hit in the development of diffuse gliomas. Recently, Lazova et al.15 detected upregulation of IDH1 in an exome sequencing study in their spitzoid melanoma samples, their conventional melanoma samples, and in one of their common nevi. However, they did not detect upregulated expression of IDH1 in their SN group, which is line with our current findings. Hence, one may speculate that IDH1 immunohistochemistry staining might be a promising tool in the differential diagnosis between various spitzoid neoplasms.
In our study, the SN group showed lower expression of insulin-like growth factor 1 receptor (IGF1R, log FC=−1.32) compared with the NCN group. IGF1R functions as a transmembrane tyrosine kinase receptor and is activiated by its related hormones insulin-like growth factors 1 and 2 (IGF-1 and IGF-2). IGF represents a polypeptide hormone similar in molecular structure to insulin and plays an important role in cell growth. Upregulation of IGF1R has been described in several human malignancies 23,24 as well as cutaneous melanomas 25. Overexpression of the IGF-1R has been reported in relation to resistance development toward the BRAF V600 inhibitor vemurafenib 26–28. Currently, we lack an explanation for this paradox of the upregulated expression of IGF1R in the NCN group versus the SN group.
The presented data show a differential expression of mRNA transcripts in the SN group compared with the NCN group. Currently, we are validating the identified top-ranked gene transcripts of SN with immunohistochemistry staining to find out whether the signature is also discriminative at the protein level. We also aim to validate this signature by NanoString nCounter gene expression analysis in a larger group of SN and NCN that were in the original list of cases. Finally, we aim to find out in the future whether the identified SN signature shows consistent changes in AST/STUMP and in SMM, and thus, whether it can contribute toward diagnosis in daily surgical pathology practice.
This is the first study using NanoString nCounter Data as a novel tool in the evaluation of Spitzoid neoplasms. We identified significantly differentially expressed genes in SN compared with NCN even when analyzing the same individuals. These findings support the hypothesis that SN most likely constitute a distinct neoplastic melanocytic group that is different from other melanocytic neoplasms. The identification of the basic molecular differences in benign Spitzoid lesions may be useful in the future to develop optimal criteria for the differential diagnosis of SN from AST/STUMP and SMM.
This work was supported financially by the academic incentive fund from the Maastricht University Hospital Center, MUMC+, the Netherlands.
Conflicts of interest
There are no conflicts of interest.
1. Requena C, Requena L, Kutzner H, Sánchez Yus E. Spitz nevus
: a clinicopathological study of 349 cases. Am J Dermatopathol 2009; 31:107–116.
2. Spitz S. Melanomas of childhood. Am J Pathol 1948; 24:591–609.
3. Weedon D, Little JH. The Spitz naevus. Aust N Z J Surg 1978; 48:21–22.
4. Lallas A, Kyrgidis A, Ferrara G, Kittler H, Apalla Z, Castagnetti F, et al. Atypical Spitz tumours and sentinel lymph node biopsy: a systematic review. Lancet Oncol 2014; 15:178–183.
5. Cerroni L, Barnhill R, Elder D, Gottlieb G, Heenan P, Kutzner H, et al. Melanocytic tumors of uncertain malignant potential: results of a tutorial held at the XXIX Symposium of the International Society of Dermatopathology in Graz, October 2008. Am J Surg Pathol 2010; 34:314–326.
6. McCormack CJ, Conyers RK, Scolyer RA, Kirkwood J, Speakman D, Wong N, et al. Atypical Spitzoid neoplasms: a review of potential markers of biological behavior including sentinel node biopsy. Melanoma Res 2014; 24:437–447.
7. Bradish JR, Cheng L. Molecular pathology of malignant melanoma: changing the clinical practice paradigm toward a personalized approach. Hum Pathol 2014; 45:1315–1326.
8. Cancer Genome Atlas Network. Genomic classification of cutaneous melanoma. Cell 2015; 161:1681–1696.
9. Wiesner T, He J, Yelensky R, Esteve-Puig R, Botton T, Yeh I, et al. Kinase fusions are frequent in Spitz tumours and spitzoid melanomas. Nat Commun 2014; 5:3116.
10. Green R, Wilkins C, Thomas S, Sekine A, Ireton RC, Ferris MT, et al. Identifying protective host gene expression signatures within the spleen during West Nile virus infection in the collaborative cross model. Genom Data 2016; 10:114–117.
11. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 2015; 43:47.
12. Law CW, Chen Y, Shi W, Smyth GK. voom: Precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol 2014; 15:29.
13. Wu D, Smyth GK. Camera: a competitive gene set test accounting for inter-gene correlation. Nucleic Acids Res 2012; 40:133.
14. Bastian BC. The molecular pathology of melanoma: an integrated taxonomy of melanocytic neoplasia. Annu Rev Pathol 2014; 9:239–271.
15. Lazova R, Pornputtapong N, Halaban R, Bosenberg M, Bai Y, Chai H, et al. Spitz nevi and Spitzoid melanomas: exome sequencing and comparison with conventional melanocytic nevi and melanomas. Mod Pathol 2017; 30:640–649.
16. Garcia PL, Miller AL, Kreitzburg KM, Council LN, Gamblin TL, Christein JD, et al. The BET bromodomain inhibitor JQ1 suppresses growth of pancreatic ductal adenocarcinoma in patient-derived xenograft models. Oncogene 2016; 35:833–845.
17. Brasó-Maristany F, Filosto S, Catchpole S, Marlow R, Quist J, Francesch-Domenech E, et al. PIM1 kinase regulates cell death, tumor growth and chemotherapy response in triple-negative breast cancer. Nat Med 2016; 22:1303–1313.
18. Suzuki Y, Ng SB, Chua C, Leow WQ, Chng J, Liu SY, et al. Multiregion ultra-deep sequencing reveals early intermixing and variable levels of intratumoral heterogeneity in colorectal cancer. Mol Oncol 2017; 11:124–139.
19. Weedon D, Little JH. Spindle and epithelioid cell nevi in children and adults. A review of 211 cases of the Spitz nevus
. Cancer 1977; 40:217–225.
20. Piepkorn M. On the nature of histologic observations: the case of the Spitz nevus
. J Am Acad Dermatol 1995; 32 (Pt 1):248–254.
21. Birck A, thor Straten P, Li L, Hou-Jensen K, Sugár J, Zeuthen J. Analysis of T cell receptor AV and BV chain gene expression by infiltrating lymphocytes in Spitz naevi and in halo naevi. Melanoma Res 1997; 7:49–57.
22. Parsons DW, Jones S, Zhang X, Lin JC, Leary RJ, Angenendt P, et al. An integrated genomic analysis of human glioblastoma multiforme. Science 2008; 321:1807–1812.
23. Samani AA, Yakar S, LeRoith D, Brodt P. The role of the IGF system in cancer growth and metastasis: overview and recent insights. Endocr Rev 2007; 28:20–47.
24. Werner H. Tumor suppressors govern insulin-like growth factor signaling pathways: implications in metabolism and cancer. Oncogene 2012; 31:2703–2714.
25. Capoluongo E. Insulin-like growth factor system and sporadic malignant melanoma. Am J Pathol 2011; 178:26–31.
26. Villanueva J, Vultur A, Lee JT, Somasundaram R, Fukunaga-Kalabis M, Cipolla AK, et al. Acquired resistance to BRAF inhibitors mediated by a RAF kinase switch in melanoma can be overcome by cotargeting MEK and IGF-1R/PI3K. Cancer Cell 2010; 18:683–695.
27. Villanueva J, Vultur A, Herlyn M. Resistance to BRAF inhibitors: unraveling mechanisms and future treatment options. Cancer Res 2011; 71:7137–7140.
28. Wang J, Sinnberg T, Niessner H, Dölker R, Sauer B, Kempf WE, et al. PTEN regulates IGF-1R-mediated therapy resistance in melanoma. Pigment Cell Melanoma Res 2015; 28:572–589.