Abstract
Purpose
Initiatives such as The Cancer Genome Atlas and International Cancer Genome Consortium have generated high-quality, multiplatform molecular data from thousands of frozen tumor samples. Although these initiatives have provided invaluable insight into cancer biology, a tremendous potential resource remains largely untapped in formalin-fixed, paraffin-embedded (FFPE) samples that are more readily available but which can present technical challenges because of crosslinking of fragile molecules such as RNA.
Materials and Methods
We extracted RNA from FFPE primary melanomas and assessed two gene expression platforms—genome-wide RNA sequencing and targeted NanoString—for their ability to generate coherent biologic signals. To do so, we generated an improved approach to quantifying gene expression pathways. We refined pathway scores through correlation-guided gene subsetting. We also make comparisons to The Cancer Genome Atlas and other publicly available melanoma datasets.
Results
The comparison of the gene expression patterns to each other, to established biologic modules, and to clinical and immunohistochemical data confirmed the fidelity of biologic signals from both platforms using FFPE samples to known biology. Moreover, correlations with patient outcome data were consistent with previous frozen-tissue–based studies.
Conclusion
FFPE samples from previously difficult-to-access cancer types, such as small primary melanomas, represent a valuable and previously unexploited source of analyte for RNA sequencing and NanoString platforms. This work provides an important step toward the use of such platforms to unlock novel molecular underpinnings and inform future biologically driven clinical decisions.
INTRODUCTION
Whole transcriptome RNA sequencing (RNA-seq) has drastically improved the fidelity, resolution, and comprehensiveness of global RNA assessment compared with gene expression microarray technology. In addition to providing higher-quality assessment, particularly of low-expressing genes, it is unbound by hybridization constraints, meaning that microRNAs and noncoding RNAs, as well as multiple gene isoforms, exon-exon junctions, and fusion genes can also be assessed. Moreover, RNA-seq provides uniform measurements across genes, whereas hybridization technology can suffer from uneven individual probe quality. Therefore, gene expression can now more accurately be compared across genes and isoforms. RNA-seq is also less prone to batch effects. These advantages have been leveraged to great effect in large-scale efforts such as The Cancer Genome Atlas (TCGA)1 and International Cancer Genome Consortium,2 where thousands of samples have been assessed together with high confidence and accuracy.
One of the limitations of implementing RNA-seq technology thus far has been the requirement for analytes prepared from frozen tissues.3 Formalin-fixed, paraffin-embedded (FFPE) samples are much more plentiful but are also more susceptible to nucleic acid fragmentation and general sample deterioration, making the extraction of well-preserved RNA for sequencing a difficult challenge.4 If, however, FFPE-derived RNA-seq data could be optimized and standardized, it would exponentially expand the number of biospecimens available and streamline sample processing. Moreover, in many cancer types, limiting the analyses to only frozen tissues may result in skewing of samples toward, for example, surgically resectable and/or bulky tumors, resulting in a sample population biased by selection. The use of readily collected FFPE samples may reduce sample biases and expand the potential for longitudinal sample collection. Finally, FFPE samples are technically easier to handle, particularly if specific regions of interest are to be microdissected from slides. Primary melanomas offer a rich illustration of these important principles. In the melanoma TCGA dataset, which used frozen samples only, primary melanomas were heavily skewed toward very thick tumors (median tumor thickness, 11 mm) that correspond to an American Joint Committee on Cancer T4 primary tumor.1,5,6 In contrast, the vast majority of cutaneous melanomas diagnosed in the United States have a tumor thickness ≤ 4 mm, most of which are ≤ 1 mm, and all of which generally undergo FFPE processing for diagnostic purposes. Tumor thickness is a well-established, independent predictor of melanoma clinical outcome and the principle primary tumor factor for treatment decision making6; therefore, the lack of thinner, more clinically relevant primary melanomas in the TCGA dataset limits a full molecular understanding of the disease and optimal potential translation to the clinic. FFPE samples offer the opportunity to include tumors across all tumor thicknesses in RNA-seq datasets, for which we demonstrate feasibility in this study.
To date, only 10 studies have assessed the technological feasibility of transcriptome-scale RNA-seq from analytes prepared from human FFPE tumor tissue.7-17 Although these studies conducted technical quality controls of the data, including direct comparison of frozen and FFPE from the same samples,7-10,15 the question of whether the data accurately reflected the underlying biology relied mainly on detecting large differences between either normal and tumor tissues8 or between transcriptionally distinct tumor subtypes,8-14,16 leaving open the question of fidelity for more subtle gene expression differences between tumors. Therefore, the goal of the current study was to generate biologically relevant approaches to validating FFPE gene expression data. Using 38 primary melanoma samples, we show, through multiple approaches, that RNA-seq is capable of producing valid gene expression data from archival FFPE samples with sufficient fidelity to quantify biologically relevant pathway-level differences between tumors.
MATERIALS AND METHODS
Patient Cohort
Patients were selected from The University of Texas MD Anderson Cancer Center Melanoma, Clinical Informatics, Tissue Resource, and Translational Pathology Core database according to specific criteria (Data Supplement). The resulting cohort (N = 357) was narrowed down to a pilot set of 38 patients. For full patient information, see the Data Supplement.
RNA Extraction
The High Pure miRNA Isolation Kit (Roche, Basel, Switzerland) was used to extract RNA, and the Zymo-Spin Column (Zymo Research, Irvine, CA) was used to remove melanin from all samples, regardless of melanin content, for consistency. Whole-exome sequencing, RNA-seq, and two NanoString gene expression panels (NanoString Technologies, Seattle, WA) were performed. Additional details, including RNAseq–run metrics, are reported in the Data Supplement.
Pathology-Specific Assessment
We conducted pathologic assessment to quantify the tumor surface area before RNA extraction, and we used this to predict the required number of 10 μM slides to yield the minimum amount of total RNA per case required. An online calculator is available at http://odin.mdacc.tmc.edu/~ayjoon/MD_Anderson_Analyte_Yield_APP_For_Melanoma/index.png. Immunohistochemistry was also performed (Data Supplement).
Defining Pathway Scores
Our iterative subsetting approach to generating pathway scores depends on the quality of the initial input gene set. It makes two assumptions: highly cocorrelating genes largely represent shared biologic functionality, and high-quality input gene sets contain sufficiently strong signal from “true biology” genes such that when the initial crude score is calculated (Fig 1A), the true biology signal will emerge from the noise (Figs 1B and 1C). Therefore, selection of the proper input gene set is critical. We have provided the R program “it.r,” its manual, and additional methods and rationales in the Data Supplement.
Fig 1.
Iterative subsetting of gene sets to identify top cocorrelating genes. (A) Schematic of the algorithm. The Cancer Genome Atlas (TCGA) immune gene set is shown as a representative. The algorithm stops when the final iteration’s median equals that of the preceding iteration, having thus reached the fixed point. This final median becomes the pathway score. The orange box denotes the final “subsetted” genes that define the score. (B) Median iteration values for the TCGA immune gene set. Iteration 3 equals iteration 2 and so is not shown. (C) Median iteration values for the TCGA melanocyte gene set. (D) Final immune gene set, TCGA data. (E) Final melanocyte gene set, TCGA data.
RESULTS
RNA Extraction From FFPE Primary Melanoma Tumor Samples
We extracted RNA from 38 FFPE biopsy specimens of primary melanoma, selected on the basis of eventual clinical outcome (Table 1). RNA aliquots were prepared from each sample, and analytes were processed for transcriptome-scale RNA sequencing and a custom immune-enriched NanoString platform panel (Fig 2A; Data Supplement).
Table 1.
Patient Disease Characteristics Stratified by Disease Recurrence Status for Corresponding Study Specimens (N = 38)
Fig 2.
RNA sequencing (RNA-seq) and NanoString compare favorably on the same samples. (A) Schematic of the sample processing workflow. (B) Spearman correlation values for all 1,362 genes shared by the RNA-seq and NanoString datasets, as well as stratified by expression and dynamic range. Probes with low expression were assigned to the “low expression” category even if they had low dynamic range. Student t test was performed for nonoverlapping categories. (*)P < .001. (C) Topmost highly and poorly correlating genes between the platforms, presented as a heatmap of the RNA-seq data. (D) Median absolute deviation divided by the median (MAD/M), a measurement of dynamic range, for the RNA-seq data. Each dot is a gene corresponding to one in (C). Avg, average; Corr., correlation.
Assessing the Biologic Fidelity of FFPE RNA-seq by Individual Gene Correlation Analysis to NanoString
We first compared the FFPE, matched RNA analyte-derived RNA-seq, and NanoString datasets on a direct, gene-by-gene basis. Of 1,362 genes shared by both platforms, 924 (68%) had a Spearman correlation value of ρ > 0.5 (Fig 2B). Initially, this seemed to indicate a limited overall correlation between platforms, but we reasoned that certain genes may seem to correlate poorly for causes unrelated to cross-platform fidelity. Therefore, we next asked what properties separated the best-correlating (ρ > 0.8) from the poorest-correlating (ρ < 0.4) genes. No significant differences were seen for gene length, guanine–cytosine content, or chromosome location. Instead, we discovered two properties that sorted with correlation: absolute gene expression level and dynamic range. Genes with low absolute expression (n = 221 genes, 16% of total) showed poor correlation across platforms, suggesting the existence of a limit of resolution (Fig 2B). For dynamic range, we calculated the median absolute deviation divided by the median (MAD/M) for all genes. The best interplatform-correlating genes had a significantly higher MAD/M compared with the poorest interplatform correlating genes (P < .001; Figs 2C and 2D). This reflects the fact that sets of numbers with low variance are assessed poorly by correlation analysis. After removing low-expressing (< 1.0 average reads per kilobase million) and low-dynamic–range genes (< 0.25 MAD/M; n = 251 genes, 18% of total; Fig 2B), the percentage of genes with a Spearman correlation value of ρ > 0.5 improved from 68% to 84% (746 of 890). Overall, the individual gene-level assessment preliminarily suggests that NanoString and RNA-seq on FFPE samples provide a strong degree of agreement.
Assessing the Biologic Fidelity of FFPE RNA-seq by Pathway Correlation Analysis
We next explored pathway-level correlation. Genes with similar functions tend to share expression patterns.18 Correlation between specific related genes, therefore, can be a measure of the biologic fidelity of a given dataset. For example, BUB1 and TOP2A, both cell cycle–related genes, are consistently coexpressed across a wide variety of datasets, including that of the current study (Appendix Fig A1). When this gene-to-gene cocorrelation analysis is extended to a larger scale to identify cocorrelating pathways, the overall strength of the correlations can gauge how well the underlying biology is captured by the quality of the dataset. We chose to use TCGA melanoma data as a reference ground truth for identifying cocorrelating core pathway genes, on the basis of the high quality and quantity of data available, as well as the high degree of uniformity of the TCGA pipeline.1
To simplify pathway analyses, we sought to create a pathway score for each pathway that accurately describes its relative activation state in each sample. One obstacle to this approach is that genes within existing defined biologic gene sets (eg, from Molecular Signatures Database [MSigDb]; http://software.broadinstitute.org/gsea/msigdb)19 are not necessarily all cocorrelated with one another, potentially leading to a noisy score. We reasoned that within each gene set is a subset of highly cocorrelating genes that, together, more faithfully captures the true biology of the pathway. Therefore, we developed a method that repetitively iterates the score calculation on the basis of gene subsetting, until the score converges on a subset of highly cocorrelated genes. This subsetting is designed to lower the noise-to-signal ratio and more faithfully reflect the true biologic signal. Figure 1A illustrates this iterative process using the example of starting with the GO_Immune_System_Process from MSigDb to create an immune score; a detailed description of the steps is presented in the Data Supplement. Figures 1B and 1C show examples of how the TCGA immune and melanocyte scores change during each score iteration, and their final subset gene lists are shown in Figures 1D and 1E.
This high degree of cocorrelation of the final gene subset implies functional similarity within a broad pathway or cell type; indeed, the final immune subset of 407 genes (of an initial 1,984) was highly enriched for known true biology genes, including CD3, CD8, and CD4 genes (Fig 1D; Data Supplement). Moreover, our immune score matched well with the immune signature described in the original TCGA publication1 (Appendix Fig A2). We then used these 407 genes as a gold standard with which to assess the ability of our FFPE RNA datasets to similarly capture true immune biology. Indeed, these 407 genes preserved a high degree of cocorrelation to each other in our FFPE RNA-seq dataset (Appendix Fig A2). We also used another approach to quantify this cocorrelation: We leveraged the sample-specific dynamic range across the 407 genes’ median-centered values. Theoretically, any given sample should have fairly consistently low, mid, or high expression of all 407 genes according to whether they have a low, mid, or high presence of immune cells, respectively. Samples with high discordance would suggest that the sample may have reduced fidelity. Somewhat surprisingly, FFPE samples had a significantly lower discordance than TCGA samples (Appendix Fig A2), suggesting overall high sample quality.
We next compared our FFPE RNA-seq and NanoString datasets. Of the 407 core immune genes we identified, 190 were present on both platforms and included core canonical immune genes such as CD3D/E/G, CD4, CD8A/B, and CD86, again retaining strong cocorrelation within each platform (Figs 3A and 3B). We then assessed each of these 190 genes individually to determine how well each platform captured true immune biology as operationally defined previously in this section. We correlated each gene to the immune score within each platform and plotted the platform results against each other (Fig 3B). The majority of genes (154 of 190; 81%) had a correlation of ρ > 0.5 on both platforms, which further increased (140 of 163; 86%) after removal of low-expressing genes. No gene had a low dynamic range. Direct comparison of the final immune scores showed strong concordance between RNA-seq and NanoString (ρ = 0.91) on a per-sample basis (Fig 3C). These results suggest that, after controlling for expression level, both platforms were able to successfully capture true positive immune genes while preserving biologic fidelity.
Fig 3.
Pathway-level correlations within and between platforms. (A) Heatmap of immune genes derived from Fig 1D for the RNA-seq and NanoString datasets, with the immune score on top. (B) Intraplatform correlation to the immune score for individual immune genes, plotted as RNA-seq versus NanoString. Several key immune genes are highlighted. (C-E) Correlation of the (C) immune, (D) cell cycle, and (E) skin scores between platforms. (F, G) Correlation of the EMT and pigment scores within (F) the TCGA primary melanomas and (G) the current FFPE RNA-seq dataset. EMT, epithelial-mesenchymal transition; RNA-seq, RNA sequencing; TCGA, The Cancer Genome Atlas.
We generated scores for additional signatures: cell cycle, skin, pigment, and epithelial-mesenchymal transition (EMT; the list of starting gene sets and final gene subsets is given in the Data Supplement). The pigment gene set contains the master melanocyte differentiation regulator MITF and known downstream MITF effectors. Similar to our results for the immune gene set, there was a high degree of intraplatform correlation (Appendix Fig A3). Interplatform correlation was also observed for the two pathways with representative genes on the NanoString platform: cell cycle and skin (Figs 3D and 3E). Moreover, genes not within these five final pathways still showed strong cross-platform correlation (Appendix Fig A3). Finally, we compared the intragene-set correlation for each of the five pathways across four datasets of primary melanomas only—TCGA, the current FFPE RNA-seq, GSE7553, and GSE15605—providing additional confirmation that they are robust markers of their respective pathways (Appendix Fig A4).
Next, we took four orthogonal approaches to further support the biologic fidelity of the FFPE RNA-seq data. First, we performed a hierarchical clustering of our dataset with 21 TCGA tumor types using the most variable pan-cancer TCGA genes and showed that our FFPE dataset clustered tightly with both cutaneous and uveal melanoma, primarily by the high expression of melanocyte-specific genes (Appendix Fig A4). Second, we performed hierarchical clustering within our FFPE RNA-seq dataset using the topmost-variable genes and obtained results that closely match those of the TCGA melanoma analysis,20 namely, the identification of clusters enriched for nervous-specific, skin, and immune genes (Appendix Fig A5). Third, we compared our five pathway genes to three gene sets not expected to be highly expressed in melanoma. This analysis confirmed that our five pathways were highly expressed in the FFPE RNA-seq dataset, whereas the three nonmelanoma gene sets were largely < 1.0 average reads per kilobase million. Finally, we examined a known pathway correlation among our five gene sets, shown in multiple cell-line studies: the EMT and pigment signatures anticorrelate.21 In vivo, multiple stromal cell types express the EMT signature, weakening the correlation. Nevertheless, in the TCGA dataset, among primary melanoma samples (n = 103), the EMT and pigment signatures anticorrelated (ρ = −0.47; Fig 3F). Similar results were observed in our FFPE dataset (ρ = −0.41; Fig 3G). Taken together, these results suggest that at the pathway level, both FFPE-derived RNA-seq and NanoString datasets accurately capture established biologic correlations.
Assessing the Biologic Fidelity of FFPE RNA-seq by Comparison With Histology and Patient Data
We next asked whether our signatures aligned with immunohistochemistry (IHC) performed on the same samples. The cell cycle signature from RNA-seq correlated well with IHC for MART1/pHH322 (ie, cells double positive for MART1, a melanoma marker, and pHH3, a marker of dividing cells; Fig 4A) and the observed mitotic rate (Fig 4B). These positive correlations were also detected in the NanoString dataset (Appendix Fig A6). Moreover, individual gene expression values for CD3 and CD8A correlated with the corresponding IHC quantifications. This correlation was stronger for positive staining in the center of the tumor (Figs 4C and 4D) compared with the periphery (Appendix Fig A6).
Fig 4.
Correlation between pathway scores and histopathology. (A, B) Correlation of the RNA-sequencing cell cycle score to (A) pHH3 immunohistochemistry (IHC) and (B) mitotic rate. (C-F) Correlation of the RNA-sequencing CD8 or CD3 gene expression data to their respective IHC at either the (C, D) center or (E, F) periphery of the tumor.
We next sought to determine how pathway scores correlated with other clinical data. We found that the skin score, which quantitates normal epidermal genes (eg, involucrin and keratin 1), anticorrelated with both Breslow thickness and tumor surface area (Figs 5A and 5B), suggesting that thinner and smaller tumor biopsy specimens may have a slightly larger percentage of normal skin content. In addition, the melanocyte score negatively correlated with the chronic sun-damage score (Fig 5C), which is a measure of the degree of solar elastosis in normal skin adjacent to the primary melanoma lesions in hematoxylin-and-eosin–stained tissue sections.23
Fig 5.
Correlation between pathway scores and patient data. (A, B) Correlation of the RNA-seq skin score to (A) Breslow thickness or (B) tumor surface area. (C) Anticorrelation of the RNA-seq pigment score to the chronic sun-damage score. (D) Gene-set enrichment analysis output showing mitotic gene sets as enriched in samples from patients in whom disease eventually recurred. (E) Correlation of RNA-seq or NanoString cell cycle scores to samples stratified by patient-level recurrence status. RNA-seq, RNA sequencing.
Finally, we performed an unbiased gene-set enrichment analysis to identify pathways that correlate with patient outcome. Consistent with findings of previous gene expression studies of primary melanoma,24-26 an increase in cell cycle gene expression was significantly enriched in samples of patients in whom melanoma eventually recurred (Fig 5D). To further substantiate the biologic fidelity of our gene-set scores, we next compared the cell cycle score between primary tumors from the recurrence versus no recurrence cohorts. We observed a significant difference (P = .02) between the two groups (Fig 5E). Together, these patient-oriented assessments further support the biologic fidelity of the FFPE RNA-seq data.
DISCUSSION
We provide evidence that transcriptome-scale RNA-seq from archival FFPE samples yields biologically coherent signals that match well with clinicopathological data. In this study, we created an approach to assessing pathway scores that subsets gene lists by enriching for genes that together form functionally related sets of well-correlated expression patterns. This approach provides a clean pathway-level score for rapid comparisons across platforms and to IHC and patient-outcome data. Importantly, our findings demonstrate that NanoString and RNA-seq generate well-correlated data once low-expressing and low-dynamic–range genes are removed from further analyses, rules which can potentially be applied to other datasets. The high quality of data obtained from both platforms opens several opportunities. For example, the use of unbiased whole-transcriptome RNA-seq allows for robust discovery of unanticipated pathways of interest, using a vastly expanded set of archival FFPE samples.
To our knowledge, our study is the first to assess the biologic fidelity of continuous differences in gene expression in FFPE-derived RNA-seq data rather than bulk differences between tumor subtypes and/or normal tissue.7-17 It is also the first to directly compare FFPE-derived RNA-seq data with NanoString data and with quantitative IHC findings. Importantly, pathway-level scores largely retained ranked correlations across platforms. Applied in a clinical setting, this approach may greatly facilitate biologically meaningful categorization of samples. In the case of primary melanoma, cell cycle scores may help stratify patients with early-stage melanoma into those likely to relapse or not; more broadly, the unbiased nature of RNA-seq will enable uncovering novel biomarkers in a variety of settings. Our results also show that FFPE RNA-seq analyses preserve the same key intergene correlations as the high-quality, high-sample-number, TCGA frozen-tumor dataset. If further validated in additional datasets and appropriately deployed, our findings suggest that FFPE RNA-seq can identify relatively subtle fold-changes associated with underlying biology.
Overall, our data strongly support that FFPE-derived RNA expression data can be robustly assessed to help unlock clinically relevant biologic insights. Such studies will likely be further enabled by the continued implementation of high-quality, well-curated tissue banks that include clinically relevant patient-outcome data. We further suggest that our approach can be widely used to more accurately calculate pathway scores, because current methods process all genes in a gene set regardless of degree of cocorrelation. Going forward, we envision future large-scale archival FFPE RNA-seq-based analyses for discovery and clinical FFPE RNA analysis for validation and, ultimately, clinical application, with the added potential for applying gene-set subsetting to produce scores and stratify patients according to biologic and clinical relevance. In the future, as findings are validated, it follows that such approaches can be introduced into contemporary clinical practice to facilitate clinical decision making.
ACKNOWLEDGMENT
We thank John van Arnam for useful discussions.
Appendix
Fig. A1.
Example of preserved gene–gene correlation across datasets. Correlation of TOP2A and BUB1 in The Cancer Genome Atlas, GSE19234, GSE22155, and current formalin-fixed paraffin-embedded tissue datasets, all assessing patients’ melanoma samples.
Fig. A2.
Additional validation of the immune gene set. (A) Correlation of our The Cancer Genome Atlas (TCGA)-derived immune score and the official TCGA immune score. (B) Venn diagram illustrating the overlap in gene membership between our TCGA-derived immune gene set and the official TCGA immune gene set. (C) Top: Heatmap of the 38 formalin-fixed paraffin-embedded (FFPE) RNA-seq dataset for all 407 TCGA-derived immune gene-set genes. Bottom: The median absolute deviation divided by the median (MAD/M) for all 407 immune genes for each sample, a measure of correlative discordance within the immune signature. (D) Comparison of the immune MAD/M for the TCGA and FFPE datasets.
Fig. A3.
Heatmaps showing the final fixed-point gene sets for (A) epithelial-mesenchymal transition (EMT), (B) cell cycle, and (C) skin for RNA sequencing (RNA-seq) and NanoString. For each gene set, the RNA-seq and NanoString samples have the same order, from highest to lowest RNA-seq score. The melanocyte score genes are not available for NanoString because no melanocyte genes were present on the panel. (D) Formalin-fixed paraffin-embedded (FFPE) NanoString versus FFPE RNA-seq correlations, comparing the final gene members of the five pathways with all other genes shared between the two platforms.
Fig. A4.
Intragene-set correlations for the five pathways. Each gene-set member is correlated with its respective overall gene-set score, for primary melanomas only in the (A) The Cancer Genome Atlas (TCGA) and current formalin-fixed paraffin-embedded (FFPE) RNA-seq and (B) GSE7553 and GSE15605 datasets. The percentages at the top indicate that of genes in FFPE tissue with ρ > 0.5. (C) Hierarchical clustering of the top 697 most-variable pan-cancer genes for 21 cancer types plus our FFPE melanoma dataset. Shaded triangles highlight clusters. The yellow box indicates melanocyte-specific genes enriched across melanoma samples, and includes such genes as TYR, TYRP1, and MLANA. ACC, adrenocortical carcinoma; BLCA, bladder carcinoma; BRCA, breast carcinoma; COADREAD, colorectal adenocarcinoma; EMT, epithelial-mesenchymal transition; FFP, this study; GBM: glioblastoma multiforme; HNSC, head and neck squamous carcinoma; LIHC, hepatocellular carcinoma; KIRC, renal clear-cell carcinoma; KIRP, renal papillary carcinoma; LAML, acute myeloid leukemia; LGG, low-grade glioma; LUAD, lung adenocarcinoma; LUSC, lung squamous carcinoma; Mets, metastases; MESO, mesothelioma; PAAD, pancreatic adenocarcinoma; PRAD, prostate adenocarcinoma; OV, ovarian; SKCM met, metastatic cutaneous melanoma; SKCM pri, primary cutaneous melanoma; THCA, thyroid carcinoma; UCEC, uterine carcinoma; UVM, uveal melanoma.
Fig. A5.
(A) Hierarchical clustering of the top 5% most variable genes. Enriched gene classes are shown for three clusters on the basis of Metacore analyses. The final cluster did not have a clear enrichment. We note that the MITF/melanocyte pathway was not recovered, because of the smaller dataset (n = 38 samples here v n = 333 in TCGA) and the lower variability in melanocyte gene expression, resulting in those genes not being among the top 5%. (B) Average reads per kilobase million (RPKMs) for all genes within the five gene sets used in this study and for three gene sets not expected to be highly expressed in melanoma. Percentages on top indicate the percentage of genes within each gene set above the 1.0 average RPKM threshold (dashed line). The three control gene sets were derived from the GTEx normal tissue database. Error bars refer to median plus interquartile range. The y-axis scale is log10.
Fig. A6.
Correlation between pathway scores and histopathology for NanoString data. (A, B) Correlation of the NanoString cell cycle score with (A) pHH3 immunohistochemistry (IHC) and (B) mitotic rate. (C-F) Correlation of the NanoString CD8 or CD3 gene expression data with their respective IHC at either the (C, D) center or (E, F) periphery of the tumor.
Footnotes
Supported in part by the National Institutes of Health (NIH) Specialized Program of Research Excellence Melanoma Grant No. P50 CA93459 to The University of Texas MD Anderson Cancer Center; the NIH National Cancer Institute through The University of Texas MD Anderson Cancer Center Support Grant No. P30CA016672; a Melanoma Research Alliance Team Science Award; generous philanthropic contributions to The University of Texas MD Anderson Melanoma Moon Shots Program; the Dr Miriam and Sheldon G. Adelson Medical Research Foundation; the Robert and Lynne Grossman Family Foundation; the Michael and Patricia Booker Melanoma Research Endowment; AIM at Melanoma; the University of Texas Rising STARS Award and the Melanoma Research Alliance Young Investigator Award 508743, both to L.N.K.
AUTHOR CONTRIBUTIONS
Conception and design: Lawrence N. Kwong, Lauren Haydu, Aron Y. Joon, Tiffany L. Calderone, Michael A. Davies, Alexander J. Lazar, Jeffrey E. Gershenwald
Financial support: Michael A. Davies, Alexander J. Lazar, Jeffrey E. Gershenwald
Administrative support: Alexander J. Lazar, Jeffrey E. Gershenwald
Provision of study material or patients: Lawrence N. Kwong, Michael T. Tetzlaff, Man Kam Kwong, Alexander J. Lazar, Jeffrey E. Gershenwald
Collection and assembly of data: Lawrence N. Kwong, Mariana Petaccia De Macedo, Lauren Haydu, Aron Y. Joon, Michael T. Tetzlaff, Tiffany L. Calderone, Chiang-Jun Wu, Michael A. Davies, Alexander J. Lazar, Jeffrey E. Gershenwald
Data analysis and interpretation: Lawrence N. Kwong, Lauren Haydu, Aron Y. Joon, Michael T. Tetzlaff, Chiang-Jun Wu, Man Kam Kwong, Jason Roszik, Kenneth Hess, Michael A. Davies, Alexander J. Lazar, Jeffrey E. Gershenwald
Manuscript writing: All authors
Final approval of manuscript: All authors
AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST
The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.
Lawrence N. Kwong
Stock and Other Ownership Interests: Sarepta Therapeutics
Research Funding: Array BioPharma
Mariana Petaccia De Macedo
No relationship to disclose
Lauren Haydu
No relationship to disclose
Aron Y. Joon
No relationship to disclose
Michael T. Tetzlaff
Consulting or Advisory Role: Myriad Genetics, Novartis
Tiffany L. Calderone
No relationship to disclose
Chiang-Jun Wu
No relationship to disclose
Man Kam Kwong
No relationship to disclose
Jason Roszik
No relationship to disclose
Kenneth R. Hess
No relationship to disclose
Michael A. Davies
Consulting or Advisory Role: GlaxoSmithKline, Roche, Novartis, Sanofi, Vaccinex, Bristol-Myers Squibb, Syndax, NanoString Technologies
Research Funding: GlaxoSmithKline (Inst), Roche (Inst), AstraZeneca (Inst), Merck (Inst), Oncothyreon (Inst), Myriad Genetics (Inst), Sanofi
Alexander J. Lazar
Employment: GE Healthcare (I)
Leadership: Beta Cat Pharmaceuticals, Archer Biosciences
Stock and Other Ownership Interests: Archer Biosciences, Beta Cat Pharmaceuticals
Honoraria: Novartis, Bristol-Myers Squibb, Janssen Oncology, Roche
Consulting or Advisory Role: Novartis, Illumina, GE Healthcare
Research Funding: MedImmune, AstraZeneca, Roche, Novartis
Patents, Royalties, Other Intellectual Property: Elsevier
Travel, Accommodations, Expenses: Bristol-Myers Squibb, Novartis
Jeffrey E. Gershenwald
Consulting or Advisory Role: Castle Biosciences, Merck
Patents, Royalties, Other Intellectual Property: Mercator Therapeutics
REFERENCES
- 1.Cancer Genome Atlas Network Genomic classification of cutaneous melanoma. Cell. 2015;161:1681–1696. doi: 10.1016/j.cell.2015.05.044. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Zhang J, Baran J, Cros A, et al. International Cancer Genome Consortium Data Portal—A one-stop shop for cancer genomics data. Database (Oxford) 2011;2011 doi: 10.1093/database/bar026. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Curry JL, Davies MA, Calderone TL, et al. Tissue resources for clinical use and marker studies in melanoma. Methods Mol Biol. 2014;1102:679–695. doi: 10.1007/978-1-62703-727-3_37. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Stewart JP, Richman S, Maughan T, et al. Standardising RNA profiling based biomarker application in cancer—The need for robust control of technical variables. Biochim Biophys Acta. 2017;1868:258–272. doi: 10.1016/j.bbcan.2017.05.005. [DOI] [PubMed] [Google Scholar]
- 5.Gershenwald JE, Scolyer RA, Hess KR, et al. Melanoma staging: Evidence-based changes in the American Joint Committee on Cancer eighth edition Cancer Staging Manual. CA Cancer J Clin. 2017;67:472–492. doi: 10.3322/caac.21409. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Gershenwald JE, Scolyer RA, Hess KR, et al. Melanoma of the skin. In: Amin M, Edge S, Greene F, et al., editors. AJCC Cancer Staging Manual (ed. 8) Basel, Switzerland: Springer; 2017. pp. 563–585. [Google Scholar]
- 7.Eikrem O, Beisland C, Hjelle K, et al. Transcriptome sequencing (RNAseq) enables utilization of formalin-fixed, paraffin-embedded biopsies with clear cell renal cell carcinoma for exploration of disease biology and biomarker development. PLoS One. 2016;11:e0149743. doi: 10.1371/journal.pone.0149743. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Graw S, Meier R, Minn K, et al. Robust gene expression and mutation analyses of RNA-sequencing of formalin-fixed diagnostic tumor samples. Sci Rep. 2015;5:12335. doi: 10.1038/srep12335. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Jovanović B, Sheng Q, Seitz RS, et al. Comparison of triple-negative breast cancer molecular subtyping using RNA from matched fresh-frozen versus formalin-fixed paraffin-embedded tissue. BMC Cancer. 2017;17:241. doi: 10.1186/s12885-017-3237-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Liu Y, Noon AP, Aguiar Cabeza E, et al. Next-generation RNA sequencing of archival formalin-fixed paraffin-embedded urothelial bladder cancer. Eur Urol. 2014;66:982–986. doi: 10.1016/j.eururo.2014.07.045. [DOI] [PubMed] [Google Scholar]
- 11.Long Q, Xu J, Osunkoya AO, et al. Global transcriptome analysis of formalin-fixed prostate cancer specimens identifies biomarkers of disease recurrence. Cancer Res. 2014;74:3228–3237. doi: 10.1158/0008-5472.CAN-13-2699. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Morton ML, Bai X, Merry CR, et al. Identification of mRNAs and lincRNAs associated with lung cancer progression using next-generation RNA sequencing from laser micro-dissected archival FFPE tissue specimens. Lung Cancer. 2014;85:31–39. doi: 10.1016/j.lungcan.2014.03.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Royse KE, Zhi D, Conner MG, et al. Differential gene expression landscape of co-existing cervical pre-cancer lesions using RNA-seq. Front Oncol. 2014;4:339. doi: 10.3389/fonc.2014.00339. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Sharron Lin X, Hu L, Sandy K, et al. Differentiating progressive from nonprogressive T1 bladder cancer by gene expression profiling: Applying RNA-sequencing analysis on archived specimens. Urol Oncol. 2014;32:327–336. doi: 10.1016/j.urolonc.2013.06.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Esteve-Codina A, Arpi O, Martinez-García M, et al. A comparison of RNA-seq results from paired formalin-fixed paraffin-embedded and fresh-frozen glioblastoma tissue samples. PLoS One. 2017;12:e0170632. doi: 10.1371/journal.pone.0170632. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Nikitina AS, Sharova EI, Danilenko SA, et al. Novel RNA biomarkers of prostate cancer revealed by RNA-seq analysis of formalin-fixed samples obtained from Russian patients. Oncotarget. 2017;8:32990–33001. doi: 10.18632/oncotarget.16518. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Li J, Fu C, Speed TP, et al. Accurate RNA sequencing from formalin-fixed cancer tissue to represent high-quality transcriptome from frozen tissue. JCO Precis Oncol. doi: 10.1200/PO.17.00150. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Stuart JM, Segal E, Koller D, et al. A gene-coexpression network for global discovery of conserved genetic modules. Science. 2003;302:249–255. doi: 10.1126/science.1087447. [DOI] [PubMed] [Google Scholar]
- 19.Liberzon A, Subramanian A, Pinchback R, et al. Molecular signatures database (MSigDB) 3.0. Bioinformatics. 2011;27:1739–1740. doi: 10.1093/bioinformatics/btr260. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Cancer Genome Atlas Network Comprehensive molecular characterization of human colon and rectal cancer. Nature. 2012;487:330–337. doi: 10.1038/nature11252. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Konieczkowski DJ, Johannessen CM, Abudayyeh O, et al. A melanoma cell state distinction influences sensitivity to MAPK pathway inhibitors. Cancer Discov. 2014;4:816–827. doi: 10.1158/2159-8290.CD-13-0424. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Tetzlaff MT, Curry JL, Ivan D, et al. Immunodetection of phosphohistone H3 as a surrogate of mitotic figure count and clinical outcome in cutaneous melanoma. Mod Pathol. 2013;26:1153–1160. doi: 10.1038/modpathol.2013.59. [DOI] [PubMed] [Google Scholar]
- 23.Curtin JA, Fridlyand J, Kageshita T, et al. Distinct sets of genetic alterations in melanoma. N Engl J Med. 2005;353:2135–2147. doi: 10.1056/NEJMoa050092. [DOI] [PubMed] [Google Scholar]
- 24.Badal B, Solovyov A, Di Cecilia S, et al. Transcriptional dissection of melanoma identifies a high-risk subtype underlying TP53 family genes and epigenome deregulation. JCI Insight. doi: 10.1172/jci.insight.92102. [epub ahead of print on May 4, 2017] [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Harbst K, Staaf J, Lauss M, et al. Molecular profiling reveals low- and high-grade forms of primary melanoma. Clin Cancer Res. 2012;18:4026–4036. doi: 10.1158/1078-0432.CCR-12-0343. [Erratum: Clin Cancer Res 15;18:5829, 2012] [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Jewell R, Elliott F, Laye J, et al. The clinicopathological and gene expression patterns associated with ulceration of primary melanoma. Pigment Cell Melanoma Res. 2015;28:94–104. doi: 10.1111/pcmr.12315. [DOI] [PMC free article] [PubMed] [Google Scholar]