NCT/DKFZ MASTER handbook of interpreting whole-genome, transcriptome, and methylome data for precision oncology

Abstract

Analysis of selected cancer genes has become an important tool in precision oncology but cannot fully capture the molecular features and, most importantly, vulnerabilities of individual tumors. Observational and interventional studies have shown that decision-making based on comprehensive molecular characterization adds significant clinical value. However, the complexity and heterogeneity of the resulting data are major challenges for disciplines involved in interpretation and recommendations for individualized care, and limited information exists on how to approach multilayered tumor profiles in clinical routine. We report our experience with the practical use of data from whole-genome or exome and RNA sequencing and DNA methylation profiling within the MASTER (Molecularly Aided Stratification for Tumor Eradication Research) program of the National Center for Tumor Diseases (NCT) Heidelberg and Dresden and the German Cancer Research Center (DKFZ). We cover all relevant steps of an end-to-end precision oncology workflow, from sample collection, molecular analysis, and variant prioritization to assigning treatment recommendations and discussion in the molecular tumor board. To provide insight into our approach to multidimensional tumor profiles and guidance on interpreting their biological impact and diagnostic and therapeutic implications, we present case studies from the NCT/DKFZ molecular tumor board that illustrate our daily practice. This manual is intended to be useful for physicians, biologists, and bioinformaticians involved in the clinical interpretation of genome-wide molecular information.

Fig. 1. Impact of patient characteristics and tissue context on the clinical interpretation of molecular alterations.

Cases were selected to exemplify how to approach, following current guidelines for oncogenicity classification, somatic variants that have not been curated and to emphasize the therapeutic impact of the histologic context. In addition, we included a structural variant, i.e., an insertion, to illustrate curation challenges in daily routine. a Gastrointestinal stromal tumor (GIST) studied by WES and RNA-seq of formalin-fixed and paraffin-embedded tumor tissue (histopathologic tumor cell content, 50%). A KIT exon 11 insertion (p.P585_R586insSPYDHKWEFP), whose expression was verified by RNA-seq, was nominated as a candidate driver because in-frame indels in KIT exon 11, encoding the KIT juxtamembrane domain, are known oncogenic events in GIST and rarely occur in other cancers (www.cancerhotspots.org). A literature search revealed that a similar variant (p.P585_586insLPYDHKWEFP) was detected in a previous study but has not been functionally characterized to date. Application of the VICC guideline for interpreting somatic variants in tumors (www.cancervariants.org) resulted in a score of seven points, classifying the variant as likely oncogenic, which was composed of evidence from the following categories: “Oncogenicity Moderate 1” (OM1; two points): variant located in a critical and well-established part of a functional domain; OM2 (two points): variant associated with protein length changes because of in-frame indels in a known oncogene or tumor suppressor gene or stop-loss variants in a known tumor suppressor gene; “Oncogenicity Supporting 3” (OP3; one point): variant absent from controls or occurring at an extremely low frequency in the Genome Aggregation Database (gnomAD; https://gnomad.broadinstitute.org); OP4 (one point): variant located in a mutation hotspot listed in Cancer Hotspots (www.cancerhotspots.org) and associated with an amino acid change count in Cancer Hotspots below 10 (resources such as cBioPortal [www.cbioportal.org], COSMIC [https://cancer.sanger.ac.uk/cosmic], or an entity’s published genetic landscape to be used for variants occurring in tumor types not covered well by Cancer Hotspots). Furthermore, we added OP2 evidence (one point; variant in a gene in a malignancy with a single genetic etiology) because KIT mutations drive the vast majority of GIST, and exon 11 indels are among the recurrent alterations. Based on this evaluation, the MTB recommended therapy with imatinib with an NCT evidence level of m1a, because KIT exon 11-mutant GIST is particularly sensitive to this agent. CI, confidence interval; SBS, single-base substitution. b RAF- and NRAS-wildtype acral melanoma studied by WGS and RNA-seq of fresh-frozen tissue (histopathologic tumor cell content, 65%) after progressing on immuncheckpoint inhibition. The KIT gene was affected by a p.K642E missense mutation with loss of heterozygosity and an allele frequency (AF) of 80%, whose expression was verified by RNA-seq (AF, 100%), and a DNA copy number of 8 (average ploidy, 3). Activating KIT mutations occur in approximately 3% of melanomas, with enrichment in the acral subtype, and include mainly missense mutations affecting exons 9, 11, 13, 17, and 18, with up to 60% occurring in exons 11 or 13. The p.K642E and p.L576 variants account for approximately one-quarter of KIT mutations in melanoma and provide a rationale for therapy with imatinib. However, the objective response and disease control rates of these patients (24.4% and 66.7%, respectively;) are lower than those of patients with KIT-mutant GIST (80% and >90%, respectively), context-specific differences whose basis remains to be elucidated. APOBEC, apolipoprotein B mRNA editing enzyme, catalytic polypeptide; UV, ultraviolet. Take-home messages: (i) Current VICC guidelines should be applied when evaluating somatic variants of unclear biological significance. Certain alteration types remain difficult to annotate and may require case-by-case assessment, which should take place in multidisciplinary MTBs whenever possible. (ii) The clinical actionability of a driver alteration, determined by, e.g., the probability and duration of response to molecularly guided therapy, can vary widely depending on the histologic context, which must always be considered when selecting and prioritizing treatment options.

Fig. 2. Complex biomarkers derived from WES of a peritoneal metastasis in a patient with ovarian cancer.

a Fractions of mutational signatures identified in the tumor. DSB, double-strand break. hom. recomb., homologous recombination. b Somatic DNA copy number profiles of the tumor and a matched normal control. The tumor exhibits segmental gains and losses of all chromosomes as well as a high HRD-LOH score and numerous LSTs (19 and 23, respectively), corresponding to a highly rearranged genome. Consistent with the genomic “scars” of HRD, the SBS3 mutational signature explained 50% of all SNVs. Of note, no germline or somatic mutations in BRCA1/2 were detected. Chromosomes 1 to X are indicated.

Fig. 3. Clinical implications of DNA methylation analysis.

Patient with undifferentiated pleomorphic sarcoma of the lung according to histopathology. Immunohistochemistry: Melan A, HMB45, CD34, MyoD, CD30, SOX10, CD68, CD117, cytokeratin 7/8, CD123, CD1a, and CD21 negative; CD56, S100, and PD-L1 (90%, Cologne Score 5) positive; proliferation rate (MIB-1), 80%. Treatment course: surgery followed by adjuvant chemotherapy with doxorubicin and ifosfamide; switch to pazopanib due to liver metastases, stable disease; switch to trabectedin and pembrolizumab due to lung metastases after six months, complete metabolic response; brain metastases with continuing remission at peripheral sites after six months. WGS and RNA-seq revealed gene expression similarity to melanoma, a high tumor mutational burden (891 SNVs and 8 indels), and a highly prevalent SBS7 mutational signature associated with UV light exposure. DNA methylation profiling showed a match score of 0.95 with cutaneous melanoma^,. The figure shows a projection of the index case and several melanomas (MEL) on a DNA methylation-based sarcoma reference cohort (n = 1077;), in which undifferentiated sarcomas (USARC) are highlighted (t-distributed stochastic neighbor embedding [t-SNE] using the 10,000 most variable probes according to standard deviation via the R package Rtsne (version 0.16) using 3000 iterations and a perplexity value of 30). This finding prompted recommendations to reevaluate the diagnosis and modify further treatment if applicable. Take-home messages: (i) New multiomics layers, such as genome-wide DNA methylation profiles, can help refine diagnosis, especially for cancer types without distinct morphologic features or pathognomonic molecular alterations. (ii) Multiomics-guided diagnostic reclassification can inform therapeutic decision-making.

Fig. 4. DNA copy number profiles as diagnostic biomarkers.

a Few DNA copy number changes in a patient with SS18::SSX2-positive synovial sarcoma. Consistent with the “silent” genomes of many fusion-driven cancers, a low tumor mutational burden was found with 23 SNVs and indels, corresponding to 0.6 non-synonymous mutations per coding megabase. b Multiple DNA copy number alterations in a patient with genomically unstable leiomyosarcoma. c Pathognomonic amplification of MDM2 and CDK4 on chromosome 12q14-q15 (red arrow), which are targeted by small-molecule inhibitors^,, and few other genomic imbalances in a patient with well-differentiated liposarcoma. d Higher genomic complexity with multiple DNA copy number alterations in a patient with dedifferentiated liposarcoma (DDLS). Take-home messages: (i) The extent of DNA copy number alterations varies considerably among tumor types. (ii)The patterns of genomic imbalances can aid in the diagnostic categorization of various cancer entities. (iii) Recurrent amplicons may harbor genes that can be targeted therapeutically, such as CDK4 and MDM2 in DDLS.

Fig. 5. DNA copy number gains as actionable biomarkers.

a Squamous cell carcinoma of an unknown primary site in the neck region studied by WGS and RNA-seq. MMR, mismatch repair. TCN, total copy number. b Evidence of numerous DNA copy number alterations, including amplification (total copy number, 8; average tumor ploidy, 2) of a region on chromosome 11q13.3 containing the oncogenes CCND1, FGF3, and FGF4 (red arrow). A query of the OnkoKB precision oncology knowledgebase (www.onkokb.org) showed that data on the oncogenicity of FGF3 and FGF4 amplification are inconclusive. RNA-seq analysis showed decreased transcription of FGF3 and FGF4, indicating that their amplification is a passenger alteration. In contrast, CCND1 was expressed, suggesting that it functions as a driver. The finding of homozygous deletion of CDKN2A/B on chromosome 9p21.3 further supported the role of the CCND1–CDK4/6 axis in the pathogenesis of this tumor. However, the clinical efficacy of CDK4/6 inhibition in this scenario varies and appears to be dependent on histology^,. Based on two clinical trials of the CDK4/6 inhibitor palbociclib in combination with cetuximab in CDKN2A-negative squamous cell carcinoma of the head and neck region^,, the MTB recommended this treatment with an NCT evidence level of m2a. Chromosomes 1 to Y are indicated. Take-home messages: (i) An amplicon can harbor dozens to thousands of genes that can act as drivers or passengers in a given histologic context. (ii) Integrating WGS/WES and RNA-seq data can guide the selection of driver genes and inform treatment.

Fig. 6. Copy number loss as actionable target.

a Advanced esophageal cancer studied by WGS and RNA-seq. b DNA copy number plot showing a heterozygous deletion of chromosome 10q associated with loss of one PTEN allele, which, together with a focal deletion of exons 6 to 8 of the other allele, is predicted to result in loss of PTEN function, providing a rationale for therapeutic inhibition of constitutively active PI3K–AKT–mTOR signaling. Chromosomes 1 to Y are indicated. c Intrahepatic cholagiocarcinoma studied by WGS and RNA-seq. In addition to a FGFR2::DBP fusion, potentially actionable DNA copy number alterations affecting tumor suppressor genes within the PI3K–AKT–mTOR pathway (average tumor ploidy, 4; PTEN copy number, 3; TSC1 copy number, 2; FBXW7 copy number, 2) were detected; however, none of these loci was affected by alterations of the remaining allele that would result in complete inactivation, leaving the functional consequences of the copy number losses unclear. RNA-seq showed that all genes were expressed, which, without information on protein expression, argued against treatment recommendations based, at best, on partial inactivation of negative regulators of PI3K–AKT–mTOR signaling. Future integration of proteomic profiling and RNA-based pathway activity estimation into our workflow will provide a more accurate assessment of the impact of DNA copy number alterations on gene function, i.e., their influence on protein synthesis in a given tumor environment. Take-home messages: (i) Heterozygous deletions of tumor suppressor genes without a second “hit” affecting the remaining allele should be interpreted with caution and require further validation, e.g., by immunohistochemistry. (ii) In the case of a deletion affecting one copy of a tumor suppressor gene, examination of the remaining allele, RNA expression, and the integrity of other genes relevant to the respective pathway support the evaluation of functional impact and, thus, clinical actionability.

Fig. 7. Clinical interpretation of gene fusions.

a FGFR1::ADAM9 fusion generated by an interstitial deletion on chromosome 8p linking FGFR1 exons 1–12 to ADAM9 exons 12−1 in a patient with chondroblastic osteosarcoma of the femur. As this fusion was out of frame, retained only part of the FGFR1 kinase domain, and was supported by only a few reads, the MTB classified it as non-functional and without therapeutic implications. b FGFR2::WDR65 fusion linking FGFR2 exons 1−17 to WDR65 exons 12−23 in a patient with intrahepatic cholangiocarcinoma (iCCC). The chimeric transcript, which is supported by multiple reads for both partners, is characterized by a recurrent breakpoint in FGFR2 exon 17 and retains the FGFR2 kinase domain. FGFR2 fusions are identified in approximately 15% of iCCC and targeted by the recently approved FGFR inhibitor pemigatinib. c Oncogenic FGFR1::PLAG1 fusion in a patient with myoepithelial carcinoma. The genomic breakpoints are located between the promoter and the transcription start site of both FGFR1 and PLAG1, resulting in the expression of full-length PLAG1 regulated by the FGFR1 promoter. Accordingly, PLAG1 but not FGFR1 was highly expressed, as indicated by the different number of reads. PLAG1 fusions are characteristic of myoepithelial carcinoma, an aggressive form of salivary gland cancer, and are not yet amenable to targeted therapies. d Detection of 184 gene fusions in a patient with DDLS, originating primarily from the alteration of chromosome 12q characteristic of this entity. Gene fusions may be a source of immunogenic neoantigens that can mediate a response to immunotherapy even in tumors with low mutational burden. Take-home messages: (i) RNA-seq is the most accurate method to detect functional gene fusions. (ii) The oncogenicity of a gene fusion does not automatically render the fusion a druggable target. (iii) To date, druggable fusions are mainly restricted to chimeric proteins containing a kinase domain that is constitutively active and triggers downstream signaling.

Fig. 8. Implications of transcriptome data for guiding cancer therapies.

Rationale: These two cases from the same entity highlight the therapeutic value derived from integrating transcriptomic analysis for the emerging list of antibody-drug conjugates and cellular immunotherapy strategies. a Adenoid cystic carcinoma (ACC) studied by WGS. The only treatment recommendation was PARP inhibition based on a dominant SBS3 mutational signature. b ACC studied by WGS and RNA-seq. In contrast to the previous case, transcriptomic information yielded several treatment recommendations, i.e., sacituzumab govitecan based on overexpression of TACSTD2; multi-tyrosine kinase inhibition based on overexpression of DDR1, FGFR2, IGF1R, PDGFA, PDGFB, and NTRK3; T-cell based immunotherapy within a phase 1 clinical trial (ClinicalTrials.gov Identifier: NCT03441100) based on MAGEA1 overexpression and a HLA-A*02 genotype; and enfortumab vedotin based on PVRL4 (also called NECTIN4) overexpression. Take-home messages: (i) RNA-seq enables the detection of targets for antibody-drug conjugates or cellular immunotherapies. (ii) The selection of kinase inhibitors can be guided through the assessment of their target landscape by RNA-seq.

Fig. 9. Assignment and ranking of biomarker-drug response associations.

The number of molecular alterations identified by multiomics is steadily increasing, which entails two main challenges, i.e., determining the biological significance and clinical relevance of a candidate biomarker. a The evidence supporting a molecularly informed therapy ranges from preclinical studies to phase 3 clinical trials or meta-analyses. The underlying histologic entity must always be considered when assigning evidence to a molecular biomarker. In the case presented, therapy A is supported by clinical data from other entities (m2a), whereas for therapy B, there are retrospective data from the same entity supporting the recommendation (m1b). b The final evidence attributed to treatment can be listed as a range between the lowest and highest level, (e.g., m2a–m3). The number of references listed varies among curators, but we advise careful judgment and restriction to the most contextually relevant ones. c, d Prioritization of therapies is a complex process and depends on several variables that must be considered in a patient’s specific clinical and socioeconomic situation at a given time. For example, a therapy may be available in a clinical trial, but exclusion criteria beyond molecular features preclude enrollment, and another treatment option must be pursued.

Fig. 10. Structure of an MTB report.

a Main components of a therapy recommendation block as used in the MASTER MTB report. b Summary and ranking of therapy recommendations with their respective evidence levels. Several factors influence the final choice of therapy, such as its availability, patient preference, side effects, and approval status.

Fig. 11. Graphical representation of multidimensional tumor characterization as performed in the MASTER trial.

Tumor DNA and RNA obtained from tumor tissue are analyzed by DNA methylation profiling and WGS/WES or RNA-seq, respectively. DNA derived from blood serves as a matched normal control for WGS/WES. Created with Biorender.com.