Advertisement

Spatial and Temporal Heterogeneity of Panel-Based Tumor Mutational Burden in Pulmonary Adenocarcinoma: Separating Biology From Technical Artifacts

Open ArchivePublished:July 23, 2019DOI:https://doi.org/10.1016/j.jtho.2019.07.006

      Abstract

      Background

      Tumor mutational burden (TMB) is an emerging biomarker used to identify patients who are more likely to benefit from immuno-oncology therapy. Aside from various unsettled technical aspects, biological variables such as tumor cell content and intratumor heterogeneity may play an important role in determining TMB.

      Methods

      TMB estimates were determined applying the TruSight Oncology 500 targeted sequencing panel. Spatial and temporal heterogeneity was analyzed by multiregion sequencing (two to six samples) of 24 pulmonary adenocarcinomas and by sequencing a set of matched primary tumors, locoregional lymph node metastases, and distant metastases in five patients.

      Results

      On average, a coding region of 1.28 Mbp was covered with a mean read depth of 609x. Manual validation of the mutation-calls confirmed a good performance, but revealed noticeable misclassification during germline filtering. Different regions within a tumor showed considerable spatial TMB variance in 30% (7 of 24) of the cases (maximum difference, 14.13 mut/Mbp). Lymph node–derived TMB was significantly lower (p = 0.016). In 13 cases, distinct mutational profiles were exclusive to different regions of a tumor, leading to higher values for simulated aggregated TMB. Combined, intratumor heterogeneity and the aggregated TMB could result in divergent TMB designation in 17% of the analyzed patients. TMB variation between primary tumor and distant metastases existed but was not profound.

      Conclusions

      Our data show that, in addition to technical aspects such as germline filtering, the tumor content and spatially divergent mutational profiles within a tumor are relevant factors influencing TMB estimation, revealing limitations of single-sample–based TMB estimations in a clinical context.

      Keywords

      Introduction

      Tumor mutational burden (TMB) has emerged as a novel biomarker to identify patients more likely to respond to immune checkpoint inhibitor therapy targeting the programmed cell death protein 1 axis or cytotoxic T-lymphocyte associated protein 4 (CTLA-4).
      • Rizvi N.A.
      • Hellmann M.D.
      • Snyder A.
      • et al.
      Cancer immunology. Mutational landscape determines sensitivity to PD-1 blockade in non–small cell lung cancer.
      • Rosenberg J.E.
      • Hoffman-Censits J.
      • Powles T.
      • et al.
      Atezolizumab in patients with locally advanced and metastatic urothelial carcinoma who have progressed following treatment with platinum-based chemotherapy: a single-arm, multicentre, phase 2 trial.
      • Snyder A.
      • Makarov V.
      • Merghoub T.
      • et al.
      Genetic basis for clinical response to CTLA-4 blockade in melanoma.
      • Van Allen E.M.
      • Miao D.
      • Schilling B.
      • et al.
      Genomic correlates of response to CTLA-4 blockade in metastatic melanoma.
      For many tumor entities, assessment of programmed death ligand 1 (PD-L1) expression on tumor and/or immune cells by immunohistochemistry (IHC) is the approved companion diagnostic.
      • Herbst R.S.
      • Soria J.C.
      • Kowanetz M.
      • et al.
      Predictive correlates of response to the anti–PD-L1 antibody MPDL3280A in cancer patients.
      TMB can potentially identify — independently from PD-L1 expression status — different patient cohorts likely to respond and in conjunction with PD-L1 status help to predict nonresponders and exceptional responders.
      • Carbone D.P.
      • Reck M.
      • Paz-Ares L.
      • et al.
      First-line nivolumab in stage IV or recurrent non–small-cell lung cancer.
      • Rizvi H.
      • Sanchez-Vega F.
      • La K.
      • et al.
      Molecular determinants of response to anti–programmed cell death (PD)-1 and anti–programmed death-ligand 1 (PD-L1) blockade in patients with non–small-cell lung cancer profiled with targeted next-generation sequencing.
      • Hellmann M.D.
      • Ciuleanu T.E.
      • Pluzanski A.
      • et al.
      Nivolumab plus ipilimumab in lung cancer with a high tumor mutational burden.
      • Hellmann M.D.
      • Nathanson T.
      • Rizvi H.
      • et al.
      Genomic features of response to combination immunotherapy in patients with advanced non–small-cell lung cancer.
      • Goodman A.M.
      • Kato S.
      • Bazhenova L.
      • et al.
      Tumor mutational burden as an independent predictor of response to immunotherapy in diverse cancers.
      • Budczies J.
      • Seidel A.
      • Christopoulos P.
      • et al.
      Integrated analysis of the immunological and genetic status in and across cancer types: impact of mutational signatures beyond tumor mutational burden.
      Immuno-oncology (IO) therapy has the potential to overcome tumor-mediated immune suppression. However, such therapies are not only associated with significant costs, but also potentially severe adverse side effects.
      • Brahmer J.R.
      • Tykodi S.S.
      • Chow L.Q.
      • et al.
      Safety and activity of anti–PD-L1 antibody in patients with advanced cancer.
      Because only a minority of patients benefit from this strategy, it is of utmost importance to establish biomarkers to guide therapy decisions. However, the cancer immune system interaction is complex and multilayered.
      • Thorsson V.
      • Gibbs D.L.
      • Brown S.D.
      • et al.
      The immune landscape of cancer.
      As a basis for immunogenicity, it has been hypothesized that tumors with a high number of coding mutations are more likely to generate tumor-specific neoantigens that will be recognized by the immune system.
      • Gubin M.M.
      • Artyomov M.N.
      • Mardis E.R.
      • et al.
      Tumor neoantigens: building a framework for personalized cancer immunotherapy.
      Recent data support a predictive potential of TMB for checkpoint inhibitor therapy (single substance and combinations) in various cancer types.
      • Hellmann M.D.
      • Ciuleanu T.E.
      • Pluzanski A.
      • et al.
      Nivolumab plus ipilimumab in lung cancer with a high tumor mutational burden.
      • Rittmeyer A.
      • Barlesi F.
      • Waterkamp D.
      • et al.
      Atezolizumab versus docetaxel in patients with previously treated non–small-cell lung cancer (OAK): a phase 3, open-label, multicentre randomised controlled trial.

      Peters S, Cho BC, Reinmuth N, et al. CT074 Tumor mutational burden (TMB) as a biomarker of survival in metastatic non–small cell lung cancer (mNSCLC): Blood and tissue TMB analysis from MYSTIC, a phase III study of first-line durvalumab ± tremelimumab vs chemotherapy. Paper presented at American Association for Cancer Research (AACR) Annual Meeting. March 29–April 3, 2019; Atlanta, Georgia.

      Whole-exome sequencing (WES)–based TMB assessment was initially used in clinical trials. Many clinical sites and recent clinical trials have implemented targeted next-generation sequencing (NGS) of widely accessible formalin-fixed, paraffin-embedded (FFPE) tissue samples to approximate TMB as an alternative to WES-based analysis with its higher costs, longer turn-around time, and limited availability of suitable specimens. Definition of cutpoints, critical panel size, and various technical and bioinformatical aspects are important to consider.
      • Allgauer M.
      • Budczies J.
      • Christopoulos P.
      • et al.
      Implementing tumor mutational burden (TMB) analysis in routine diagnostics-a primer for molecular pathologists and clinicians.
      • Chan T.A.
      • Yarchoan M.
      • Jaffee E.
      • et al.
      Development of tumor mutation burden as an immunotherapy biomarker: utility for the oncology clinic.
      • Buchhalter I.
      • Rempel E.
      • Endris V.
      • et al.
      Size matters: dissecting key parameters for panel-based tumor mutational burden analysis.
      • Endris V.
      • Buchhalter I.
      • Allgauer M.
      • et al.
      Measurement of tumor mutational burden (TMB) in routine molecular diagnostics: in silico and real-life analysis of three larger gene panels.
      Here, we investigate intratumoral heterogeneity (ITH) as a biological factor affecting TMB determination using panel-based NGS. ITH is a well-described phenomenon in lung adenocarcinoma (ADC) on a radiologic, histopathologic, genetic, epigenetic, and tumor-microenvironmental level.
      • Ma Y.
      • Feng W.
      • Wu Z.
      • et al.
      Intra-tumoural heterogeneity characterization through texture and colour analysis for differentiation of non-small cell lung carcinoma subtypes.
      • Park S.
      • Ha S.
      • Lee S.H.
      • et al.
      Intratumoral heterogeneity characterized by pretreatment PET in non-small cell lung cancer patients predicts progression-free survival on EGFR tyrosine kinase inhibitor.
      • Cadioli A.
      • Rossi G.
      • Costantini M.
      • et al.
      Lung cancer histologic and immunohistochemical heterogeneity in the era of molecular therapies: analysis of 172 consecutive surgically resected, entirely sampled pulmonary carcinomas.
      • Warth A.
      • Muley T.
      • Meister M.
      • et al.
      The novel histologic International Association for the Study of Lung Cancer/American Thoracic Society/European Respiratory Society classification system of lung adenocarcinoma is a stage-independent predictor of survival.
      • Jamal-Hanjani M.
      • Wilson G.A.
      • McGranahan N.
      • et al.
      tracking the evolution of non–small-cell lung cancer.
      • McGranahan N.
      • Swanton C.
      Clonal heterogeneity and tumor evolution: past, present, and the future.
      • Kazdal D.
      • Harms A.
      • Endris V.
      • et al.
      Subclonal evolution of pulmonary adenocarcinomas delineated by spatially distributed somatic mitochondrial mutations.
      • Dietz S.
      • Lifshitz A.
      • Kazdal D.
      • et al.
      Global DNA methylation reflects spatial heterogeneity and molecular evolution of lung adenocarcinomas.
      • Alizadeh A.A.
      • Aranda V.
      • Bardelli A.
      • et al.
      Toward understanding and exploiting tumor heterogeneity.
      • Mony J.T.
      • Schuchert M.J.
      prognostic implications of heterogeneity in intra-tumoral immune composition for recurrence in early stage lung cancer.
      According to the most widely accepted theory, ITH is mainly the result of subclonal evolution during natural tumor progression and therapeutic interventions.
      • de Bruin E.C.
      • McGranahan N.
      • Mitter R.
      • et al.
      Spatial and temporal diversity in genomic instability processes defines lung cancer evolution.
      • Warth A.
      • Endris V.
      • Stenzinger A.
      • et al.
      Genetic changes of non–small cell lung cancer under neoadjuvant therapy.
      The great molecular variability associated with high levels of ITH is seen as one of the leading causes for lack of response or development of resistance under therapy.
      • Lovly C.M.
      • Salama A.K.
      • Salgia R.
      Tumor heterogeneity and therapeutic resistance.
      • Dagogo-Jack I.
      • Shaw A.T.
      Tumour heterogeneity and resistance to cancer therapies.
      In addition, ITH can interfere with molecular diagnostic testing for prognostication or selection of optimal systemic therapy. Although some actionable targets (e.g., truncal sensitizing mutations in EGFR) are generally present across all tumor sites, other alterations such as tumor protein p53 (TP53) mutations or the expression of PD-L1 were described to exhibit a heterogeneous distribution.
      • Dietz S.
      • Harms A.
      • Endris V.
      • et al.
      Spatial distribution of EGFR and KRAS mutation frequencies correlates with histological growth patterns of lung adenocarcinomas.
      • Zhang L.L.
      • Kan M.
      • Zhang M.M.
      • et al.
      Multiregion sequencing reveals the intratumor heterogeneity of driver mutations in TP53-driven non-small cell lung cancer.
      • Casadevall D.
      • Clave S.
      • Taus A.
      • et al.
      Heterogeneity of tumor and immune cell PD-L1 expression and lymphocyte counts in surgical NSCLC samples.
      • Liu Y.
      • Dong Z.
      • Jiang T.
      • et al.
      Heterogeneity of PD-L1 expression among the different histological components and metastatic lymph nodes in patients with resected lung adenosquamous carcinoma.
      Given this molecular heterogeneity, TMB counts might be influenced by ITH.
      • Rosenthal R.
      • Cadieux E.L.
      • Salgado R.
      • et al.
      Neoantigen-directed immune escape in lung cancer evolution.
      First, adding to previous in silico data from our group, we validated the TruSight Oncology 500 (TSO500) targeted sequencing panel (Illumina Inc., San Diego, California) for TMB estimation using a cohort of patients with known WES-based TMB counts.
      • Buchhalter I.
      • Rempel E.
      • Endris V.
      • et al.
      Size matters: dissecting key parameters for panel-based tumor mutational burden analysis.
      • Endris V.
      • Buchhalter I.
      • Allgauer M.
      • et al.
      Measurement of tumor mutational burden (TMB) in routine molecular diagnostics: in silico and real-life analysis of three larger gene panels.
      Then we used this panel to perform multiregion sequencing of a well-characterized cohort of pulmonary ADC and further compared a set of primary tumors to their locoregional and distant metastases.
      Our data show that regional variability of TMB is significant in lung ADC; it can alter TMB classification of individual patients and thus influence therapeutic decisions.

      Materials and Methods

      Samples

      All patient ADC specimens analyzed were obtained from surgical procedures at the Thoraxklinik at University Hospital Heidelberg and diagnosed according to the criteria of the 2015 WHO classification of lung tumors at the Institute of Pathology, at University Hospital Heidelberg.
      • Travis W.D.
      • Brambilla E.
      • Nicholson A.G.
      • et al.
      The 2015 World Health Organization classification of lung tumors: impact of genetic, clinical and radiologic advances since the 2004 classification.
      FFPE tissue sections were supplied by the tissue bank of the National Center for Tumor Diseases (NCT; project: # 1746, # 2015) in accordance with its ethical regulations approved by the local ethics committee.
      To validate panel sequencing–based bTMB (psTMB) estimation with the TSO500 panel against the gold standard of WES-based TMB calculations, FFPE samples of 16 NSCLC specimens (biopsy and resection specimens) were obtained from the Heidelberg Lung Biobank, member of the BioMaterialbank Heidelberg and the Biobank Platform of the German Center for Lung Research (ethical approval S-270/2001, S-206/2011) of which corresponding WES data were available, derived from the DKFZ HIPO and the NCT MASTER programs.
      • Endris V.
      • Buchhalter I.
      • Allgauer M.
      • et al.
      Measurement of tumor mutational burden (TMB) in routine molecular diagnostics: in silico and real-life analysis of three larger gene panels.
      • Horak P.
      • Klink B.
      • Heining C.
      • et al.
      Precision oncology based on omics data: the NCT Heidelberg experience.
      For the evaluation of ITH, a cohort of 24 patients with ADC, each consisting of two to four multiregional samples (see Table 1 for clinicopathologic details) was constructed as described before, but excluding tumors with clinically targetable driver mutations.
      • Kazdal D.
      • Harms A.
      • Endris V.
      • et al.
      Prevalence of somatic mitochondrial mutations and spatial distribution of mitochondria in non–small cell lung cancer.
      In short, a central section of each tumor was fixed in formalin and subsequently cut into 5 × 5 mm segments according to a Cartesian grid. Ink marks maintained the original orientation of each segment during histologic processing. Tumor regions considered for sequencing were selected in accordance with the tumor size (the larger the tumor the more regions), different histologic growth patterns, as well as sufficient tumor cell content (≥10%) and DNA concentration (≥4 ng/μL). The predominant histologic growth pattern in each segment (defined as the pattern with the highest percentage) was determined by an experienced pathologist. Additionally, FFPE samples of locoregional lymph node metastases were analyzed if present.
      Table 1Clinicopathologic Data of Spatial Heterogeneity (1-24) and Temporal Heterogeneity (I-V) Cohorts
      CaseSexAge, yearsSmoking statuspTNM ClassificationHistologic Pattern
      Minor component shown in parentheses.
      Driver MutationTumor Area, cm2
      1f68NApT4, pN2 (17/31), pMXA, MP, (S)n.d.3.75
      2m84formerpT2b, pN2 (6/24), pMXSKRAS:p.Gly12Cys11.50
      3f75NApT2a, pN0 (0/23), pMXA, Sn.d.4.00
      4m63NApT2a, pN0 (0/39), pMXA, PKRAS:p.Gly12Cys8.75
      5m77activepT4, pN2 (4/37), pMXA, P, (MP)n.d.7.50
      6f66NApT2a, pN1 (6/30), pMXA, S, (L)KRAS:p.Gly12Cys7.25
      7m74formerpT3, pN0 (0/32), pMXA, S, (MP)KRAS:p.Gly12Cys6.50
      8m70activepT1a (mi), pN0 (0/31), pMXL, AKRAS:p.Gly12Cys6.25
      9m54activepT2a, pN1 (1/10), pMXA, P, S, LKRAS:p.Gly13Cys6.25
      10m57formerpT3, pN0 (0/25), pMXA, L, Pn.d.5.50
      11f54NApT3, pN2 (1/17), pMXSKRAS:p.Gly12Cys8.00
      12m60activepT2b, pN1 (4/38), pMXA, S, (MP)KRAS:p.Gly12Asp3.25
      13m63NApT3, pN2 (7/35), pMXS, (A)KRAS:p.Gly12Val3.00
      14m60formerpT2a, pN1 (3/51), pMXL, AKRAS:p.Gly13Cys3.50
      15m51neverpT2a, pN0 (0/21), pMXPKRAS:p.Gly12Asp4.75
      16m66neverpT3, pN1 (1/48), pMXA, SERBB2:p.Ser310Phe4.50
      17m60formerpT2a, pN0 (0/37), pMXAn.d.2.00
      18m72activepT3, pN2 (5/40), pMXA, (L, M, P)KRAS:p.Gly12Asp14.50
      19m66activepT1a, pN0 (0/22), pMXL, (A)KRAS:p.Gly12Val1.5
      20f74NApT1a, pN0 (0/28), pMXL, Pn.d.2.25
      21f66NApT1b, pN0 (0/22), pMXM, S, (P, L)KRAS:p.Gly12Ser3.75
      22f65NApT3, pN2 (4/52), pMXSn.d.13.50
      23f80NApT1b, pN1 (3/32), pMXMn.d.3.75
      24f59NApT2a(m), pN2 (12/18), pMXS, M (A)KRAS:p.Gly12Cys12.25
      If74NApT2a, pN2 (3/34), pM(ADR, BRA)ABRAF:p.Gly466ValNA
      IIm61NApT3, pN2 (14/35), pM(ADR)An.d.NA
      IIIm52NApT4, pN2 (9/24), pM (OTH
      Case III with metastases to pancreas.
      , HEP)
      AKRAS:p.Gly12CysNA
      IVm60NApT3, pN2 (12/29), pM(ADR)Sn.dNA
      Vm58NApT3, pN2 (15/28), pM(ADR)Mn.d.NA
      f, female; m, male; A, acinar; L, lepidic, MP, micropapillary; P, papillary; S, solid; ADR, adrenal; BRA, brain; HEP, liver; OTH, other; NA, not available; n.d., not detected.
      a Minor component shown in parentheses.
      b Case III with metastases to pancreas.
      For the assessment of TMB over time, a cohort of five patients with ADC and local as well as distant metastases was investigated. For each patient, one sample of the primary tumor plus one locoregional lymph node site (available in four cases) and one to two distant metastatic sites were tested (see Table 1 for clinicopathologic details).

      In Silico TMB Computation

      TMB was defined as the total number of missense mutations from WES data generated in the DKFZ HIPO and the NCT MASTER programs. The mean sequencing depth of the WES data set ranged from 180 to 200×. Additionally, a simulated panel-sequencing–based TMB (sim-psTMB) was calculated as the number of missense mutations detected by WES within the coding region covered by the TSO500 panel divided by the size of this region (1.34 Mbp). The TSO500 panel has a total size of 1.95 Mbp and covers 1.34 Mbp of coding region. psTMB and sim-psTMB levels and cutpoints were calibrated against WES-based TMB using linear regression fits.

      DNA Extraction and Quantification

      For DNA extraction, six consecutive 10-μm thick FFPE sections of each sample were pooled, deparaffinized, and digested with proteinase K overnight. Subsequently, DNA was extracted automatically using a Maxwell 16 Research system and the Maxwell 16 FFPE Tissue LEV DNA Purification Kit (both Promega, Madison, Wisconsin). DNA concentrations were determined with the Qubit HS DNA assay (Thermo Fisher Scientific, Waltham, Massachusetts). All assays for DNA extraction and quantification were performed according to the manufacturers’ protocols.

      Library Preparation and Massive Parallel Sequencing

      In the initial step of the library preparation for the capture-based TruSight Oncology 500 panel (Illumina), the grade of DNA integrity of a sample was assessed using the Genomic DNA ScreenTape Analysis on a 4150 TapeStation System (both Agilent, Santa Clara, California). To fragment the DNA strands to a length of 90 to 250 bp, 80 ng DNA of each sample were sheared according to their degradation level for 50 to 78 seconds using a focused ultrasonicator ME220 (Covaris, Woburn, Massachusetts). Following two-target capture and purification steps, the enriched libraries were amplified (15 cycles polymerase chain reaction [PCR]) and subsequently quality controlled using the KAPA SYBR Library Quantification Kit on a StepOnePlus quantitative PCR system (both Thermo Fisher Scientific). Up to eight libraries were sequenced simultaneously on a NextSeq 500 (Illumina) using high-output cartridge and v2 chemistry. All assays were performed according to the manufacturers’ protocols.

      NGS Data Analysis and TMB Determination

      Procession of raw sequencing data and variant calling was carried out using the TruSight Oncology 500 Local App (Illumina, pipeline version 1.3.0.39). All variants considered for TMB estimation were manually validated by visual inspection in the integrative genome viewer.
      • Robinson J.T.
      • Thorvaldsdottir H.
      • Winckler W.
      • et al.
      Integrative genomics viewer.
      Further, the presence of a variation called in one sample of a respective patient was checked in all associated samples. Polymorphisms/germline mutations identified by the TSO500 germline filter were evaluated by the comparison of multiple samples of the same tumor with varying tumor cell content and in nine cases by sequencing matched adjacent non-neoplastic lung tissue for nine specific cases in addition. TMB counts were calculated as the number of synonymous and nonsynonymous mutations divided by the covered coding region. For the multiregion sequencing approach, additionally an aggregated TMB was calculated to simulate a pooling of samples. To this end, the number of individual mutations considering all samples of a patient was divided by the average covered coding region of a patient. For this study, considerable ITH was defined as a variation of 5 mut/Mbp as it represents 50% of the most widely used cutpoint (10 mut/Mbp) applied in clinical trials that use tissue-based TMB testing.
      • Hellmann M.D.
      • Ciuleanu T.E.
      • Pluzanski A.
      • et al.
      Nivolumab plus ipilimumab in lung cancer with a high tumor mutational burden.

      IHC

      For the determination of tumor cell content, immune cell composition, and PD-L1 status, IHC stainings for thyroid transcription factor 1 (TTF-1), cluster of differentiation (CD) 45, CD8, and PD-L1 (see Supplementary Material 1 for antibody details) were prepared using an autostainer (BenchMark ULTRA, Ventana Medical Systems, Tucson, Arizona) according to the manufacturer’s instructions. The IHC sections were digitalized with a slide scanner (Aperio CS2, Leica Biosystems, Wetzlar, Germany) and evaluated with QuPath (v.0.1.2; Queen’s University, Belfast, United Kingdom) applying standard settings for cell detection. For automated cell categorization, specific classifiers were trained and verified by an experienced pathologist.
      • Bankhead P.
      • Loughrey M.B.
      • Fernandez J.A.
      • et al.
      QuPath: open source software for digital pathology image analysis.
      The tumor cell content as the histologic tumor purity (ratio of tumor cells to total cell number) was determined based on digital evaluation of the TTF1-IHC. PD-L1 positivity was defined as linear membranous PD-L1 staining greater than or equal to 1% of tumor cells.

      Statistical Data Analysis and Plot Generation

      For statistical analyses, the R software (v.3.3.0; R Core Team, 2016) was used with the following functions of the “stats” package (v.3.3.0): chisq.test() for chi-squared contingency table tests; cor.test() to test for association between paired samples using Pearson's product moment correlation coefficient; fisher.test() for Fisher's exact test; lm() to perform linear regression; and wilcox.test() for the Mann-Whitney U test.
      For plot generation the “ggplot2” (v2.1.0) and the “waffle” (v.0.7) package or Microsoft Excel 2013 (Microsoft, Redmond, Washington) and the “Daniel's XL Toolbox NG” (7.1.4, https://www.xltoolbox.net) add-in were used.

      Results

      Study Outline

      In the present study (Fig. 1), we investigated the spatial distribution of TMB estimates in a multiregional sample set of ADC addressing ITH and subsequently the temporal impact on TMB counts in a sample set of primary tumor and local as well as distant metastases. Considering both cohorts, 109 samples were sequenced with a mean read depth of 609×, covering an average coding region of 1.28 Mbp. Initially, the correlation of psTMB with WES data was examined and the mutations called for TMB measurement were manually validated.
      Figure thumbnail gr1
      Figure 1Study outline. Following an initial validation of the TSO500 TMB panel, we investigated the spatial distribution of TMB estimates in a multiregional ADC sample set and subsequently the temporal impact on TMB counts in a sample set of primary tumor and local as well as distant metastases. ADC, adenocarcinoma; TMB, tumor mutational burden.

      Agreement of TMB Measurement Based on WES and the TSO500 Panel

      psTMB estimated by sequencing with the TSO500 panel was compared to TMB determined by WES sequencing in a cohort of 16 NSCLC cases (Fig. 2). A strong Pearson correlation of R = 0.9 (p < 0.01) was observed between the two approaches, which is in line with in silico data from our group.
      • Buchhalter I.
      • Rempel E.
      • Endris V.
      • et al.
      Size matters: dissecting key parameters for panel-based tumor mutational burden analysis.
      WES TMB cutpoints of 158 (from clinical trial CheckMate 012), 199 (CM227), and 243 (CM026) somatic mutations were converted to psTMB cutpoints of 10.5, 13.5, and 16.7 mut/Mbp using a linear regression curve, which served for subsequent individual calibration of the TSO500 panel.
      • Carbone D.P.
      • Reck M.
      • Paz-Ares L.
      • et al.
      First-line nivolumab in stage IV or recurrent non–small-cell lung cancer.
      • Hellmann M.D.
      • Ciuleanu T.E.
      • Pluzanski A.
      • et al.
      Nivolumab plus ipilimumab in lung cancer with a high tumor mutational burden.
      • Hellmann M.D.
      • Nathanson T.
      • Rizvi H.
      • et al.
      Genomic features of response to combination immunotherapy in patients with advanced non–small-cell lung cancer.
      Using these predefined thresholds, classification as TMB-high versus TMB-low was in agreement for 13 (81%), 14 (88%), and 16 (100%) of the 16 investigated tumors, respectively. To decompose different sources of the deviations of psTMB from WES TMB, we investigated the impact of the limited panel size in more detail. To this end, we simulated psTMB by counting the number of mutations detected by WES in the regions captured by the panel, resulting in a sim-psTMB estimate. Correlations between sim-psTMB and WES TMB as well as between psTMB and sim-psTMB were significant (both p < 0.01) and higher than the correlation between psTMB and WES TMB (Figs. 2B and C). Differences between psTMB and WES TMB were typically (cohort median) 2.7 times higher than differences between sim-psTMB and WES TMB (Supplementary Material 2).
      Figure thumbnail gr2
      Figure 2Comparison of the performance of panel sequencing and whole exome sequencing (WES) for the estimation of tumor mutational burden (TMB) in NSCLC. A, Strong correlation of panel sequencing–based TMB (psTMB) and WES TMB (R = 0.9). For the WES TMB cutpoints of 158, 199, and 243 mutations corresponding to the psTMB cutpoints of 10.5, 13.5, and 16.7 mut/Mbp classification as TMB-high versus TMB-low was in agreement for 81%, 88%, and 100% of the tumors, respectively. Correlations between a simulated panel sequencing–based TMB (sim-psTMB) and WES TMB (B) as well as between psTMB and sim-psTMB (C) were higher than the correlation between psTMB and WES TMB. R = Pearson correlation.

      Validation of Mutations Called for TMB Measurement

      All mutations identified by the TSO500 mutation calling pipeline were manually validated, in nine cases also including a comparison to matched non-neoplastic tissue. In 21% (23 of 109) of the analyzed samples, mutation calling could be confirmed, whereas one to two and even more than two mutations were either missed (false-negative) or unjustified (false-positive) in 38% or 41% of samples, respectively (Fig. 3, left). The main reasons for the miss or nonconsideration of a mutation were the assumption of a single-nucleotide polymorphism/germline mutation (72%) and borderline allele frequencies (21%). An incorrect mutation call resulted mostly from a misclassification as somatic (84%) or the annotation of a complex mutation as two distinct mutations (14%). There was no association between the numbers of missed or unjustified mutations to the estimated TMB (Fig. 3, left gray line). In total, 583 somatic mutations (Supplementary Material 3; of which 581 were case specific) could be validated and were considered for TMB estimation (by design excluding recurrently mutated genes such as KRAS to avoid overestimation of TMB). Thus, the discordant number of mutations affected pipeline-automated TMB calculations. For the great majority of samples (72%) the absolute difference to the manually validated TMB was smaller than 1 mut/Mbp (Fig. 3, right). However, a switch of the TMB designation from high to low (n = 2) or vice versa (n = 2) was observed in 3.6% of the samples.
      Figure thumbnail gr3
      Figure 3Left, Manual validation of mutations considered for tumor mutational burden (TMB) estimation; red, missed mutations; blue, unjustified mutations; gray line gives the respective curated TMB estimation of each sample. Right, Resulting differences of the TMB estimation following manual validation of the called mutations.

      Spatial TMB Heterogeneity: Multiregional Analysis

      The spatial distribution of TMB counts was investigated assessing central tumor sections in a multiregional approach as well as locoregional lymph node metastases (Fig. 4A). Following segmentation, two to four samples of each tumor and up to two lymph node metastases were selected based on sufficient tumor cell content (>10%; Supplementary Material 4), and DNA yield (>4 ng/μL), with differing histologic growth patterns and distances to each other, if applicable. In total, TMB was determined with the TSO500 panel in 69 tumor segments and 23 locoregional lymph node metastases derived from 24 patients. TMB counts of the analyzed samples ranged from 0 to 52.55 mut/Mbp (Fig. 4B) and had a median value of 7.04 mut/Mbp. Considerable ITH of TMB counts, defined by us as a variation of at least 5 mut/Mbp between different regions of a given tumor, was detected in a third (7 of 24) of the analyzed cases (#3, #4, #6, #9, #13, #16, and #20). The highest TMB estimates in these tumors were 5.5, 7.8, 17.2, 14.1, 6.3, 52.6, and 14.8 mut/Mbp, respectively. Mean absolute deviations ranged from 2.43 to 6.11 with a maximum difference of 14.13 mut/Mbp in case #16. The variation of TMB counts within individual cases was even greater when lymph node metastases were included in the analysis (12 cases with ±5 mut/Mbp; maximum difference: 14.21 mut/Mbp; mean absolute deviations: 1.57 to 5.67).
      Figure thumbnail gr4
      Figure 4A, Representative sample (case #14): central tumor section, before (left) and after segmentation (middle), as well as the tumor cell content, DNA content, and histologic growth pattern (blue indicates lepidic; green indicates acinar; and gray indicates non-neoplastic) determined for each tumor segment separately. Segments selected for tumor mutational burden (TMB) measurement are circled in red. B, Overview of the multiregional TMB analysis of 24 adenocarcinoma (ADC) samples (two to six samples per tumor) using the TSO500 panel. Black dot indicates tumor segment, white square indicates lymph node metastasis, red line indicates aggregated TMB value, considering the mutations detected in all samples of a tumor. Bottom panel, TMB status considering 10 mut/Mbp as cutpoint for tumor segments/lymph nodes/aggregated TMB; green indicates positive, red indicates negative, yellow indicates positive and negative results, and gray indicates not available. Programmed death ligand 1 (PD-L1) status of tumor cells is shown in homologous darker color code. Upper left inset, Paired analyses of the average TMB values of tumor segments and lymph node metastases of respective cases. Red indicates decrease; black indicates increase. C, Showcases illustrating different factors influencing intratumor heterogeneity (ITH) of TMB counts. Left, case #4 low tumor cell content; #20 ITH, subclonal development; red indicates mutation present; blue indicates mutation not present/detection not valid; numbers indicate allele frequencies.
      Besides intratumoral variation of the mutation numbers, in 13 cases (#6 through #9, #11, #14, #16 through #20, #22, and #24) distinct mutations were exclusive to different regions within a tumor, also indicating branched tumor evolution. We also simulated pooling of DNA from a patient’s various tissue samples by calculating an aggregated TMB count of all detected mutations (Figure 4B, red horizontal bars) which resulted in 0.79 to 7.03 mut/Mbp higher TMB values for these tumors.
      Because universally accepted cutpoints regarding the classification of psTMB counts are not yet established and highly controversial, we applied various cutpoints from recent clinical trials. In Figure 4B, the TMB status of tumor segments, lymph node metastases, or the aggregated TMB counts was determined applying a cutpoint of 10 mut/Mbp, a clinically prospectively validated threshold. Respective corresponding analyses for additional cutpoints (10.5, 13.5, and 16.7 mut/Mbp) derived from WES data (158, 199, and 243 mutations) as referred to above, are given in Supplementary Material 5.
      Despite substantial intratumoral variability, in 71% (17 of 24) of the analyzed tumors, estimated TMB status was consistent in different tumor regions and lymph node metastases, with 12 cases (#1 through #5, #8, #10, #13 through 15, #18, and #23) found to be TMB-low and 5 cases (#11, #16, #17, #19, #21) to be TMB-high. Cases #12, #22, and #24 were TMB-high in all tumor segments and consequently was their aggregated TMB, but had at least one lymph node metastasis classified as TMB-low. In three cases (#6, #9, and #20) inconsistent TMB approximations were observed in different segments of the same tumor. Here, the analyzed lymph node metastases were TMB-low. In case #7, only the aggregated TMB would justify a TMB-high classification, whereas all individual tumor segments were found to be TMB-low.
      The occurrence of lower TMB values in lymph node metastases compared to corresponding primary tumors was significant (p = 0.016) in a paired analysis of the average TMB values of tumor segments and lymph nodes of respective cases (Fig. 4B, upper left inset). Only 6 of 23 analyzed lymph node metastases (#9-N1, #14-N1, #22-N2, #23-N1 + N2, and #24-N1) had a private mutation that was not detectable in the corresponding tumor. Except for one, these mutations had allele frequencies below 10%.
      In an IHC analysis, six tumors (25%) were found to be PD-L1–positive in all analyzed regions, whereas one case (#21) showed intratumoral variation of PD-L1 status. TMB status or TMB count did not correlate with PD-L1 status nor did it correlate with immune cell infiltrates (CD8 and CD45). There was merely a significant correlation (p < 0.01; R = 0.49) of tumor-infiltrating cytotoxic T cells (CD8) to the total number of present leucocytes (CD45), and both levels were significantly (p < 0.05) higher in PD-L1–positive tumors (Supplementary Material 6).
      An in-depth analysis of the seven cases with substantial intratumoral differences of TMB estimates revealed varying tumor cell content (4 of 7) and the development of distinct mutational profiles (3 of 7) as the two primary contributing factors (Fig. 4C). For example, in case #4, the tumor cell content of segment D4 was considerably lower (18%) when compared to regions B6 (29%) and F2 (42%). Here, 7 of 10 somatic mutations that were present in the two other segments could not be called due to allele frequencies (2% to 4%) below the assay’s detection limit of 5%. The three analyzed tumor segments of case #20 shared nine somatic alterations. Additionally, two to five private alterations were detected in each region as well as five alterations shared only between segments A2 and B1.

      TMB Heterogeneity in Tumor Progression: Comparison of Primary Tumor and Metachronous Distant Metastasis

      Next, we investigated the potential variation of TMB during tumor progression. Therefore, we assessed TMB in a smaller cohort (n = 5) of matched primary tumors, locoregional lymph node metastases (n = 4) resected at the time of surgery, and one to two distant metastases per case resected (n = 5) or biopsied (n = 3) several months (median = 11 months; range = 4 – 58 months) after initial surgery (Fig. 5). Supporting our previous observation, two of four lymph node metastases had lower TMB values compared to the primary tumor. TMB estimates for the other two lymph node metastases and for the distant metastases were generally in a similar range as for the respective primary tumor. Despite similar TMB estimates, we detected distinct private mutations in different samples. In case IV (Fig. 5, bottom), the great majority of mutations (56) were detectable in all samples, but up to four mutations were exclusively found in the primary tumor (n = 4), the lymph node metastasis (n = 1), and in the first (n = 4) or in the second distant metastasis (n = 1), respectively. Additionally, two mutations were shared between the tumor and the lymph node metastasis, one between lymph node and second metastases, two between tumor and both distant metastasis, and four between both distant metastases.
      Figure thumbnail gr5
      Figure 5Top, Tumor mutatioinal burden (TMB) estimation in matched samples of the primary tumor, locoregional lymph node metastases and distant metastases: t1 indicates time point 1, t2 indicates time point 2; NA indicates not available. Bottom, Detailed illustration of the detected mutations found in the samples of case IV, revealing shared and private mutations between the samples. Red indicates mutation present; blue indicates mutation not present/detection not valid. Numbers indicate allele frequencies. Metastasis: t1a indicates adrenal right, t1b indicates adrenal left.

      Discussion

      In this study, we comprehensively analyzed the applicability of a 523-gene–spanning targeted sequencing panel for estimation of TMB. Following recent in silico and now wet-lab assay validation, we investigated TMB estimates in multiregional and in temporally separated sample sets using two different ADC cohorts.
      • Buchhalter I.
      • Rempel E.
      • Endris V.
      • et al.
      Size matters: dissecting key parameters for panel-based tumor mutational burden analysis.
      Our data show considerable ITH of TMB estimates which could impact clinical decision-making. We uncovered critical technical and biological aspects that significantly influence diagnostic TMB assessment.
      We observed a strong correlation of TMB levels estimated with the TSO500 panel and TMB levels determined by WES (R = 0.9). Agreement of the two methods in classification of TMB as high or low increased when using higher TMB cutpoints. In a recent comprehensive theoretical analysis of psTMB estimates, we showed that the relative error of psTMB levels decreased proportionally to both the square root of the panel size and the square root of the TMB level.
      • Budczies J.
      • Allgauer M.
      • Litchfield K.
      • et al.
      Optimizing panel-based tumor mutational burden (TMB) measurement [e-pub ahead of print].
      Thus, relative errors are lower for high TMB levels, which is in line with a better classification performance of psTMB for high cutpoints. In terms of correlation with WES, the TSO500 panel performed similar to a panel of comparable size, but better than panels of smaller size, again in line with the theoretical analysis and with the observation that size matters.
      • Buchhalter I.
      • Rempel E.
      • Endris V.
      • et al.
      Size matters: dissecting key parameters for panel-based tumor mutational burden analysis.
      • Endris V.
      • Buchhalter I.
      • Allgauer M.
      • et al.
      Measurement of tumor mutational burden (TMB) in routine molecular diagnostics: in silico and real-life analysis of three larger gene panels.
      Finally, we showed that a substantial part of the deviation of psTMB from WES TMB is connected to the evaluation of mutations in only a restricted region of the exome, underscoring the relevance of our earlier work on simulations of psTMB. Although differences between sim-psTMB and WES-based TMB can be explained with the smaller sequence covered by the panel, the comparison of sim-psTMB and psTMB indicated that additional factors (e.g., different sequencing technologies, FFPE versus fresh frozen tissue, sampling bias, ITH, tumor purity, and the analysis of matched normal tissue) may influence TMB assessment.
      Upon manual evaluation of mutations called by the TSO500 local app, we could confirm these as confident and reliable mutational calls with an excellent reduction of artifacts from formalin fixation or misalignment. With using an increasing panel size for TMB estimation, correct germline filtering becomes eminently important. The algorithm applied here appeared sufficient in most instances, but leaves room for improvement considering a switch in the TMB designation in 3.6% of the analyzed samples after manual validation. Our matched germline analysis of non-neoplastic tissue revealed that several somatic mutations were considered germline or vice versa and that filtering was inconsistent between different samples of the same tumor based on, for example, varying allele frequencies. Although optimized algorithms for germline variant filtering without matched non-neoplastic tissue are described and may have their strengths, our data indicate that neither querying polymorphism databases nor approximations based on allele frequencies are sufficient for this task.
      • Sun J.X.
      • He Y.
      • Sanford E.
      • et al.
      A computational approach to distinguish somatic vs. germline origin of genomic alterations from deep sequencing of cancer specimens without a matched normal.
      Future studies are warranted to investigate whether concurrent germline sequencing for panel-based TMB estimation is as essential as it is for WES approaches.
      • Cheng D.T.
      • Mitchell T.N.
      • Zehir A.
      • et al.
      Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT): a hybridization capture-based next-generation sequencing clinical assay for solid tumor molecular oncology.
      In this regard, legal restrictions of germline analysis as well as available laboratory capacity or economic feasibility might impede rapid implementation in routine diagnostics.
      The multiregional assessment of psTMB estimates revealed that ITH was considerable and would have a critical impact on IO therapy decision making in 12.5% (3 of 24) of the analyzed cases. In these, a designation as TMB-high versus TMB-low was dependent on the tumor area sampled. Our findings are consistent with recently reported findings based on the TRACERx cohort, where 21% of the tumors had inconsistent TMB designations using WES data and a cutpoint of 10 mut/Mbp.
      • Rosenthal R.
      • Cadieux E.L.
      • Salgado R.
      • et al.
      Neoantigen-directed immune escape in lung cancer evolution.
      In this study, variations in tumor purity between sampled regions were found to be an important factor influencing TMB estimates.
      Additionally, the multiregional approach revealed distinct mutational profiles in different regions of the same tumor. Incorporating all TMB estimates of a tumor into an aggregated TMB, a post hoc pooling of samples, led to an increase of TMB estimates posing the intriguing question of whether a single sample approach is valid for estimation of a tumor’s TMB. Assuming that an aggregated TMB above the cutpoint would predict IO therapy response or treatment decisions similar to a single sample TMB, this would have even led to a different classification in one case. Tumors with high TMB were more likely to have multiple mutation sets, as seen in a recent study by Zhang et al.
      • Zhang Y.
      • Chang L.
      • Yang Y.
      • et al.
      The correlations of tumor mutational burden among single-region tissue, multi-region tissues and blood in non-small cell lung cancer.
      However, the predictive value of an aggregated TMB remains to be evaluated given that so far clinical trials have used tissue-based TMB estimates by sequencing single samples. ITH of the TMB status and TMB-high designation for the aggregated TMB, but not the individual tumor samples were seen for all applied cutpoints.
      Obtaining sufficient tissue for molecular testing in advanced-stage lung cancer patients can be challenging. Mostly, only little biopsy material is available precluding multiregional analysis. Despite its known limitations (e.g., variable DNA shedding), analyzing cell-free DNA derived from a liquid biopsy might have the potential to provide a more holistic picture of the present mutations, hence blood derived TMB (bTMB) estimation is in the focus of current clinical trials.
      • Volckmar A.L.
      • Sultmann H.
      • Riediger A.
      • et al.
      A field guide for cancer diagnostics using cell-free DNA: from principles to practice and clinical applications.
      • Gandara D.R.
      • Paul S.M.
      • Kowanetz M.
      • et al.
      Blood-based tumor mutational burden as a predictor of clinical benefit in non–small-cell lung cancer patients treated with atezolizumab.
      A recent exploratory analysis in a phase III trial was able to prove the feasibility of large-scale bTMB assessment and suggests that bTMB is a predictive biomarker for the combination of anti–CTLA-4 and anti–PD-L1 drugs in naive, EGFR and ALK wild type, advanced NSCLC.

      Peters S, Cho BC, Reinmuth N, et al. CT074 Tumor mutational burden (TMB) as a biomarker of survival in metastatic non–small cell lung cancer (mNSCLC): Blood and tissue TMB analysis from MYSTIC, a phase III study of first-line durvalumab ± tremelimumab vs chemotherapy. Paper presented at American Association for Cancer Research (AACR) Annual Meeting. March 29–April 3, 2019; Atlanta, Georgia.

      In this consideration, investigating the correlation between bTMB estimates and aggregated tumor TMB (multiple regions) estimates with therapy response would be of great importance.
      In lymph node metastases, we observed mostly lower TMB estimates. This might be due to lower tumor cell content because of surrounding lymphocytes or more likely, reflecting the oligoclonal nature of metastases, with less subclonal diversity when compared to a heterogeneous primary tumor which harbors multiple intermixed distinct subclones. This finding questions the use of lymph node metastases for TMB estimation of the tumor, or at least suggests that different cutpoints might need to be further evaluated.
      Interestingly, we also detected ITH in distant metastases, but this did not affect TMB estimates considerably. Although the relatively small sample size limits definite conclusions and futher studies are required to fully understand the impact on clinical decision-making, our results indicate that biopsy samples of (potentially easily accessible) metastatic sites can provide similar TMB estimates as the primary tumor.
      In conclusion, we show crucial factors influencing TMB estimation. Besides technical aspects such as tumor cell content, sufficient coverage, and germline filtering the subclonal development of tumors and subsequent ITH also affect TMB estimation. From a clinical point of view, ITH of the TMB status or a switch to a TMB-high designation when considering an aggregated TMB for the whole tumor could lead to misclassification of TMB greater than or equal to 17% of the cases, indicating limitations of single-sample based TMB estimations. In this regard, our findings point to open questions considering the definition of TMB as a predictive biomarker when used in clinical trials or therapeutic workflows.

      Acknowledgments

      The authors thank Diana Stürmer, Katja Lorenz, Tina Uhrig, Lidia Jost, Kathrin Ridinger, and Christiane Zgorzelski, as well as the Tissue Bank of the National Center for Tumor Diseases for excellent technical assistance.

      Supplementary Data

      • Supplementary 2

        Comparison of whole exome sequencing (WES)-based tumor mutational burden (TMB) to TMB estimations based on panel sequencing (psTMB) and on simulated panel sequencing (sim-psTMB)

      Figure thumbnail figs1
      Supplementary 5Overview of the multi-regional TMB analysis of 24 ADC (2-6 samples per tumor) using the TSO500 panel. Black dot = tumor segment, white square = lymph node metastasis, red line = aggregated TMB value, considering the mutations detected in all samples of a tumor. Bottom panel: TMB status considering 10.5, 13.5, and 16.7 mut/Mbp as cut-point for tumor segments / lymph nodes / aggregated TMB; “green” positive, “red” negative, “yellow” positive and negative results, “gray” not available. PD-L1 status of tumor cells: homologous darker color code. Upper left inset: Paired analyses of the average TMB values of tumor segments and lymph node metastases of respective cases; red = decrease; black = increase
      Figure thumbnail figs2
      Supplementary 6Left: correlation of CD8 and CD45 positive cells in all analyzed tumor samples. Proportion of CD8 (middle) and CD45 (right) positive cells in PD-L1 negative and positive tumor samples.

      References

        • Rizvi N.A.
        • Hellmann M.D.
        • Snyder A.
        • et al.
        Cancer immunology. Mutational landscape determines sensitivity to PD-1 blockade in non–small cell lung cancer.
        Science. 2015; 348: 124-128
        • Rosenberg J.E.
        • Hoffman-Censits J.
        • Powles T.
        • et al.
        Atezolizumab in patients with locally advanced and metastatic urothelial carcinoma who have progressed following treatment with platinum-based chemotherapy: a single-arm, multicentre, phase 2 trial.
        Lancet. 2016; 387: 1909-1920
        • Snyder A.
        • Makarov V.
        • Merghoub T.
        • et al.
        Genetic basis for clinical response to CTLA-4 blockade in melanoma.
        N Engl J Med. 2014; 371: 2189-2199
        • Van Allen E.M.
        • Miao D.
        • Schilling B.
        • et al.
        Genomic correlates of response to CTLA-4 blockade in metastatic melanoma.
        Science. 2015; 350: 207-211
        • Herbst R.S.
        • Soria J.C.
        • Kowanetz M.
        • et al.
        Predictive correlates of response to the anti–PD-L1 antibody MPDL3280A in cancer patients.
        Nature. 2014; 515: 563-567
        • Carbone D.P.
        • Reck M.
        • Paz-Ares L.
        • et al.
        First-line nivolumab in stage IV or recurrent non–small-cell lung cancer.
        N Engl J Med. 2017; 376: 2415-2426
        • Rizvi H.
        • Sanchez-Vega F.
        • La K.
        • et al.
        Molecular determinants of response to anti–programmed cell death (PD)-1 and anti–programmed death-ligand 1 (PD-L1) blockade in patients with non–small-cell lung cancer profiled with targeted next-generation sequencing.
        J Clin Oncol. 2018; 36: 633-641
        • Hellmann M.D.
        • Ciuleanu T.E.
        • Pluzanski A.
        • et al.
        Nivolumab plus ipilimumab in lung cancer with a high tumor mutational burden.
        N Engl J Med. 2018; 378: 2093-2104
        • Hellmann M.D.
        • Nathanson T.
        • Rizvi H.
        • et al.
        Genomic features of response to combination immunotherapy in patients with advanced non–small-cell lung cancer.
        Cancer Cell. 2018; 33: 843-852 e844
        • Goodman A.M.
        • Kato S.
        • Bazhenova L.
        • et al.
        Tumor mutational burden as an independent predictor of response to immunotherapy in diverse cancers.
        Mol Cancer Ther. 2017; 16: 2598-2608
        • Budczies J.
        • Seidel A.
        • Christopoulos P.
        • et al.
        Integrated analysis of the immunological and genetic status in and across cancer types: impact of mutational signatures beyond tumor mutational burden.
        Oncoimmunology. 2018; 7: e1526613
        • Brahmer J.R.
        • Tykodi S.S.
        • Chow L.Q.
        • et al.
        Safety and activity of anti–PD-L1 antibody in patients with advanced cancer.
        N Engl J Med. 2012; 366: 2455-2465
        • Thorsson V.
        • Gibbs D.L.
        • Brown S.D.
        • et al.
        The immune landscape of cancer.
        Immunity. 2018; 48: 812-830 e814
        • Gubin M.M.
        • Artyomov M.N.
        • Mardis E.R.
        • et al.
        Tumor neoantigens: building a framework for personalized cancer immunotherapy.
        J Clin Invest. 2015; 125: 3413-3421
        • Rittmeyer A.
        • Barlesi F.
        • Waterkamp D.
        • et al.
        Atezolizumab versus docetaxel in patients with previously treated non–small-cell lung cancer (OAK): a phase 3, open-label, multicentre randomised controlled trial.
        Lancet. 2017; 389: 255-265
      1. Peters S, Cho BC, Reinmuth N, et al. CT074 Tumor mutational burden (TMB) as a biomarker of survival in metastatic non–small cell lung cancer (mNSCLC): Blood and tissue TMB analysis from MYSTIC, a phase III study of first-line durvalumab ± tremelimumab vs chemotherapy. Paper presented at American Association for Cancer Research (AACR) Annual Meeting. March 29–April 3, 2019; Atlanta, Georgia.

        • Allgauer M.
        • Budczies J.
        • Christopoulos P.
        • et al.
        Implementing tumor mutational burden (TMB) analysis in routine diagnostics-a primer for molecular pathologists and clinicians.
        Transl Lung Cancer Res. 2018; 7: 703-715
        • Chan T.A.
        • Yarchoan M.
        • Jaffee E.
        • et al.
        Development of tumor mutation burden as an immunotherapy biomarker: utility for the oncology clinic.
        Ann Oncol. 2018; 30: 44-56
        • Buchhalter I.
        • Rempel E.
        • Endris V.
        • et al.
        Size matters: dissecting key parameters for panel-based tumor mutational burden analysis.
        Int J Cancer. 2019; 144: 848-858
        • Budczies J.
        • Allgauer M.
        • Litchfield K.
        • et al.
        Optimizing panel-based tumor mutational burden (TMB) measurement [e-pub ahead of print].
        Ann Oncol. 2019; 30: 1496-1506
        • Endris V.
        • Buchhalter I.
        • Allgauer M.
        • et al.
        Measurement of tumor mutational burden (TMB) in routine molecular diagnostics: in silico and real-life analysis of three larger gene panels.
        Int J Cancer. 2019; 144: 2303-2312
        • Ma Y.
        • Feng W.
        • Wu Z.
        • et al.
        Intra-tumoural heterogeneity characterization through texture and colour analysis for differentiation of non-small cell lung carcinoma subtypes.
        Phys Med Biol. 2018; 63: 165018
        • Park S.
        • Ha S.
        • Lee S.H.
        • et al.
        Intratumoral heterogeneity characterized by pretreatment PET in non-small cell lung cancer patients predicts progression-free survival on EGFR tyrosine kinase inhibitor.
        PLoS One. 2018; 13: e0189766
        • Cadioli A.
        • Rossi G.
        • Costantini M.
        • et al.
        Lung cancer histologic and immunohistochemical heterogeneity in the era of molecular therapies: analysis of 172 consecutive surgically resected, entirely sampled pulmonary carcinomas.
        Am J Surg Pathol. 2014; 38: 502-509
        • Warth A.
        • Muley T.
        • Meister M.
        • et al.
        The novel histologic International Association for the Study of Lung Cancer/American Thoracic Society/European Respiratory Society classification system of lung adenocarcinoma is a stage-independent predictor of survival.
        J Clin Oncol. 2012; 30: 1438-1446
        • Jamal-Hanjani M.
        • Wilson G.A.
        • McGranahan N.
        • et al.
        tracking the evolution of non–small-cell lung cancer.
        N Engl J Med. 2017; 376: 2109-2121
        • McGranahan N.
        • Swanton C.
        Clonal heterogeneity and tumor evolution: past, present, and the future.
        Cell. 2017; 168: 613-628
        • Kazdal D.
        • Harms A.
        • Endris V.
        • et al.
        Subclonal evolution of pulmonary adenocarcinomas delineated by spatially distributed somatic mitochondrial mutations.
        Lung Cancer. 2018; 126: 80-88
        • Dietz S.
        • Lifshitz A.
        • Kazdal D.
        • et al.
        Global DNA methylation reflects spatial heterogeneity and molecular evolution of lung adenocarcinomas.
        Int J Cancer. 2019; 144: 1061-1072
        • Alizadeh A.A.
        • Aranda V.
        • Bardelli A.
        • et al.
        Toward understanding and exploiting tumor heterogeneity.
        Nat Med. 2015; 21: 846-853
        • Mony J.T.
        • Schuchert M.J.
        prognostic implications of heterogeneity in intra-tumoral immune composition for recurrence in early stage lung cancer.
        Front Immunol. 2018; 9: 2298
        • de Bruin E.C.
        • McGranahan N.
        • Mitter R.
        • et al.
        Spatial and temporal diversity in genomic instability processes defines lung cancer evolution.
        Science. 2014; 346: 251-256
        • Warth A.
        • Endris V.
        • Stenzinger A.
        • et al.
        Genetic changes of non–small cell lung cancer under neoadjuvant therapy.
        Oncotarget. 2016; 7: 29761-29769
        • Lovly C.M.
        • Salama A.K.
        • Salgia R.
        Tumor heterogeneity and therapeutic resistance.
        Am Soc Clin Oncol Educ Book. 2016; 35: e585-e593
        • Dagogo-Jack I.
        • Shaw A.T.
        Tumour heterogeneity and resistance to cancer therapies.
        Nat Rev Clin Oncol. 2018; 15: 81-94
        • Dietz S.
        • Harms A.
        • Endris V.
        • et al.
        Spatial distribution of EGFR and KRAS mutation frequencies correlates with histological growth patterns of lung adenocarcinomas.
        Int J Cancer. 2017; 141: 1841-1848
        • Zhang L.L.
        • Kan M.
        • Zhang M.M.
        • et al.
        Multiregion sequencing reveals the intratumor heterogeneity of driver mutations in TP53-driven non-small cell lung cancer.
        Int J Cancer. 2017; 140: 103-108
        • Casadevall D.
        • Clave S.
        • Taus A.
        • et al.
        Heterogeneity of tumor and immune cell PD-L1 expression and lymphocyte counts in surgical NSCLC samples.
        Clin Lung Cancer. 2017; 18: 682-691 e685
        • Liu Y.
        • Dong Z.
        • Jiang T.
        • et al.
        Heterogeneity of PD-L1 expression among the different histological components and metastatic lymph nodes in patients with resected lung adenosquamous carcinoma.
        Clin Lung Cancer. 2018; 19: e421-e430
        • Rosenthal R.
        • Cadieux E.L.
        • Salgado R.
        • et al.
        Neoantigen-directed immune escape in lung cancer evolution.
        Nature. 2019; 567: 479-485
        • Travis W.D.
        • Brambilla E.
        • Nicholson A.G.
        • et al.
        The 2015 World Health Organization classification of lung tumors: impact of genetic, clinical and radiologic advances since the 2004 classification.
        J Thorac Oncol. 2015; 10: 1243-1260
        • Horak P.
        • Klink B.
        • Heining C.
        • et al.
        Precision oncology based on omics data: the NCT Heidelberg experience.
        Int J Cancer. 2017; 141: 877-886
        • Kazdal D.
        • Harms A.
        • Endris V.
        • et al.
        Prevalence of somatic mitochondrial mutations and spatial distribution of mitochondria in non–small cell lung cancer.
        Br J Cancer. 2017; 117: 220-226
        • Robinson J.T.
        • Thorvaldsdottir H.
        • Winckler W.
        • et al.
        Integrative genomics viewer.
        Nat Biotechnol. 2011; 29: 24-26
        • Bankhead P.
        • Loughrey M.B.
        • Fernandez J.A.
        • et al.
        QuPath: open source software for digital pathology image analysis.
        Sci Rep. 2017; 7: 16878
        • Sun J.X.
        • He Y.
        • Sanford E.
        • et al.
        A computational approach to distinguish somatic vs. germline origin of genomic alterations from deep sequencing of cancer specimens without a matched normal.
        PLoS Comput Biol. 2018; 14: e1005965
        • Cheng D.T.
        • Mitchell T.N.
        • Zehir A.
        • et al.
        Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT): a hybridization capture-based next-generation sequencing clinical assay for solid tumor molecular oncology.
        J Mol Diagn. 2015; 17: 251-264
        • Zhang Y.
        • Chang L.
        • Yang Y.
        • et al.
        The correlations of tumor mutational burden among single-region tissue, multi-region tissues and blood in non-small cell lung cancer.
        J Immunother Cancer. 2019; 7: 98
        • Volckmar A.L.
        • Sultmann H.
        • Riediger A.
        • et al.
        A field guide for cancer diagnostics using cell-free DNA: from principles to practice and clinical applications.
        Genes Chromosomes Cancer. 2018; 57: 123-139
        • Gandara D.R.
        • Paul S.M.
        • Kowanetz M.
        • et al.
        Blood-based tumor mutational burden as a predictor of clinical benefit in non–small-cell lung cancer patients treated with atezolizumab.
        Nat Med. 2018; 24: 1441-1448