Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The AURORA pilot study for molecular screening of patients with advanced breast cancer–a study of the breast international group

The AURORA pilot study for molecular screening of patients with advanced breast cancer–a study of... www.nature.com/npjbcancer ARTICLE OPEN The AURORA pilot study for molecular screening of patients with advanced breast cancer–a study of the breast international group 1 1 2 3 4 5 6 Marion Maetens , David Brown , Alexandre Irrthum , Philippe Aftimos , Giuseppe Viale , Sibylle Loibl , Jean-François Laes , 7 8,9 10,11 5 9 10,11 1 Peter J. Campbell , Alastair Thompson , Javier Cortes , Sabine Seiler , Sara Vinnicombe , Mafalda Oliveira , Françoise Rothé , 1 2 2 1 3 12 1 Yacine Bareche , Debora Fumagalli , Dimitrios Zardavas , Christine Desmedt , Martine Piccart , Sherene Loi and Christos Sotiriou Several studies have demonstrated the feasibility of molecular screening of tumour samples for matching patients with cancer to targeted therapies. However, most of them have been carried out at institutional or national level. Herein, we report on the pilot phase of AURORA (NCT02102165), a European multinational collaborative molecular screening initiative for advanced breast cancer patients. Forty-one patients were prospectively enroled at four participating centres across Europe. Metastatic tumours were biopsied and profiled using an Ion Torrent sequencing platform at a central facility. Sequencing results were obtained for 63% of the patients in real-time with variable turnaround time stemming from delays between patient consent and biopsy. At least one clinically actionable mutation was identified in 73% of patients. We used the Illumina sequencing technology for orthogonal validation and achieved an average of 66% concordance of substitution calls per patient. Additionally, copy number aberrations inferred from the Ion Torrent sequencing were compared to single nucleotide polymorphism arrays and found to be 59% concordant on average. Although this study demonstrates that powerful next generation genomic techniques are logistically ready for international molecular screening programs in routine clinical settings, technical challenges remain to be addressed in order to ensure the accuracy and clinical utility of the genomic data. npj Breast Cancer (2017) 3:23 ; doi:10.1038/s41523-017-0026-6 INTRODUCTION The Breast International Group launched AURORA–Aiming to Understand the Molecular Aberrations in Metastatic Breast Cancer, Several efforts have advanced our understanding of the altera- 1, 2 a pan-European molecular screening programme whose main tions characterizing cancer genomes. Coupled to recent goal is to deepen our knowledge of the genomic landscape of successes of targeted therapies in patients with molecularly 3, 4 advanced breast cancer. Herein, we report on the pilot phase of profiled tumours and the decreasing costs of massively parallel this study whereby the primary objective was to investigate the sequencing, this has motivated several studies, albeit of limited feasibility with four European recruitment sites and central size, to investigate the implementation of personalised molecular 5–7 pathological and sequencing facilities. Secondary aims were to screening in the clinical settings. assess the concordance of somatic mutations between two Most of these studies were focused on primary tumours and targeted next generation sequencing (NGS) platforms and of despite growing evidence that distant metastases may harbour somatic copy number aberrations (CNA) obtained from NGS and additional molecular alterations absent from their matched 8–12 single nucleotide polymorphism (SNP) arrays. primaries, genomic information about metastatic disease remains limited. Even though the clinical relevance of many of these alterations remains to be established, it is increasingly RESULTS recognised that molecular profiling of advanced disease could Patient recruitment and logistics help elucidate the biological underpinnings of phenomena such as distant recurrence and the emergence of de novo resistance to A total of 41 patients provided informed consent and were therapy. Lastly, in order to find applications in routine clinical enroled in this pilot study between February 2013 and September practice, it is essential to assess the reliability and robustness of 2014. Fig. 1a and b illustrates the study design and Table 1 the chosen sequencing platform using orthogonal sequencing summarises the clinical and pathological characteristics of the 14, 15 strategies. patients. Formalin fixed paraffin embedded (FFPE) biopsies of 1 2 J.-C. Heuson Breast Cancer Translational Research Laboratory, Institut Jules Bordet, Université Libre de Bruxelles, Brussels, Belgium; Breast International Group, Brussels, Belgium; 3 4 5 Department of Medical Oncology, Institut Jules Bordet, Université Libre de Bruxelles, Brussels, Belgium; European Institute of Oncology, University of Milan, Milan, Italy; Sana 6 7 Klinikum, Offenbach, Germany and German Breast Group, Neu-Isenburg, Germany; OncoDNA, Gosselies, Belgium; Cancer Genome Project, Wellcome Trust Sanger Institute, 8 9 Hinxton, UK; Department of Breast Surgical Oncology, University of Texas MD Anderson Cancer Center, Houston, USA; Dundee Cancer Centre, University of Dundee, Dundee, 10 11 12 UK; Ramon y Cajal University Hospital, Madrid, Spain; Vall d´Hebron Institute of Oncology, Barcelona, Spain and Division of Clinical Medicine and Research, Peter MacCallum Cancer Centre, Melbourne, Victoria, Australia Correspondence: Marion Maetens (marion.maetens@bordet.be) Marion Maetens, David Brown, Sherene Loi and Christos Sotiriou contributed equally to this work. Received: 24 February 2017 Revised: 4 May 2017 Accepted: 1 June 2017 Published in partnership with the Breast Cancer Research Foundation AURORA—screening patients with advanced breast cancer M Maetens et al. Fig. 1 Logistics and feasibility of the study. a logistic workflow of the study, b inclusion criteria and number of patients with successful results, c anatomical distribution of biopsied lesions with breakdown by recruiting centre, d global distribution of patients by recruiting centre, e turnaround time for each patient with breakdown by recruiting centre, f median global turnaround time by recruiting centre, and g median delivery turnaround time by recruiting centre. In a, the Illumina sequencing and Affymetrix OncoScan SNP arrays were done as batch processes. 19 patients were sequenced by Illumina targeted NGS and 18 patients were genotyped using the Affymetrix SNP arrays. However, these are not fully overlapping subsets and only 14 (54%) patients had the full set of all three data types. In a and b, the Ion Torrent sequencing results were obtained in real-time and in (e), the darker shades indicate delivery turnaround time (TAT) whilst the lighter shades indicate global turnaround time. The colour codes for the recruiting centres in c to g are indicated in (d) metastatic lesions and whole blood were prospectively collected Mutation detection from the Ion Torrent OncoDEEP clinical cancer panel and CNA using SNP arrays from 35 (85%) of them. 20 (57%) patients had ER + metastatic disease, 7 (20%) were HER2 + and 8 (23%) were triple negative Quality control metrics for the sequencing data are provided in breast cancer. Fig. 1c and d represent the distribution of the Supplementary Figure S1. The target regions were covered on different anatomical sites and the breakdown of patients’ average at 1758X (494–3075X) sequencing depth. Non- recruitment by participating centre. Following central pathological synonymous somatic mutations were called from the OncoDEEP review, a tumour content below 10% was recorded for 8 (23%) clinical cancer panel in exons covered by at least 100X sequencing patients whilst the median cellularity was 50% (range 10–85). One depth and a fixed threshold of 10% variant allele fraction (VAF). In additional sample had an insufficient amount of extracted DNA total, 128 unique genes harboured at least one mutation whilst the median yield was 3.6 µg (range 0.075–59.2). Overall, representing one third of the panel (Fig. 2a). The median number real-time Ion Torrent sequencing results were obtained for 26 of mutations indexed per patient was 6 (range 0–35) and the overall mutation detection rate i.e., percentage of patients where (74%) of the 35 patients (Fig. 1b). The median global turnaround time (TAT) from patient biopsy to sequencing results was 51 at least one mutation could be indexed, was 96%. Only 8 (4.15%) of the 193 mutations had previously been described in release working days (range 16–146) whilst the median delivery TAT v.76 of COSMIC in any type of cancer whilst 178 (92.23%) from sample reception at the sequencing facility to final NGS report was 9 working days (range 5–17) (Fig. 1e). There were no mutations had never been reported in the literature. The most frequently mutated genes were PIK3CA (50%), TP53 (31%), SYNE1 significant differences in TAT between the recruiting centres (19%), and NF2 (15%). (Fig. 1f and g). npj Breast Cancer (2017) 23 Published in partnership with the Breast Cancer Research Foundation AURORA—screening patients with advanced breast cancer M Maetens et al. patients harboured at least one actionable mutation, such as Table 1. Clinical and pathological characteristics of patients and PIK3CAp.H1047R or ESR1p.Y537N, whilst 8 (31%) had only one biopsies (n = 35) such mutation and six (23%) had at least one biologically relevant mutation and none that were clinically actionable (Supplementary Characteristics Number of patients (%) Figure S5). Comparatively, 16 of 18 (89%) patients for whom the Age (years) SNP arrays were available harboured at least one actionable CNA, 17, 18 such as EGFR amplification, whilst 2 (11%) had only one such Median 56 aberration and an equal number had at least one biologically Range 23–73 relevant CNA and none that were clinically actionable. Using the ECOG performance status combined information from the SNP arrays and the targeted 0 16 (45) sequencing, and focusing on the subset of strictly actionable 1 17 (49) alterations, 23 of 26 (88%) patients had at least one clinically actionable alteration whilst only 3 (12%) had only one such 2 2 (6) alteration. Overall, of the initial 7 patients without actionable Breast cancer subtype mutations, four were found to harbour at least one actionable ER+ 20 (57) CNA. HER2+ 7 (20) ER-/HER2- 8 (23) Orthogonal cross-testing of substitution calls using Illumina NGS Number of prior lines of therapy The substitutions obtained from the Ion Torrent sequencing were 0 6 (17) compared to data generated from the Illumina NGS platform. The same DNA samples were sequenced and substitutions were called 1 5 (14) in overlapping regions using similar filtering criteria and a 2 4 (11) combination of three established mutation callers. Patient 3 6 (17) matched data were available for 19 (73%) patients and all three >3 14 (41) mutation callers were mostly consistent (Fig. 3a). All but one Number of metastatic sites actionable substitution, NOTCH1p.D2082E in cluster 1 (see below), were concordant. 1 8 (23) The substitutions were categorised into three clusters and the 2 10 (28) error rates of each mutation caller and several combinations 3 8 (23) thereof were benchmarked using this framework. Substitutions in >3 9 (26) clusters 1 and 3 are exclusive to one of the sequencing platforms and have 0% VAF in the alternate data, whereas substitutions in ECOG; Eastern Cooperative Oncology Group, ER; oestrogen receptor cluster 2 are those found by either or both NGS platforms and evaluated by IHC, HER2; human epidermal growth factor receptor 2 evaluated by IHC and FISH. The breast cancer subtypes are based on the have non-zero VAF in both sequencing data. The definite calls characteristics of the metastatic lesions were made for each substitution based on a majority vote of any two of the three mutation callers (Fig. 3b and c). This choice was guided by a low global Illumina specific false negative error rate We evaluated two established methods to call CNA from SNP on cluster 2 substitutions whilst maintaining a relatively low global arrays and obtained similar results (Fig. 2c and d) thereby attesting Illumina specific false positive error rate on cluster 2 and 3 to the robustness of SNP arrays for the estimation of CNA. All the substitutions. Using this approach, a global concordance rate of patients presented with at least one CNA and the CNA frequencies 40.9% was achieved. were comparable to those reported in the literature (Fig. 2d–g). Substitutions in cluster 1 cannot be reconciled. There was a For instance, 13 (72%) patients harboured a gain or an statistically significant difference in coverage between the Ion amplification of chromosome 8q where the MYC oncogene Torrent and Illumina NGS data for these substitutions (Fig. 3d). resides whilst 9 (50%) had a deletion of 17p where TP53 is located. However, all but one of the substitutions were covered in excess Two patients presented with a higher than expected number of of 100X in the Illumina NGS data and the residual coverage, which mutations. The integration of the sequencing and CNA data is the difference between the observed sequencing depth and the obtained from these two outlier patients is shown in Supplemen- expected value at 99% statistical power to detect a mutation for tary Figure S2. In general, the distributions of VAF and cancer cell given CCF and copy numbers, were positive for all except two fractions (CCF) were high and tightly clustered, indicative of substitutions (Fig. 3e). Similarly, two substitutions were called in genuine somatic mutations. For the whole cohort, ERBB2 cluster 3 which constitute Illumina specific false positives and were amplification status was assessed centrally using FISH whilst PTEN covered at 239X and 1668X using the Ion Torrent NGS platform. In loss was evaluated by immunohistochemistry (IHC). These single total, only 2.7% of the substitutions in cluster 2 were Illumina assay data were compared with CNA obtained from the SNP arrays specific false positives using the Ion Torrent platform as standard and the results are shown in Supplementary Figures S3 and S4.No whilst 5.41% were Illumina specific false negatives. There were no false negative calls were recorded for ERBB2 amplification status statistically significant differences in sequencing coverage comparing SNP arrays to FISH data. However, 7 of 18 patients between these groups of substitutions and the true positives (39%) effectively harboured supernumerary copies of ERBB2 (n ≤ 4 called by both platforms (Fig. 3f and g). Furthermore, similar to copies) which were missed by FISH assays owing to centromere 17 cluster 1 substitutions, all the Illumina specific false negatives had co-amplification or low level 17q polysomy. There was no concrete positive residual coverage (Fig. 3e). For comparison, the sequen- evidence of a correspondence between PTEN IHC staining results cing coverage and the percentage of non-reference bases at these and the corresponding copy number level as determined by SNP particular loci in the normal matched samples are shown in array. Supplementary Figure S6. Actionable alterations from targeted NGS and SNP arrays Comparison of CNA from targeted NGS and SNP arrays Supplementary Tables S1 and S2 provide the definitions of The CNA obtained from the Ion Torrent sequencing were also actionable and biologically relevant alterations. 19 of 26 (73%) compared to the profiles obtained from SNP arrays and Illumina Published in partnership with the Breast Cancer Research Foundation npj Breast Cancer (2017) 23 AURORA—screening patients with advanced breast cancer M Maetens et al. Fig. 2 Mutation detection from the OncoDEEP clinical cancer panel and CNA from SNP arrays. a genes for which at least one mutation was indexed across the 26 patients, b and c comparison of cancer cell fraction and genomic mass respectively obtained from two widely used algorithms for estimating CNA from SNP arrays. In a, asterisk indicates potentially hypermutated patients. In b and c, each dot represents a sample. In d–g, each dot represents a gene from the list of clinically actionable or biologically relevant targets and is the mean of 100 bootstrap replicates such that for each replicate, the external cohorts are matched for ER and HER2 status. The size of each dot is proportional to the standard error of the mean frequency estimate NGS. Fig. 4a–c show the Log ratios from the SNP array and both bimodal for the Ion Torrent platform with 5 (29%) samples NGS platforms for a case patient. The segmented Log ratios from displaying a poor correlation of ρ < 0.5 (Fig. 5a). The median the SNP array and Illumina sequencing clustered, as expected, into correlation coefficients were ρ = 0.615 and ρ = 0.745 for the Ion canonical copy number genotypes whilst the segmented Log Torrent and Illumina NGS platforms respectively. ratios obtained from the Ion Torrent NGS platform were spread In order to further evaluate the ability of both NGS platforms to between those canonical genotypes leading to relatively poor call CNA, the segmented Log ratios were grouped into three correlation in Log ratio data space (Fig. 4d and e). Nonetheless, categories i.e. deletion (−1), copy neutral (0), gain/amplification (1) this approach was applied to each copy number platform for 14 and further compared using the accuracy which is the sum of (54%) patients for whom all three data types were available and concordant calls relative to the total number of aberrations. Fig. 5b the segmented Log ratios were compared genome-wide using shows the distribution of accuracy values evaluated genome-wide the Spearman’s correlation. The distribution of correlation for each patient. The average accuracy of CNA call was 59.1% for coefficients comparing the SNP array to the NGS data was the Ion Torrent and 74.0% for the Illumina NGS platforms. Both npj Breast Cancer (2017) 23 Published in partnership with the Breast Cancer Research Foundation AURORA—screening patients with advanced breast cancer M Maetens et al. Fig. 3 Orthogonal cross-testing of substitution calls using Illumina NGS. a distribution of % concordance for single nucleotide substitutions indexed from the Ion Torrent and Illumina NGS platforms using different mutation callers, b % error rate of the different mutation callers over three substitution clusters compared to the Ion Torrent sequencing, c correlation of VAF for substitutions indexed from the Ion Torrent and Illumina NGS using a majority rule of any two of three mutation callers, d comparison of sequencing coverage between the Ion Torrent and Illumina NGS for substitutions in cluster 1, e empirical cumulative distribution of theoretical residual coverage from the Illumina sequencing for substitutions in cluster 1 and cluster 2 false negatives, f comparison of sequencing coverage from the Ion Torrent NGS for substitutions in cluster 2 false positives, and g comparison of sequencing coverage from the Illumina NGS for substitutions in cluster 2 false negatives. In a, the different mutation callers and any combination thereof are colour coded and indicated at the bottom. The leftmost panel gives the number of substitutions called. In b–g, the cluster numbers are relative to (c). Substitutions in clusters 1 and 3 are exclusive to one of the sequencing platforms and have 0% VAF in the alternate data. Substitutions in cluster 2 are those found by either or both NGS platforms and have non-zero VAF in both sequencing data. False negatives in cluster 2 are substitutions indexed by Ion Torrent NGS only whilst false positives are substitutions indexed by Illumina NGS only using a given mutation caller or any combination thereof. The size of each dots is proportional to the difference in coverage between the two sequencing platforms. In e, the residual coverage is obtained by subtracting the theoretical coverage required to achieve 99% power for indexing a substitution given one mutated copy out of n total copies from the observed value of sequencing depth. Only two substitutions in cluster 1 failed the criteria of positive residual for detection and are non-callable loci distributions were bimodal with 6 (35%) and 5 (29%) patients 86.8–93.3) for MYC, 63.8% (95% CI: 61.4–66.2) for CCND1 and showing low values of α < 0.5 and α < 0.6 for the Ion Torrent and 57.4% (95% CI: 52.3–62.5) for ERBB2 whilst the corresponding Illumina NGS platforms, respectively. Lastly, the ability of both values for the same genes were 29.5% (95% CI: 27.5–31.8), 40.2% sequencing platforms to call CNA in particular genes of interest (95% CI: 37.7–42.7) and 28.3% (95% CI: 25.6–30.9), respectively, was evaluated on 15 clinically actionable or biologically relevant using the Ion Torrent sequencing platform. There was a significant genes. For each gene, the CNA calls were pooled and the cohort- negative correlation between the accuracy and the number of wise concordance measured as the accuracy of concordant calls. aberrations per patient for the Ion Torrent data (Fig. 5d) but not Fig. 5c contrasts the distribution of accuracy values observed for the Illumina NGS platform (Fig. 5e). There were no associations across these 15 genes for both NGS platforms. We observed between the accuracy measured cohort-wise and the frequency of relatively high concordance rates using the Illumina NGS platform aberrations affecting the 15 genes of interest for either of the two for genes frequently aberrant in breast cancers e.g., 90.1% (95% CI: NGS platforms (Fig. 5f and g). Published in partnership with the Breast Cancer Research Foundation npj Breast Cancer (2017) 23 AURORA—screening patients with advanced breast cancer M Maetens et al. Fig. 4 Genome-wide copy number profiles inferred from SNP arrays and targeted NGS data. Genome-wide Log ratio profiles of patient IJB0021 obtained using a SNP array b Ion Torrent and c Illumina targeted NGS data. The solid vertical lines represent chromosome boundaries whilst the dashed horizontal lines indicate canonical copy numbers inferred from the cancer cell fraction and ploidy of the sample, both estimated from the SNP array. The solid horizontal lines represent the segmented Log ratios. Correlation of segmented Log ratios between d 2 2 the SNP array and the Ion Torrent platform, e the SNP arrays and the Illumina platform, and f the two NGS platforms for the same case patient. In a–c, the loci are sorted according to their coordinate on the human genome reference hg19/GRCh37. For ease of representation, they are plotted by indices DISCUSSION communication). 73% of the patients harboured one or more actionable mutations and this increased to 88% when considering The growing number of targeted anticancer therapies either the underlying CNA. These numbers are encouraging on account approved or under clinical development has led to a wide interest 19, 20 of the availability of alterations that can be theoretically targeted in personalised treatment approaches in clinical practice. despite limited clinical trial based evidence of benefit to patient However, many of these cancer molecular screening initiatives survival from targeted therapy in breast cancer. Yet, the decision have had limitations. For instance, several of these screening to prioritise a particular molecular target for trial allocation given programs used different sequencing techniques and very often several such alterations within the same patient constitutes a lacked orthogonal validation. Together with varying concepts of considerable challenge for which specific algorithms and com- what constitutes an actionable mutation, this makes the results bined expertise will be required in the future. difficult to compare. The current study is a pilot phase undertaken Regarding the secondary endpoint and considering clinically to evaluate the feasibility of AURORA, a pan-European molecular actionable mutations, our results are reassuring. All but one screening programme for advanced breast cancer patients. substitution was positive using both the Ion Torrent and Illumina The success rate for the primary endpoint of 63% compares 5–7, 21 sequencing platforms. Furthermore, from a pragmatic viewpoint, 5 favourably with existing literature. The median global TAT of (26%) sample pairs showed complete agreement. However, no 51 working days was due to unforeseen delays either between single mutation caller was able to reproduce globally with 100% patient consent and surgical biopsy or between collection of concordance the results obtained in real-time from the Ion Torrent biological samples and shipment to the central pathology laboratory. In part, due to the above findings, these parameters NGS platform. The Illumina specific false negative and false are now rigorously monitored in the parent programme where the positive rates over somatic substitutions with non-zero VAF in global TAT does not exceed 20 working days (personal both data types can be reasonably contained whilst substitutions npj Breast Cancer (2017) 23 Published in partnership with the Breast Cancer Research Foundation AURORA—screening patients with advanced breast cancer M Maetens et al. Fig. 5 Comparison of copy number profiles from SNP arrays and targeted sequencing data. a distribution of Spearman’s correlation coefficient comparing the segmented Log ratios from the SNP arrays and the NGS data across patients with all three data types. The segmented Log 2 2 ratios were further categorised into copy number aberration calls i.e., −1, 0 and +1 and compared patient wise between the SNP arrays and the NGS data. The resulting distributions of accuracy values are depicted in b. For a small set of 15 clinically actionable or biologically relevant genes, the accuracy values comparing SNP arrays and NGS data were computed using all available samples and displayed individually for each gene in c. d, e correlation of accuracy values measured genome-wide for each patient as in b vs. the number of aberrations as determined by the corresponding NGS platform. f, g correlation of accuracy for each of the 15 clinically actionable or biologically relevant genes vs. the frequency of aberrations for the same genes measured using the SNP array. In c, each value of accuracy was generated by 100 bootstrap replicates. The values displayed represent the mean of the replicates and the error bars represent the standard error of this estimate private to either sequencing technology despite adequate cover- sensitivity to a given drug. In the broader context of AURORA, all age in both data types were more frequent in the Ion Torrent data. the results are reported to investigators unaltered. However, our Thus, from a technical perspective, our results are somewhat analysis shows the major limitations of this approach outside sobering and whilst the higher background error rate of the Ion known mutation hotspots and in anticipation, the AURORA study Torrent NGS technology could be a contributing factor, the small design allows biobanking of frozen and residual FFPE material for number of samples and mutations assessed here do not allow for later research purposes using alternative high-throughput tech- a full exploration of the possible causes of these discrepancies. nologies such as whole genome or exome sequencing. Massively parallel sequencing using targeted gene panels has To the best of our knowledge, no other studies have compared emerged as a technology that can be applied within a clinically prospectively CNA obtained from targeted sequencing data to meaningful TAT to identify known biomarkers of resistance or SNP arrays in a clinical setting. Our results show that the gold Published in partnership with the Breast Cancer Research Foundation npj Breast Cancer (2017) 23 AURORA—screening patients with advanced breast cancer M Maetens et al. standard FISH assay for ERBB2 is fully concordant with SNP arrays. shipped to (1) OncoDNA (Gosselies, Belgium) for sequencing using Ion Torrent NGS technology, (2) the Wellcome Trust Sanger Institute (Hinxton, However, the Illumina NGS platform outperforms the Ion Torrent UK) for sequencing using Illumina NGS technology, and (3) the J.-C. Heuson technology when compared to the same SNP arrays. This is not Breast Cancer Translational Research Laboratory (Brussels, Belgium) where unexpected since the Illumina NGS panel accommodates inter- the Affymetrix OncoScan FFPE Express arrays were performed. The Illumina genic targets which, combined with the fact that hybrid capture NGS and Affymetrix SNP arrays were performed as batch processes whilst methods for sequencing library preparation often carry over off- the sequencing and delivery of results from OncoDNA were monitored in target reads, increases the effective number of regions where real time. 22, 23 copy numbers can be evaluated. Several algorithms take advantage of this, benefiting downstream CNA calling and Targeted gene screen using Ion Torrent NGS translating into higher concordance with the denser SNP arrays. Somatic mutations were assessed using the OncoDEEP clinical cancer The combination of assays, such as mutation profiling and panel which is a validated AmpliSeq design panel targeting the exonic assessment of CNA, into one cost effective package is an attractive regions of 409 cancer related genes (Supplementary Table S3). The same idea. Together with a low requirement of input DNA, as is often protocol was applied to DNA extracted from FFPE tumour and whole blood obtainable from clinical samples, makes for a compelling normal matched samples. Briefly, the targeted sequencing libraries were argument in favour of the Ion Torrent NGS platform. However, generated using the Ion AmpliSeq library kit 2.0 according to the given the poor agreement with solid benchmarks such as SNP manufacturer’s instructions (Life Technologies) using 80 ng of genomic DNA. The primers used for amplification were partially digested by Pfu arrays, the results presented here do not support, at least for now, restriction enzyme and the digestion products were ligated to barcoded the targeted Ion Torrent NGS platform as a truly multipurpose adaptors and purified using Ampure Beads. The purified products were assay in a clinical setting. Within AURORA, central pathology amplified for five cycles and purified once more using Ampure Beads. The results reported to investigators in real-time include ERBB2 IHC as quality of the libraries was assessed using a qPCR following which 10 pM of well as FISH when applicable. This is clinically meaningful since each library underwent emulsion PCR using an IonChef system. The chips ERBB2 gene amplification remains, to date, the sole approved CNA were loaded on an Ion PGM and were sequenced at a target coverage of biomarker in breast cancer. Sub-studies undertaken in the 500X. framework of the main AURORA programme include real-time SNP arrays and retrospective whole genome sequencing to Targeted gene screen using Illumina NGS alleviate the technical limitations of identifying CNA using the The exonic regions of 371 cancer related genes (Supplementary Table S4) Ion Torrent technology and will allow the study of the clonal were enriched using a custom design of SureSelect RNA baits following the evolution of breast cancer. manufacturer’s instructions (Agilent). The same protocol was applied to Overall, our study contrasts the benefits and pitfalls of DNA extracted from tumour and normal matched samples. Briefly, 500 ng personalised molecular screening for patients with advanced of genomic DNA was fragmented to an average insert size of 145 bp cancer involving the combination of multiple high-throughput (75–300 bp) and subjected to Illumina DNA sequencing library preparation using the Bravo automated liquid handling platform. Individual samples genomic techniques. In view of these results, greater effort is were indexed using a unique DNA barcode via six cycles of PCR. Equimolar being devoted to improving the concordance of mutation calls pools of 16 libraries were prepared and hybridised to the custom RNA baits between different NGS platforms and to harmonising CNA calls and sequenced using an Illumina HiSeq device in 75 bp paired-end mode with SNP arrays in the ongoing AURORA programme. With recent at a target coverage of 200X. 11, 24 findings on the clinical utility of circulating tumour DNA and its implementation into the AURORA study design, it becomes Copy number aberration profiling using SNP arrays imperative to thoroughly evaluate the technical feasibility of Copy number aberration profiling using the SNP arrays was performed profiling circulating biomarkers to support clinical decision according to the manufacturer’s instructions (Affymetrix). In short, the making. molecular inversion probes (MIP) were incubated with the FFPE extracted DNA at 58°C overnight after an initial denaturation at 95°C for 5 min. Each sample was then split into two aliquots and a gap fill reaction was METHODS performed. Uncircularised MIP and genomic DNA were digested using a Patients and samples cocktail of exonucleases. The remaining circular MIP were then linearised Patients were enroled in four European centres (Institut Jules Bordet, using a cleavage enzyme and amplified by PCR. Following a second round Belgium; Hospital Vall d’Hebron, Spain; Sana Klinikum Offenbach, Germany; of PCR amplification, the 120 bp amplicons were cleaved into two and Dundee Cancer Centre, United Kingdom). The study was approved by fragments with the HaeIII enzyme. The samples were then mixed with the respective ethical committees from the named institutions and was the hybridisation buffer and injected into the arrays where they were performed in accordance with relevant guidelines and procedures. Patients allowed to hybridise at 49°C for 16–18 h. At the end of the hybridisation considered eligible for this study were those (1) with histologically proven period, the arrays were stained and washed using the GeneChip Fluidics distant metastatic or locally recurrent invasive breast cancer (2) with an Station 450 and loaded into the GeneChip Scanner 3000 where array Eastern Cooperative Oncology Group Performance Status equal to or less fluorescence intensity was scanned to generate binary CEL files using the than two, (3) with clinical and laboratory parameters safe for tumour Affymetrix GeneChip Command Console. biopsy, (4) for whom FFPE tumour tissue from a locally recurrent or metastatic lesion and whole blood for research purposes were available. Bioinformatics analyses The samples were centralised in real time at the European Institute of Mutation calling from Ion Torrent targeted gene screen. Sequence reads Oncology (Milan, Italy) where H&E slides were reviewed by a board- from the tumour and matched normal samples were aligned against the certified pathologist for the evaluation of cellularity. At the central human genome reference version hg19/GRCh37 using the Ion Torrent laboratory, IHC for ER, PR, HER2, Ki67 and PTEN were performed using the TMAP aligner with default parameter settings. Mutations were called from ER pharmDx kit (Dako), the HercepTest kit (Dako), the mouse mAb anti h- the resulting BAM files using the Torrent Suite variant caller (Life Ki-67, the clone MIB-1 (Dako) and the rabbit mAb anti h-PTEN, clone 138G6 Technologies) with the default settings of the ‘Somatic High Stringency’ (Cell Signalling) respectively. FISH for ERBB2 was performed using the HER2 pipeline and cross-checked using the NextGENe software (Softgenetics) FISH pharmDx kit (Dako) and scored according to the ASCO/CAP using the ‘Ion Torrent’ predefined pipeline. Germline mutations were guidelines. Wherever applicable, macrodissections to enrich for tumour filtered by subtracting variants found in the matched normal sample from cells were performed. DNA was extracted from the tissue and blood those called in the corresponding tumour sample. The resulting somatic samples using either the Qiagen DNA FFPE tissue kit and the QIAamp mutation calls were further filtered to exclude variants (1) that were not DNeasy blood and tissue kit, respectively, following the manufacturer’s sequenced in both sense with a minimum ratio of 10/90%, (2) with less instructions. The cut-off values for tumour content and DNA quantity were than 100 read depth, and (3) variant allele fractions lower than 10% in the 10% and 400 ng, respectively. DNA concentrations were measured using tumour sample. For an alpha list of target genes (Supplementary Table S5), the Qubit fluorometer (Life Technologies) following which, aliquots were npj Breast Cancer (2017) 23 Published in partnership with the Breast Cancer Research Foundation AURORA—screening patients with advanced breast cancer M Maetens et al. mutations occurring below 10% VAF were accepted. Manual evaluation and genomic mass were obtained using ABSOLUTE and non-rounded was done for indels using IGV. estimates of copy numbers and copy number calls were obtained as described above for SNP arrays. Unless otherwise stated, all parameter settings were kept at default values. Mutation calling from Illumina targeted gene screen. Sequence reads from the tumour and matched normal samples were pre-processed according to the GATK best practices. The raw reads were aligned against the human Statistical analyses genome reference hg19/GRCh37 using the BWA aligner. Duplicate reads Integration of the VAF of mutations indexed by the Ion Torrent NGS were marked using Picard following which the data were filtered to keep platform and CNA profiles from the matched SNP arrays to compute the only properly paired and mapped reads with mapping quality greater than Log-Likelihood and individual CCF of each mutation was done using 60. Sequence reads around potential indels were locally realigned and the ABSOLUTE. Similarly, the sequencing depth required to achieved 99% base quality scores were recalibrated using GATK. Mutations were called 28 29 statistical power to call a given base was computed using ABSOLUTE from the resulting BAM files using MuTect2, SomaticSniper and with an error rate E =2E-03 and an FDR = 5E-07. The frequencies of Strelka in matched tumour/normal mode using the default parameter aberration calls and the percentage accuracy were obtained by boot- settings for each of the aforementioned algorithms. The resulting somatic strapping. The values displayed are mean estimates and confidence mutation calls were further filtered to include only variants (1) occurring in intervals are standard errors. The concordance of mutation calls for regions common to both Ion Torrent and Illumina sequencing panels, (2) substitutions was evaluated using the simple matching coefficient. All having a VAF above 10% in the tumour sample, and (3) having a read correlations were measured using the non-parametric Spearman’s ρ depth greater than 50X. Similar to the Ion Torrent NGS, mutations coefficient and all statistical hypothesis tests were done using the non- occurring in the alpha list (Supplementary Table S5) were accepted if they parametric Wilcoxon’s rank sum test. P-values were two sided and paired occurred with a VAF below 10%. The resulting lists of mutation calls were or unpaired as appropriate. Unless otherwise stated, all computations were considered individually for each mutation caller or combined by voting done in R/Bionconductor. using (1) a simple rule of a mutation indexed by at least one of the three, (2) a majority rule of a mutation called by at least two of the three, and (3) a strict consensus of all three mutation callers. In downstream comparison, Data availability we considered only substitutions on account of the known issues of calling The sequencing and SNP arrays data have been deposited at the European indels with confidence from Ion Torrent NGS data. Genome-Phenome Archive (http://www.ebi.ac.uk/ega/) which is hosted by the European Bioinformatics Institute, under accession number Copy number aberration analysis using SNP arrays. The raw intensity EGAD00001000870. values were normalised to obtained Log ratios, B Allele Frequencies (BAF) and genotyping calls (AA/AB/BB) using Affymetrix Power Tools. We used release NA.33 of the NetAffx library for the reference model and ACKNOWLEDGEMENTS annotation. We computed the Median Absolute Pairwise Deviation and The authors would like to extend their gratitude to the patients who participated in the Median Auto-Correlation from the Log ratios as quality control metrics this study, the BIG HQ, Phuong Dinh and Roberto Salgado for contributing to project and used a threshold of 0.40 and 0.45, respectively, to discard failed arrays. set-up and management and to Delphine Vincent and Samira Majjaj for technical We used two parallel approaches involving (a) allele specific copy number assistance. D.B. is supported by the Belgian Fonds National de la Recherche analysis using heterozygous SNP probes and (b) total copy number Scientifique (F.R.S.-FNRS). This work was supported by the Breast Cancer Research analysis using the full set of 200 K markers and parameters from (a) to Foundation (BCRF). control for the cancer cell fraction and genomic mass. From the BAF and genotyping calls, only informative SNP probes displaying heterozygous genotype (AB) and 0.1 < BAF < 0.9 were kept for analysis at (a). The Log AUTHOR CONTRIBUTIONS ratios and BAF were smoothed using the median absolute deviation and segmented jointly using a multitrack segmentation algorithm from the M.P., Sh.L. and C.S. designed the study. M.M. and D.B. analysed the data, interpreted library copynumber to determine common breakpoints. Estimates for the the results and wrote the manuscript. P.A., A.I., F.R., Y.B., D.F., D.Z. and C.D. were cancer cell fraction and genomic mass were obtained using GAP and involved in data interpretation. P.A., S.L., A.T., J.C., S.S., S.V. and M.O. were involved in compared to the results obtained from ABSOLUTE . Samples with a cancer ethics submission, patient consent and recruitment. G.V. performed the central cell fraction lower than 30% were further excluded. For analysis at (b), the histopathological review of the samples. J.L. and P.C. generated the sequencing data. Log ratios for the same samples analysed at (a) were segmented by All the authors read and approved the final manuscript. penalised least square regression as above and non-rounded estimates of copy numbers were obtained as ADDITIONAL INFORMATION 1 x y ¼ 2ðÞ ψα þ21ðÞ  α21ðÞ  α Supplementary Information accompanies the paper on the npj Breast Cancer website (doi:10.1038/s41523-017-0026-6). where α is the CCF and ψ is the genomic mass, both estimated at (a). c = 0.8 is a constant representing the compression ratio of the array and finally Competing interests: The authors declare that they have no competing financial x is the observed Log ratio of a given segment. The copy numbers were interests. categorised as deletions (−1) if y < ψ−0.5, gains (+1) if y > ψ + 0.5, amplifications (+2) if y > ψ + 2.5, and copy neutral (0) otherwise. Unless Publisher’s note: Springer Nature remains neutral with regard to jurisdictional otherwise stated, all parameter settings were kept at default values and all claims in published maps and institutional affiliations. computations were done using R/Bioconductor. Copy number aberration analysis using targeted NGS. The read counts from aligned and sorted BAM files of the Ion Torrent sequencing were REFERENCES processed using ONCOCNV with default parameter settings to correct for 1. Pereira, B. et al. The somatic mutation profiles of 2433 breast cancers refines their library size, GC content and amplicon length. The pool of normal samples genomic and transcriptomic landscapes. Nat. Commun. 7 Article number: 11479 was used as baseline control to capture the technology specific bias (2016). against which each tumour sample was normalised to obtain an estimate 2. Nik-Zainal, S. et al. Landscape of somatic mutations in 560 breast cancer whole- of Log ratio. The Log ratios were segmented using the circular binary 2 2 genome sequences. Nature 534,47–54 (2016). segmentation and CNA calls were obtained using the default clustering 3. Flaherty, K. T. et al. Inhibition of mutated, activated BRAF in metastatic mela- approached implemented in ONCOCNV. The read counts from the aligned noma. N. Engl. J. Med. 363, 809–819 (2010). and sorted BAM files of the Illumina sequencing were processed using 4. Schwaederle, M. et al. Association of biomarker-based treatment strategies with cnvkit to obtain an estimate of Log ratio for on and off target regions. response rates and progression-free survival in refractory malignant neoplasms: a The pool of normal samples was used as baseline control against which meta-analysis. JAMA Oncol. 2, 1452–1459 (2016). each tumour sample was normalised. The Log ratios were smoothed by 5. Roychowdhury, S. et al. Personalized oncology through integrative high- median absolute deviation and segmented by penalised least square throughput sequencing: a pilot study. Sci. Transl. Med. 3, 111ra121 (2011). regression using the library copynumber. Estimates of cancer cell fraction Published in partnership with the Breast Cancer Research Foundation npj Breast Cancer (2017) 23 AURORA—screening patients with advanced breast cancer M Maetens et al. 6. Ong, M. et al. Validation and utilisation of high-coverage next-generation 24. Garcia-Murillas, I. et al. Mutation tracking in circulating tumor DNA predicts sequencing to deliver the pharmacological audit trail. Br. J. Cancer 111, 828–836 relapse in early breast cancer. Sci. Transl. Med. 7, 302ra133 (2015). (2014). 25. Thorvaldsdóttir, H. et al. Integrative genomics viewer (IGV): high-performance 7. Boland, G. M. et al. Clinical next generation sequencing to identify actionable genomics data visualization and exploration. Brief. Bioinform. 14, 178–192 aberrations in a phase I program. Oncotarget 6, 20099–20110 (2015). (2013). 8. Ding, L. et al. Genome remodelling in a basal-like breast cancer metastasis and 26. Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows-Wheeler xenograft. Nature 464, 999–1005 (2010). transform. Bioinformatics 25, 1754–1760 (2009). 9. Navin, N. et al. Tumour evolution inferred by single-cell sequencing. Nature 472, 27. McKenna, A. et al. The genome analysis toolkit: a MapReduce framework for 90–94 (2011). analyzing next-generation DNA sequencing data. Genome Res. 20, 1297–1303 10. Gerlinger, M. et al. Intratumor heterogeneity and branched evolution revealed by (2010). multiregion sequencing. N. Engl. J. Med. 366, 883–892 (2012). 28. Cibulskis, K. et al. Sensitive detection of somatic point mutations in impure and 11. Murtaza, M. et al. Multifocal clonal evolution characterized using circulating heterogeneous cancer samples. Nat. Biotechnol. 31, 213–219 (2013). tumour DNA in a case of metastatic breast cancer. Nat. Commun. 6 Article 29. Larson, D. E. et al. SomaticSniper: identification of somatic point mutations in number: 8760 (2015). whole genome sequencing data. Bioinformatics 28, 311–317 (2012). 12. Yates, L. R. et al. Subclonal diversification of primary breast cancer revealed by 30. Saunders, C. T. et al. Strelka: accurate somatic small-variant calling multiregion sequencing. Nat. Med. 21, 751–759 (2015). from sequenced tumor–normal sample pairs. Bioinformatics 28, 1811–1817 13. de Duenas, E. M. et al. Prospective evaluation of the conversion rate in the (2012). receptor status between primary breast cancer and metastasis: results from 31. Nilsen, G. et al. Copynumber: Efficient algorithms for single- and multi-track copy the GEICAM 2009-03 ConvertHER study. Breast Cancer Res. Treat. 143, 507–515 number segmentation. BMC Genomics 13, 591 (2012). (2014). 32. Popova, T. et al. Genome alteration print (GAP): a tool to visualize and mine 14. Misyura, M. et al. Comparison of next-generation sequencing panels and plat- complex cancer genomic profiles obtained by SNP arrays. Genome Biol. 10, R128 forms for detection and verification of somatic tumor variants for clinical diag- (2009). nostics. J. Mol. Diagn. 18, 842–850 (2016). 33. Carter, S. L. et al. Absolute quantification of somatic DNA alterations in human 15. Kuderer, N. M. et al. Comparison of 2 commercially available next-generation cancer. Nat. Biotechnol. 30, 413–421 (2012). sequencing platforms in oncology. JAMA Oncol. Published online December 15, 34. Boeva, V. et al. Multi-factor data normalization enables the detection of copy 2016. doi:10.1001 (2016). number aberrations in amplicon sequencing data. Bioinformatics 30, 3443–3450 16. Zardavas, D. et al. The AURORA initiative for metastatic breast cancer. Br. J. Cancer (2014). 111, 1881–1887 (2014). 35. Olshen, A. B. et al. Circular binary segmentation for the analysis of array-based 17. Thatcher, N. et al. Necitumumab plus gemcitabine and cisplatin versus gemci- DNA copy number data. Biostatistics 5, 557–572 (2004). tabine and cisplatin alone as first-line therapy in patients with stage IV squamous non-small-cell lung cancer (SQUIRE): an open-label, randomised, controlled phase 3 trial. Lancet Oncol. 16, 763–774 (2015). Open Access This article is licensed under a Creative Commons 18. Park, H. S. et al. High EGFR gene copy number predicts poor outcome in triple- Attribution 4.0 International License, which permits use, sharing, negative breast cancer. Mod. Pathol. 27, 1212–1222 (2014). adaptation, distribution and reproduction in any medium or format, as long as you give 19. Zardavas, D. et al. Emerging targeted agents in metastatic breast cancer. Nat. Rev. appropriate credit to the original author(s) and the source, provide a link to the Creative Clin. Oncol. 10, 191–210 (2013). Commons license, and indicate if changes were made. The images or other third party 20. Zardavas, D. et al. Personalized therapy for breast cancer: a dream or a reality? material in this article are included in the article’s Creative Commons license, unless Future Oncol. 9, 1105–1119 (2013). indicated otherwise in a credit line to the material. If material is not included in the 21. André, F. et al. Comparative genomic hybridisation array and DNA sequencing to article’s Creative Commons license and your intended use is not permitted by statutory direct treatment of metastatic breast cancer: a multicentre, prospective trial regulation or exceeds the permitted use, you will need to obtain permission directly (SAFIR01/UNICANCER). Lancet Oncol. 15, 267–274 (2014). from the copyright holder. To view a copy of this license, visit http://creativecommons. 22. Talevich, E. et al. CNVkit: genome-wide copy number detection and visualization org/licenses/by/4.0/. from targeted DNA sequencing. PLoS Comput. Biol. 12, e1004873 (2016). 23. Kuilman, T. et al. CopywriteR: DNA copy number detection from off-target © The Author(s) 2017 sequence data. Genome. Biol. 16, 49 (2015). npj Breast Cancer (2017) 23 Published in partnership with the Breast Cancer Research Foundation http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png npj Breast Cancer Springer Journals

The AURORA pilot study for molecular screening of patients with advanced breast cancer–a study of the breast international group

Download PDF
10 pages

Loading next page...
 
/lp/springer-journals/the-aurora-pilot-study-for-molecular-screening-of-patients-with-bg26S5l4cn

References (41)

Publisher
Springer Journals
Copyright
Copyright © 2017 by The Author(s)
Subject
Biomedicine; Biomedicine, general; Cancer Research; Oncology; Human Genetics; Cell Biology
eISSN
2374-4677
DOI
10.1038/s41523-017-0026-6
Publisher site
See Article on Publisher Site

Abstract

www.nature.com/npjbcancer ARTICLE OPEN The AURORA pilot study for molecular screening of patients with advanced breast cancer–a study of the breast international group 1 1 2 3 4 5 6 Marion Maetens , David Brown , Alexandre Irrthum , Philippe Aftimos , Giuseppe Viale , Sibylle Loibl , Jean-François Laes , 7 8,9 10,11 5 9 10,11 1 Peter J. Campbell , Alastair Thompson , Javier Cortes , Sabine Seiler , Sara Vinnicombe , Mafalda Oliveira , Françoise Rothé , 1 2 2 1 3 12 1 Yacine Bareche , Debora Fumagalli , Dimitrios Zardavas , Christine Desmedt , Martine Piccart , Sherene Loi and Christos Sotiriou Several studies have demonstrated the feasibility of molecular screening of tumour samples for matching patients with cancer to targeted therapies. However, most of them have been carried out at institutional or national level. Herein, we report on the pilot phase of AURORA (NCT02102165), a European multinational collaborative molecular screening initiative for advanced breast cancer patients. Forty-one patients were prospectively enroled at four participating centres across Europe. Metastatic tumours were biopsied and profiled using an Ion Torrent sequencing platform at a central facility. Sequencing results were obtained for 63% of the patients in real-time with variable turnaround time stemming from delays between patient consent and biopsy. At least one clinically actionable mutation was identified in 73% of patients. We used the Illumina sequencing technology for orthogonal validation and achieved an average of 66% concordance of substitution calls per patient. Additionally, copy number aberrations inferred from the Ion Torrent sequencing were compared to single nucleotide polymorphism arrays and found to be 59% concordant on average. Although this study demonstrates that powerful next generation genomic techniques are logistically ready for international molecular screening programs in routine clinical settings, technical challenges remain to be addressed in order to ensure the accuracy and clinical utility of the genomic data. npj Breast Cancer (2017) 3:23 ; doi:10.1038/s41523-017-0026-6 INTRODUCTION The Breast International Group launched AURORA–Aiming to Understand the Molecular Aberrations in Metastatic Breast Cancer, Several efforts have advanced our understanding of the altera- 1, 2 a pan-European molecular screening programme whose main tions characterizing cancer genomes. Coupled to recent goal is to deepen our knowledge of the genomic landscape of successes of targeted therapies in patients with molecularly 3, 4 advanced breast cancer. Herein, we report on the pilot phase of profiled tumours and the decreasing costs of massively parallel this study whereby the primary objective was to investigate the sequencing, this has motivated several studies, albeit of limited feasibility with four European recruitment sites and central size, to investigate the implementation of personalised molecular 5–7 pathological and sequencing facilities. Secondary aims were to screening in the clinical settings. assess the concordance of somatic mutations between two Most of these studies were focused on primary tumours and targeted next generation sequencing (NGS) platforms and of despite growing evidence that distant metastases may harbour somatic copy number aberrations (CNA) obtained from NGS and additional molecular alterations absent from their matched 8–12 single nucleotide polymorphism (SNP) arrays. primaries, genomic information about metastatic disease remains limited. Even though the clinical relevance of many of these alterations remains to be established, it is increasingly RESULTS recognised that molecular profiling of advanced disease could Patient recruitment and logistics help elucidate the biological underpinnings of phenomena such as distant recurrence and the emergence of de novo resistance to A total of 41 patients provided informed consent and were therapy. Lastly, in order to find applications in routine clinical enroled in this pilot study between February 2013 and September practice, it is essential to assess the reliability and robustness of 2014. Fig. 1a and b illustrates the study design and Table 1 the chosen sequencing platform using orthogonal sequencing summarises the clinical and pathological characteristics of the 14, 15 strategies. patients. Formalin fixed paraffin embedded (FFPE) biopsies of 1 2 J.-C. Heuson Breast Cancer Translational Research Laboratory, Institut Jules Bordet, Université Libre de Bruxelles, Brussels, Belgium; Breast International Group, Brussels, Belgium; 3 4 5 Department of Medical Oncology, Institut Jules Bordet, Université Libre de Bruxelles, Brussels, Belgium; European Institute of Oncology, University of Milan, Milan, Italy; Sana 6 7 Klinikum, Offenbach, Germany and German Breast Group, Neu-Isenburg, Germany; OncoDNA, Gosselies, Belgium; Cancer Genome Project, Wellcome Trust Sanger Institute, 8 9 Hinxton, UK; Department of Breast Surgical Oncology, University of Texas MD Anderson Cancer Center, Houston, USA; Dundee Cancer Centre, University of Dundee, Dundee, 10 11 12 UK; Ramon y Cajal University Hospital, Madrid, Spain; Vall d´Hebron Institute of Oncology, Barcelona, Spain and Division of Clinical Medicine and Research, Peter MacCallum Cancer Centre, Melbourne, Victoria, Australia Correspondence: Marion Maetens (marion.maetens@bordet.be) Marion Maetens, David Brown, Sherene Loi and Christos Sotiriou contributed equally to this work. Received: 24 February 2017 Revised: 4 May 2017 Accepted: 1 June 2017 Published in partnership with the Breast Cancer Research Foundation AURORA—screening patients with advanced breast cancer M Maetens et al. Fig. 1 Logistics and feasibility of the study. a logistic workflow of the study, b inclusion criteria and number of patients with successful results, c anatomical distribution of biopsied lesions with breakdown by recruiting centre, d global distribution of patients by recruiting centre, e turnaround time for each patient with breakdown by recruiting centre, f median global turnaround time by recruiting centre, and g median delivery turnaround time by recruiting centre. In a, the Illumina sequencing and Affymetrix OncoScan SNP arrays were done as batch processes. 19 patients were sequenced by Illumina targeted NGS and 18 patients were genotyped using the Affymetrix SNP arrays. However, these are not fully overlapping subsets and only 14 (54%) patients had the full set of all three data types. In a and b, the Ion Torrent sequencing results were obtained in real-time and in (e), the darker shades indicate delivery turnaround time (TAT) whilst the lighter shades indicate global turnaround time. The colour codes for the recruiting centres in c to g are indicated in (d) metastatic lesions and whole blood were prospectively collected Mutation detection from the Ion Torrent OncoDEEP clinical cancer panel and CNA using SNP arrays from 35 (85%) of them. 20 (57%) patients had ER + metastatic disease, 7 (20%) were HER2 + and 8 (23%) were triple negative Quality control metrics for the sequencing data are provided in breast cancer. Fig. 1c and d represent the distribution of the Supplementary Figure S1. The target regions were covered on different anatomical sites and the breakdown of patients’ average at 1758X (494–3075X) sequencing depth. Non- recruitment by participating centre. Following central pathological synonymous somatic mutations were called from the OncoDEEP review, a tumour content below 10% was recorded for 8 (23%) clinical cancer panel in exons covered by at least 100X sequencing patients whilst the median cellularity was 50% (range 10–85). One depth and a fixed threshold of 10% variant allele fraction (VAF). In additional sample had an insufficient amount of extracted DNA total, 128 unique genes harboured at least one mutation whilst the median yield was 3.6 µg (range 0.075–59.2). Overall, representing one third of the panel (Fig. 2a). The median number real-time Ion Torrent sequencing results were obtained for 26 of mutations indexed per patient was 6 (range 0–35) and the overall mutation detection rate i.e., percentage of patients where (74%) of the 35 patients (Fig. 1b). The median global turnaround time (TAT) from patient biopsy to sequencing results was 51 at least one mutation could be indexed, was 96%. Only 8 (4.15%) of the 193 mutations had previously been described in release working days (range 16–146) whilst the median delivery TAT v.76 of COSMIC in any type of cancer whilst 178 (92.23%) from sample reception at the sequencing facility to final NGS report was 9 working days (range 5–17) (Fig. 1e). There were no mutations had never been reported in the literature. The most frequently mutated genes were PIK3CA (50%), TP53 (31%), SYNE1 significant differences in TAT between the recruiting centres (19%), and NF2 (15%). (Fig. 1f and g). npj Breast Cancer (2017) 23 Published in partnership with the Breast Cancer Research Foundation AURORA—screening patients with advanced breast cancer M Maetens et al. patients harboured at least one actionable mutation, such as Table 1. Clinical and pathological characteristics of patients and PIK3CAp.H1047R or ESR1p.Y537N, whilst 8 (31%) had only one biopsies (n = 35) such mutation and six (23%) had at least one biologically relevant mutation and none that were clinically actionable (Supplementary Characteristics Number of patients (%) Figure S5). Comparatively, 16 of 18 (89%) patients for whom the Age (years) SNP arrays were available harboured at least one actionable CNA, 17, 18 such as EGFR amplification, whilst 2 (11%) had only one such Median 56 aberration and an equal number had at least one biologically Range 23–73 relevant CNA and none that were clinically actionable. Using the ECOG performance status combined information from the SNP arrays and the targeted 0 16 (45) sequencing, and focusing on the subset of strictly actionable 1 17 (49) alterations, 23 of 26 (88%) patients had at least one clinically actionable alteration whilst only 3 (12%) had only one such 2 2 (6) alteration. Overall, of the initial 7 patients without actionable Breast cancer subtype mutations, four were found to harbour at least one actionable ER+ 20 (57) CNA. HER2+ 7 (20) ER-/HER2- 8 (23) Orthogonal cross-testing of substitution calls using Illumina NGS Number of prior lines of therapy The substitutions obtained from the Ion Torrent sequencing were 0 6 (17) compared to data generated from the Illumina NGS platform. The same DNA samples were sequenced and substitutions were called 1 5 (14) in overlapping regions using similar filtering criteria and a 2 4 (11) combination of three established mutation callers. Patient 3 6 (17) matched data were available for 19 (73%) patients and all three >3 14 (41) mutation callers were mostly consistent (Fig. 3a). All but one Number of metastatic sites actionable substitution, NOTCH1p.D2082E in cluster 1 (see below), were concordant. 1 8 (23) The substitutions were categorised into three clusters and the 2 10 (28) error rates of each mutation caller and several combinations 3 8 (23) thereof were benchmarked using this framework. Substitutions in >3 9 (26) clusters 1 and 3 are exclusive to one of the sequencing platforms and have 0% VAF in the alternate data, whereas substitutions in ECOG; Eastern Cooperative Oncology Group, ER; oestrogen receptor cluster 2 are those found by either or both NGS platforms and evaluated by IHC, HER2; human epidermal growth factor receptor 2 evaluated by IHC and FISH. The breast cancer subtypes are based on the have non-zero VAF in both sequencing data. The definite calls characteristics of the metastatic lesions were made for each substitution based on a majority vote of any two of the three mutation callers (Fig. 3b and c). This choice was guided by a low global Illumina specific false negative error rate We evaluated two established methods to call CNA from SNP on cluster 2 substitutions whilst maintaining a relatively low global arrays and obtained similar results (Fig. 2c and d) thereby attesting Illumina specific false positive error rate on cluster 2 and 3 to the robustness of SNP arrays for the estimation of CNA. All the substitutions. Using this approach, a global concordance rate of patients presented with at least one CNA and the CNA frequencies 40.9% was achieved. were comparable to those reported in the literature (Fig. 2d–g). Substitutions in cluster 1 cannot be reconciled. There was a For instance, 13 (72%) patients harboured a gain or an statistically significant difference in coverage between the Ion amplification of chromosome 8q where the MYC oncogene Torrent and Illumina NGS data for these substitutions (Fig. 3d). resides whilst 9 (50%) had a deletion of 17p where TP53 is located. However, all but one of the substitutions were covered in excess Two patients presented with a higher than expected number of of 100X in the Illumina NGS data and the residual coverage, which mutations. The integration of the sequencing and CNA data is the difference between the observed sequencing depth and the obtained from these two outlier patients is shown in Supplemen- expected value at 99% statistical power to detect a mutation for tary Figure S2. In general, the distributions of VAF and cancer cell given CCF and copy numbers, were positive for all except two fractions (CCF) were high and tightly clustered, indicative of substitutions (Fig. 3e). Similarly, two substitutions were called in genuine somatic mutations. For the whole cohort, ERBB2 cluster 3 which constitute Illumina specific false positives and were amplification status was assessed centrally using FISH whilst PTEN covered at 239X and 1668X using the Ion Torrent NGS platform. In loss was evaluated by immunohistochemistry (IHC). These single total, only 2.7% of the substitutions in cluster 2 were Illumina assay data were compared with CNA obtained from the SNP arrays specific false positives using the Ion Torrent platform as standard and the results are shown in Supplementary Figures S3 and S4.No whilst 5.41% were Illumina specific false negatives. There were no false negative calls were recorded for ERBB2 amplification status statistically significant differences in sequencing coverage comparing SNP arrays to FISH data. However, 7 of 18 patients between these groups of substitutions and the true positives (39%) effectively harboured supernumerary copies of ERBB2 (n ≤ 4 called by both platforms (Fig. 3f and g). Furthermore, similar to copies) which were missed by FISH assays owing to centromere 17 cluster 1 substitutions, all the Illumina specific false negatives had co-amplification or low level 17q polysomy. There was no concrete positive residual coverage (Fig. 3e). For comparison, the sequen- evidence of a correspondence between PTEN IHC staining results cing coverage and the percentage of non-reference bases at these and the corresponding copy number level as determined by SNP particular loci in the normal matched samples are shown in array. Supplementary Figure S6. Actionable alterations from targeted NGS and SNP arrays Comparison of CNA from targeted NGS and SNP arrays Supplementary Tables S1 and S2 provide the definitions of The CNA obtained from the Ion Torrent sequencing were also actionable and biologically relevant alterations. 19 of 26 (73%) compared to the profiles obtained from SNP arrays and Illumina Published in partnership with the Breast Cancer Research Foundation npj Breast Cancer (2017) 23 AURORA—screening patients with advanced breast cancer M Maetens et al. Fig. 2 Mutation detection from the OncoDEEP clinical cancer panel and CNA from SNP arrays. a genes for which at least one mutation was indexed across the 26 patients, b and c comparison of cancer cell fraction and genomic mass respectively obtained from two widely used algorithms for estimating CNA from SNP arrays. In a, asterisk indicates potentially hypermutated patients. In b and c, each dot represents a sample. In d–g, each dot represents a gene from the list of clinically actionable or biologically relevant targets and is the mean of 100 bootstrap replicates such that for each replicate, the external cohorts are matched for ER and HER2 status. The size of each dot is proportional to the standard error of the mean frequency estimate NGS. Fig. 4a–c show the Log ratios from the SNP array and both bimodal for the Ion Torrent platform with 5 (29%) samples NGS platforms for a case patient. The segmented Log ratios from displaying a poor correlation of ρ < 0.5 (Fig. 5a). The median the SNP array and Illumina sequencing clustered, as expected, into correlation coefficients were ρ = 0.615 and ρ = 0.745 for the Ion canonical copy number genotypes whilst the segmented Log Torrent and Illumina NGS platforms respectively. ratios obtained from the Ion Torrent NGS platform were spread In order to further evaluate the ability of both NGS platforms to between those canonical genotypes leading to relatively poor call CNA, the segmented Log ratios were grouped into three correlation in Log ratio data space (Fig. 4d and e). Nonetheless, categories i.e. deletion (−1), copy neutral (0), gain/amplification (1) this approach was applied to each copy number platform for 14 and further compared using the accuracy which is the sum of (54%) patients for whom all three data types were available and concordant calls relative to the total number of aberrations. Fig. 5b the segmented Log ratios were compared genome-wide using shows the distribution of accuracy values evaluated genome-wide the Spearman’s correlation. The distribution of correlation for each patient. The average accuracy of CNA call was 59.1% for coefficients comparing the SNP array to the NGS data was the Ion Torrent and 74.0% for the Illumina NGS platforms. Both npj Breast Cancer (2017) 23 Published in partnership with the Breast Cancer Research Foundation AURORA—screening patients with advanced breast cancer M Maetens et al. Fig. 3 Orthogonal cross-testing of substitution calls using Illumina NGS. a distribution of % concordance for single nucleotide substitutions indexed from the Ion Torrent and Illumina NGS platforms using different mutation callers, b % error rate of the different mutation callers over three substitution clusters compared to the Ion Torrent sequencing, c correlation of VAF for substitutions indexed from the Ion Torrent and Illumina NGS using a majority rule of any two of three mutation callers, d comparison of sequencing coverage between the Ion Torrent and Illumina NGS for substitutions in cluster 1, e empirical cumulative distribution of theoretical residual coverage from the Illumina sequencing for substitutions in cluster 1 and cluster 2 false negatives, f comparison of sequencing coverage from the Ion Torrent NGS for substitutions in cluster 2 false positives, and g comparison of sequencing coverage from the Illumina NGS for substitutions in cluster 2 false negatives. In a, the different mutation callers and any combination thereof are colour coded and indicated at the bottom. The leftmost panel gives the number of substitutions called. In b–g, the cluster numbers are relative to (c). Substitutions in clusters 1 and 3 are exclusive to one of the sequencing platforms and have 0% VAF in the alternate data. Substitutions in cluster 2 are those found by either or both NGS platforms and have non-zero VAF in both sequencing data. False negatives in cluster 2 are substitutions indexed by Ion Torrent NGS only whilst false positives are substitutions indexed by Illumina NGS only using a given mutation caller or any combination thereof. The size of each dots is proportional to the difference in coverage between the two sequencing platforms. In e, the residual coverage is obtained by subtracting the theoretical coverage required to achieve 99% power for indexing a substitution given one mutated copy out of n total copies from the observed value of sequencing depth. Only two substitutions in cluster 1 failed the criteria of positive residual for detection and are non-callable loci distributions were bimodal with 6 (35%) and 5 (29%) patients 86.8–93.3) for MYC, 63.8% (95% CI: 61.4–66.2) for CCND1 and showing low values of α < 0.5 and α < 0.6 for the Ion Torrent and 57.4% (95% CI: 52.3–62.5) for ERBB2 whilst the corresponding Illumina NGS platforms, respectively. Lastly, the ability of both values for the same genes were 29.5% (95% CI: 27.5–31.8), 40.2% sequencing platforms to call CNA in particular genes of interest (95% CI: 37.7–42.7) and 28.3% (95% CI: 25.6–30.9), respectively, was evaluated on 15 clinically actionable or biologically relevant using the Ion Torrent sequencing platform. There was a significant genes. For each gene, the CNA calls were pooled and the cohort- negative correlation between the accuracy and the number of wise concordance measured as the accuracy of concordant calls. aberrations per patient for the Ion Torrent data (Fig. 5d) but not Fig. 5c contrasts the distribution of accuracy values observed for the Illumina NGS platform (Fig. 5e). There were no associations across these 15 genes for both NGS platforms. We observed between the accuracy measured cohort-wise and the frequency of relatively high concordance rates using the Illumina NGS platform aberrations affecting the 15 genes of interest for either of the two for genes frequently aberrant in breast cancers e.g., 90.1% (95% CI: NGS platforms (Fig. 5f and g). Published in partnership with the Breast Cancer Research Foundation npj Breast Cancer (2017) 23 AURORA—screening patients with advanced breast cancer M Maetens et al. Fig. 4 Genome-wide copy number profiles inferred from SNP arrays and targeted NGS data. Genome-wide Log ratio profiles of patient IJB0021 obtained using a SNP array b Ion Torrent and c Illumina targeted NGS data. The solid vertical lines represent chromosome boundaries whilst the dashed horizontal lines indicate canonical copy numbers inferred from the cancer cell fraction and ploidy of the sample, both estimated from the SNP array. The solid horizontal lines represent the segmented Log ratios. Correlation of segmented Log ratios between d 2 2 the SNP array and the Ion Torrent platform, e the SNP arrays and the Illumina platform, and f the two NGS platforms for the same case patient. In a–c, the loci are sorted according to their coordinate on the human genome reference hg19/GRCh37. For ease of representation, they are plotted by indices DISCUSSION communication). 73% of the patients harboured one or more actionable mutations and this increased to 88% when considering The growing number of targeted anticancer therapies either the underlying CNA. These numbers are encouraging on account approved or under clinical development has led to a wide interest 19, 20 of the availability of alterations that can be theoretically targeted in personalised treatment approaches in clinical practice. despite limited clinical trial based evidence of benefit to patient However, many of these cancer molecular screening initiatives survival from targeted therapy in breast cancer. Yet, the decision have had limitations. For instance, several of these screening to prioritise a particular molecular target for trial allocation given programs used different sequencing techniques and very often several such alterations within the same patient constitutes a lacked orthogonal validation. Together with varying concepts of considerable challenge for which specific algorithms and com- what constitutes an actionable mutation, this makes the results bined expertise will be required in the future. difficult to compare. The current study is a pilot phase undertaken Regarding the secondary endpoint and considering clinically to evaluate the feasibility of AURORA, a pan-European molecular actionable mutations, our results are reassuring. All but one screening programme for advanced breast cancer patients. substitution was positive using both the Ion Torrent and Illumina The success rate for the primary endpoint of 63% compares 5–7, 21 sequencing platforms. Furthermore, from a pragmatic viewpoint, 5 favourably with existing literature. The median global TAT of (26%) sample pairs showed complete agreement. However, no 51 working days was due to unforeseen delays either between single mutation caller was able to reproduce globally with 100% patient consent and surgical biopsy or between collection of concordance the results obtained in real-time from the Ion Torrent biological samples and shipment to the central pathology laboratory. In part, due to the above findings, these parameters NGS platform. The Illumina specific false negative and false are now rigorously monitored in the parent programme where the positive rates over somatic substitutions with non-zero VAF in global TAT does not exceed 20 working days (personal both data types can be reasonably contained whilst substitutions npj Breast Cancer (2017) 23 Published in partnership with the Breast Cancer Research Foundation AURORA—screening patients with advanced breast cancer M Maetens et al. Fig. 5 Comparison of copy number profiles from SNP arrays and targeted sequencing data. a distribution of Spearman’s correlation coefficient comparing the segmented Log ratios from the SNP arrays and the NGS data across patients with all three data types. The segmented Log 2 2 ratios were further categorised into copy number aberration calls i.e., −1, 0 and +1 and compared patient wise between the SNP arrays and the NGS data. The resulting distributions of accuracy values are depicted in b. For a small set of 15 clinically actionable or biologically relevant genes, the accuracy values comparing SNP arrays and NGS data were computed using all available samples and displayed individually for each gene in c. d, e correlation of accuracy values measured genome-wide for each patient as in b vs. the number of aberrations as determined by the corresponding NGS platform. f, g correlation of accuracy for each of the 15 clinically actionable or biologically relevant genes vs. the frequency of aberrations for the same genes measured using the SNP array. In c, each value of accuracy was generated by 100 bootstrap replicates. The values displayed represent the mean of the replicates and the error bars represent the standard error of this estimate private to either sequencing technology despite adequate cover- sensitivity to a given drug. In the broader context of AURORA, all age in both data types were more frequent in the Ion Torrent data. the results are reported to investigators unaltered. However, our Thus, from a technical perspective, our results are somewhat analysis shows the major limitations of this approach outside sobering and whilst the higher background error rate of the Ion known mutation hotspots and in anticipation, the AURORA study Torrent NGS technology could be a contributing factor, the small design allows biobanking of frozen and residual FFPE material for number of samples and mutations assessed here do not allow for later research purposes using alternative high-throughput tech- a full exploration of the possible causes of these discrepancies. nologies such as whole genome or exome sequencing. Massively parallel sequencing using targeted gene panels has To the best of our knowledge, no other studies have compared emerged as a technology that can be applied within a clinically prospectively CNA obtained from targeted sequencing data to meaningful TAT to identify known biomarkers of resistance or SNP arrays in a clinical setting. Our results show that the gold Published in partnership with the Breast Cancer Research Foundation npj Breast Cancer (2017) 23 AURORA—screening patients with advanced breast cancer M Maetens et al. standard FISH assay for ERBB2 is fully concordant with SNP arrays. shipped to (1) OncoDNA (Gosselies, Belgium) for sequencing using Ion Torrent NGS technology, (2) the Wellcome Trust Sanger Institute (Hinxton, However, the Illumina NGS platform outperforms the Ion Torrent UK) for sequencing using Illumina NGS technology, and (3) the J.-C. Heuson technology when compared to the same SNP arrays. This is not Breast Cancer Translational Research Laboratory (Brussels, Belgium) where unexpected since the Illumina NGS panel accommodates inter- the Affymetrix OncoScan FFPE Express arrays were performed. The Illumina genic targets which, combined with the fact that hybrid capture NGS and Affymetrix SNP arrays were performed as batch processes whilst methods for sequencing library preparation often carry over off- the sequencing and delivery of results from OncoDNA were monitored in target reads, increases the effective number of regions where real time. 22, 23 copy numbers can be evaluated. Several algorithms take advantage of this, benefiting downstream CNA calling and Targeted gene screen using Ion Torrent NGS translating into higher concordance with the denser SNP arrays. Somatic mutations were assessed using the OncoDEEP clinical cancer The combination of assays, such as mutation profiling and panel which is a validated AmpliSeq design panel targeting the exonic assessment of CNA, into one cost effective package is an attractive regions of 409 cancer related genes (Supplementary Table S3). The same idea. Together with a low requirement of input DNA, as is often protocol was applied to DNA extracted from FFPE tumour and whole blood obtainable from clinical samples, makes for a compelling normal matched samples. Briefly, the targeted sequencing libraries were argument in favour of the Ion Torrent NGS platform. However, generated using the Ion AmpliSeq library kit 2.0 according to the given the poor agreement with solid benchmarks such as SNP manufacturer’s instructions (Life Technologies) using 80 ng of genomic DNA. The primers used for amplification were partially digested by Pfu arrays, the results presented here do not support, at least for now, restriction enzyme and the digestion products were ligated to barcoded the targeted Ion Torrent NGS platform as a truly multipurpose adaptors and purified using Ampure Beads. The purified products were assay in a clinical setting. Within AURORA, central pathology amplified for five cycles and purified once more using Ampure Beads. The results reported to investigators in real-time include ERBB2 IHC as quality of the libraries was assessed using a qPCR following which 10 pM of well as FISH when applicable. This is clinically meaningful since each library underwent emulsion PCR using an IonChef system. The chips ERBB2 gene amplification remains, to date, the sole approved CNA were loaded on an Ion PGM and were sequenced at a target coverage of biomarker in breast cancer. Sub-studies undertaken in the 500X. framework of the main AURORA programme include real-time SNP arrays and retrospective whole genome sequencing to Targeted gene screen using Illumina NGS alleviate the technical limitations of identifying CNA using the The exonic regions of 371 cancer related genes (Supplementary Table S4) Ion Torrent technology and will allow the study of the clonal were enriched using a custom design of SureSelect RNA baits following the evolution of breast cancer. manufacturer’s instructions (Agilent). The same protocol was applied to Overall, our study contrasts the benefits and pitfalls of DNA extracted from tumour and normal matched samples. Briefly, 500 ng personalised molecular screening for patients with advanced of genomic DNA was fragmented to an average insert size of 145 bp cancer involving the combination of multiple high-throughput (75–300 bp) and subjected to Illumina DNA sequencing library preparation using the Bravo automated liquid handling platform. Individual samples genomic techniques. In view of these results, greater effort is were indexed using a unique DNA barcode via six cycles of PCR. Equimolar being devoted to improving the concordance of mutation calls pools of 16 libraries were prepared and hybridised to the custom RNA baits between different NGS platforms and to harmonising CNA calls and sequenced using an Illumina HiSeq device in 75 bp paired-end mode with SNP arrays in the ongoing AURORA programme. With recent at a target coverage of 200X. 11, 24 findings on the clinical utility of circulating tumour DNA and its implementation into the AURORA study design, it becomes Copy number aberration profiling using SNP arrays imperative to thoroughly evaluate the technical feasibility of Copy number aberration profiling using the SNP arrays was performed profiling circulating biomarkers to support clinical decision according to the manufacturer’s instructions (Affymetrix). In short, the making. molecular inversion probes (MIP) were incubated with the FFPE extracted DNA at 58°C overnight after an initial denaturation at 95°C for 5 min. Each sample was then split into two aliquots and a gap fill reaction was METHODS performed. Uncircularised MIP and genomic DNA were digested using a Patients and samples cocktail of exonucleases. The remaining circular MIP were then linearised Patients were enroled in four European centres (Institut Jules Bordet, using a cleavage enzyme and amplified by PCR. Following a second round Belgium; Hospital Vall d’Hebron, Spain; Sana Klinikum Offenbach, Germany; of PCR amplification, the 120 bp amplicons were cleaved into two and Dundee Cancer Centre, United Kingdom). The study was approved by fragments with the HaeIII enzyme. The samples were then mixed with the respective ethical committees from the named institutions and was the hybridisation buffer and injected into the arrays where they were performed in accordance with relevant guidelines and procedures. Patients allowed to hybridise at 49°C for 16–18 h. At the end of the hybridisation considered eligible for this study were those (1) with histologically proven period, the arrays were stained and washed using the GeneChip Fluidics distant metastatic or locally recurrent invasive breast cancer (2) with an Station 450 and loaded into the GeneChip Scanner 3000 where array Eastern Cooperative Oncology Group Performance Status equal to or less fluorescence intensity was scanned to generate binary CEL files using the than two, (3) with clinical and laboratory parameters safe for tumour Affymetrix GeneChip Command Console. biopsy, (4) for whom FFPE tumour tissue from a locally recurrent or metastatic lesion and whole blood for research purposes were available. Bioinformatics analyses The samples were centralised in real time at the European Institute of Mutation calling from Ion Torrent targeted gene screen. Sequence reads Oncology (Milan, Italy) where H&E slides were reviewed by a board- from the tumour and matched normal samples were aligned against the certified pathologist for the evaluation of cellularity. At the central human genome reference version hg19/GRCh37 using the Ion Torrent laboratory, IHC for ER, PR, HER2, Ki67 and PTEN were performed using the TMAP aligner with default parameter settings. Mutations were called from ER pharmDx kit (Dako), the HercepTest kit (Dako), the mouse mAb anti h- the resulting BAM files using the Torrent Suite variant caller (Life Ki-67, the clone MIB-1 (Dako) and the rabbit mAb anti h-PTEN, clone 138G6 Technologies) with the default settings of the ‘Somatic High Stringency’ (Cell Signalling) respectively. FISH for ERBB2 was performed using the HER2 pipeline and cross-checked using the NextGENe software (Softgenetics) FISH pharmDx kit (Dako) and scored according to the ASCO/CAP using the ‘Ion Torrent’ predefined pipeline. Germline mutations were guidelines. Wherever applicable, macrodissections to enrich for tumour filtered by subtracting variants found in the matched normal sample from cells were performed. DNA was extracted from the tissue and blood those called in the corresponding tumour sample. The resulting somatic samples using either the Qiagen DNA FFPE tissue kit and the QIAamp mutation calls were further filtered to exclude variants (1) that were not DNeasy blood and tissue kit, respectively, following the manufacturer’s sequenced in both sense with a minimum ratio of 10/90%, (2) with less instructions. The cut-off values for tumour content and DNA quantity were than 100 read depth, and (3) variant allele fractions lower than 10% in the 10% and 400 ng, respectively. DNA concentrations were measured using tumour sample. For an alpha list of target genes (Supplementary Table S5), the Qubit fluorometer (Life Technologies) following which, aliquots were npj Breast Cancer (2017) 23 Published in partnership with the Breast Cancer Research Foundation AURORA—screening patients with advanced breast cancer M Maetens et al. mutations occurring below 10% VAF were accepted. Manual evaluation and genomic mass were obtained using ABSOLUTE and non-rounded was done for indels using IGV. estimates of copy numbers and copy number calls were obtained as described above for SNP arrays. Unless otherwise stated, all parameter settings were kept at default values. Mutation calling from Illumina targeted gene screen. Sequence reads from the tumour and matched normal samples were pre-processed according to the GATK best practices. The raw reads were aligned against the human Statistical analyses genome reference hg19/GRCh37 using the BWA aligner. Duplicate reads Integration of the VAF of mutations indexed by the Ion Torrent NGS were marked using Picard following which the data were filtered to keep platform and CNA profiles from the matched SNP arrays to compute the only properly paired and mapped reads with mapping quality greater than Log-Likelihood and individual CCF of each mutation was done using 60. Sequence reads around potential indels were locally realigned and the ABSOLUTE. Similarly, the sequencing depth required to achieved 99% base quality scores were recalibrated using GATK. Mutations were called 28 29 statistical power to call a given base was computed using ABSOLUTE from the resulting BAM files using MuTect2, SomaticSniper and with an error rate E =2E-03 and an FDR = 5E-07. The frequencies of Strelka in matched tumour/normal mode using the default parameter aberration calls and the percentage accuracy were obtained by boot- settings for each of the aforementioned algorithms. The resulting somatic strapping. The values displayed are mean estimates and confidence mutation calls were further filtered to include only variants (1) occurring in intervals are standard errors. The concordance of mutation calls for regions common to both Ion Torrent and Illumina sequencing panels, (2) substitutions was evaluated using the simple matching coefficient. All having a VAF above 10% in the tumour sample, and (3) having a read correlations were measured using the non-parametric Spearman’s ρ depth greater than 50X. Similar to the Ion Torrent NGS, mutations coefficient and all statistical hypothesis tests were done using the non- occurring in the alpha list (Supplementary Table S5) were accepted if they parametric Wilcoxon’s rank sum test. P-values were two sided and paired occurred with a VAF below 10%. The resulting lists of mutation calls were or unpaired as appropriate. Unless otherwise stated, all computations were considered individually for each mutation caller or combined by voting done in R/Bionconductor. using (1) a simple rule of a mutation indexed by at least one of the three, (2) a majority rule of a mutation called by at least two of the three, and (3) a strict consensus of all three mutation callers. In downstream comparison, Data availability we considered only substitutions on account of the known issues of calling The sequencing and SNP arrays data have been deposited at the European indels with confidence from Ion Torrent NGS data. Genome-Phenome Archive (http://www.ebi.ac.uk/ega/) which is hosted by the European Bioinformatics Institute, under accession number Copy number aberration analysis using SNP arrays. The raw intensity EGAD00001000870. values were normalised to obtained Log ratios, B Allele Frequencies (BAF) and genotyping calls (AA/AB/BB) using Affymetrix Power Tools. We used release NA.33 of the NetAffx library for the reference model and ACKNOWLEDGEMENTS annotation. We computed the Median Absolute Pairwise Deviation and The authors would like to extend their gratitude to the patients who participated in the Median Auto-Correlation from the Log ratios as quality control metrics this study, the BIG HQ, Phuong Dinh and Roberto Salgado for contributing to project and used a threshold of 0.40 and 0.45, respectively, to discard failed arrays. set-up and management and to Delphine Vincent and Samira Majjaj for technical We used two parallel approaches involving (a) allele specific copy number assistance. D.B. is supported by the Belgian Fonds National de la Recherche analysis using heterozygous SNP probes and (b) total copy number Scientifique (F.R.S.-FNRS). This work was supported by the Breast Cancer Research analysis using the full set of 200 K markers and parameters from (a) to Foundation (BCRF). control for the cancer cell fraction and genomic mass. From the BAF and genotyping calls, only informative SNP probes displaying heterozygous genotype (AB) and 0.1 < BAF < 0.9 were kept for analysis at (a). The Log AUTHOR CONTRIBUTIONS ratios and BAF were smoothed using the median absolute deviation and segmented jointly using a multitrack segmentation algorithm from the M.P., Sh.L. and C.S. designed the study. M.M. and D.B. analysed the data, interpreted library copynumber to determine common breakpoints. Estimates for the the results and wrote the manuscript. P.A., A.I., F.R., Y.B., D.F., D.Z. and C.D. were cancer cell fraction and genomic mass were obtained using GAP and involved in data interpretation. P.A., S.L., A.T., J.C., S.S., S.V. and M.O. were involved in compared to the results obtained from ABSOLUTE . Samples with a cancer ethics submission, patient consent and recruitment. G.V. performed the central cell fraction lower than 30% were further excluded. For analysis at (b), the histopathological review of the samples. J.L. and P.C. generated the sequencing data. Log ratios for the same samples analysed at (a) were segmented by All the authors read and approved the final manuscript. penalised least square regression as above and non-rounded estimates of copy numbers were obtained as ADDITIONAL INFORMATION 1 x y ¼ 2ðÞ ψα þ21ðÞ  α21ðÞ  α Supplementary Information accompanies the paper on the npj Breast Cancer website (doi:10.1038/s41523-017-0026-6). where α is the CCF and ψ is the genomic mass, both estimated at (a). c = 0.8 is a constant representing the compression ratio of the array and finally Competing interests: The authors declare that they have no competing financial x is the observed Log ratio of a given segment. The copy numbers were interests. categorised as deletions (−1) if y < ψ−0.5, gains (+1) if y > ψ + 0.5, amplifications (+2) if y > ψ + 2.5, and copy neutral (0) otherwise. Unless Publisher’s note: Springer Nature remains neutral with regard to jurisdictional otherwise stated, all parameter settings were kept at default values and all claims in published maps and institutional affiliations. computations were done using R/Bioconductor. Copy number aberration analysis using targeted NGS. The read counts from aligned and sorted BAM files of the Ion Torrent sequencing were REFERENCES processed using ONCOCNV with default parameter settings to correct for 1. Pereira, B. et al. The somatic mutation profiles of 2433 breast cancers refines their library size, GC content and amplicon length. The pool of normal samples genomic and transcriptomic landscapes. Nat. Commun. 7 Article number: 11479 was used as baseline control to capture the technology specific bias (2016). against which each tumour sample was normalised to obtain an estimate 2. Nik-Zainal, S. et al. Landscape of somatic mutations in 560 breast cancer whole- of Log ratio. The Log ratios were segmented using the circular binary 2 2 genome sequences. Nature 534,47–54 (2016). segmentation and CNA calls were obtained using the default clustering 3. Flaherty, K. T. et al. Inhibition of mutated, activated BRAF in metastatic mela- approached implemented in ONCOCNV. The read counts from the aligned noma. N. Engl. J. Med. 363, 809–819 (2010). and sorted BAM files of the Illumina sequencing were processed using 4. Schwaederle, M. et al. Association of biomarker-based treatment strategies with cnvkit to obtain an estimate of Log ratio for on and off target regions. response rates and progression-free survival in refractory malignant neoplasms: a The pool of normal samples was used as baseline control against which meta-analysis. JAMA Oncol. 2, 1452–1459 (2016). each tumour sample was normalised. The Log ratios were smoothed by 5. Roychowdhury, S. et al. Personalized oncology through integrative high- median absolute deviation and segmented by penalised least square throughput sequencing: a pilot study. Sci. Transl. Med. 3, 111ra121 (2011). regression using the library copynumber. Estimates of cancer cell fraction Published in partnership with the Breast Cancer Research Foundation npj Breast Cancer (2017) 23 AURORA—screening patients with advanced breast cancer M Maetens et al. 6. Ong, M. et al. Validation and utilisation of high-coverage next-generation 24. Garcia-Murillas, I. et al. Mutation tracking in circulating tumor DNA predicts sequencing to deliver the pharmacological audit trail. Br. J. Cancer 111, 828–836 relapse in early breast cancer. Sci. Transl. Med. 7, 302ra133 (2015). (2014). 25. Thorvaldsdóttir, H. et al. Integrative genomics viewer (IGV): high-performance 7. Boland, G. M. et al. Clinical next generation sequencing to identify actionable genomics data visualization and exploration. Brief. Bioinform. 14, 178–192 aberrations in a phase I program. Oncotarget 6, 20099–20110 (2015). (2013). 8. Ding, L. et al. Genome remodelling in a basal-like breast cancer metastasis and 26. Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows-Wheeler xenograft. Nature 464, 999–1005 (2010). transform. Bioinformatics 25, 1754–1760 (2009). 9. Navin, N. et al. Tumour evolution inferred by single-cell sequencing. Nature 472, 27. McKenna, A. et al. The genome analysis toolkit: a MapReduce framework for 90–94 (2011). analyzing next-generation DNA sequencing data. Genome Res. 20, 1297–1303 10. Gerlinger, M. et al. Intratumor heterogeneity and branched evolution revealed by (2010). multiregion sequencing. N. Engl. J. Med. 366, 883–892 (2012). 28. Cibulskis, K. et al. Sensitive detection of somatic point mutations in impure and 11. Murtaza, M. et al. Multifocal clonal evolution characterized using circulating heterogeneous cancer samples. Nat. Biotechnol. 31, 213–219 (2013). tumour DNA in a case of metastatic breast cancer. Nat. Commun. 6 Article 29. Larson, D. E. et al. SomaticSniper: identification of somatic point mutations in number: 8760 (2015). whole genome sequencing data. Bioinformatics 28, 311–317 (2012). 12. Yates, L. R. et al. Subclonal diversification of primary breast cancer revealed by 30. Saunders, C. T. et al. Strelka: accurate somatic small-variant calling multiregion sequencing. Nat. Med. 21, 751–759 (2015). from sequenced tumor–normal sample pairs. Bioinformatics 28, 1811–1817 13. de Duenas, E. M. et al. Prospective evaluation of the conversion rate in the (2012). receptor status between primary breast cancer and metastasis: results from 31. Nilsen, G. et al. Copynumber: Efficient algorithms for single- and multi-track copy the GEICAM 2009-03 ConvertHER study. Breast Cancer Res. Treat. 143, 507–515 number segmentation. BMC Genomics 13, 591 (2012). (2014). 32. Popova, T. et al. Genome alteration print (GAP): a tool to visualize and mine 14. Misyura, M. et al. Comparison of next-generation sequencing panels and plat- complex cancer genomic profiles obtained by SNP arrays. Genome Biol. 10, R128 forms for detection and verification of somatic tumor variants for clinical diag- (2009). nostics. J. Mol. Diagn. 18, 842–850 (2016). 33. Carter, S. L. et al. Absolute quantification of somatic DNA alterations in human 15. Kuderer, N. M. et al. Comparison of 2 commercially available next-generation cancer. Nat. Biotechnol. 30, 413–421 (2012). sequencing platforms in oncology. JAMA Oncol. Published online December 15, 34. Boeva, V. et al. Multi-factor data normalization enables the detection of copy 2016. doi:10.1001 (2016). number aberrations in amplicon sequencing data. Bioinformatics 30, 3443–3450 16. Zardavas, D. et al. The AURORA initiative for metastatic breast cancer. Br. J. Cancer (2014). 111, 1881–1887 (2014). 35. Olshen, A. B. et al. Circular binary segmentation for the analysis of array-based 17. Thatcher, N. et al. Necitumumab plus gemcitabine and cisplatin versus gemci- DNA copy number data. Biostatistics 5, 557–572 (2004). tabine and cisplatin alone as first-line therapy in patients with stage IV squamous non-small-cell lung cancer (SQUIRE): an open-label, randomised, controlled phase 3 trial. Lancet Oncol. 16, 763–774 (2015). Open Access This article is licensed under a Creative Commons 18. Park, H. S. et al. High EGFR gene copy number predicts poor outcome in triple- Attribution 4.0 International License, which permits use, sharing, negative breast cancer. Mod. Pathol. 27, 1212–1222 (2014). adaptation, distribution and reproduction in any medium or format, as long as you give 19. Zardavas, D. et al. Emerging targeted agents in metastatic breast cancer. Nat. Rev. appropriate credit to the original author(s) and the source, provide a link to the Creative Clin. Oncol. 10, 191–210 (2013). Commons license, and indicate if changes were made. The images or other third party 20. Zardavas, D. et al. Personalized therapy for breast cancer: a dream or a reality? material in this article are included in the article’s Creative Commons license, unless Future Oncol. 9, 1105–1119 (2013). indicated otherwise in a credit line to the material. If material is not included in the 21. André, F. et al. Comparative genomic hybridisation array and DNA sequencing to article’s Creative Commons license and your intended use is not permitted by statutory direct treatment of metastatic breast cancer: a multicentre, prospective trial regulation or exceeds the permitted use, you will need to obtain permission directly (SAFIR01/UNICANCER). Lancet Oncol. 15, 267–274 (2014). from the copyright holder. To view a copy of this license, visit http://creativecommons. 22. Talevich, E. et al. CNVkit: genome-wide copy number detection and visualization org/licenses/by/4.0/. from targeted DNA sequencing. PLoS Comput. Biol. 12, e1004873 (2016). 23. Kuilman, T. et al. CopywriteR: DNA copy number detection from off-target © The Author(s) 2017 sequence data. Genome. Biol. 16, 49 (2015). npj Breast Cancer (2017) 23 Published in partnership with the Breast Cancer Research Foundation

Journal

npj Breast Cancer – Springer Journals

Published: Jun 29, 2017

There are no references for this article.