Université Libre de Bruxelles
FACULTE DE MEDECINE
Debora Fumagalli, MD
Thèse présentée en vue de l’obtention du grade académique
de Docteur en Sciences Médicales
Promoteur de thèse: Professeur Christos Sotiriou
Année académique 2015-2016
This thesis has been written under the supervision of Prof. Christos
Sotiriou.
The members of the Jury are:
Prof. Gilbert Vassart (Président ; Université Libre de Bruxelles, Belgium) Prof. Carine Maenhaut (Secrétaire ; Université Libre de Bruxelles, Belgium) Prof. Christos Sotiriou (Promoteur ; Institut Jules Bordet, Université Libre de
Bruxelles, Belgium)
Prof. Cédric Blanpain (Université Libre de Bruxelles, Belgium)
Prof. Ahmad Awada (Institut Jules Bordet, Université Libre de Bruxelles, Belgium)
The external experts are:
Acknowledgments
I would like to thank all the people who helped and supported me in these last years, both in my professional and in my personal journey. In particular:
My supervisors Christos Sotiriou, for the great opportunities he gave me.
Professor Martine Piccart, for the possibility of starting this fruitful international experience.
My colleagues at the BCTL, in particular: Delphine, Samira, Ghizlane, Naima, Sandy, Pierre-Yves for their help with the experiments; Vinu, Sylvan, David, for their help with bioinformatics & statistics & other; Christine, Norman, Roberto, Sherene, Stefan, Michalis, for what they taught me; Dominique and Jeanne, for their great logistics support; and last, but not least, Françoise and Marion, for their friendship, their constant support throughout these years, and for their craziness.
The fellows who shared this experience with me, in particular Otto, Hatem, Felipe, Kamal, Carmen, Ivana, Marta.
My colleagues at Bordet, in particular Lissandra, Andrea, Philippe, Evandro, Karen.
My colleagues at ULB, in particular Vincent, David, Danai, Tomasz.
My former Italian colleagues, in particular Professor L. Gianni, Diego, and Giampi, and my former American colleagues, in particular Dr S. Paik, Kay, Patrick, Megan, Melany.
My lifelong friends, in particular Dafna, Sara, Alessandra, and my more recent, great friends Marcella, Stella and Giulia, for everything they did and keep doing for me.
My mum, dad, brother and sister in-law, for all the moral and material support.
My family in-law, for all they do for me.
Table of Contents
Acknowledgments ... 3
Table of Contents ... 5
List of Tables ... 7
List of Figures ... 9
Abbreviations ... 13
Summary in French ... 16
Introduction ... 19
Inter-tumor heterogeneity ... 21
Intra-tumor heterogeneity ... 26
Spatial heterogeneity ... 26 Temporal heterogeneity ... 28Origins of tumor heterogeneity ... 31
The cell of origin ... 31
The cancer stem cell hypothesis & the clonal evolution model ... 32
Beyond the Cancer Genome ... 34
The influence of the tumor microenvironment ... 34
The emerging role of RNA editing ... 35
Breast cancer heterogeneity: the clinical challenges ... 39
Aims of my research works ... 40
CHAPTER 1: Intra-tumor heterogeneity ... 43
PART A: Multifocal breast cancers ... 43
Introduction ... 45
Materials and Methods ... 47
Results ... 57
Discussion ... 73
PART B: Metastatic ER positive/HER2 negative breast cancer
patients. ... 76
Abstract ... 76
Introduction ... 78
Material and Methods ... 80
Results ... 89
Discussion ... 108
Conclusions and Perspectives ... 113
CHAPTER 2: Inter-tumor heterogeneity ... 115
The role of RNA editing in breast cancer ... 115
Abstract ... 115
Introduction ... 117
Material and Methods ... 119
Results ... 127
Discussion ... 164
Conclusions and Perspectives ... 171
Concluding remarks ... 173
References ... 177
List of Tables
Table 1. Overview of the first main NGS-based studies published in breast
cancer. ... 25
Table 2. Summary of patient and tumor characteristics. ... 50
Table 3. List of genes interrogated by targeting sequencing. ... 54
Table 4. Summary of patient and tumor characteristics. ... 57
Table 5. List of genes and hotspot mutations assessed in the mutation panel ... 83
Table 6. List of genes assessed in the 400-gene expression panel ... 85
Table 7. List of genes assessed in the copy number panel ... 86
Table 8. Patient and tumor characteristics ... 90
Table 9. Aberrations prevalence in primary vs metastatic samples ... 91
Table 10. Concordance rate of aberrations in matched primary and metastatic samples ... 94
Table 11. Demographics and molecular profiles of patients with ESR1 mutations ... 97
Table 12. Subtype prevalence in primary vs metastatic samples ... 100
Table 13. Subtype prevalence in matched primary vs metastatic samples ... 100
Table 14. Prevalence of aberrations in luminal A vs luminal B primary tumors . 101 Table 15. Prevalence of aberrations in luminal A vs luminal B metastatic tumors ... 102
Table 16. Patient and sample characteristics. ... 127
Table 17. Studies comparison ... 135
List of Figures
Figure 1. Outcome of gene expression-based breast cancer subtypes. ... 22
Figure 2. Overview of the A-to-I RNA editing process. ... 36
Figure 3. Validation of the mutations using the alternative sequencing platform. 59 Figure 4A and B. Boxplots of mutational burden per patient in terms of (A) the number of lesions and (B) samples that have been interrogated. ... 60
Figure 5. Distribution of non-silent substitutions and indels in the "homogeneous" MFBC group. ... 62
Figure 6. Distribution of non-silent substitutions and indels in the "intermediate" MFBC group. ... 63
Figure 7. Distribution of non-silent substitutions and indels in the "heterogeneous" MFBC group. ... 65
Figure 8A and B. Boxplots of the number of (A) interrogated lesions and (B) samples per patient in terms of the group of MFBC. ... 66
Figure 9. Boxplot of the mutational burden per patient in terms of the group of MFBC. ... 67
Figure 10. Inter-lesion heterogeneity and inter-lesion distance. ... 68
Figure 11. Genome-wide alterations. ... 69
Figure 12. Genome-wide CNAs. ... 72
Figure 13. Consort diagram. ... 81
Figure 14. Landscape of molecular aberrations in primary and metastatic ER positive/HER2 negative breast cancers. ... 92
Figure 15. Evolution of somatic mutations and CN alterations in matched primary and metastatic samples. ... 95
Figure 34. Validation of our A-to-I editing model. ... 144 Figure 35. Sequencing coverage, ADAR expression and number of detected sites. ... 145 Figure 36. Editing of individual mRNA molecules. ... 146 Figure 37. Dose-response curves for experiment in cell line BT474. ... 147 Figure 38. Example of a fit of the logistic model (line) to experimental points (dots). ... 148 Figure 39A and B. Distribution of (A) and (B) across the 81 sites. ... 149 Figure 40. Correlation between i estimates in vitro and editing frequency in vivo.
Figure 51. ADAR amplification and the IFN response predict ADAR expression in human cancers. ... 163 Figure 52. Overview of the TRACERx study. ... 174 Figure 53. Overview of the AURORA program. ... 175 Figure 54. Kaplan-Meier plots showing OS of the 18 genes identified in Figure
18. Appendix 5.
Figure 55. Additional controls associated with Figure 32. Appendix 8.
Figure 56. Figure referring to the modeling of editing frequency with the logistic function. Appendix 8.
Figure 57. Gen set analysis & analysis of ADAR isoforms. Appendix 8.
Abbreviations
BC: Breast Cancer bp: base pairCAP: College of American Pathologists
CGH: Comparative Genomic Hybridization CI: Confidence Interval
CNA: Copy Number Aberration CNV: Copy Number Variation CSC: Cancer Stem Cell CT: ChemoTherapy
CTC: Circulating Tumor Cells ctDNA: circulating tumor DNA DCIS: Ductal Carcinoma In Situ DNA-seq: DNA Sequencing dsRNA: double stranded RNA
EC: Ethical Committee ER: Estrogen Receptor ET: Endocrine Treatment
FFPE: Formalin-Fixed, Paraffin-Embedded FISH: Fluorescence In Situ Hybridization
FPKM: Fragment Per Kilobase per Million aligned reads
G: Grade
HER2: Human Epidermal Growth Factor Receptor 2
IHC: ImmunoHistoChemistry
IDC: Invasive Ductal Carcinoma
IDFS: Invasive Disease Free Survival IFN: Interferon
ILC: Invasive Lobular Carcinoma
lncRNA: long non-coding RNA LVI: Lympho Vascular Invasion
MFBC: MultiFocal Breast Cancer
mRNA: messenger RNA miRNA: micro RNA
NGS: Next-Generation Sequencing OS: Overall Survival
PR: Progesterone Receptor
RDD: RNA-DNA Difference RNA-seq: RNA sequencing
RPPA: Reverse Phase Protein Array
shRNA: small hairpin RNA siRNA: small-interfering RNA
SNP: Single Nucleotide Polymorphism
TILs: Tumor Infiltrating Lymphocytes TME: Tumor MicroEnvironment
UK: UnKnown
UTR: UnTranslated Region
WES: Whole-Exome Sequencing WGS: Whole-Genome Sequencing
Summary in French
Le cancer du sein est le cancer le plus fréquent chez la femme et représente la principale cause de mortalité liée au cancer. Le décés est habituellement causé par le développement de résistance aux traitements et la propagation métastatique de la maladie. Malgré la pertinence clinique, la complexité moléculaire de la maladie et sa dynamique restent à ce jour peu connues.
Depuis longtemps, l’hétérogénéité du cancer du sein a été observée au niveau histologique et du profil évolutif clinique, et ces différences ont servi de base pour la classification de la maladie. Avec le développement des technologies à haut débit, telles que les puces à damier (microarrays) et le séquençage à haut débit, cette classification a été affinée et une complexité génétique jusqu'alors inconnue a été révélée.
complexité pourrait avoir un impact important sur la façon dont les patientes atteintes d’un cancer du sein sont prises en charge et traitées.
La recherche que j’ai menée dans le Breast Cancer Translational Research Laboratory sous la direction du Professeur Christos Sotiriou avait deux objectifs principaux. Le premier était de déterminer l'ampleur et les implications cliniques de l'hétérogénéité intra-tumorale dans deux scénarios cliniques courants, à savoir: les cancers du sein multifocaux (MFBCs) et les cancers du sein métastatiques ER positif / HER2 négatif. Le deuxième était d'étudier l'impact de l'édition de l'ARN dans la détermination de l'hétérogénéité inter-tumorale, phénomène encore peu caractérisé.
Notre recherche a notamment montré que:
1) Les lésions de tous les MFBCs que l’on a étudiés partagent une origine commune. Malgré cela, et malgré des caractéristiques pathologiques similaires, chez un tiers des patientes, les lésions multifocales d’une même patiente ne partageaient aucune substitution et aucune insertion/déletion. De plus, l’hétérogénéité inter-lésion a été observée pour des mutations oncogéniques dans des gènes tels que PIK3CA, TP53, GATA3 et PTEN;
été observée entre les lésions primaires et métastatiques appariées de cancers du sein ER positif / HER2 négatif. Des différences entre les lésions appariées ont cependant été trouvées pour les niveaux d’expression de certains gènes. Dans les lésions primaires, seuls les niveaux d’expression de quelques gènes et un niveau élevé d'amplification de FGFR1 ont été associés à la survie;
3) L'édition de l’ARN est une source généralisée de variation du transcriptome dans le cancer du sein. Dans ce cancer, et potentiellement dans tous les cancers, l'édition de l’ARN est principalement contrôlée par deux facteurs, à savoir l'amplification de 1q et l'inflammation, qui sont toutes deux très répandues parmi les cancers humains. La magnitude de l'édition de l’ARN, en combinaison avec la conservation des sites d'édition détectés dans les tissus et les patientes, suggère qu'il pourrait y avoir des implications cliniques et thérapeutiques pour un large éventail de patientes atteintes d’un cancer.
Introduction
Breast cancer is the most frequently diagnosed cancer and the leading cause of cancer-related mortality in women worldwide (Jemal et al., 2011). In most instances, death is caused by the metastatic spread of the disease and its uncontrolled growth at distant sites determined by the development of resistance to available treatment strategies. Despite the clinical relevance, our understanding of the molecular complexity of the disease and of the dynamics underling its behavior is still limited.
Since long time, pathologists have noted that breast cancers are heterogeneous at the morphological level, and comprise different types of cells and distinct stromal composition between different regions (Bloom & Richardson., 1957; Tan et al., 2013). Differences among breast cancer patients, as well as between different lesions of the same patient, have also been reported for the expression levels of the estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2), the three biomarkers used in standard practice for the characterization of the disease and for clinical decision
-making (Aurilio et al., 2014; Pusztai et al.,2010).
perception of the disease (Sotiriou & Pusztai., 2009; Desmedt et al., 2012; Bedard et al., 2013). Gene expression profiling studies have shown that breast cancers can be classified into the so-defined intrinsic molecular subtypes, which bear both predictive and prognostic values (Sotiriou et Piccart., 2007). Genome-wide sequencing studies have further unraveled the complex and diverse genomic landscape characterizing each breast lesion (Desmedt et al., 2012). These studies have shown that molecular differences exist not only between different breast cancer patients (inter-tumor heterogeneity), but also within the same patient (intra-tumor heterogeneity). Furthermore, intra-tumor heterogeneity could occur either between different geographical regions of a tumor (spatial intra-tumor heterogeneity), or as molecular evolution of a tumor over time (temporal intra-tumor heterogeneity) (Zardavas et al., 2015).
Heterogeneity in breast cancer
Inter-tumor heterogeneity
Diversity in breast cancer histology is a known and well-characterized phenomenon, as reflected in the World Health Organization (WHO) classification of the disease into several categories based on microscopic features (Tan et al., 2013). This classification has, however, limited clinical relevance as the great majority of breast cancers (>70%) are classified as ‘invasive ductal carcinoma not otherwise specified’ (IDC NOS) despite showing distinct clinical behaviors and diverse sensitivity to treatments (Li CI et al., 2005).
in the same amplicon), and are usually associated with a high histologic grade. Basal-like breast cancers do not express hormone receptors or HER2 (triple negative phenotype), express markers of mammary basal cells, such as basal cytokeratins, and are associated with aggressive behavior and poor prognosis (Figure 1).
Figure 1. Outcome of gene expression-based breast cancer subtypes.
Outcome predictions according to the four gene expression defined breast cancer subtypes in a set of 710 node-negative, untreated breast cancer patients (taken from Parker et al., 2009)
and the ‘molecular apocrine’ (Farmer et al., 2005), were identified. The analysis of gene-expression profiles from about 600 TNBCs revealed that these tumors could be clustered in six subtypes with distinct sensitivity to anticancer treatment (Lehmann et al., 2011). More recently, in the context of the METABRIC (Molecular Taxonomy of Breast Cancer International Consortium) initiative, the combined analysis of gene-expression and copy-number profiles of 2000 breast cancers identified 10 different molecular subgroups, defined as integrative clusters, that showed different clinical outcomes and distinct molecular features, and that divided each intrinsic subtype into separate groups (Curtis et al., 2012). Even though further validation is warranted, these findings suggest that patients classified as belonging to the same subtype can display divergent clinical behaviors.
and colleagues identified so many as 73 distinct combinations of mutations in cancer-related genes (Stephens et al., 2012). Banerji and colleagues, analyzing 103 breast cancers of diverse subtypes, confirmed known somatic mutations and identified new recurrent aberrations, among which a recurrent fusion in TNBCs that warrants further investigation as treatment target (Banerji et al., 2012).
Study Breast Cancer Subtype
Technology and number of
samples
Findings
Stephens et al. (2012)
ER+/ER- WES (100) * Substantial variation in
number and patterns of aberrations Banerji et al. (2012) All major expression subtypes WES (103) WGS (22) * Confirmation of previously known, recurrent somatic mutations * Identification of new alterations * Identification of recurrent fusion gene in TNBC Shah SP et al. (2012) TNBC Affymetrix SNP6.0(104) RNA sequencing (80) WES/WGS (65) * Varying mutational aberrations within TNBC
TCGA (2012) All major expression subtypes Whole-exome sequencing (507) DNA methylation (802) SNP arrays (773) mRNA microarrays (547) miRNA sequencing (697) RPPA (403)
* Molecular landscape of the main breast cancer subtypes * Four main breast cancer phenotypic classes recapitulate diverse genetic and epigenetic alterations
Intra-tumor heterogeneity
Intra-tumor heterogeneity is determined by the co-existence of cancer cell populations differing in their genomic, phenotypic or behavioral traits either across different regions of a single tumor lesion (spatial heterogeneity), or in the course of the evolution of the disease (temporal heterogeneity). Considering that these cell populations might be driven by distinct genetic aberrations (Shah SP et al., 2012), it is evident that intra-tumor heterogeneity represents a significant challenge in terms of biomarkers characterization and treatment selection.
Spatial heterogeneity
morphologically distinct components of these lesions. Similar observations were made in a case of a TNBC in which regions of apocrine differentiation harbored genomic gains on chromosome arms 9p and genomic losses on chromosome arms 9q that were not present in non-apocrine areas of the same cancer (Patani et al; 2011).
having at least one aberration confined to a restricted number of adjacent regions.
Even though the most objective way to assess spatial tumor heterogeneity might be represented by single cell sequencing (Navin et al., 2011; 2015), for the time being the prohibitive costs and the time required to perform these experiments limit their clinical applicability.
Temporal heterogeneity
The evolution of genetic aberrations over time has been observed in three differ-ent instances: the transition from in situ to invasive breast cancer, the evolution of primary breast cancers over time, and the progression of primary breast cancers to metastatic disease.
CNAs and a limited number of gene mutations, the second using single-cell FISH analyses for multiple cancer genes. In both studies, despite the genomic similarities described between the synchronous lesions, a shift in the abundance of clonal composition was observed during progression. In the above-mentioned study, using whole genome sequencing Yates and colleagues detected the presence of several subclones in the DCIS lesion of a patient, two of which gave rise to distinct lesions of a multifocal carcinoma (Yates et al., 2015).
Beyond the natural evolution of the disease, temporal intra-tumor heterogeneity can be observed as a consequence of external selective pressures such as the ones exerted by anticancer treatments. A good model to study this phenomenon is represented by the profiling of cancer lesions before and after the administration of neoadjuvant treatments. In a recent work published by Balko and colleagues, the comprehensive molecular analyses of 74 patients with residual TNBC after neoadjuvant treatment showed that the residual tumor tissue had a different genomic landscape compared with matched pre-treatment tumor and, in the majority of the cases, contained genetic alterations potentially treatable with targeted drugs (Balko et al, 2014).
Origins of tumor heterogeneity
The origins of inter- and intra-tumor heterogeneity are yet not fully understood. Several hypotheses have been postulated; the most accredited will be summarized below.
The cell of origin
of origin and the genetic alteration that drove its development (Van Keymeulen et al., 2015, Koren et al., 2015). In a recent report, Van Keymeulen and colleagues have in fact shown that induction of PIK3CAH1047R mutations into either luminal or basal restricted progenitors resulted in de-differentiation, and that the phenotype of the transduced cell determined the phenotype of the resulting tumor, ie PIK3CAH1047R expression in luminal cells gave rise to more aggressive basal-like tumors, while its expression in basal cells gave rise to less aggressive luminal tumors (Van Keymeulen et al., 2015).
The cancer stem cell hypothesis & the clonal evolution model
Two are the most accredited hypotheses attempting to explain the generation and maintenance of intra-tumor heterogeneity: the cancer stem cell (CSC) and the clonal evolution hypotheses.
into mice (Al-Hajj et al., 2003). Additional markers such as ALDH1 (Ginestier et al., 2007) have subsequently been reported useful to characterize this cell population.
The clonal evolution model was first described in a landmark paper by P. Nowell in 1976 (Nowell., 1976). According to this model, the majority of cells within a tumor has the potential to be tumorigenic, and could contribute to progression and resistance to treatment following Darwinian evolutionary rules. Being genetically unstable, tumor cells accumulate genetic alteration; only the subpopulation of cells with a biological fitness advantage could keep growing and survive the selective pressure imposed by the (micro)environment and/or by external factors such as therapies. The co-existence of multiple clones with distinct genotype, phenotype, metastatic potential, and sensitivity to therapeutics would contribute to intra-tumor heterogeneity. Important features of this model are also the concepts of ‘driver’ and ‘passenger’ somatic mutations. While the former increase the fitness of the cells, the latter are neutral (Greaves & Maley., 2012).
Beyond the Cancer Genome
The influence of the tumor microenvironment
The emerging role of RNA editing
While intense effort is currently being dedicated to cancer genome sequencing, comparatively little attention has been devoted at understanding how faithful RNA sequences are to the DNA sequences from which they were derived. Messenger RNA (mRNA) is the target of a series of post-transcriptional modifications that can affect its structure and stability, one of the most relevant being RNA editing (Bass, 2002; Levanon et al., 2004; Nishikura, 2010).
Figure 2. Overview of the A-to-I RNA editing process.
The substitution of a single base in a section of mRNA alters the genetic information transcribed from the DNA before it is translated into a protein (taken from Hayden., 2011).
ADAR enzymes are essential in mammals (Higuchi et al., 2000; Wang Q et al., 2000) and exist in three forms: ADAR (also known as ADAR1), which is ubiquitous and has two isoforms—p110 is constitutive and p150 is inducible; ADARB1 (also known as ADAR2), principally expressed in the brain; and ADARB2 (also known as ADAR3), which contrary to ADAR and ADARB1 seems to be enzymatically inactive (Chen et al., 2000; Savva et al., 2012).
micro RNAs (miRNAs), small-interfering RNAs (siRNAs) and long non-coding RNAs (lncRNAs), affecting both their structure and activities (Blow et al., 2006; Hundley and Bass, 2010; Kapusta et al., 2013; Kawahara et al., 2007).
A-to-I editing has been shown to occur predominantly in highly repetitive Alu sequences, likely because their frequency (>106) in the human genome makes their arrangement in quasi-palindrome configurations prone to RNA duplex formation highly probable (Athanasiadis et al., 2004; Bazak et al., 2014a; Kim et al., 2004; Levanon et al., 2004). High-throughput sequencing studies suggest that tens of thousands to millions of positions are targeted by A-to-I editing in the human transcriptome (Bahn et al., 2012; Ju et al., 2011; Li JB et al., 2009; Park et al., 2012; Peng et al., 2012; Ramaswami et al., 2012, 2013), and a recent publication reports that potentially all adenosines in specific Alu repeats undergo A-to-I editing (Bazak et al., 2014b).
Breast cancer heterogeneity: the clinical challenges
Our limited knowledge of the molecular mechanisms that drive the origin, development, and progression of breast cancer have so far precluded the advancements of curative treatment strategies.
The presence and extent of inter- and intra-tumor heterogeneity represent an important clinical challenge. Inter-tumor heterogeneity implies that each breast cancer patient carries a lesion with a unique genomic make-up, phenotype and behavior, precluding therapeutic approaches based on the ‘one-size-fits-all’ paradigm. Intra-tumor heterogeneity is even more critical. The co-existence of multiple sub-clones characterized by diverse molecular aberrations and with different drug sensitivities implies that therapeutic strategies targeting predominant aberrations can’t be effective against the whole tumor mass. Instead, treatment could select the resistant clones that, beyond limiting the success of the treatment, will lead to tumor progression and relapse.
Aims of my research works
The three research works that I carried out with and under the supervision and direction of Prof Christos Sotiriou in the Breast Cancer Translational Research Laboratory (BCTL) at Institut Jules Bordet had two main aims, which are described in the two main chapters of my dissertation. Each of these research works led to a manuscript.
CHAPTER 1: Intra-tumor heterogeneity
The first aim was to determine the extent and the clinical implications of intra-tumor heterogeneity in two common clinical scenarios, namely: multifocal breast cancers (MFBCs) and metastatic ER positive/HER2 negative breast cancers.
PART A (Manuscript n° 1)
discrepancy, respectively). The main purposes of this first research work were 1), to assess if the distinct clinical behavior of MFBCs might be explained by the genomic differences between ductal MFBC lesions with similar phenotype based on classical pathological parameters such histological grade, ER, and HER2 status; and 2) if the different lesions of MFBCs share a common origin. This study was carrier out using a series of 36 patients with tumor material available for at least two foci of their multifocal cancers.
PART B (Manuscript n° 2)
2) to explore the relationship between molecular aberrations and survival. This study was carrier out in a population of 182 ER positive/HER2 negative metastatic breast cancer patients with tumor tissue from primary lesions available for all, and tumor material from matched metastatic lesions available for 88 of them, respectively, and with long term follow-up data.
CHAPTER 2: Inter-tumor heterogeneity (Manuscript n° 3)
Currently, very little is known of the role of RNA editing in the development and progression of cancer and of its impact on the degree of tumor heterogeneity observed between patients. Very few studies on RNA editing have been conducted so far; moreover, these studies usually included few patients and/or investigated few editing sites, limiting the applicability of their findings. The main aims of this third research work were 1) to investigate the magnitude of this phenomenon in breast cancer, and 2) to identify the mechanisms that govern RNA editing in breast and, potentially, all cancers. For this purpose, we characterized the tumor and matched normal tissues obtained from a cohort 58 breast cancer patients equally distributed among the main breast cancer subtypes (luminal A, luminal B, HER2 positive, triple negative) using both RNA and DNA sequencing, and explored the biological factors that defined their RNA editing profiles.
CHAPTER 1: Intra-tumor heterogeneity
PART A: Multifocal breast cancers
Abstract
Multifocal breast cancer (MFBC), defined as multiple synchronous unilateral lesions of invasive breast cancer, is relatively frequent and has been associated with more aggressive features than unifocal cancer. Here, we aimed to investigate the genomic heterogeneity between MFBC lesions sharing similar histopathological parameters.
Characterization of different lesions from 36 patients with ductal MFBC involved the identification of non-silent coding mutations in 360 protein-coding genes (171
This research work is related to the following publication (Appendix 1):
C. Desmedt,* D. Fumagalli, * E. Pietri * G. Zoppoli, D. Brown, S. Nik-Zainal, G. Gundem, F. Rothé, S. Majjaj, A. Garuti, E. Carminati, S. Loi, T. Van Brussel, B. Boeckx, M. Maetens, L. Mudie, D. Vincent, N. Kheddoumi, L. Serra, I. Massa, A. Ballestrero, D. Amadori, R. Salgado, A. de Wind, D. Lambrechts, M. Piccart, D. Larsimont, P.J. Campbell, C. Sotiriou.
* These authors contributed equally to the present work.
Uncovering the genomic heterogeneity of multifocal breast cancer
tumour and 36 matched normal samples). We selected only patients with lesions presenting the same grade, ER, and HER2 status. Mutations were classified as ‘oncogenic’ in the case of recurrent substitutions reported in COSMIC or truncating mutations affecting tumour suppressor genes. All mutations identified in a given patient were further interrogated in all samples from that patient through deep resequencing using an orthogonal platform. Whole-genome rearrangement screen was further conducted in 8/36 patients.
Twenty-four patients (67%) had substitutions/indels shared by all their lesions, of which 11 carried the same mutations in all lesions, and 13 had lesions with both common and private mutations. Three-quarters of those 24 patients shared oncogenic variants. The remaining 12 patients (33%) did not share any substitution/indels, with inter-lesion heterogeneity observed for oncogenic mutation(s) in genes such as PIK3CA, TP53, GATA3, and PTEN. Genomically heterogeneous lesions tended to be further apart in the mammary gland than homogeneous lesions. Genome-wide analyses of a limited number of patients identified a common somatic background in all studied MFBCs, including those with no mutation in common between the lesions.
Introduction
respectively) (Buggi et al., 2012; Choi et al., 2012; Pekar et al., 2014).
The main objectives of this research work were:
To characterize the genomic profile of a series of 36 ductal MFBC lesions with concordant histological grade, ER, and HER2 status using targeted gene screen;
To investigate the potential inter-lesions genomic heterogeneity;
If existing, to investigate the association of inter-lesion heterogeneity with clinical and/or histopathological characteristics;
To interrogate the phylogenetic relationship between the various lesions.
My contribution to this project involves:
- Identifying the candidate patients for the study;
- Samples preparation and pathological characterization; - Analyzing the re-sequencing experiments;
- Data and results interpretation; - Manuscript preparation.
Materials and Methods
Patient selection and samples characterization
Patients were retrospectively selected on the basis of the following criteria: 1) documented MFBC, defined in the pathology report as the presence of multiple, synchronous, ipsilateral invasive lesions in the surgical specimen separated by benign breast tissue;
2) ductal histology;
3) tumour cellularity greater than 40%;
4) availability of two or more MFBC lesions, from which a minimum amount of 700 ng of double-stranded DNA (dsDNA) could be extracted;
5) similar histological grade, ER, and HER2 status following central pathology review of the different lesions of the same patient;
6) no neo-adjuvant treatment received;
7) availability of germline DNA, derived from whole blood or a tumour-free axillary lymph node.
From these patients, multiple formalin-fixed, paraffin-embedded (FFPE) tumor samples and 36 matched normal samples were obtained. Eight of the 36 patients were selected for a low-coverage whole-genome rearrangement screen based on the availability of a frozen sample of at least two invasive lesions, from which a minimum of 3 μg of dsDNA could be extracted.
The histological grade and the ER, HER2, Ki67, and PTEN status of tumor samples was defined as described in Appendix 2. Samples characteristics are summarized in Table 2; additional details can be found in Table 19 of Appendix 3.
DNA and RNA extraction
Patient ID Total of lesions Size largest lesion (cm) Aggregate size (cm)* Size of the characterized lesions (cm) Largest distance between lesions
Subtype Grade** DCIS LVI
PD13790 2 2,2 3,3 2.2; 1.1 2 ER+HER2- G1;G2 Y NA PD13791 2 1,2 2,2 1.2; 1 1.5 ER+HER2- G1 Y N PD13792 3 1.5* 3,7 1.1; 1.1 UK ER+HER2- G1 Y N PD13793 3 1,4 2,8 1.2; 1.4 5 ER+HER2- G1 Y N PD13794 2 1,2 2 1.2; 0.8 UK ER+HER2- G3 Y N PD13796 2 2,2 3,2 1; 2.2 6 HER2+ G1 Y N PD13798 2 1,3 2,5 1.2; 1.3 1.7 HER2+ G3 Y Y PD13799 3 1,2 3,2 1; 1; 1.2 3 ER-HER2- G3 N N PD13800 2 1,2 2,2 1.2; 1 0.9 ER-HER2- G3 Y Y PD13801 3 2,3 5,3 2.3; 2 0.3 HER2+ G3 Y Y PD13802 2 2,5 4 1.5; 2.5 1 HER2+ G1 Y Y PD13804 3 2,2 4,6 2.2; 1.5; 0.9 2.5 HER2+ G3 N N PD13805 2 1,7 2,5 1.7; 0.8 2 HER2+ G3 Y N
Table 2. Summary of patient and tumor characteristics.
*Comprising all the foci, also the ones for which no material was available
Targeted gene screen
The exonic regions of 360 cancer-related genes (Table 3) were enriched using in-solution RNA baits (SureSelect, Agilent, UK) and sequenced on an Illumina HiSeq 2000 instrument at the Wellcome Trust Sanger Institute, following the manufacturer’s instruction. Samples from eight patients were processed with an earlier version of the bait design, lacking 46 of those 360 genes. Somatic base substitutions and small insertions or deletions were identified by comparison with the matched normal sample using established bioinformatic algorithms (Verela et al., 2011; Ye et al., 2009).
Gene Common or private to version 2 Gene Common or private to version 2 Gene Common or private to version 2
EIF5A2 common MDM4 common SRSF2 common ELK3 common MED12 common STAT3 V2 EP300 common MED12L V2 STK11 common
EP400 V2 MED13 V2 SUFU common
EPHA3 common MED29 common TBX22 common EPHA5 common MEN1 common TBX3 common EPHA6 common MET common TERT common EPHA7 common MITF common TET2 common EPHB1 common MLH1 common TGFBR2 common EPHB4 common MLL common TNFAIP3 common EPHB6 common MLL2 common TOP1 common ERBB2 common MLL3 common TP53 common ERBB3 common MPL common TP63 V2 ERBB4 common MRAS common TP73 common ERCC2 common MRE11A common TRAF2 common ERCC3 common MSH2 common TSC1 common ERCC4 common MSH6 common TSC2 common ERCC5 common MST1 V2 TSHR common ESR1 common MTDH common U2AF1 common ETV1 V2 MTOR common USP9X common EZH2 common MUTYH common VEGFA common
FADD common MYB V2 VHL common
FAM123B common MYC common WHSC1L1 V2 FANCA common MYCL1 common WSB1 common FANCC common MYCN common WT1 common FANCD2 common MYD88 common XPA common FANCE common MYO3A V2 XPC common FANCF common MYO5B V2 XPO1 common FANCG common MYOC common YAP1 common FAS common NBN common YWHAB common FBXO11 common NCOA2 V2 YWHAQ common FBXW7 common NCOA3 common YWHAZ common FGFR1 common NF1 common ZNF217 common FGFR2 common NF2 common ZRSR2 common
Identification of oncogenic mutations
Mutations were classified in one of the following categories according to the following definition, slightly modified from Papaemmanuil et al., 2013:
1) Oncogenic: non-synonymous substitutions or in-frame mutations in canonical oncogenes at recurrent hotspots; non-synonymous substitutions recurrent in two or more confirmed samples in COSMIC; non-synonymous substitutions recurrent in two or more samples in our own dataset; and nonsense, frameshifting insertions, and deletions in known tumour suppressors;
2) Putative oncogenic: previously unreported non-synonymous substitutions in a known cancer gene within ±3 amino acids of a mutation recurrent in two or more samples in COSMIC; more than two non-synonymous substitutions within three amino acids of each other (mutation clusters);
3) Possible oncogenic: non-synonymous substitutions confirmed somatic in one sample in COSMIC; mutations close to mutation clusters in COSMIC;
4) Unknown significance: all remaining mutations.
For the sake of simplicity, all oncogenic, putative, and possible oncogenic mutations are referred to as ‘oncogenic’ here.
Low coverage whole-genome sequencing
to the standard protocol (Illumina Inc, San Diego, CA, USA). Short reads were mapped back to the reference genome (GRCh37) and discordantly mapping reads were identified as pairs that did not match with the expected insert size, mapped in the wrong orientation, or mapped to different genomic regions, as previously described (Campbell et al., 2010; Stephens et al., 2009). Putative rearrangements were validated and breakpoints annotated using capillary sequencing. CNAs were assessed by QDNAseq (version 1.0.5) (Scheinin et al., 2014); absolute estimates of copy numbers were obtained using ABSOLUTE (version 1.0.6) (Carter et al., 2012) in ‘total’ copy number mode. Additional details can be found in Appendix 2.
Statistical analyses
Results
Targeted sequencing of cancer genes in MFBC
The clinico-pathologic characteristics of the 36 MFBC patients involved in this study and of their lesions are summarized in Table 4.
Patient characteristic Number of
patients Age, years < 40 2 40-49 11 50-69 17 > 70 6 Tumor size, cm 1-2 17 2-5 18 > 5 1 N° of positive nodes None 16 1-3 16 4-9 1 > 10 3 Tumor Grade G1 8 G2 7 G3 21 Molecular Subtype ER+/HER2- 26 ER-/HER2- 4 HER2+ 6 DCIS Absent 6 Present 30 LVI Absent 21 Present 14 Unknown 1 Inter-lesion distance*, cm < 2 10 ≥ 2 16 Unknown 10 Number of lesions 2 22 > 2 14
Table 4. Summary of patient and tumor characteristics.
*When there were more than 2 lesions, the largest inter-lesion distance was taken into
Whenever lesion size allowed, we interrogated multiple geographically distinct samples per lesion, leading to a total of 171 investigated tumor samples. Sequencing was carried out at a median exonic coverage of 178X. Overall, 474 somatic mutations were identified, corresponding to 145 and 55 unique coding non-silent somatic substitutions and indels, respectively, across all samples. The list of identified aberrations can be found in Table S4 of the online version of the manuscript (http://onlinelibrary.wiley.com/doi/10.1002/path.4540/suppinfo )
Figure 3. Validation of the mutations using the alternative sequencing platform.
Fifty mutations could not be detected above the predefined cut-off despite sufficient coverage at that specific genomic location. Of interest, this additional sequencing step also identified the presence of 52 additional mutations in samples where the mutation was previously undetected though present in other sample(s) from the same patient (referred to as ‘present’ in Table S4 of the online version of the manuscript).
We observed a median of three mutations (range 1–27) per patient. The number of mutations detected per patient correlated neither with the number of lesions nor with the number of samples sequenced (Figures 4A and 4B). Of note, among the 141 validated unique mutations, 62 (44%) were identified as oncogenic and thus susceptible to having contributed to the development of the cancer.
A B
Figure 4A and B. Boxplots of mutational burden per patient in terms of (A) the number of lesions and (B)
Identification of inter-lesion heterogeneity
By comparing the mutations identified in the various samples and lesions of a single patient, we arbitrarily classified the patients into three groups: those for which all samples from all lesions carried the same mutations (11/36 patients, 31%, Figure 5); those with both common and private mutations (13/36 patients, 36%, Figure 6); and those with no single mutation in common among all samples from the investigated lesions (12/36 patients, 33%, Figure 7). These groups will further be referred to as ‘homogeneous’, ‘intermediate’, and ‘heterogeneous’, respectively.
Figure 5. Distribution of non-silent substitutions and indels in the "homogeneous" MFBC group.
Figure 6. Distribution of non-silent substitutions and indels in the "intermediate" MFBC group.
In the heterogeneous group, most cancers were characterized by the presence of different oncogenic mutations in the investigated lesions. In six patients, oncogenic PIK3CA mutations were only present in one of the lesions. A similar observation was found for oncogenic TP53 (n=3), GATA3 (n=3), and PTEN (n=2) mutations. We further demonstrated that the PTEN mutation resulted in the loss of PTEN staining by immunohistochemistry (IHC) only in the lesion carrying the mutation, as exemplified for patient PD13774 in Figure 7B. Finally, we observed a case of possible convergent evolution for patient PD4877, with each lesion carrying a different TP53 oncogenic mutation (all three samples from the first lesion carried the R196* mutation, whereas one sample from the other lesion was characterized by the R273H variant).
Figure 7. Distribution of non-silent substitutions and indels in the "heterogeneous" MFBC group.
We also explored the association of the group in which the patients were classified with the number of lesions, samples and mutations. There was no significant association between the number of samples and lesions that were sequenced per patient and the group to which they were categorized (Figures 8A and B).
A B
Figure 8A and B. Boxplots of the number of (A) interrogated lesions and (B) samples per patient
in terms of the group of MFBC.
Figure 9. Boxplot of the mutational burden per patient in terms of the group of MFBC.
Association of inter-lesion heterogeneity with clinico-pathological variables
A B
Figure 10. Inter-lesion heterogeneity and inter-lesion distance.
(A) Boxplot of inter-lesion heterogeneity in terms of oncogenic mutations and largest inter-lesion distance. Here patients were classified in two groups: those sharing oncogenic mutations between their lesions and those only having oncogenic mutations private to some of their lesions. Patients without identified oncogenic mutations were not considered here. (B) Boxplot of inter-lesion heterogeneity in terms of the three groups identified according to the targeted sequencing data considering all mutations, and largest inter-lesion distance.
Low coverage genome-wide comparison of multifocal lesions
heterogeneous group of MFBC tumours, as defined by targeted sequencing.
Figure 11. Genome-wide alterations.
Interestingly, we observed common rearrangements and CNAs between the lesions for all patients, even for those belonging to the heterogeneous groups, implying that different lesions from the same patient are genetically related. Nevertheless, we cannot exclude that some of the common rearrangements might have arisen during mammary gland development and/or ageing. Although the numbers are too small to draw any statistical conclusion, we further observed a higher proportion of private rearrangements and CNAs for patients belonging to the heterogeneous group, suggesting an early divergent parallel evolution of the lesions. Most rearrangements did not involve a known cancer gene. Nevertheless, we observed a tandem duplication involving the oncogene MYC in the second lesion of patient PD4877. The list of validated private and common rearrangements can be found in Table S6 of the online version of the manuscript
(http://onlinelibrary.wiley.com/doi/10.1002/path.4540/suppinfo).
At the copy number level, inter-lesion differences involving cancer-related genes were present, such as, for example, PTEN loss in only one of the lesions of patients PD4877, PD4878, and PD11773, and MYC amplification in one lesion from PD11776 (Figure 12; for additional details, please refer to Tables S7 and S8 of the online version of the manuscript
Figure 12. Genome-wide CNAs.
Log2 based estimate of copy number (Log2 Ratio) aberrations, represented in red, across the patients with
Discussion
Although the numbers in our study are too small to draw definitive conclusions, we did not observe any significant associations between inter-lesion heterogeneity and commonly used clinico-pathological features, with the exception of the inter-lesion distance. Indeed, MFBCs whose lesions shared oncogenic variants were closer to each other than lesions not sharing any oncogenic variant. This observation supports the concept underlying the historical definition of multicentric and multifocal tumours – namely, that lesions in close proximity to each other are more likely to be biologically similar than lesions that are far apart (Salgado et al., 2015).
The fact that lesions from one-third of the MFBCs that we studied harbored distinct oncogenic mutations may also have substantial therapeutic implications in the context of genotype-driven trials. Although these trials are mainly running in the metastatic setting, most of them allow identification of the mutation(s) to be performed in the primary tumour (Zardavas et al., 2014). In the case of MFBC, the identification of the molecular targets, for example PIK3CA and PTEN, would differ depending on the lesion interrogated. Our results therefore suggest that ideally all lesions from patients with MFBC should be evaluated, in particular when the lesions are relatively distant from each other.
support a common origin in at least two-thirds of MFBCs. Our low-coverage genome-wide analyses, which are limited to a subset of the patients that we studied, suggest that an even greater proportion of MFBCs, including those without any common oncogenic mutations, are clonally related as shown by the presence of some common structural rearrangements. This suggests that multifocal lesions may arise either through intra-mammary spread of the tumor cells or or via hematogenous self-seeding route (Norton., 2011). The intrinsically invasive nature of the cells making up MFBCs might explain the worse prognostic features associated with multifocal compared with unifocal cancers.
PART B: Metastatic ER positive/HER2 negative breast cancer
patients.
Abstract
ER+/HER2- breast cancers constitute the most frequent breast cancer subtype. The molecular landscape of ER+/HER2- relapsed disease is not well characterized. In this study, we aimed to describe the genomic evolution between primary and matched metastatic ER+/HER2- breast cancers after adjuvant therapy. 182 ER+/HER2- metastatic breast cancers patients with long-term follow-up were identified from a single institution (Institut Jules Bordet, Brussels,
This research work is related to the following paper under review at Annals of Oncology:
D. Fumagalli*, T. R. Wilson*, R. Salgado, X. Lu, J. Yu, C. O’Brien, K. Walter, L. Yuh-Huw, C. Criscitiello, I. Laios, V. Jose, D.N. Brown, F. Rothé, M. Maetens, D. Zardavas, P. Savas, D. Larsimont, M. J. Piccart-Gebhart, S. Michiels, M. R. Lackner, C. Sotiriou$, S. Loi$
* These authors contributed equally to the present work. $ Senior co-authors
Belgium). Tumor tissue from primary lesions was available for all patients, with 88 having material available from matched metastatic lesions. Samples were characterized using a 120 mutational hotspot, a 29 gene copy number and a 400 gene expression panels. ESR1 mutations were assayed by droplet digital PCR. Molecular alterations were correlated with overall survival using Cox proportional hazards regression models.
The genomic analysis of primary tumors revealed somatic mutations in PIK3CA, KRAS, AKT1, FGFR3, HRAS and BRAF at frequencies of 41%, 6%, 5%, 2%, 1% and 2%, respectively, and amplification of CCND1, ZNF703, FGFR1, RSF1 and PAK1 at 23%, 19%, 17%, 12% and 11% respectively. Mutations and gene amplifications were largely concordant between primaries and matched metastatses (>84%). ESR1 mutations were found in 9.8% of the metastases but none of the primaries. Thirteen genes, among which ESR1, FOXA1, and HIF1A showed significant differential expression between primary and matched metastases. In primary tumors, the differential expression of 18 genes, among which IDO1, was significantly associated with overall survival (FDR < 0.1).
Introduction
ER positive/HER2 negative breast cancers are the most frequently diagnosed breast cancer subtype and are treated with therapies that impair estrogen signaling, such as tamoxifen, fulvestrant and aromatase inhibitors (Puhalla S et al., 2012). Whilst such therapies are quite effective, a fraction of early-stage patients ultimately relapse.
Gene expression profiling and large-scale genomic analysis have shown that this breast cancer subtype is heterogeneous, both in terms of prognosis and molecular profile (Sotiriou & Pusztai., 2009; Cancer Genome Atlas Network., 2012; Curtis et al., 2012; Ellis et al., 2012; Stephens et al., 2012). As these studies are based mostly on primary cancers, little is known about the changes that occur at the molecular level under the pressure of anti-cancer treatment. In clinical practice, patients are not routinely re-biopsied in metastatic disease, and treatment decisions are often made on the primary, pre-treatment tissue. Considering that novel targeted therapies are under development in breast cancer, a change in the biology of the tumor during adjuvant treatment could limit the validity of such approach.
The main objectives of this research work were:
To characterize the mutation, copy number and expression profiles of cancer-associated genes in a series of 182 ER positive/HER2 negative
metastatic breast cancer patients with available primary (182) and matched metastatic (88) lesions and with long term follow-up data;
To compare the concordance between paired primary and metastatic lesions for the identified aberrations;
To explore the relationship between molecular aberrations and survival.
My contribution to this project involves: - Identification of the study population; - Preparation of the clinical database;
- Samples preparation and pathological characterization; - Coordination of the molecular characterization of the lesions; - Data and results interpretation;
- Manuscript preparation.
Material and Methods
Patient selection
FFPE tumor samples were obtained retrospectively from 182 ER positive/HER2 negative metastatic breast cancer patients who underwent both primary breast cancer surgery and metastatic biopsy at a single center (Institute Jules Bordet, Brussels, Belgium) between 1982 and 2008. The study received local Ethical Committee (EC) approval (EC number: 1772). For all the patients, clinico-pathologic, treatment and long-term follow-up information were available. Tissue from the primary tumor was available for all patients, whilst sufficient material was available for 88 (48%) of the matched metastases.
Figure 13. Consort diagram.
The diagram depicts depicts the molecular characterization and data availability for the study population.
Somatic mutation analysis
Hematoxylin and eosin (H&E) sections were prepared for all samples and were reviewed by a breast pathologist to confirm diagnosis and assess tumor content.
Five-μm sections were cut from each FFPE tumor block and microdissected. DNA was extracted using the QIAamp DNA FFPE tissue kit (Qiagen) after
each sample, 160 ng of DNA was profiled using an in house (Genentech Inc),
high throughput, microfluidic chip-based qPCR somatic hotspot mutation assay (MUT-MAP) run on the BioMark platform (Fluidigm Corp) that detects hotspot mutations in AKT1, BRAF, EGFR, FGFR3, FLT3, HRAS, KIT, KRAS, MET, NRAS and PIK3CA genes(Table 5).
The following Estrogen Receptor 1 (ESR1) common mutations within the ER ligand-binding domain were instead evaluated using digital droplet PCR: Y537C (1980A>G), Y537N (1979 T>A), Y537S (1980 A>C), and D538G (1983 A>G). All assays were performed on the QX200™ Droplet Digital PCR System (BioRad) according to the manufacturer’s instructions.
Additional details can be found in Appendix 4.
PTEN IHC staining
Table 5. List of genes and hotspot mutations assessed in the mutation panel Gene Exon AA mutation Pos Gene Exon AA mutation Pos
EGFR 18 G719S KRAS 2 G12A G719C G12C G719A G12D 19 K745_E749del G12S E746_A750>IP G12R E746_A750del G12V E746_T751>IP G13D E746_T751>I G12F E746_S752>I G13A E746_A750>RP G13C E746_A750del G13R E746_T751del G13S E746_T751>A G13V E746_T751>V 3 Q61H E746_T751>VA Q61H E746_S752>A Q61K E746_S752>V Q61L E746_P753>VS Q61R
L747_A750>P BRAF 15 V600E L747_T751del NRAS 2 G13D L747_T751>Q 3 Q61K E746_S752>D Q61R L747_E749del Q61L
Gene expression profiling
FFPE sections were macrodissected to enrich for neoplastic tissue followed by RNA extraction using the High Pure FFPE RNA Micro Kit (Roche Applied Sciences) according to the manufacturer’s instructions. Gene expression was subsequently determined using the NanoString nCounter Analysis System on a custom designed 400-gene panel tailored for breast cancer (Table 6). Gene expression was dichotomized by median gene expression. Molecular subtypes were defined using the PAM50 predictors as described in more details in the Appendix 4.
BAG2 CD86 ENTPD3 HMGCS2 LY6E PDGFRA SERPINB13 TTL BCL-XL CD8b ENTPD5 HPCAL1 LYN PDGFRB SERPINE2 TUBB2A BCL2 CDC20 EPHA4 HPK1 MAML2 PDK1 SFRP1 TWIST1 BDCA1 CDC25A EPPK1 HRAS MAP2K1 PDZK1IP1 SFRP4 TYMS BDCA2 CDC45L ERBB2 HSPA4L MAP2K2 PFDN4 SHCBP1 UBE2C BF CDC6 ERBB3 ICAM1 MAP2K4 PFN2 SLC39A6 UBE2T BGN CDCA1 ERBB4 ICOS MAP3K1 PGR SNAI1 VAV3 BIM CDCA5 EREG IDO MAPK1 PHF19 SNAI2 VEGFA BIRC5 CDCA7 ESR1 IFI27 MAPK3 PHGDH SORL1 VIM BIT1 CDCA7L ETS1 IFNg MAPT PHLDA1 SPQR1214 XBP1 BLVRA CDCP1 EXO1 IGF1 MASTL PHLPP1 SPRY2 YPEL2 BMF CDH1 FANCA IGF1R MCL1 PIK3CA SQLE ZBTB10 BMP2 CDH2 FANCI IGF2 MCM10 PIK3CB SRC ZEB1 BRCA1 CDH3 FAP IGFBP2 MDM2 PIK3IP1 SRD5A1 ZEB2 BRCA2 CDK1 FASN IL10 MELK PIK3R1 ST3GAL2 ZNF703 BTC CDK4 FGF1 IL17 MET PIK3R2 STAT1 ACTB BTG3 CDK6 FGF2 IL4 MIA PIM1 STAT3 AL-137727-1 BUB1 CDKN1A FGFR1 IL6 MKI67 PIP STEAP1 CLTC CAPN2 CDKN1B FGFR2 INPP4B MLPH PLEK2 STMN1 GAPDH CAV1 CDKN1C FGFR3 IRF8 MME PMAIP1 T-bet GUSb CAV2 CDKN2C FGFR4 IRS1 MMP11 PMM2 TACSTD2 PPIA CCL18 CENPF FLJ10587 IRS2 MMP2 PNKP TCF3 RPLP0 CCL19 CENPM FN1 ITGA5 MTAP PPM1D TCF4 TFRC CCL2 CEP55 FOSL1 ITK MTHFD1 PPP2R2A TDP1 TUBB CCL21 CHEK1 FOXA1 JAK1 MTHFD1L PSD3 TDP2 UBC CCL22 CLDN1 FOXC1 KIAA1102 MTHFD2 PSIP1 TEK VPS33B
Table 6. List of genes assessed in the 400-gene expression panel
Copy Number Alterations
CDKN1B STK11 PIK3CA AKT1 MYC CCND1 FGFR2 ZNF703 FGFR1 HIF1A IGF1R RPS6KB1 SOX2 PPP2R2A ERBB2 AURKA MAP2K4 CCNE1 RB1 AKT3 MET INPP4B RSF1 PAK1 MDM2 EGFR PHLPP1 PHLPP2 AKT2
KIF18B Housekeeping gene ATF2 Housekeeping gene RPPH1 Housekeeping gene GPR107 Housekeeping gene MRPL50 Housekeeping gene AR Y chromosome control gene RPP30 Housekeeping gene MURC Housekeeping gene GPR15 Housekeeping gene ZNF80 Housekeeping gene OR4X1 Housekeeping gene HSPA5 Housekeeping gene ADRB2 Housekeeping gene
Identification of genes differentially expressed between primary and metastatic tumors
Expression of individual genes was compared between primary and metastatic tumors using a linear mixed effect model to compare the log-transformed expression level of each gene, with subject as a random effect to adjust for within-subject variability. The Benjamini–Hochberg procedure (Benjamini et al., 1995) was used to adjust for multiple comparisons by controlling the false discovery rate. Genes with an adjusted p-value < 0.05 were considered significant.
Comparison of biomarker prevalence between primary and metastatic tumors and between molecular subtypes
Identification of genes associated with overall survival
Results
Patient and sample characteristics
Characteristics N (%) Characteristics N (%) Clinical Parameters at Diagnosis Age Site Metastatic Biopsy Bone 23 (26.1) < 50 53 (29.1) Skin 18 (20.5) ≥ 50 129 (70.9) Lymph Nodes 12 (13.6) Menopause Liver 10 (11.4) Pre- 50 (27.5) Ovary 8 (9) Post- 119 (65.4) Pleura 4 (4.5) Unknown 13 (7.1) Soft Tissue 5 (5.7)
Tumor Size Other 8 (9)
≤ 2cm 62 (34.1) Time Metastatic Biopsy At first relapse 55 (62.5) 2 - 5 cm 91 (50) On treatment 31 (35.2) >5cm 15 (8.2) Unknown 2 (22.7) Unknown 14 (7.7) Pathology of Metastatic Tumor Ki67 Node <14 38 (43.2%) Negative 42 (23.1) ≥14 35 (39.7) Positive 133 (73.1) Unknown 15 (17) Unknown 7 (3.8) ER Pathology at Diagnosis (Primary Tumor) Histotype Positive 80 (90.9) IDC 141 (77.5) Negative 7 (8) ILC 26 (14.3) Unknown 1 (1.1) IDC + ILC 12 (6.6) HER2
Other 2 (1.1) Positive 7 (8) Unknown 1 (0.5) Negative 80 (90.9)
Grade Unknown 1 (1.1)
G1 17 (9.3) Treatment Adjuvant Treatment G2 105 (57.7) Untreated 8 (4.4%) G3 46 (25.3) Only ET 50 (27.5) Unknown 14 (7.7) Only CT 22 (12.1) Ki67 ET+CT 89 (48.9) <14 113 (62.1) Not Applicable 9 (4.9) ≥14 52 (28.6) Unknown 4 (2.2) Unknown 17 (9.3) First 2 lines of met treatment
ER Untreated 4 (2.2) Positive 180 (98.9) Only ET 57 (31.3) Negative 2 (1.1) Only CT 25 (13.7) HER2 ET –> CT 45 (24.7) Negative 179 (98.4) CT –> ET 17 (9.3) Unknown 3 (1.6) Other/UK 34 (18.6)
Mutation profiling and PTEN status of primary and metastatic ER positive breast cancers
The 11-genes hotspot mutation panel was assessed in all the available samples (Table 9, Figure 14). The most frequently mutated gene was PIK3CA with 41.5% of the primary and 33.3% of the metastatic samples carrying a mutation. In this series, few genes were mutated at frequencies higher than previously reported (Forbes et al., 2011). KRAS, AKT1, FGFR3, HRAS and BRAF were in fact respectively mutated in 5.8%, 5.4%, 2.4%, 1.5% and 1.9% of the primary and in 6.5%, 3%, 1.8%, 1.7% and 0% of the metastatic samples.
Variable Type Primary Metastatic raw p-value FDR p-value
ER positive IHC 98.9% (180/182) 92% (80/87) 0.006 0.11 Ki67>14% IHC 30.9% (51/165) 48.6% (35/72) 0.012 0.12 PTEN null IHC 4.6% (8/175) 3.5% (3/85) >0.99 >0.99 PIK3CA Mutation 41.5% (68/164) 33.3% (24/72) 0.25 0.59 KRAS Mutation 5.8% (8/139) 6.5% (4/62) >0.99 >0.99 ESR1 Mutation 0% (0/37) 10.8% (4/37) 0.11 0.55 AKT1 Mutation 5.4% (8/148) 3% (2/67) 0.73 0.99 FGFR3 Mutation 2.4% (3/126) 1.8% (1/57) >0.99 >0.99 HRAS Mutation 1.5% (2/134) 1.7% (1/58) >0.99 >0.99 BRAF Mutation 1.9% (3/155) 0% (0/67) 0.56 0.88 CCND1 CNV 22.7% (34/150) 14.3% (10/70) 0.2 0.59 ZNF703 CNV 19.3% (29/150) 12.9% (9/70) 0.26 0.59 FGFR1 CNV 17.3% (26/150) 15.7% (11/70) 0.85 >0.99 PAK1 CNV 11.3% (17/150) 8.6% (6/70) 0.64 0.94 RSF1 CNV 12% (18/150) 5.7% (4/70) 0.23 0.59 ERBB2 CNV 4.7% (7/150) 7.1% (5/70) 0.53 0.88 CDKN1B CNV 7.3% (11/150) 0% (0/70) 0.018 0.12 RPS6KB1 CNV 5.3% (8/150) 2.9% (2/70) 0.51 0.88 MYC CNV 5.3% (8/150) 1.4% (1/70) 0.28 0.59
Figure 14. Landscape of molecular aberrations in primary and metastatic ER positive/HER2 negative breast cancers.
Comparing matched primary and metastatic pairs, an overall concordance rate of 86.8% was observed for PIK3CA mutations (Table 10, Figure 15). In five cases (7.4%), the mutation was not found in the metastatic sample and in four cases (5.9%) the mutation was found in the metastasis but was undetected in the primary. Consistent with previous reports, loss of PTEN was observed in 8 out of 175 primary (4.6%) (Schleifman et al., 2014 a) and in 3 out of 85 metastatic (3.5%) samples. In 4 cases, this alteration co-existed with a PIK3CA mutation. In matched pairs, the concordance rate for PTEN loss was 95%.
Figure 15. Evolution of somatic mutations and CN alterations in matched primary and metastatic samples.
Red denotes the acquisition of a mutation or CN amplification in the metastatic tumor. Dark blue denotes the loss of a mutation or a CN amplification in the metastatic tumor. Light blue denotes both primary and metastatic tissue being mutation positive or CN amplified. White
Detection of ESR1 mutations in metastatic tumors
