The long and winding road to biomarkers for immunotherapy: a retrospective analysis of samples from patients with triple-negative breast cancer treated with pembrolizumab

BACKGROUND

Immune checkpoint blockade (ICB) in combination with chemotherapy improves outcome of patients with triple-negative breast cancer (TNBC) in metastatic and early settings. The identification of predictive biomarkers able to guide treatment decisions is challenging and currently limited to programmed death-ligand 1 (PD-L1) expression and high tumor mutational burden (TMB) in the advanced setting, with several limitations.

MATERIALS AND METHODS

We carried out a retrospective analysis of clinical-pathological and molecular characteristics of tumor samples from 11 patients with advanced TNBC treated with single-agent pembrolizumab participating in two early-phase clinical trials: KEYNOTE-012 and KEYNOTE-086. Clinical, imaging, pathological [i.e. tumor-infiltrating lymphocytes (TILs), PD-L1 status], RNA sequencing, and whole-exome sequencing data were analyzed. We compared our results with publicly available transcriptomic data from TNBC cohorts from TCGA and METABRIC.

RESULTS

Response to pembrolizumab was heterogeneous: two patients experienced exceptional long-lasting responses, six rapid progressions, and three relatively slower disease progression. Neither PD-L1 nor stromal TILs were significantly associated with response to treatment. Increased TMB values were observed in tumor samples from exceptional responders compared to the rest of the cohort (P = 3.4 × 10-4). Tumors from exceptional responders were enriched in adaptive and innate immune cell signatures. Expression of regulatory T-cell markers (FOXP3, CCR4, CCR8, TIGIT) was mainly observed in tumors from responders except for glycoprotein-A repetitions predominant (GARP), which was overexpressed in tumors from rapid progressors. GARP RNA expression in primary breast tumors from the public dataset was significantly associated with a worse prognosis.

CONCLUSIONS

The wide spectrum of clinical responses to ICB supports that TNBC is a heterogeneous disease. Tumors with high TMB respond better to ICB. However, the optimal cut-off of 10 mutations (mut)/megabase (Mb) may not reflect the complexity of all tumor subtypes, despite its approval as a tumor-agnostic biomarker. Further studies are required to better elucidate the relevance of the tumor microenvironment and its components as potential predictive biomarkers in the context of ICB.

Abstract P3-07-04: Molecular characterization of mucinous breast cancers

Introduction

Mucinous carcinoma, a special histological subtype of breast cancer (BC) which accounts for ˜2% of all invasive breast tumors, is characterized by the presence of extracellular mucin, typically expresses the estrogen receptor (ER), and lacks HER2 amplification. The majority of patients with mucinous BC are older at diagnosis, rarely present axillary lymph node metastases, and are associated with a better prognosis as compared to invasive ductal cancer patients (IDC, formally referred to as “breast cancer of no special type”). So far, it is unknown what is driving the mucinous phenotype of these tumors. Here we interrogated the genomics, transcriptomics, immune infiltration and epigenetics profiles of these tumors, using a retrospective institutional series and a publicly available dataset.

Patients and methods

After central pathology review and DNA extraction, a total of 31 pure mucinous cases from Institut J. Bordet Biobank were included (referred to as IJB cohort). Tumor infiltrating lymphocytes (TILs) were assessed on hematoxylin and eosin (H&E). Low pass whole genome sequencing was conducted to assess ploidy and to detect copy number aberrations (CNAs). A second cohort was analyzed from a publicly available dataset (referred to as BRCA560 cohort), with available centrally reviewed histology subtyping and DNA methylation profiles (Nik-Zainal et al. Nature 2016). 207 ER+ primary BC, 14 of which were mucinous, had HumanMethylation450K methylation profiles. RNAseq was available for a subset of 145 cancers, 13 of which were mucinous. DNA methylation data was processed in R using the minfi package. Beta-values were normalized (preprocessQuantile) and dmpFinder was used to identify differentially methylated positions.

Results

In the IJB cohort, whole genome sequencing revealed that all but two tumors were diploid. The most frequently deleted cancer genes were RB1 (38,1%), CDH1 (23,8%), BRCA2 (38,1%),TP53 (23,8%), MAP2K4 (23,8%), EGFR (28,6%) and PGR (23,8%). In terms of amplifications, only ZNF217 (19,4%) and FGFR1/ZNF703 (9,5%) were observed. These mucinous tumors generally displayed low TIL levels (median=5%). In the BRCA560 cohort, there was no significant difference in the clinico-pathological features of ER+ IDC vs mucinous subtypes. At the genomic level, we identified a lower frequency of PIK3CA mutations as compared to ER+ IDC from the same cohort. At the DNA methylation level, we identified 8013 differentially methylated CpGs (q-val < .05) between IDC and mucinous tumors. The top differentially methylated CpG mapped to MUC2, an extracellular mucin. MUC2 was significantly demethylated in mucinous tumors as compared to IDC (q-val <.001). There was a negative association between methylation level of this CpG and MUC2 expression (rho = - 0.27, p< .001). Finally, MUC2 expression was significantly increased in mucinous tumors as compared to IDC (p< .001).

Conclusion

To the best of our knowledge, this study is the first to report hypomethylation of MUC2 in mucinous carcinoma of the breast, as a possible mechanism of extracellular production of mucin in these tumors. Furthermore, it highlights the minor immune infiltration of these cancers, possibly prevented by the mucin, and sheds light on the few genomic alterations present in these overall stable cancer genomes.

Citation Format: Desmedt C, Veys I, Nguyen B, Leduc S, Bareche Y, Majjaj S, Boeckx B, Lambrechts D, Sotiriou C. Molecular characterization of mucinous breast cancers [abstract]. In: Proceedings of the 2018 San Antonio Breast Cancer Symposium; 2018 Dec 4-8; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2019;79(4 Suppl):Abstract nr P3-07-04.

Abstract P1-05-17: Interrogating the impact of pregnancy on breast cancer biology using DNA copy number profiling

Background

Epidemiological evidence indicates a clear relationship between pregnancy and breast cancer (BC) risk. However, little is known regarding the impact of pregnancy on BC biology. DNA copy number aberrations (CNAs) play an important role in breast carcinogenesis. BC during pregnancy is a rare disease but yet could serve as a good model to study the impact of pregnancy on BC biology.

Methods

We retrospectively included 54 pregnant and 113 non-pregnant BC patients matched for age and stage with complete clinico-pathological, gene expression and 5-year follow-up data. CNAs were assessed using Affymetrix OncoScan FFPE arrays. We identified the CNAs associated with pregnancy using a multivariate logistic regression adjusted for classical clinico-pathological features. We further evaluated their impact on gene expression.

Results

After quality control, CNA profiles were obtained for 38 pregnant and 87 non-pregnant BC patients. We identified 13 regions with copy number gains, 11 of which were more frequently gained in pregnant compared to non-pregnant controls and 5 regions with copy number loss, 3 of which were more frequently lost in pregnant patients (p≤0.05). Of interest, we identified 4 genes previously identified as driver event associated with CNAs in breast cancer (S. Nik-Zainal et al, Nature 2016). AKT1 and CDKN2A/B were more frequently gained in the pregnant compared to the non-pregnant (23.7% vs. 8.0%, p=0.068 and 18.4% vs. 4.6% p=0.036) and ARID1B was less frequently gained in the pregnant cohort (2.6% vs. 13.8%, p=0.02). Interestingly, PAPPA which had been previously identified as a pregnancy-dependent oncogene (Takabatake Y. et al, EMBO Mol Med. 2016) was also more frequently gained in the pregnant compared to the non-pregnant patients (21.1% vs 5.8%, p=0.03). We next evaluated the effect of these CNAs on their own gene expression levels and found that AKT1 and CDKN2A/B CNAs were affected by gene-dosage effect.

Conclusions

In this study, we were able to identify several genomic alterations associated with pregnancy that could further elucidate the impact of pregnancy on BC risk. Moreover, by combining CNAs with gene expression, we were able to identify genes whose expression were associated with CNAs and therefore could be considered potential drivers of this rare disease.

Citation Format: Nguyen B, Brown DN, Rothé F, Desmedt C, Majjaj S, Pruneri G, Peccatori F, Azim Jr HA, Sotiriou C. Interrogating the impact of pregnancy on breast cancer biology using DNA copy number profiling [abstract]. In: Proceedings of the 2016 San Antonio Breast Cancer Symposium; 2016 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2017;77(4 Suppl):Abstract nr P1-05-17.

Abstract P6-07-14: Mutational and transcriptomic characterization of breast cancer (BC) arising in young patients (pts) and during pregnancy and their associations with long-term outcome

Background: BC arising in young age is biologically distinct. Whether diagnosis during pregnancy has an impact on prognosis and tumor biology remains to be elucidated. We report for the first time mutational and transcriptomic profiling of BC arising in young pts and the impact of diagnosis during pregnancy. We also correlate these findings with clinical outcome.

Methods: 65 pts with BC during pregnancy diagnosed at the European Institute of Oncology in the period 1999–2009 were matched to 130 BC pts who were diagnosed and treated at the same period/institute. We screened for 84 somatic hotspot mutations on 17 cancer-related genes using mass spectroscopy-based sequencing (Sequenom). We evaluated the pattern of mutations in the two cohorts and according to BC subtype defined using central immunohistochemistry as follows: Luminal A (ER+, HER2−, Ki67 <14%), Luminal-B (ER+, HER2−, Ki67 >14%), HER2+ (HER2+ irrespective of ER), and triple negative (ER−, PgR−, HER2−). Survival endpoints included distant relapse free survival (DRFS) and overall survival (OS).

Results: Median age at diagnosis was 36 years (range: 28–47). At a median follow-up of 74 months (IQR: 42–96), 44 (23%) and 29 (15%) pts developed a DRFS and OS event respectively. Pts diagnosed during pregnancy had inferior DRFS (HR: 3.2 [1.5–6.7]) and OS (HR: 2.9 [1.1–7.9]) after adjusting for pT, pN, grade, BC subtype, and therapy. Mutational profiling was successful in 97% of pts. A total of 57 hotspot mutations (30%) were detected in 51 pts (15 [23%] pregnant and 36 [28%] controls). The differences in mutations between the two groups are summarized in the table.

Abstract P3-04-10: Comparison between RNA-Seq and Affymetrix gene expression data

Background: Microarrays have revolutionized breast cancer research by enabling gene expression comparisons on a transcriptome-wide scale. They have provided clinically useful information such as the development of the intrinsic classification as well as several gene prognostic signatures. Recently, with the advent of next generation sequencing technology, RNA-Seq has emerged as a promising technology for digital read-out of the transcriptome. Here, we will present a comparison of microarray and RNA-Seq data on a series of 58 breast cancer samples.

Methods: 58 tumors representing the different molecular subtypes [17 triple negative (ER−, PgR−, HER2−); 14 HER2 positive (HER2+); 17 luminal A (ER+, HER2−, histological grade 1), 10 luminal B (ER+, HER2−, histological grade 3)] were obtained from patients recruited between 2007 and 2011 at Jules Bordet Institute. Their RNA was profiled using the Affymetrix HG-U133 Plus 2.0 chips and sequenced with the ILLUMINA platform, producing ∼30 million 2*50 bp paired-end reads per sample. The reads alignment and expression quantification were performed using Tophat 2 and Cufflinks 2, respectively, on the basis of the Ensembl genome annotations hg19. Affymetrix microarray data were normalized using fRMA and probesets were selected using JetSet. A selection of 7 clinically relevant gene signatures was evaluated.

Results: About 16000 genes were retained for analysis. Comparison of Affymetrix and RNA-Seq data revealed that 50%, 25% and 5% of these genes had a Spearman correlation superior to 0.7, 0.82 and 0.92, respectively. The genes with the highest correlation coefficients between Affymetrix and RNA-Seq tended to have medium/high expression levels (median Affymetrix log2 expression of 7.5) consistent with the fact that microarrays have a limited dynamic range compared to RNA-Seq. We observed an excellent correlation between Affymetrix and RNA-Seq for the clinically relevant genes ER (0.97), PgR (0.95) and HER2 (0.90) respectively. We observed, however, a few discordances between ER and PgR quantification by immunohistochemistry and RNA-Seq/Affymetrix. We next measured the consistency of two molecular classifiers, SCMGENE (Haibe-Kains JNCI2012) and PAM50 (Parker JCO2009), between Affymetrix arrays and RNA-seq. A kappa coefficient of 0.93 (CI 0.85–1) and 0.79 (CI 0.66–0.91) was observed for SCMGENE and PAM50, respectively, with the majority of the discrepancies for PAM50 concerning the HER2 subtype. The first generation proliferation-based gene expression signatures showed an excellent correlation between Affymetrix and RNA-Seq: Genomic Grade Index (0.98), Mammaprint (0.98) and Oncotype (0.97). The immune and stroma gene signatures (Desmedt et al. 2008) showed a correlation of 0.97 and 0.79 respectively. The lower correlation for the stroma signature could partly be explained by the fact that RNA-Seq and Affymetrix are measuring different splice variants.

Conclusions: This is to our knowledge the first study to report a systematic comparison between RNA-Seq and Affymetrix HG-U133 Plus2.0 data in breast cancer for individual genes and several prognostic gene signatures. According to these results, RNA-Seq can reliably be used for the quantification of breast cancer transcriptome.

Citation Information: Cancer Res 2012;72(24 Suppl):Abstract nr P3-04-10.

Use of quantitative RT-PCR assay to predict metastases size of sentinel node from breast cancer patients

621

Background: A RT-PCR Assay (GeneSearch, Veridex, LLC), FDA approved and CE marked to detect metastases > 0.2 mm in sentinel lymph nodes (SLNs) of breast cancer patients, has been in clinical use in our institute for 25 months. The assay gives qualitative (negative/positive) results by applying cutoff values to cycle times (Cts) of mammaglobin (MG) and cytokeratin 19 (CK19) (cutoff values = 30 / 31 Cts). This study evaluates if quantitative Ct values can be used to estimate the probable size of nodal metastases. Methods: To date 384 patients have been clinically tested with the BLN Assay at Institut Jules Bordet. An additional 74 patients were tested in a pilot study. Both sets had SLNs of breast cancer patients divided in ∼2 mm pieces. Alternating pieces were homogenized and intra-operatively processed by the assay. The remaining pieces were used for histological analysis (positive when > 0.2 mm). During clinical use, same-surgery complete axillary dissection was performed when the assay was positive (n=59). Results: Performance of the assay's qualitative results against permanent section H&E was similar for both groups (total N = 458), for a total sensitivity 89% (68/76), specificity 95% (364/382) and overall agreement 94%. The marker Ct values are correlated with metastases size as determined by H&E on adjacent node pieces (r = -0.74 for MG and -0.77 for CK19, p < 0.001). Lower Ct values indicate increased likelihood of macrometastases reported by H&E, as described in the table. Additionally, Assay Ct values in the SLN were predictive of metastases in non-SLN nodes. For example when CK19 and MG < 24 Cts, the non-SLN positivity rate jumps from 29% to 58%. Conclusions: The BLN Assay's high intra- operative qualitative performance minimizes the need for second surgeries for complete axillary dissections. Results from this investigational study examining the marker Ct values suggest that the assay may provide valuable individual tumor volume data intra- operatively or post-operatively.

[Table: see text]

[Table: see text]

Clinical validation of a molecular assay for intra-operative detection of metastases in breast sentinel lymph nodes

BACKGROUND

In breast cancer patients, the status of the sentinel lymph nodes (SLNs) has been shown to accurately reflect the presence of metastases in the axillary lymph nodes (ALNs). Intra-operative SLN evaluation by frozen section histology may miss positive cases, leading to a second surgery for complete ALN dissection. Permanent section histology itself has tissue sampling limitations and is partially dependent on pathologist expertise.

METHODS

A prospective study (N=78) was conducted in our institution to validate a new intra-operative molecular assay, the GeneSearch breast lymph node (BLN) assay. This assay quantifies the expression of mammaglobin and cytokeratin-19 genes using quantitative RT-PCR technology to determine SLN status. Fresh SLN sections (2 mm thick) were analyzed alternatively by BLN assay or post-operative histology (haematoxylin-eosin and immunohistochemistry). The subject was considered positive when histology revealed a focus >0.2 mm.

RESULTS

BLN assay results corroborated with histologic results in 75 out of 78 patients for an overall agreement of 96%, a sensitivity of 92%, and a specificity of 97%. The positive and negative predictive values of the BLN assay were of 86% (12/14) and 98% (63/64), respectively. Interestingly, a statistically significant correlation was observed between the metastases' histologic size and both assay markers' expression levels as represented by cycle time to positivity (rho > or = 0.71, all p<0.0001).

CONCLUSIONS

The performance of the BLN assay in identifying nodal metastases >0.2 mm was similar to that of permanent section histology, with the added advantages of an objective and rapid output that could be used for intra-operative decision to remove additional ALN.

Expression cloning of an interferon-inducible 17-kDa membrane protein implicated in the control of cell growth

Interferon-inducible membrane proteins of approximately 17 kDa have been suggested to play a role in the antiproliferative activity of interferons based on (1) their pattern of induction in interferon-sensitive and -resistant cell lines and (2) the ability of a membrane fraction enriched in 17-kDa proteins to inhibit cell growth. To gain insight into the nature of the proteins that mediate the antiproliferative activity of interferons, a monoclonal antibody, 13A5, was generated that reacted specifically with a 17-kDa interferon-inducible cell surface protein. The expression pattern of this 17-kDa protein by human cell lines correlated with sensitivity to the antiproliferative activity of interferons. To obtain information regarding the structure of this protein, the 13A5 antibody was used to screen COS cells transfected with a human cDNA expression library. Sequence analysis of a full-length cDNA clone revealed identity with the 9-27 cDNA, previously isolated on the basis of its interferon inducibility by differential screening. In addition, the 17-kDa protein encoded by the 9-27 gene was shown to be identical to the Leu-13 antigen. Leu-13 was previously identified as a 16-kDa interferon-inducible protein in leukocytes and endothelial cells and is a component of a multimeric complex involved in the transduction of antiproliferative and homotypic adhesion signals. These results suggest a novel level of cellular regulation by interferons involving a membrane protein, encoded by the interferon-inducible 9-27 gene, which associates with other proteins at the cell surface, forming a complex relaying growth inhibitory and aggregation signals.