BIOLOGICALLY MEANINGFUL AND CLINICALLY RELEVANT GENE EXPRESSION PROFILE FOR OPTIMAL TREATMENT PLANNING IN BREAST CANCER. [Tesi di dottorato]
Università degli Studi di Milano, 2013-01-31

ABSTRACT Breast cancer is a highly heterogeneous disease from the molecular and clinical point of view. Optimal treatment with the currently available drugs depends therefore on the ability to individually predict treatment. Presently available outcome prediction models are however suboptimal, single predictive variables have limited accuracy, and the actual clinical outcomes remain heterogeneous in any given prognostic group. ER status and HER-2 status are helpful in identifying patients who are not eligible for endocrine or trastuzumab therapies by virtue of their high negative predictive values (NPVs) and high sensitivities. However, only a minority of ER-positive or HER-2-positive patients respond to receptor-targeted therapy. The positive predictive values (PPVs) of these tests are <50%. Moreover, currently there are no accepted molecular predictors of response to various chemotherapeutic drugs. These limitations have driven biomarker research to develop more accurate molecular predictors of clinical outcome. In the present thesis work we report data on an innovative strategy to predict treatment response to conventional therapies by combining in a hierarchical way genomic predictors related to prognosis and to treatment sensitivity and resistance. The strategy takes into account the intrinsic molecular heterogeneity of breast tumors by distinctly developing genomic predictors for each of the three main tumor subtypes defines as: ER+/Her2-, Her2+ and ER-/Her2-. The considered genomic predictors are built based on literature data and results previously obtained at the INT (as in the case of the prognostic role of ISG genes) and are treated as metagenes mainly to facilitate cross-platform comparisons and to stick to pathways with s clear biological role. An important part of the development of the prediction strategy deals with analytical approaches which were optimized to gain more accurate information from FFPE samples. Gene expression profiles used in such approach were obtained from public database, but also from expression studies carried out at INT on two types of samples; fresh frozen and formalin fixed samples. A detailed comparative analysis of technical solutions (Illumina HT 12, Illumina Ref8, Illumina DASL and Affymetrix HG Plus2.0 chips) for optimizing gene expression profiles in FFPE samples with heavily degraded and chemically-modified RNA is reported. A new robust protocol was developed based on linear amplification of RNA under conditions minimizing rRNA amplification, and on use of the Affymetrix HG Plus 2.0 chips. The protocol was tested in a pilot study on 60 samples to estimate the actual percentage of archived FFPE clinical samples which could yield technically acceptable gene expression profiles and to evaluate the biological reliability of gene expression profiles obtained from fixed samples. Reliability was tested by comparing ER-status classifiers developed using FF-derived expression data and testing the classifier on predicting ER status in FFPE and FF dataset. Prediction accuracy between the two types of samples was comparable (FF Cohen’s k 0,92, FFPE Cohen’s k 0.89). With the various tools developed as described above, it was possible to identify a priori in the ER+/Her2- a subset of patients who were predicted (and confirmed by external validation) to have 95% 5-year disease free survival. Such accurate (and validated) prediction was achieved by combining optimized metagenes with clear biological roles in proliferation, ER signaling and immunity and separately analyzing prognostic and predictive information. Work is still in progress for the other molecular subtypes (Her2+ and ER-/Her2-). The role of immune genes was particularly interesting as it definitely added important information to refine prognosis in our strategy, but at the same way raised some questions due to its paradoxical role encompassing both tumor promotion and tumor inhibition. In our gene expression profiling data in node negative breast cancer patients, ISG expression was associated to likelihood to develop distant metastases in patients with ER+/Her2-tumors, but upon validation in larger data set it clearly emerged that the ISG prognostic role was strongly dependent on the molecular subtypes of the tumor (protective in patients with Her2+ tumors, risk-associated in patients with ER+/Her2- tumors, and neutral in those with ER-/Her2- tumors). Furthermore in the clinical setting a high ISG expression did not reflect stronger lymphocyte infiltration, and upon correction for genuine T cell associated genes ISG genes were found to be associated to proliferation. Interestingly in our clinical samples the ISG genes were demonstrated to be expressed by the cancer cells rather than the stroma. To further search for biological mechanism justifying the prognostic role of ISG expression we switched to in vitro co-cultures of epithelial cells with normal fibroblasts of different origin or CAFs. Many different experimental setting either 2D or 3D were adopted and described in detail. The experimental setting played a major role on the experimental results. On the average basal-like cells were more sensitive to fibroblasts promoted effects in terms of cytokine secretion in the culture medium and up-regulation of ISG genes in the epithelial cells. Those cells were characterized by high intrinsic migratory and invasive ability which did not change much upon co-cultures. On the contrary luminal cells were stimulated to growth and gained invasive and migratory abilities after stimulation with fibroblasts and CAFs. Our experiments showed a complicated and sometimes controversial interplay between fibroblasts and epithelial cells which fully justifies the paradoxical role of the immune system in clinical tumors. In conclusion FFPE samples derived from clinical trials are non more an inaccessible source of biological and clinical information, and can be exploited to build true genomic predictor allowing a true personalized cancer treatment. While our prediction model proofed to be effective the role of immune genes and of stroma, although recognized as very important, is still elusive despite the development of different types of heterotypic cells co-culture models.

diritti: info:eu-repo/semantics/openAccess
coordinatore: A. Gianni ; relatore: C. Carlo-Stella ; tutor: V. Cappelletti
CARLO STELLA, CARMELO
GIANNI, ALESSANDRO
Settore BIO/13 - - Biologia Applicata


Tesi di dottorato. | Lingua: Inglese. | Paese: | BID: TD16001014