Glioblastoma (GB) is the most common and aggressive primary brain malignancy, with poor prognosis and a lack of effective therapeutic options. Accumulating evidence suggests that intratumor heterogeneity likely is the key to understanding treatment failure. However, the extent of intratumor heterogeneity as a result of tumor evolution is still poorly understood. To address this, we developed a unique surgical multisampling scheme to collect spatially distinct tumor fragments from 11 GB patients. We present an integrated genomic analysis that uncovers extensive intratumor heterogeneity, with most patients displaying different GB subtypes within the same tumor. Moreover, we reconstructed the phylogeny of the fragments for each patient, identifying copy number alterations in EGFR and CDKN2A/B/p14ARF as early events, and aberrations in PDGFRA and PTEN as later events during cancer progression. We also characterized the clonal organization of each tumor fragment at the singlemolecule level, detecting multiple coexisting cell lineages. Our results reveal the genome-wide architecture of intratumor variability in GB across multiple spatial scales and patient-specific patterns of cancer evolution, with consequences for treatment design.tumor progression | high grade glioma

Summary. In this article, we propose a generalized estimating equations (GEE) approach for correlated ordinal or nominal multinomial responses using a local odds ratios parameterization. Our motivation lies upon observing that: (i) modeling the dependence between correlated multinomial responses via the local odds ratios is meaningful both for ordinal and nominal response scales and (ii) ordinary GEE methods might not ensure the joint existence of the estimates of the marginal regression parameters and of the dependence structure. To avoid (ii), we treat the so-called "working" association vector α as a "nuisance" parameter vector that defines the local odds ratios structure at the marginalized contingency tables after tabulating the responses without a covariate adjustment at each time pair. To estimate α and simultaneously approximate adequately possible underlying dependence structures, we employ the family of association models proposed by Goodman. In simulations, the parameter estimators with the proposed GEE method for a marginal cumulative probit model appear to be less biased and more efficient than those with the independence "working" model, especially for studies having time-varying covariates and strong correlation. Key words:Association models; Generalized estimating equations; Local odds ratios; Longitudinal data analysis; Multinomial responses. IntroductionLiang and Zeger (1986) originally proposed the generalized estimating equations (GEE) method as an extension of generalized linear models to handle longitudinal data. In contrast to ordinary maximum likelihood approaches, the GEE method provides consistent estimators of the marginal regression parameter vector β and of the covariance matrix of those estimates even if α, the parameter vector that describes the correlation/association pattern within the subjects, has been misspecified.Application of the GEE method for correlated multinomial responses with at least three response categories has been in need of further development. One reason relates to difficult issues in parameterizing the association structure in a way that is sensible for categorical response variables, and in particular, is suitable for both nominal and ordinal variables. In the relevant literature, a correlation coefficient Miller, Davis, and Landis, 1993;Parsons, Edmondson, and Gilmour, 2006) and a global odds ratios (Williamson, Kim, and Lipsitz, 1995;Lumley, 1996) parameterization for α have been proposed. The correlation coefficient parameterization is severely restricted by the marginal model even for bivariate multinomial responses, as we show in Section 2, while the use of a global odds ratios parameterization is limited to ordinal responses. To this end, note that the use of the GEE approach of Parsons et al. (2006) is restricted to ordinal responses under a marginal cumulative logistic model. Another difficult issue is

Glioblastoma, the most common and aggressive adult brain tumor, is characterized by extreme phenotypic diversity and treatment failure. Through fluorescence-guided resection, we identified fluorescent tissue in the sub-ependymal zone (SEZ) of patients with glioblastoma. Histologic analysis and genomic characterization revealed that the SEZ harbors malignant cells with tumor-initiating capacity, analogous to cells isolated from the fluorescent tumor mass (T). We observed resistance to supramaximal chemotherapy doses along with differential patterns of drug response between T and SEZ in the same tumor. Our results reveal novel insights into glioblastoma growth dynamics, with implications for understanding and limiting treatment resistance. Cancer Res; 75(1); 194-202. Ó2014 AACR.