The em p /em -values are from linear regression for qPCR data (-dCt y-axis) and from DESeq2 (vsd y-axis) for sequencing data.(394K, pptx) Additional file 3. of the genes whose expression is modulated by estrogens. Indeed, the presence of functional estrogen response elements in the gene [8C10] underpins the transcriptional activation of by estrogens [11]. Tumor-released VEGF recruits stromal cells and promotes a desmoplastic microenvironment; stromal cells, in turn, provide mitogenic and angiogenic growth factors stimulating both tumor and stromal cell growth [4]. VEGF secreted by stromal cells and acting cooperatively with LTBP1 other factors can substitute for estrogens and foster hormone-independent growth of luminal tumors [5]. At the clinical level, in hormone receptor-positive breast cancer, elevated intratumoral levels of VEGF have been associated with suboptimal responses to hormonal therapies and poorer clinical outcomes [12C14] lending support to the hypothesis that VEGF and angiogenesis may contribute to resistance to endocrine therapies. These preclinical studies set the stage for a pilot, single-institution, single-arm study of preoperative letrozole in combination with bevacizumab in postmenopausal women with hormone receptor-positive breast cancer [15]. In that study (test was used to examine the correlation between response and CTC and CEC values at baseline and at each time point for all patients and patients within each arm. Response was categorized into binary variables, 0 for stable and progressive disease and 1 for partial and complete response while CTC and CEC values were tested as continuous variables. Correlation between response and the changes in CTC and CEC numbers between baseline and other time points and the changes between time points were also examined. Correlations with a value ?0.05 were considered significant. Genomic data analyses Raw sequencing reads were analyzed as described previously [26]. Briefly, reads were aligned to the human whole genome (hg19) requiring perfect matches. Features were created by merging overlapping alignments and total read counts reported for each. To generate a small RNA-based Cetylpyridinium Chloride classifier of treatment response, patients were categorized as responders if they had achieved a pathologic treatment response ?30% and non-responders if they had stable or progressive disease. Pathologic response was assessed by comparing the maximum cumulative diameter of the target lesion(s) at the time of diagnosis as assessed by imaging studies with the size of the tumor in the final surgical pathology. Due to Cetylpyridinium Chloride the small number of patients who achieved pCR or microscopic residual disease, a genomic classifier on the basis of achievement of pCR or microscopic residual disease could not be generated. Differential expression of feature counts was assessed using DESeq2 [27]. From previously generated full-length RNA-seq data on these samples (data not presented), we had quality Cetylpyridinium Chloride control (QC) metrics (fraction of reads mapping to mRNA and cDNA concentration). Significance was assessed using a likelihood ratio test between the full (response variable + QC metrics) and null (QC metrics) models as implemented in DESeq2. We predicted the agreement between qPCR and sequencing data using previously described methods [26]. We evaluated the relative proportions of 3 ends of small RNA features, predicting that those with many, equal proportioned 3 ends would not yield concordant data between qPCR and sequencing measurements. We then selected small RNA features that had low values from the likelihood ratio test and were predicted to yield concordant qPCR and sequencing measurements. We employed LASSO regression on this subset as implemented in the R package glmnet [28], in order to find small RNA features with optimal ability to classify responder vs. non-responder status. A binomial regression model to predict responder status was created from these optimal classifiers, operating on values were calculated using the R package verification. Results Patients and dispositions Patient demographics and tumor characteristics are shown in Table?1. In the study, 75 patients Cetylpyridinium Chloride were randomly assigned, 50 in the Let/Bev arm and 25 in the Let arm (2:1 ratio). All patients received at least 1?cycle of therapy. The median age for the patients enrolled in the Let/Bev arm was 61.4?years (range, 50.4 to 81.9) and 65 for the Let arm (range, 50.4 to 86.3). All patients had an ECOG performance status of 0. The protocol arms were well balanced for race, stage, nodal status, and tumor type. The proportion of patients with grade 2 tumors was higher in the Let arm (76% vs. 58% respectively), while no patient with grade 3 tumors was randomized to the Let arm (16% vs. 0% respectively). Randomization was not stratified for any demographic or disease parameter. Table 1 Patient characteristics value(%)?White20 (80)42 (84)NSD?Black3 (12)6 (12)NSD?Hispanic1 (4)2 (4)NSD?Others1Asian (4)CClinical tumor stage, (%)?IIA9 (36)16 (32)NSD?IIB7 (28)21 (42)NSD?IIIA9 (36)8 (16)NSD?IIIBC5 (10)NSDNodal status, (%)?N09 (36)21 (42)NSD?N+16 (64)29 (58)NSDTumor type, n (%)?Invasive ductal carcinoma16 (64)29 (58)NSD?Invasive lobular carcinoma5 (20)13* (26)NSD?Invasive mixed carcinoma4 (16)7 (14)NSD?OthersC1mucinous (2)NSDHistologic grade, (%)?Grade I6 (24)13 (26)NSD?Grade II19 (76)29 (58)0.042?Grade IIIC8 (16)0.025 Open in a separate window *Includes one patient Cetylpyridinium Chloride with concurrent separate invasive.