Deep learning for automated scoring of immunohistochemically stained tumour tissue sections - Validation across tumour types based on patient outcomes.

We aimed to develop deep learning (DL) models to detect protein expression in immunohistochemically (IHC) stained tissue-sections, and to compare their accuracy and performance with manually scored clinically relevant proteins in common cancer types. Five cancer patient cohorts (colon, two prostate, breast, and endometrial) were included. We developed separate DL models for scoring IHC-stained tissue-sections with nuclear, cytoplasmic, and membranous staining patterns. For training, we used images with annotations of cells with positive and negative staining from the colon cohort stained for Ki-67 and PMS2 (nuclear model), the prostate cohort 1 stained for PTEN (cytoplasmic model) and ß-catenin (membranous model). The nuclear DL model was validated for MSH6 in the colon, MSH6 and PMS2 in the endometrium, Ki-67 and CyclinB1 in prostate, and oestrogen and progesterone receptors in the breast cancer cohorts. The cytoplasmic DL model was validated for PTEN and Mapre2, and the membranous DL model for CD44 and Flotillin1, all in prostate cohorts. When comparing the results of manual and DL scores in the validation sets, using manual scores as the ground truth, we observed an average correct classification rate of 91.5 % (76.9-98.5 %) for the nuclear model, 85.6 % (73.3-96.6 %) for the cytoplasmic model, and 78.4 % (75.5-84.3 %) for the membranous model. In survival analyses, manual and DL scores showed similar prognostic impact, with similar hazard ratios and p-values for all DL models. Our findings demonstrate that DL models offer a promising alternative to manual IHC scoring, providing efficiency and reproducibility across various data sources and markers.

The NEOLETRIB trial: neoadjuvant treatment with Letrozole and Ribociclib in ER-positive, HER2-negative breast cancer.

Chemotherapy is used as neoadjuvant therapy for all subgroups of breast cancer, including ER-positive, and HER2-negative cases. However, studies have suggested that using aromatase inhibitors combined with CDK4/6-inhibitors might be an appropriate alternative in selected patients. Thus, the NEOLETRIB trial evaluates the response of ER-positive, HER2-negative luminal A/B breast cancer to the combination of letrozole and ribociclib in the neoadjuvant setting. Comprehensive molecular biology procedures, including sequential single-cell RNA-sequencing of tumor biopsies, are performed during 6 months of treatment with extensive biobanking of blood samples, tumor biopsies and gut microbiome specimens. Our findings will hopefully contribute to an improved selection of patients who may benefit from this drug combination and give new insights into the intra-tumoral changes during this treatment.Trial registration number: NCT05163106 (ClinicalTrials.gov).

[Box: see text].

Computational model predicts patient outcomes in Luminal B breast cancer treated with endocrine therapy and CDK4/6 inhibition.

PURPOSE: Development of a computational biomarker to predict, prior to treatment, the response to CDK4/6 inhibition (CDK4/6i) in combination with endocrine therapy in patients with breast cancer. EXPERIMENTAL DESIGN: A mechanistic mathematical model that accounts for protein signaling and drug mechanisms of action was developed and trained on extensive, publicly available data from breast cancer cell lines. The model was built to provide a patient-specific response score based on the expression of six genes (CCND1, CCNE1, ESR1, RB1, MYC and CDKN1A). The model was validated in five independent cohorts of 148 patients in total with early-stage or advanced breast cancer treated with endocrine therapy and CDK4/6i. Response was measured either by evaluating Ki67 levels and PAM50 risk of relapse (ROR) after neoadjuvant treatment or by evaluating progression-free survival (PFS). RESULTS: The model showed significant association with patient´s outcomes in all five cohorts. The model predicted high Ki67 (area under the curve; AUC (95% confidence interval) of 0.80 (0.64 - 0.92), 0.81 (0.60 - 1.00) and 0.80 (0.65 - 0.93)) and high PAM50 ROR (AUC of 0.78 (0.64 - 0.89)). This observation was not obtained in patients treated with chemotherapy. In the other cohorts, patient stratification based on the model prediction was significantly associated with PFS (hazard ratio=2.92 (95% CI 1.08 - 7.86), p=0.034 and HR=2.16 (1.02 4.55), p=0.043). CONCLUSION: A mathematical modeling approach accurately predicts patient outcome following CDK4/6i plus endocrine therapy, which marks a step towards more personalized treatments in patients with Luminal B breast cancer.

Benign breast tumors may arise on different immunological backgrounds.

Benign breast tumors are a nonthreatening condition defined as abnormal cell growth within the breast without the ability to invade nearby tissue. However, benign lesions hold valuable biological information that can lead us toward better understanding of tumor biology. In this study, we have used two pathway analysis algorithms, Pathifier and gene set variation analysis (GSVA), to identify biological differences between normal breast tissue, benign tumors and malignant tumors in our clinical dataset. Our results revealed that one-third of all pathways that were significantly different between benign and malignant tumors were immune-related pathways, and 227 of them were validated by both methods and in the METABRIC dataset. Furthermore, five of these pathways (all including genes involved in cytokine and interferon signaling) were related to overall survival in cancer patients in both datasets. The cellular moieties that contribute to immune differences in malignant and benign tumors were analyzed using the deconvolution tool, CIBERSORT. The results showed that levels of some immune cells were specifically higher in benign than in malignant tumors, and this was especially the case for resting dendritic cells and follicular T-helper cells. Understanding the distinct immune profiles of benign and malignant breast tumors may aid in developing noninvasive diagnostic methods to differentiate between them in the future.

Coagulation factor V in breast cancer: a p53-regulated tumor suppressor and predictive marker for treatment response to chemotherapy.

BACKGROUND: Patients with cancer are at an increased risk of developing coagulation complications, and chemotherapy treatment increases the risk. Tumor progression is closely linked to the hemostatic system. Breast cancer tumors express coagulation factor V (FV), an essential factor in blood coagulation. The functional role of FV during treatment with chemotherapy is poorly understood and was explored in this study. OBJECTIVES: We aimed to investigate the role of FV in breast cancer progression by exploring associations with treatment response, gene regulation, and the functional effects of FV. METHODS: The receiver operating characteristic plotter was used to explore the predictive value of FV mRNA (F5) expression for treatment with FEC (5-fluorouracil, anthracycline, and cyclophosphamide). Breast cancer cohorts were analyzed to study treatment response to FEC. The effect of chemotherapy on F5 expression, the regulation of F5, and the functional effects of FV dependent and independent of chemotherapy were studied in breast cancer cell lines. RESULTS: F5 tumor expression was significantly higher in responders to FEC than in nonresponders. In vitro experiments revealed that anthracycline treatment increased the expression of F5. Inhibition and knockdown of p53 reduced the anthracycline-induced F5 expression. Mutation of a p53 half-site (c.158+1541/158+1564) in a luciferase plasmid reduced luciferase activity, suggesting that p53 plays a role in regulating F5. FV overexpression increased apoptosis and reduced proliferation slightly during anthracycline treatment. CONCLUSION: Our study identified F5 as a p53-regulated tumor suppressor candidate and a promising marker for response to chemotherapy. FV may have functional effects that are therapeutically relevant in breast cancer.

The breast cancer coagulome in the tumor microenvironment and its role in prognosis and treatment response to chemotherapy.

BACKGROUND: The procoagulant phenotype in cancer is linked to thrombosis, cancer progression, and immune response. A novel treatment that reduces the risk of both thrombosis and cancer progression without excess bleeding risk remains to be identified. OBJECTIVES: Here, we aimed to broadly investigate the breast tumor coagulome and its relation to prognosis, treatment response to chemotherapy, and the tumor microenvironment. METHODS: Key coagulation-related genes (n = 35) were studied in a Norwegian cohort with tumor (n = 134) and normal (n = 189) tissue and in the Cancer Genome Atlas (n = 1052) data set. We performed gene set variation analysis in the Norwegian cohort, and in the Cancer Genome Atlas cohort, associations with the tumor microenvironment and prognosis were evaluated. Analyses were performed with cBioPortal, Estimation of Stromal and Immune cells in Malignant Tumors Using Expression Data, Tumor Immune Estimation Resource, the integrated repository portal for tumor-immune system interactions, Tumor Immune Single-cell Hub 2, and the receiver operating characteristic plotter. Six independent breast cancer cohorts were used to study the tumor coagulome and treatment response to chemotherapy. RESULTS: Twenty-two differentially expressed coagulation-related genes were identified in breast tumors. Several coagulome factors were correlated with tumor microenvironment characteristics and were expressed by nonmalignant cells in the tumor microenvironment. PLAT and F8 were independent predictors of better overall survival and progression-free survival, respectively. F12 and PLAU were predictors of worse progression-free survival. The PROCR-THBD-PLAT signature showed a promising predictive value (area under the curve, 0.75; 95% CI, 0.69-0.81; P = 3.6 × 10-17) for combination chemotherapy with fluorouracil, epirubicin, and cyclophosphamide. CONCLUSION: The breast tumor coagulome showed potential in prediction of prognosis and chemotherapy response. Cells within the tumor microenvironment are sources of coagulome factors and may serve as therapeutic targets of coagulation factors.

Human thymic putative CD8αα precursors exhibit a biased TCR repertoire in single cell AIRR-seq.

Thymic T cell development comprises T cell receptor (TCR) recombination and assessment of TCR avidity towards self-peptide-MHC complexes presented by antigen-presenting cells. Self-reactivity may lead to negative selection, or to agonist selection and differentiation into unconventional lineages such as regulatory T cells and CD8[Formula: see text] T cells. To explore the effect of the adaptive immune receptor repertoire on thymocyte developmental decisions, we performed single cell adaptive immune receptor repertoire sequencing (scAIRR-seq) of thymocytes from human young paediatric thymi and blood. Thymic PDCD1+ cells, a putative CD8[Formula: see text] T cell precursor population, exhibited several TCR features previously associated with thymic and peripheral ZNF683+ CD8[Formula: see text] T cells, including enrichment of large and positively charged complementarity-determining region 3 (CDR3) amino acids. Thus, the TCR repertoire may partially explain the decision between conventional vs. agonist selected thymocyte differentiation, an aspect of importance for the development of therapies for patients with immune-mediated diseases.

Analysis of Spatial Molecular Data.

Digital analysis of pathology whole-slide images has been recently gaining interest in the context of cancer diagnosis and treatment. In particular, deep learning methods have demonstrated significant potential in supporting pathology analysis, recently detecting molecular traits never before recognized in pathology H&E whole-slide images (WSIs). Alongside these advancements in the digital analysis of WSIs, it is becoming increasingly evident that both spatial and overall tumor heterogeneity may be significant determinants of cancer prognosis and treatment outcome. In this chapter, we describe methods that aim to support these two elements. We describe both an end-to-end deep learning pipeline for producing limited spatial transcriptomics from WSIs with associated bulk gene expression data, as well as an algorithm for quantifying spatial tumor heterogeneity based on the results of this pipeline.

A regulatory network comprising let-7 miRNA and SMUG1 is associated with good prognosis in ER+ breast tumours.

Single-strand selective uracil-DNA glycosylase 1 (SMUG1) initiates base excision repair (BER) of uracil and oxidized pyrimidines. SMUG1 status has been associated with cancer risk and therapeutic response in breast carcinomas and other cancer types. However, SMUG1 is a multifunctional protein involved, not only, in BER but also in RNA quality control, and its function in cancer cells is unclear. Here we identify several novel SMUG1 interaction partners that functions in many biological processes relevant for cancer development and treatment response. Based on this, we hypothesized that the dominating function of SMUG1 in cancer might be ascribed to functions other than BER. We define a bad prognosis signature for SMUG1 by mapping out the SMUG1 interaction network and found that high expression of genes in the bad prognosis network correlated with lower survival probability in ER+ breast cancer. Interestingly, we identified hsa-let-7b-5p microRNA as an upstream regulator of the SMUG1 interactome. Expression of SMUG1 and hsa-let-7b-5p were negatively correlated in breast cancer and we found an inhibitory auto-regulatory loop between SMUG1 and hsa-let-7b-5p in the MCF7 breast cancer cells. We conclude that SMUG1 functions in a gene regulatory network that influence the survival and treatment response in several cancers.

Subtype and cell type specific expression of lncRNAs provide insight into breast cancer.

Long non-coding RNAs (lncRNAs) are involved in breast cancer pathogenesis through chromatin remodeling, transcriptional and post-transcriptional gene regulation. We report robust associations between lncRNA expression and breast cancer clinicopathological features in two population-based cohorts: SCAN-B and TCGA. Using co-expression analysis of lncRNAs with protein coding genes, we discovered three distinct clusters of lncRNAs. In silico cell type deconvolution coupled with single-cell RNA-seq analyses revealed that these three clusters were driven by cell type specific expression of lncRNAs. In one cluster lncRNAs were expressed by cancer cells and were mostly associated with the estrogen signaling pathways. In the two other clusters, lncRNAs were expressed either by immune cells or fibroblasts of the tumor microenvironment. To further investigate the cis-regulatory regions driving lncRNA expression in breast cancer, we identified subtype-specific transcription factor (TF) occupancy at lncRNA promoters. We also integrated lncRNA expression with DNA methylation data to identify long-range regulatory regions for lncRNA which were validated using ChiA-Pet-Pol2 loops. lncRNAs play an important role in shaping the gene regulatory landscape in breast cancer. We provide a detailed subtype and cell type-specific expression of lncRNA, which improves the understanding of underlying transcriptional regulation in breast cancer.

Serglycin Is Involved in TGF-ß Induced Epithelial-Mesenchymal Transition and Is Highly Expressed by Immune Cells in Breast Cancer Tissue.

Serglycin is a proteoglycan highly expressed by immune cells, in which its functions are linked to storage, secretion, transport, and protection of chemokines, proteases, histamine, growth factors, and other bioactive molecules. In recent years, it has been demonstrated that serglycin is also expressed by several other cell types, such as endothelial cells, muscle cells, and multiple types of cancer cells. Here, we show that serglycin expression is upregulated in transforming growth factor beta (TGF-ß) induced epithelial-mesenchymal transition (EMT). Functional studies provide evidence that serglycin plays an important role in the regulation of the transition between the epithelial and mesenchymal phenotypes, and it is a significant EMT marker gene. We further find that serglycin is more expressed by breast cancer cell lines with a mesenchymal phenotype as well as the basal-like subtype of breast cancers. By examining immune staining and single cell sequencing data of breast cancer tissue, we show that serglycin is highly expressed by infiltrating immune cells in breast tumor tissue.

Epigenetic alterations at distal enhancers are linked to proliferation in human breast cancer.

Aberrant DNA methylation is an early event in breast carcinogenesis and plays a critical role in regulating gene expression. Here, we perform genome-wide expression-methylation Quantitative Trait Loci (emQTL) analysis through the integration of DNA methylation and gene expression to identify disease-driving pathways under epigenetic control. By grouping the emQTLs using biclustering we identify associations representing important biological processes associated with breast cancer pathogenesis including regulation of proliferation and tumor-infiltrating fibroblasts. We report genome-wide loss of enhancer methylation at binding sites of proliferation-driving transcription factors including CEBP-ß, FOSL1, and FOSL2 with concomitant high expression of proliferation-related genes in aggressive breast tumors as we confirm with scRNA-seq. The identified emQTL-CpGs and genes were found connected through chromatin loops, indicating that proliferation in breast tumors is under epigenetic regulation by DNA methylation. Interestingly, the associations between enhancer methylation and proliferation-related gene expression were also observed within known subtypes of breast cancer, suggesting a common role of epigenetic regulation of proliferation. Taken together, we show that proliferation in breast cancer is linked to loss of methylation at specific enhancers and transcription factor binding and gene activation through chromatin looping.

Multimodal human thymic profiling reveals trajectories and cellular milieu for T agonist selection.

To prevent autoimmunity, thymocytes expressing self-reactive T cell receptors (TCRs) are negatively selected, however, divergence into tolerogenic, agonist selected lineages represent an alternative fate. As thymocyte development, selection, and lineage choices are dependent on spatial context and cell-to-cell interactions, we have performed Cellular Indexing of Transcriptomes and Epitopes by sequencing (CITE-seq) and spatial transcriptomics on paediatric human thymu​​s. Thymocytes expressing markers of strong TCR signalling diverged from the conventional developmental trajectory prior to CD4+ or CD8+ lineage commitment, while markers of different agonist selected T cell populations (CD8αα(I), CD8αα(II), T(agonist), Treg(diff), and Treg) exhibited variable timing of induction. Expression profiles of chemokines and co-stimulatory molecules, together with spatial localisation, supported that dendritic cells, B cells, and stromal cells contribute to agonist selection, with different subsets influencing thymocytes at specific developmental stages within distinct spatial niches. Understanding factors influencing agonist T cells is needed to benefit from their immunoregulatory effects in clinical use.

Assessing heterogeneity in spatial data using the HTA index with applications to spatial transcriptomics and imaging.

MOTIVATION: Tumour heterogeneity is being increasingly recognized as an important characteristic of cancer and as a determinant of prognosis and treatment outcome. Emerging spatial transcriptomics data hold the potential to further our understanding of tumour heterogeneity and its implications. However, existing statistical tools are not sufficiently powerful to capture heterogeneity in the complex setting of spatial molecular biology. RESULTS: We provide a statistical solution, the HeTerogeneity Average index (HTA), specifically designed to handle the multivariate nature of spatial transcriptomics. We prove that HTA has an approximately normal distribution, therefore lending itself to efficient statistical assessment and inference. We first demonstrate that HTA accurately reflects the level of heterogeneity in simulated data. We then use HTA to analyze heterogeneity in two cancer spatial transcriptomics datasets: spatial RNA sequencing by 10x Genomics and spatial transcriptomics inferred from H&E. Finally, we demonstrate that HTA also applies to 3D spatial data using brain MRI. In spatial RNA sequencing, we use a known combination of molecular traits to assert that HTA aligns with the expected outcome for this combination. We also show that HTA captures immune-cell infiltration at multiple resolutions. In digital pathology, we show how HTA can be used in survival analysis and demonstrate that high levels of heterogeneity may be linked to poor survival. In brain MRI, we show that HTA differentiates between normal ageing, Alzheimer's disease and two tumours. HTA also extends beyond molecular biology and medical imaging, and can be applied to many domains, including GIS. AVAILABILITY AND IMPLEMENTATION: Python package and source code are available at: https://github.com/alonalj/hta. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Crosstalk between microRNA expression and DNA methylation drives the hormone-dependent phenotype of breast cancer.

BACKGROUND: Abnormal DNA methylation is observed as an early event in breast carcinogenesis. However, how such alterations arise is still poorly understood. microRNAs (miRNAs) regulate gene expression at the post-transcriptional level and play key roles in various biological processes. Here, we integrate miRNA expression and DNA methylation at CpGs to study how miRNAs may affect the breast cancer methylome and how DNA methylation may regulate miRNA expression. METHODS: miRNA expression and DNA methylation data from two breast cancer cohorts, Oslo2 (n = 297) and The Cancer Genome Atlas (n = 439), were integrated through a correlation approach that we term miRNA-methylation Quantitative Trait Loci (mimQTL) analysis. Hierarchical clustering was used to identify clusters of miRNAs and CpGs that were further characterized through analysis of mRNA/protein expression, clinicopathological features, in silico deconvolution, chromatin state and accessibility, transcription factor binding, and long-range interaction data. RESULTS: Clustering of the significant mimQTLs identified distinct groups of miRNAs and CpGs that reflect important biological processes associated with breast cancer pathogenesis. Notably, two major miRNA clusters were related to immune or fibroblast infiltration, hence identifying miRNAs associated with cells of the tumor microenvironment, while another large cluster was related to estrogen receptor (ER) signaling. Studying the chromatin landscape surrounding CpGs associated with the estrogen signaling cluster, we found that miRNAs from this cluster are likely to be regulated through DNA methylation of enhancers bound by FOXA1, GATA2, and ER-alpha. Further, at the hub of the estrogen cluster, we identified hsa-miR-29c-5p as negatively correlated with the mRNA and protein expression of DNA methyltransferase DNMT3A, a key enzyme regulating DNA methylation. We found deregulation of hsa-miR-29c-5p already present in pre-invasive breast lesions and postulate that hsa-miR-29c-5p may trigger early event abnormal DNA methylation in ER-positive breast cancer. CONCLUSIONS: We describe how miRNA expression and DNA methylation interact and associate with distinct breast cancer phenotypes.

Efficient gene expression signature for a breast cancer immuno-subtype.

MOTIVATION AND BACKGROUND: The patient's immune system plays an important role in cancer pathogenesis, prognosis and susceptibility to treatment. Recent work introduced an immune related breast cancer. This subtyping is based on the expression profiles of the tumor samples. Specifically, one study showed that analyzing 658 genes can lead to a signature for subtyping tumors. Furthermore, this classification is independent of other known molecular and clinical breast cancer subtyping. Finally, that study shows that the suggested subtyping has significant prognostic implications. RESULTS: In this work we develop an efficient signature associated with survival in breast cancer. We begin by developing a more efficient signature for the above-mentioned breast cancer immune-based subtyping. This signature represents better performance with a set of 579 genes that obtains an improved Area Under Curve (AUC). We then determine a set of 193 genes and an associated classification rule that yield subtypes with a much stronger statistically significant (log rank p-value < 2 × 10-4 in a test cohort) difference in survival. To obtain these improved results we develop a feature selection process that matches the high-dimensionality character of the data and the dual performance objectives, driven by survival and anchored by the literature subtyping.

Spatial transcriptomics inferred from pathology whole-slide images links tumor heterogeneity to survival in breast and lung cancer.

Digital analysis of pathology whole-slide images is fast becoming a game changer in cancer diagnosis and treatment. Specifically, deep learning methods have shown great potential to support pathology analysis, with recent studies identifying molecular traits that were not previously recognized in pathology H&E whole-slide images. Simultaneous to these developments, it is becoming increasingly evident that tumor heterogeneity is an important determinant of cancer prognosis and susceptibility to treatment, and should therefore play a role in the evolving practices of matching treatment protocols to patients. State of the art diagnostic procedures, however, do not provide automated methods for characterizing and/or quantifying tumor heterogeneity, certainly not in a spatial context. Further, existing methods for analyzing pathology whole-slide images from bulk measurements require many training samples and complex pipelines. Our work addresses these two challenges. First, we train deep learning models to spatially resolve bulk mRNA and miRNA expression levels on pathology whole-slide images (WSIs). Our models reach up to 0.95 AUC on held-out test sets from two cancer cohorts using a simple training pipeline and a small number of training samples. Using the inferred gene expression levels, we further develop a method to spatially characterize tumor heterogeneity. Specifically, we produce tumor molecular cartographies and heterogeneity maps of WSIs and formulate a heterogeneity index (HTI) that quantifies the level of heterogeneity within these maps. Applying our methods to breast and lung cancer slides, we show a significant statistical link between heterogeneity and survival. Our methods potentially open a new and accessible approach to investigating tumor heterogeneity and other spatial molecular properties and their link to clinical characteristics, including treatment susceptibility and survival.

Loss of progesterone receptor is associated with distinct tyrosine kinase profiles in breast cancer.

PURPOSE: The aim of this study was to assess protein tyrosine kinase profiles in primary breast cancer samples in correlation with the distinct hormone and growth receptor profiles ER, PR, and HER2. EXPERIMENTAL DESIGN: Pamchip® microarrays were used to measure the phosphorylation of 144 tyrosine kinase substrates in 29 ER+ breast cancer samples and cell lines MCF7, BT474 and ZR75-1. mRNA expression data from the METABRIC cohort and publicly available PR chip-sequencing data were used for validation purposes, together with RT-PCR. RESULTS: In ER+ breast tumors and cell lines, we observed that the loss of PR expression correlated to higher kinase activity in samples and cell lines that were HER2-. A number of kinases, representing mostly proteins within the PI3K/AKT pathway, were identified as responsible for the differential phosphorylation between PR- and PR+ in ER+/HER2- tumors. We used the METABRIC cohort to analyze mRNA expression from 977 ER+/HER2- breast cancers. Twenty four kinase-encoding genes were identified as differentially expressed between PR+ and PR-, dividing ER+/HER2- samples in two distinct clusters with significant differences in survival (p < 0.05). Four kinase genes, LCK, FRK, FGFR4, and MST1R, were identified as potential direct targets of PR. CONCLUSIONS: Our results suggest that the PR status has a profound effect on tyrosine kinases, especially for FGFR4 and LCK genes, in ER+/HER2- breast cancer patients. The influence of these genes on the PI3K/AKT signaling pathway may potentially lead to novel drug targets for ER+/PR- breast cancer patients.

Immune phenotype of tumor microenvironment predicts response to bevacizumab in neoadjuvant treatment of ER-positive breast cancer.

Antiangiogenic drugs are potentially a useful supplement to neoadjuvant chemotherapy for a subgroup of patients with human epidermal growth factor receptor 2 (HER2) negative breast cancer, but reliable biomarkers for improved response are lacking. Here, we report on a randomized phase II clinical trial to study the added effect of bevacizumab in neoadjuvant chemotherapy with FEC100 (5-fluorouracil, epirubicin and cyclophosphamide) and taxanes (n = 132 patients). Gene expression from the tumors was obtained before neoadjuvant treatment, and treatment response was evaluated by residual cancer burden (RCB) at time of surgery. Bevacizumab increased the proportion of complete responders (RCB class 0) from 5% to 20% among patients with estrogen receptor (ER) positive tumors (P = .02). Treatment with bevacizumab was associated with improved 8-year disease-free survival (P = .03) among the good responders (RCB class 0 or I). Patients treated with paclitaxel (n = 45) responded better than those treated with docetaxel (n = 21; P = .03). Improved treatment response was associated with higher proliferation rate and an immune phenotype characterized by high presence of classically activated M1 macrophages, activated NK cells and memory activated CD4 T cells. Treatment with bevacizumab increased the number of adverse events, including hemorrhage, hypertension, infection and febrile neutropenia, but despite this, the ECOG status was not affected.