A deep learning approach for staging embryonic tissue isolates with small data.

Machine learning approaches are becoming increasingly widespread and are now present in most areas of research. Their recent surge can be explained in part due to our ability to generate and store enormous amounts of data with which to train these models. The requirement for large training sets is also responsible for limiting further potential applications of machine learning, particularly in fields where data tend to be scarce such as developmental biology. However, recent research seems to indicate that machine learning and Big Data can sometimes be decoupled to train models with modest amounts of data. In this work we set out to train a CNN-based classifier to stage zebrafish tail buds at four different stages of development using small information-rich data sets. Our results show that two and three dimensional convolutional neural networks can be trained to stage developing zebrafish tail buds based on both morphological and gene expression confocal microscopy images, achieving in each case up to 100% test accuracy scores. Importantly, we show that high accuracy can be achieved with data set sizes of under 100 images, much smaller than the typical training set size for a convolutional neural net. Furthermore, our classifier shows that it is possible to stage isolated embryonic structures without the need to refer to classic developmental landmarks in the whole embryo, which will be particularly useful to stage 3D culture in vitro systems such as organoids. We hope that this work will provide a proof of principle that will help dispel the myth that large data set sizes are always required to train CNNs, and encourage researchers in fields where data are scarce to also apply ML approaches.

Oncogenic mTOR signalling recruits myeloid-derived suppressor cells to promote tumour initiation.

Myeloid-derived suppressor cells (MDSCs) play critical roles in primary and metastatic cancer progression. MDSC regulation is widely variable even among patients harbouring the same type of malignancy, and the mechanisms governing such heterogeneity are largely unknown. Here, integrating human tumour genomics and syngeneic mammary tumour models, we demonstrate that mTOR signalling in cancer cells dictates a mammary tumour's ability to stimulate MDSC accumulation through regulating G-CSF. Inhibiting this pathway or its activators (for example, FGFR) impairs tumour progression, which is partially rescued by restoring MDSCs or G-CSF. Tumour-initiating cells (TICs) exhibit elevated G-CSF. MDSCs reciprocally increase TIC frequency through activating Notch in tumour cells, forming a feedforward loop. Analyses of primary breast cancers and patient-derived xenografts corroborate these mechanisms in patients. These findings establish a non-canonical oncogenic role of mTOR signalling in recruiting pro-tumorigenic MDSCs and show how defined cancer subsets may evolve to promote and depend on a distinct immune microenvironment.

Upregulation of EGFR signaling is correlated with tumor stroma remodeling and tumor recurrence in FGFR1-driven breast cancer.

INTRODUCTION: Despite advances in early detection and adjuvant targeted therapies, breast cancer is still the second most common cause of cancer mortality among women. Tumor recurrence is one of the major contributors to breast cancer mortality. However, the mechanisms underlying this process are not completely understood. In this study, we investigated the mechanisms of tumor dormancy and recurrence in a preclinical mouse model of breast cancer. METHODS: To elucidate the mechanisms driving tumor recurrence, we employed a transplantable Wnt1/inducible fibroblast growth factor receptor (FGFR) 1 mouse mammary tumor model and utilized an FGFR specific inhibitor, BGJ398, to study the recurrence after treatment. Histological staining was performed to analyze the residual tumor cells and tumor stroma. Reverse phase protein array was performed to compare primary and recurrent tumors to investigate the molecular mechanisms leading to tumor recurrence. RESULTS: Treatment with BGJ398 resulted in rapid tumor regression, leaving a nonpalpable mass of dormant tumor cells organized into a luminal and basal epithelial layer similar to the normal mammary gland, but surrounded by dense stroma with markedly reduced levels of myeloid-derived tumor suppressor cells (MDSCs) and decreased tumor vasculature. Following cessation of treatment the tumors recurred over a period of 1 to 4 months. The recurrent tumors displayed dense stroma with increased collagen, tenascin-C expression, and MDSC infiltration. Activation of the epidermal growth factor receptor (EGFR) pathway was observed in recurrent tumors, and inhibition of EGFR with lapatinib in combination with BGJ398 resulted in a significant delay in tumor recurrence accompanied by reduced stroma, yet there was no difference observed in initial tumor regression between the groups treated with BGJ398 alone or in combination with lapatinib. CONCLUSION: These studies have revealed a correlation between tumor recurrence and changes of stromal microenvironment accompanied by altered EGFR signaling.

Fibroblast growth factor receptor signaling is essential for normal mammary gland development and stem cell function.

Fibroblast growth factor (FGF) signaling plays an important role in embryonic stem cells and adult tissue homeostasis, but the function of FGFs in mammary gland stem cells is less well defined. Both FGFR1 and FGFR2 are expressed in basal and luminal mammary epithelial cells (MECs), suggesting that together they might play a role in mammary gland development and stem cell dynamics. Previous studies have demonstrated that the deletion of FGFR2 resulted only in transient developmental defects in branching morphogenesis. Using a conditional deletion strategy, we investigated the consequences of FGFR1 deletion alone and then the simultaneous deletion of both FGFR1 and FGFR2 in the mammary epithelium. FGFR1 deletion using a keratin 14 promoter-driven Cre-recombinase resulted in an early, yet transient delay in development. However, no reduction in functional outgrowth potential was observed following limiting dilution transplantation analysis. In contrast, a significant reduction in outgrowth potential was observed upon the deletion of both FGFR1 and FGFR2 in MECs using adenovirus-Cre. Additionally, using a fluorescent reporter mouse model to monitor Cre-mediated recombination, we observed a competitive disadvantage following transplantation of both FGFR1/R2-null MECs, most prominently in the basal epithelial cells. This correlated with the complete loss of the mammary stem cell repopulating population in the FGFR1/R2-attenuated epithelium. FGFR1/R2-null MECs were partially rescued in chimeric outgrowths containing wild-type MECs, suggesting the potential importance of paracrine mechanisms involved in the maintenance of the basal epithelial stem cells. These studies document the requirement for functional FGFR signaling in mammary stem cells during development.

Activation of Wnt signaling by chemically induced dimerization of LRP5 disrupts cellular homeostasis.

Wnt signaling is crucial for a variety of biological processes, including body axis formation, planar polarity, stem cell maintenance and cellular differentiation. Therefore, targeted manipulation of Wnt signaling in vivo would be extremely useful. By applying chemical inducer of dimerization (CID) technology, we were able to modify the Wnt co-receptor, low-density lipoprotein (LDL)-receptor-related protein 5 (LRP5), to generate the synthetic ligand inducible Wnt switch, iLRP5. We show that iLRP5 oligomerization results in its localization to disheveled-containing punctate structures and sequestration of scaffold protein Axin, leading to robust ß-catenin-mediated signaling. Moreover, we identify a novel LRP5 cytoplasmic domain critical for its intracellular localization and casein kinase 1-dependent ß-catenin signaling. Finally, by utilizing iLRP5 as a Wnt signaling switch, we generated the Ubiquitous Activator of ß-catenin (Ubi-Cat) transgenic mouse line. The Ubi-Cat line allows for nearly ubiquitous expression of iLRP5 under control of the H-2K(b) promoter. Activation of iLRP5 in isolated prostate basal epithelial stem cells resulted in expansion of p63(+) cells and development of hyperplasia in reconstituted murine prostate grafts. Independently, iLRP5 induction in adult prostate stroma enhanced prostate tissue regeneration. Moreover, induction of iLRP5 in male Ubi-Cat mice resulted in prostate tumor progression over several months from prostate hyperplasia to adenocarcinoma. We also investigated iLRP5 activation in Ubi-Cat-derived mammary cells, observing that prolonged activation results in mammary tumor formation. Thus, in two distinct experimental mouse models, activation of iLRP5 results in disruption of tissue homeostasis, demonstrating the utility of iLRP5 as a novel research tool for determining the outcome of Wnt activation in a precise spatially and temporally determined fashion.

Fibroblast growth factor receptor signaling dramatically accelerates tumorigenesis and enhances oncoprotein translation in the mouse mammary tumor virus-Wnt-1 mouse model of breast cancer.

Fibroblast growth factor (FGF) cooperates with the Wnt/beta-catenin pathway to promote mammary tumorigenesis. To investigate the mechanisms involved in FGF/Wnt cooperation, we genetically engineered a model of inducible FGF receptor (iFGFR) signaling in the context of the well-established mouse mammary tumor virus-Wnt-1 transgenic mouse. In the bigenic mice, iFGFR1 activation dramatically enhanced mammary tumorigenesis. Expression microarray analysis did not show transcriptional enhancement of Wnt/beta-catenin target genes but instead showed a translational gene signature that also correlated with elevated FGFR1 and FGFR2 in human breast cancer data sets. Additionally, iFGFR1 activation enhanced recruitment of RNA to polysomes, resulting in a marked increase in protein expression of several different Wnt/beta-catenin target genes. FGF pathway activation stimulated extracellular signal-regulated kinase and the phosphorylation of key translation regulators both in vivo in the mouse model and in vitro in a human breast cancer cell line. Our results suggest that cooperation of the FGF and Wnt pathways in mammary tumorigenesis is based on the activation of protein translational pathways that result in, but are not limited to, increased expression of Wnt/beta-catenin target genes (at the level of protein translation). Further, they reveal protein translation initiation factors as potential therapeutic targets for human breast cancers with alterations in FGF signaling.