Dealing with Multi-Dimensional Data and the Burden of Annotation: Easing the Burden of Annotation.

The need for huge data sets represents a bottleneck for the application of artificial intelligence. Substantially fewer annotated target lesions than normal tissues for comparison present an additional problem in the field of pathology. Organic brains overcome these limitations by utilizing large numbers of specialized neural nets arranged in both linear and parallel fashion, with each solving a restricted classification problem. They rely on local Hebbian error corrections as compared to the nonlocal back-propagation used in most artificial neural nets, and leverage reinforcement. For these reasons, even toddlers are able to classify objects after only a few examples. Rather than provide an overview of current AI research in pathology, this review focuses on general strategies for overcoming the data bottleneck. These include transfer learning, zero-shot learning, Siamese networks, one-class models, generative networks, and reinforcement learning. Neither an extensive mathematic background nor advanced programing skills are needed to make these subjects accessible to pathologists. However, some familiarity with the basic principles of deep learning, briefly reviewed here, is expected to be useful in understanding both the current limitations of machine learning and determining ways to address them.

Structural basis for chemokine recognition by a G protein-coupled receptor and implications for receptor activation.

Chemokines orchestrate cell migration for development, immune surveillance, and disease by binding to cell surface heterotrimeric guanine nucleotide-binding protein (G protein)-coupled receptors (GPCRs). The array of interactions between the nearly 50 chemokines and their 20 GPCR targets generates an extensive signaling network to which promiscuity and biased agonism add further complexity. The receptor CXCR4 recognizes both monomeric and dimeric forms of the chemokine CXCL12, which is a distinct example of ligand bias in the chemokine family. We demonstrated that a constitutively monomeric CXCL12 variant reproduced the G protein-dependent and ß-arrestin-dependent responses that are associated with normal CXCR4 signaling and lead to cell migration. In addition, monomeric CXCL12 made specific contacts with CXCR4 that are not present in the structure of the receptor in complex with a dimeric form of CXCL12, a biased agonist that stimulates only G protein-dependent signaling. We produced an experimentally validated model of an agonist-bound chemokine receptor that merged a nuclear magnetic resonance-based structure of monomeric CXCL12 bound to the amino terminus of CXCR4 with a crystal structure of the transmembrane domains of CXCR4. The large CXCL12:CXCR4 protein-protein interface revealed by this structure identified previously uncharacterized functional interactions that fall outside of the classical "two-site model" for chemokine-receptor recognition. Our model suggests a mechanistic hypothesis for how interactions on the extracellular face of the receptor may stimulate the conformational changes required for chemokine receptor-mediated signal transduction.

MIF inhibits monocytic movement through a non-canonical receptor and disruption of temporal Rho GTPase activities in U-937 cells.

Macrophage migration inhibitory factor (MIF) is a pro-inflammatory cytokine that was initially identified by its ability to inhibit the movement of macrophages. Cell migration is a highly complex process involving changes to the cytoskeleton and cell adhesion molecules, and is regulated by the Rho GTPases. A simple model using human monocytic U-937 cells to elicit the classic MIF response was implemented to examine the mechanism of MIF-induced migration inhibition. Our results demonstrate that MIF inhibits migration of these U-937 cells through a non-canonical receptor, CXCR4, in the absence of the putative primary MIF receptor CD74. Migration inhibition is dependent upon a series of temporal perturbations of the activities of the Rho GTPases: initial activation followed by subsequent inactivation of RhoA, inactivation of Rac1, and cyclic activation of Cdc42. MIF-mediated changes in the activities of the Rho GTPases jointly contributed to migration inhibition in these cells. Collectively, these data suggest that the MIF-mediated migration inhibition is mediated by the outcome of G-protein signaling, and in less adherent cells such as those of the monocyte/macrophage lineage, RhoA directly affects net translocation through its ability to induce cell body contraction. These findings demonstrate that CXCR4 can mediate MIF signaling in the absence of CD74 in addition to serving as a MIF co-receptor along with CD74. These results correlate MIF activity to specific and sequential Rho GTPase activity perturbations, and given that CXCR4 functions in numerous processes, suggests potential roles for the modulation of cell movement in those events including development, cell survival and viral infection.

Biophysical profiling of tumor cell lines.

Despite significant differences in genetic profiles, cancer cells share common phenotypic properties, including membrane-associated changes that facilitate invasion and metastasis. The Corning Epic optical biosensor was used to monitor dynamic mass rearrangements within and proximal to the cell membrane in tumor cell lines derived from cancers of the colon, bone, cervix, lung and breast. Data was collected in real time and required no exogenously added signaling moiety (signal-free technology). Cell lines displayed unique profiles over the time-courses: the time-courses all displayed initial signal increases to maximal values, but the rate of increase to those maxima and the value of those maxima were distinct for each cell line. The rate of decline following the maxima also differed among cell lines. There were correlations between the signal maxima and the observed metastatic behavior of the cells in xenograft experiments; for most cell types the cells that were more highly metastatic in mice had lower time-course maxima values, however the reverse was seen in breast cancer cells. The unique profiles of these cell lines and the correlation of at least one profile characteristic with metastatic behavior demonstrate the potential utility of biophysical tumor cell profiling in the study of cancer biology.

The expression of neurokinin-1 and preprotachykinin-1 in breast cancer cells depends on the relative degree of invasive and metastatic potential.

Breast cancer has a predilection for metastasis to the bone marrow. The preprotachykinin-I (PPT-I) gene has a central role in the early migration of breast cancer cells into the bone marrow, making this organ a latent repository of the cancer cells. This study investigated whether the invasive and metastatic potential of breast cancer cells correlate with the expression of the PPT-I gene and the receptors for its peptides, neurokinin-1 (NK-1) and NK-2. The studies compared cells that are non-tumorigenic (MCF12A), low metastatic and invasive potential (MCF7), and sublines of MCF with increased invasive and metastatic potential (LCC1 and LCC2). LCC2, but not LCC1 is tamoxifen resistant. Quantitative RT-PCR showed increased expression of PPT-I, NK-1 and NK-2 mRNA LCC1 and LCC2. MCF7 required stimulation by phorbol ester for NK-1 induction. The levels of NK-2 mRNA were significantly increased in LCC2. Clonogenic assays with specific receptor antagonists showed a predominant role for NK-2 in the proliferation of both LCC1 and LCC2. While the growth rate of LCC1 and LCC2 were similar, the latter showed increased migration. Use of a nude mouse model confirmed higher metastatic potential of LCC2, including increased migration to regions of the endosteum. Overall, these studies show a correlation between three neuroendocrine-related genes: PPT-I, NK-1 and NK-2 and the metastatic potential of specific breast cancer cells. These cells provide a model for future studies on bone marrow metastasis.

Investigation of promoter polymorphisms in the tumor necrosis factor-alpha and interleukin-10 genes in liver transplant patients.

BACKGROUND: Cytokines such as tumor necrosis factor (TNF)-alpha and interleukin (IL)-10 play significant roles in the inflammatory and immune responses that mediate allograft rejection. The presence of a G-->A polymorphism at position -308 in the promoter region of the TNF-alpha gene increased its transcription 6- to 7-fold. A similar polymorphism at position -1082 of the IL-10 promoter results in decreased production of IL-10 protein. In this study we have determined whether the single nucleotide polymorphisms in the promoter regions of the TNF-alpha and IL-10 genes can predict the outcome of the allograft in liver recipients. METHODS: DNA was extracted from whole blood of liver recipients. The genotype of the patients was determined by polymerase chain reaction using sequence-specific primers. The level of TNF-alpha and IL-10 protein was measured by ELISA after stimulation of peripheral blood mononuclear cells with concanavalin A. RESULTS: There was significant correlation between acute cellular rejection and the presence of the -308A polymorphism (P<0.001), with 8 of 13 patients with the TNF-alpha polymorphism having evidence of acute rejection. Cell stimulation studies revealed that the level of TNF-alpha protein produced by patients with liver rejection was significantly higher than for patients without rejection (P=0.001). There were no strong associations between the presence of the IL-10 polymorphisms and rejection (P=0.71). CONCLUSIONS: This study adds to the understanding of the role of cytokine polymorphisms in liver transplants. The data suggest that cytokine promoter polymorphisms may be a risk factor associated with allograft rejection in the liver.