Reinforcement learning in dendritic structures

The discovery of binary dendritic events such as local NMDA spikes in dendritic subbranches led to the suggestion that dendritic trees could be computationally equivalent to a 2-layer network of point neurons, with a single output unit represented by the soma, and input units represented by the dendritic branches. Although this interpretation endows a neuron with a high computational power, it is functionally not clear why nature would have preferred the dendritic solution with a single but complex neuron, as opposed to the network solution with many but simple units. We show that the dendritic solution has a distinguished advantage over the network solution when considering different learning tasks. Its key property is that the dendritic branches receive an immediate feedback from the somatic output spike, while in the corresponding network architecture the feedback would require additional backpropagating connections to the input units. Assuming a reinforcement learning scenario we formally derive a learning rule for the synaptic contacts on the individual dendritic trees which depends on the presynaptic activity, the local NMDA spikes, the somatic action potential, and a delayed reinforcement signal. We test the model for two scenarios: the learning of binary classifications and of precise spike timings. We show that the immediate feedback represented by the backpropagating action potential supplies the individual dendritic branches with enough information to efficiently adapt their synapses and to speed up the learning process.

Reinforcement learning in dendritic structures

The discovery of binary dendritic events such as local NMDA spikes in dendritic subbranches led to the suggestion that dendritic trees could be computationally equivalent to a 2-layer network of point neurons, with a single output unit represented by the soma, and input units represented by the dendritic branches. Although this interpretation endows a neuron with a high computational power, it is functionally not clear why nature would have preferred the dendritic solution with a single but complex neuron, as opposed to the network solution with many but simple units. We show that the dendritic solution has a distinguished advantage over the network solution when considering different learning tasks. Its key property is that the dendritic branches receive an immediate feedback from the somatic output spike, while in the corresponding network architecture the feedback would require additional backpropagating connections to the input units. Assuming a reinforcement learning scenario we formally derive a learning rule for the synaptic contacts on the individual dendritic trees which depends on the presynaptic activity, the local NMDA spikes, the somatic action potential, and a delayed reinforcement signal. We test the model for two scenarios: the learning of binary classifications and of precise spike timings. We show that the immediate feedback represented by the backpropagating action potential supplies the individual dendritic branches with enough information to efficiently adapt their synapses and to speed up the learning process.

Alpha4beta1 integrin mediates the recruitment of immature dendritic cells across the blood-brain barrier during experimental autoimmune encephalomyelitis

Dendritic cells (DCs) within the CNS are recognized to play an important role in the effector phase and propagation of the immune response in experimental autoimmune encephalomyelitis (EAE), a mouse model for multiple sclerosis. However, the mechanisms regulating DC trafficking into the CNS still need to be characterized. In this study, we show by performing intravital fluorescence videomicroscopy of the inflamed spinal cord white-matter microvasculature in SJL mice with EAE that immature, and to a lesser extent, LPS-matured, bone marrow-derived DCs efficiently interact with the CNS endothelium by rolling, capturing, and firm adhesion. Immature but not LPS-matured DCs efficiently migrated across the wall of inflamed parenchymal microvessels into the CNS. Blocking alpha4 integrins interfered with the adhesion but not the rolling or capturing of immature and LPS-matured DCs to the CNS microvascular endothelium, inhibiting their migration across the vascular wall. Functional absence of beta1 integrins but not of beta7 integrins or alpha4beta7 integrin similarly reduced the adhesion of immature DCs to the CNS microvascular endothelium, demonstrating that alpha4beta1 but not alpha4beta7 integrin mediates this step of immature DCs interaction with the inflamed blood-brain barrier during EAE. Our study shows that during EAE, especially immature DCs migrate into the CNS, where they may be crucial for the perpetuation of the CNS-targeted autoimmune response. Thus therapeutic targeting of alpha4 integrins affects DC trafficking into the CNS and may therefore lead to the resolution of the CNS autoimmune inflammation by reducing the number of CNS professional APCs.

Porcine circovirus type 2 DNA influences cytoskeleton rearrangements in plasmacytoid and monocyte-derived dendritic cells

Functional disruption of dendritic cells (DC) is an important strategy for viral pathogens to evade host defences. In this context, porcine circovirus type 2 (PCV2), a single-stranded DNA virus, impairs plasmacytoid DC (pDC) and conventional DC activation by certain viruses or Toll-like receptor (TLR) ligands. This inhibitory capacity is associated with the viral DNA, but the impairment does not affect all signalling cascades; TLR7 ligation by small chemical molecules will still induce interleukin-6 (IL-6) and tumour necrosis factor-α secretion, but not interferon-α or IL-12. In this study, the molecular mechanisms by which silencing occurs were investigated. PP2, a potent inhibitor of the Lyn and Hck kinases, produced a similar profile to the PCV2 DNA interference with cytokine secretion by pDC, efficiently inhibiting cell activation induced through TLR9, but not TLR7, ligation. Confocal microscopy and cytometry analysis strongly suggested that PCV2 DNA impairs actin polymerization and endocytosis in pDC and monocyte-derived DC, respectively. Altogether, this study delineates for the first time particular molecular mechanisms involved in PCV2 interference with DC danger recognition, which may be responsible for the virus-induced immunosuppression observed in infected pigs.

Monocyte differentiation toward regulatory dendritic cells is not affected by respiratory syncytial virus-induced inflammatory mediators

Airway epithelial cells were shown to drive the differentiation of monocytes into dendritic cells (DCs) with a suppressive phenotype. In this study, we investigated the impact of virus-induced inflammatory mediator production on the development of DCs. Monocyte differentiation into functional DCs, as reflected by the expression of CD11c, CD123, BDCA-4, and DC-SIGN and the capacity to activate T cells, was similar for respiratory syncytial virus (RSV)-infected and mock-infected BEAS-2B and A549 cells. RSV-conditioned culture media resulted in a partially mature DC phenotype, but failed to up-regulate CD80, CD83, CD86, and CCR7, and failed to release proinflammatory mediators upon Toll-like receptor (TLR) triggering. Nevertheless, these DCs were able to maintain an antiviral response by the release of Type I IFN. Collectively, these data indicate that the airway epithelium maintains an important suppressive DC phenotype under the inflammatory conditions induced by infection with RSV.

Biomedical nanoparticles modulate specific CD4+ T cell stimulation by inhibition of antigen processing in dendritic cells

Understanding how nanoparticles may affect immune responses is an essential prerequisite to developing novel clinical applications. To investigate nanoparticle-dependent outcomes on immune responses, dendritic cells (DCs) were treated with model biomedical poly(vinylalcohol)-coated super-paramagnetic iron oxide nanoparticles (PVA-SPIONs). PVA-SPIONs uptake by human monocyte-derived DCs (MDDCs) was analyzed by flow cytometry (FACS) and advanced imaging techniques. Viability, activation, function, and stimulatory capacity of MDDCs were assessed by FACS and an in vitro CD4+ T cell assay. PVA-SPION uptake was dose-dependent, decreased by lipopolysaccharide (LPS)-induced MDDC maturation at higher particle concentrations, and was inhibited by cytochalasin D pre-treatment. PVA-SPIONs did not alter surface marker expression (CD80, CD83, CD86, myeloid/plasmacytoid DC markers) or antigen-uptake, but decreased the capacity of MDDCs to process antigen, stimulate CD4+ T cells, and induce cytokines. The decreased antigen processing and CD4+ T cell stimulation capability of MDDCs following PVA-SPION treatment suggests that MDDCs may revert to a more functionally immature state following particle exposure.