Neural network

From Wikipedia, the free encyclopedia

https://en.wikipedia.org/wiki/Neural_network

 

Simplified view of a feedforward artificial neural network

A neural network can refer to either a neural circuit of biological neurons (sometimes also called a biological neural network), or a network of artificial neurons or nodes in the case of an artificial neural network. Artificial neural networks are used for solving artificial intelligence (AI) problems; they model connections of biological neurons as weights between nodes. A positive weight reflects an excitatory connection, while negative values mean inhibitory connections. All inputs are modified by a weight and summed. This activity is referred to as a linear combination. Finally, an activation function controls the amplitude of the output. For example, an acceptable range of output is usually between 0 and 1, or it could be −1 and 1.

These artificial networks may be used for predictive modeling, adaptive control and applications where they can be trained via a dataset. Self-learning resulting from experience can occur within networks, which can derive conclusions from a complex and seemingly unrelated set of information.

Overview

A biological neural network is composed of a group of chemically connected or functionally associated neurons. A single neuron may be connected to many other neurons and the total number of neurons and connections in a network may be extensive. Connections, called synapses, are usually formed from axons to dendrites, though dendrodendritic synapses and other connections are possible. Apart from electrical signalling, there are other forms of signalling that arise from neurotransmitter diffusion.

Artificial intelligence, cognitive modelling, and neural networks are information processing paradigms inspired by how biological neural systems process data. Artificial intelligence and cognitive modelling try to simulate some properties of biological neural networks. In the artificial intelligence field, artificial neural networks have been applied successfully to speech recognition, image analysis and adaptive control, in order to construct software agents (in computer and video games) or autonomous robots.

Historically, digital computers evolved from the von Neumann model, and operate via the execution of explicit instructions via access to memory by a number of processors. On the other hand, the origins of neural networks are based on efforts to model information processing in biological systems. Unlike the von Neumann model, neural network computing does not separate memory and processing.

Neural network theory has served to identify better how the neurons in the brain function and provide the basis for efforts to create artificial intelligence.

History

The preliminary theoretical base for contemporary neural networks was independently proposed by Alexander Bain (1873) and William James (1890). In their work, both thoughts and body activity resulted from interactions among neurons within the brain.

Computer simulation of the branching architecture of the dendrites of pyramidal neurons.

For Bain, every activity led to the firing of a certain set of neurons. When activities were repeated, the connections between those neurons strengthened. According to his theory, this repetition was what led to the formation of memory. The general scientific community at the time was skeptical of Bain's theory because it required what appeared to be an inordinate number of neural connections within the brain. It is now apparent that the brain is exceedingly complex and that the same brain “wiring” can handle multiple problems and inputs.

James's theory was similar to Bain's, however, he suggested that memories and actions resulted from electrical currents flowing among the neurons in the brain. His model, by focusing on the flow of electrical currents, did not require individual neural connections for each memory or action.

C. S. Sherrington (1898) conducted experiments to test James's theory. He ran electrical currents down the spinal cords of rats. However, instead of demonstrating an increase in electrical current as projected by James, Sherrington found that the electrical current strength decreased as the testing continued over time. Importantly, this work led to the discovery of the concept of habituation.

Wilhelm Lenz (1920) and Ernst Ising (1925) created and analyzed the Ising model which is essentially a non-learning artificial recurrent neural network (RNN) consisting of neuron-like threshold elements. In 1972, Shun'ichi Amari made this architecture adaptive. His learning RNN was popularised by John Hopfield in 1982. McCulloch and Pitts (1943) also created a computational model for neural networks based on mathematics and algorithms. They called this model threshold logic. These early models paved the way for neural network research to split into two distinct approaches. One approach focused on biological processes in the brain and the other focused on the application of neural networks to artificial intelligence.

In the late 1940s psychologist Donald Hebb created a hypothesis of learning based on the mechanism of neural plasticity that is now known as Hebbian learning. Hebbian learning is considered to be a 'typical' unsupervised learning rule and its later variants were early models for long term potentiation. These ideas started being applied to computational models in 1948 with Turing's B-type machines.

Farley and Clark (1954) first used computational machines, then called calculators, to simulate a Hebbian network at MIT. Other neural network computational machines were created by Rochester, Holland, Habit, and Duda (1956).

Frank Rosenblatt (1958) created the perceptron, an algorithm for pattern recognition based on a two-layer learning computer network using simple addition and subtraction. With mathematical notation, Rosenblatt also described circuitry not in the basic perceptron, such as the exclusive-or circuit.

Some say that neural network research stagnated after the publication of machine learning research by Marvin Minsky and Seymour Papert (1969). They discovered two key issues with the computational machines that processed neural networks. The first issue was that single-layer neural networks were incapable of processing the exclusive-or circuit. The second significant issue was that computers were not sophisticated enough to effectively handle the long run time required by large neural networks. However, by the time this book came out, methods for training multilayer perceptrons (MLPs) were already known. The first deep learning MLP was published by Alexey Grigorevich Ivakhnenko and Valentin Lapa in 1965. The first deep learning MLP trained by stochastic gradient descent was published in 1967 by Shun'ichi Amari. In computer experiments conducted by Amari's student Saito, a five layer MLP with two modifiable layers learned useful internal representations to classify non-linearily separable pattern classes.

Neural network research was boosted when computers achieved greater processing power. Also key in later advances was the backpropagation algorithm. It is an efficient application of the Leibniz chain rule (1673) to networks of differentiable nodes. It is also known as the reverse mode of automatic differentiation or reverse accumulation, due to Seppo Linnainmaa (1970). The term "back-propagating errors" was introduced in 1962 by Frank Rosenblatt, but he did not have an implementation of this procedure, although Henry J. Kelley had a continuous precursor of backpropagation already in 1960 in the context of control theory. In 1982, Paul Werbos applied backpropagation to MLPs in the way that has become standard.

In the late 1970s to early 1980s, interest briefly emerged in theoretically investigating the Ising model by Wilhelm Lenz (1920) and Ernst Ising (1925) in relation to Cayley tree topologies and large neural networks. In 1981, the Ising model was solved exactly for the general case of closed Cayley trees (with loops) with an arbitrary branching ratio and found to exhibit unusual phase transition behavior in its local-apex and long-range site-site correlations.

The parallel distributed processing of the mid-1980s became popular under the name connectionism. The text by Rumelhart and McClelland (1986) provided a full exposition on the use of connectionism in computers to simulate neural processes.

Neural networks, as used in artificial intelligence, have traditionally been viewed as simplified models of neural processing in the brain, even though the relation between this model and brain biological architecture is debated, as it is not clear to what degree artificial neural networks mirror brain function.

Artificial intelligence

Main article: Artificial neural network

A neural network (NN), in the case of artificial neurons called artificial neural network (ANN) or simulated neural network (SNN), is an interconnected group of natural or artificial neurons that uses a mathematical or computational model for information processing based on a connectionistic approach to computation. In most cases an ANN is an adaptive system that changes its structure based on external or internal information that flows through the network.

In more practical terms neural networks are non-linear statistical data modeling or decision making tools. They can be used to model complex relationships between inputs and outputs or to find patterns in data.

An artificial neural network involves a network of simple processing elements (artificial neurons) which can exhibit complex global behavior, determined by the connections between the processing elements and element parameters. Artificial neurons were first proposed in 1943 by Warren McCulloch, a neurophysiologist, and Walter Pitts, a logician, who first collaborated at the University of Chicago.

One classical type of artificial neural network is the recurrent Hopfield network.

The concept of a neural network appears to have first been proposed by Alan Turing in his 1948 paper Intelligent Machinery in which he called them "B-type unorganised machines".

The utility of artificial neural network models lies in the fact that they can be used to infer a function from observations and also to use it. Unsupervised neural networks can also be used to learn representations of the input that capture the salient characteristics of the input distribution, e.g., see the Boltzmann machine (1983), and more recently, deep learning algorithms, which can implicitly learn the distribution function of the observed data. Learning in neural networks is particularly useful in applications where the complexity of the data or task makes the design of such functions by hand impractical.

Applications

Neural networks can be used in different fields. The tasks to which artificial neural networks are applied tend to fall within the following broad categories:

• Function approximation, or regression analysis, including time series prediction and modeling.
• Classification, including pattern and sequence recognition, novelty detection and sequential decision making.
• Data processing, including filtering, clustering, blind signal separation and compression.

Application areas of ANNs include nonlinear system identification and control (vehicle control, process control), game-playing and decision making (backgammon, chess, racing), pattern recognition (radar systems, face identification, object recognition), sequence recognition (gesture, speech, handwritten text recognition), medical diagnosis, financial applications, data mining (or knowledge discovery in databases, "KDD"), visualization and e-mail spam filtering. For example, it is possible to create a semantic profile of user's interests emerging from pictures trained for object recognition.

Neuroscience

Theoretical and computational neuroscience is the field concerned with the analysis and computational modeling of biological neural systems. Since neural systems are intimately related to cognitive processes and behaviour, the field is closely related to cognitive and behavioural modeling.

The aim of the field is to create models of biological neural systems in order to understand how biological systems work. To gain this understanding, neuroscientists strive to make a link between observed biological processes (data), biologically plausible mechanisms for neural processing and learning (biological neural network models) and theory (statistical learning theory and information theory).

Types of models

Many models are used; defined at different levels of abstraction, and modeling different aspects of neural systems. They range from models of the short-term behaviour of individual neurons, through models of the dynamics of neural circuitry arising from interactions between individual neurons, to models of behaviour arising from abstract neural modules that represent complete subsystems. These include models of the long-term and short-term plasticity of neural systems and its relation to learning and memory, from the individual neuron to the system level.

Connectivity

See also: Brain connectivity estimators

In August 2020 scientists reported that bi-directional connections, or added appropriate feedback connections, can accelerate and improve communication between and in modular neural networks of the brain's cerebral cortex and lower the threshold for their successful communication. They showed that adding feedback connections between a resonance pair can support successful propagation of a single pulse packet throughout the entire network.

Criticism

Historically, a common criticism of neural networks, particularly in robotics, was that they require a large diversity of training samples for real-world operation. This is not surprising, since any learning machine needs sufficient representative examples in order to capture the underlying structure that allows it to generalize to new cases. Dean Pomerleau, in his research presented in the paper "Knowledge-based Training of Artificial Neural Networks for Autonomous Robot Driving," uses a neural network to train a robotic vehicle to drive on multiple types of roads (single lane, multi-lane, dirt, etc.). A large amount of his research is devoted to (1) extrapolating multiple training scenarios from a single training experience, and (2) preserving past training diversity so that the system does not become overtrained (if, for example, it is presented with a series of right turns—it should not learn to always turn right). These issues are common in neural networks that must decide from amongst a wide variety of responses, but can be dealt with in several ways, for example by randomly shuffling the training examples, by using a numerical optimization algorithm that does not take too large steps when changing the network connections following an example, or by grouping examples in so-called mini-batches.

A. K. Dewdney, a former Scientific American columnist, wrote in 1997, "Although neural nets do solve a few toy problems, their powers of computation are so limited that I am surprised anyone takes them seriously as a general problem-solving tool."

Arguments for Dewdney's position are that to implement large and effective software neural networks, much processing and storage resources need to be committed. While the brain has hardware tailored to the task of processing signals through a graph of neurons, simulating even a most simplified form on Von Neumann technology may compel a neural network designer to fill many millions of database rows for its connections—which can consume vast amounts of computer memory and data storage capacity. Furthermore, the designer of neural network systems will often need to simulate the transmission of signals through many of these connections and their associated neurons—which must often be matched with incredible amounts of CPU processing power and time. While neural networks often yield effective programs, they too often do so at the cost of efficiency (they tend to consume considerable amounts of time and money).

Arguments against Dewdney's position are that neural nets have been successfully used to solve many complex and diverse tasks, such as autonomously flying aircraft.

Technology writer Roger Bridgman commented on Dewdney's statements about neural nets:

Neural networks, for instance, are in the dock not only because they have been hyped to high heaven, (what hasn't?) but also because you could create a successful net without understanding how it worked: the bunch of numbers that captures its behaviour would in all probability be "an opaque, unreadable table...valueless as a scientific resource".

In spite of his emphatic declaration that science is not technology, Dewdney seems here to pillory neural nets as bad science when most of those devising them are just trying to be good engineers. An unreadable table that a useful machine could read would still be well worth having.

Although it is true that analyzing what has been learned by an artificial neural network is difficult, it is much easier to do so than to analyze what has been learned by a biological neural network. Moreover, recent emphasis on the explainability of AI has contributed towards the development of methods, notably those based on attention mechanisms, for visualizing and explaining learned neural networks. Furthermore, researchers involved in exploring learning algorithms for neural networks are gradually uncovering generic principles that allow a learning machine to be successful. For example, Bengio and LeCun (2007) wrote an article regarding local vs non-local learning, as well as shallow vs deep architecture.

Some other criticisms came from believers of hybrid models (combining neural networks and symbolic approaches). They advocate the intermix of these two approaches and believe that hybrid models can better capture the mechanisms of the human mind (Sun and Bookman, 1990).^{[full citation needed]}

Recent improvements

While initially research had been concerned mostly with the electrical characteristics of neurons, a particularly important part of the investigation in recent years has been the exploration of the role of neuromodulators such as dopamine, acetylcholine, and serotonin on behaviour and learning.

Biophysical models, such as BCM theory, have been important in understanding mechanisms for synaptic plasticity, and have had applications in both computer science and neuroscience. Research is ongoing in understanding the computational algorithms used in the brain, with some recent biological evidence for radial basis networks and neural backpropagation as mechanisms for processing data.

Computational devices have been created in CMOS for both biophysical simulation and neuromorphic computing. More recent efforts show promise for creating nanodevices for very large scale principal components analyses and convolution. If successful, these efforts could usher in a new era of neural computing that is a step beyond digital computing, because it depends on learning rather than programming and because it is fundamentally analog rather than digital even though the first instantiations may in fact be with CMOS digital devices.

Between 2009 and 2012, the recurrent neural networks and deep feedforward neural networks developed in the research group of Jürgen Schmidhuber at the Swiss AI Lab IDSIA have won eight international competitions in pattern recognition and machine learning. For example, multi-dimensional long short term memory (LSTM) won three competitions in connected handwriting recognition at the 2009 International Conference on Document Analysis and Recognition (ICDAR), without any prior knowledge about the three different languages to be learned.

Variants of the back-propagation algorithm as well as unsupervised methods by Geoff Hinton and colleagues at the University of Toronto can be used to train deep, highly nonlinear neural architectures, similar to the 1980 Neocognitron by Kunihiko Fukushima, and the "standard architecture of vision", inspired by the simple and complex cells identified by David H. Hubel and Torsten Wiesel in the primary visual cortex.

Radial basis function and wavelet networks have also been introduced. These can be shown to offer best approximation properties and have been applied in nonlinear system identification and classification applications.

Deep learning feedforward networks alternate convolutional layers and max-pooling layers, topped by several pure classification layers. Fast GPU-based implementations of this approach have won several pattern recognition contests, including the IJCNN 2011 Traffic Sign Recognition Competition and the ISBI 2012 Segmentation of Neuronal Structures in Electron Microscopy Stacks challenge. Such neural networks also were the first artificial pattern recognizers to achieve human-competitive or even superhuman performance on benchmarks such as traffic sign recognition (IJCNN 2012), or the MNIST handwritten digits problem of Yann LeCun and colleagues at NYU.

Analytical and computational techniques derived from statistical physics of disordered systems, can be extended to large-scale problems, including machine learning, e.g., to analyze the weight space of deep neural networks.

Large-scale brain network

From Wikipedia, the free encyclopedia

https://en.wikipedia.org/wiki/Large-scale_brain_network

Large-scale brain networks (also known as intrinsic brain networks) are collections of widespread brain regions showing functional connectivity by statistical analysis of the fMRI BOLD signal or other recording methods such as EEG, PET and MEG. An emerging paradigm in neuroscience is that cognitive tasks are performed not by individual brain regions working in isolation but by networks consisting of several discrete brain regions that are said to be "functionally connected". Functional connectivity networks may be found using algorithms such as cluster analysis, spatial independent component analysis (ICA), seed based, and others. Synchronized brain regions may also be identified using long-range synchronization of the EEG, MEG, or other dynamic brain signals.

The set of identified brain areas that are linked together in a large-scale network varies with cognitive function. When the cognitive state is not explicit (i.e., the subject is at "rest"), the large-scale brain network is a resting state network (RSN). As a physical system with graph-like properties, a large-scale brain network has both nodes and edges and cannot be identified simply by the co-activation of brain areas. In recent decades, the analysis of brain networks was made feasible by advances in imaging techniques as well as new tools from graph theory and dynamical systems.

Large-scale brain networks are identified by their function and provide a coherent framework for understanding cognition by offering a neural model of how different cognitive functions emerge when different sets of brain regions join together as self-organized coalitions. The number and composition of the coalitions will vary with the algorithm and parameters used to identify them. In one model, there is only the default mode network and the task-positive network, but most current analyses show several networks, from a small handful to 17. The most common and stable networks are enumerated below. The regions participating in a functional network may be dynamically reconfigured.

Disruptions in activity in various networks have been implicated in neuropsychiatric disorders such as depression, Alzheimer's, autism spectrum disorder, schizophrenia, ADHD and bipolar disorder.

Core networks

An example that identified 10 large-scale brain networks from resting state fMRI activity through independent component analysis.

Because brain networks can be identified at various different resolutions and with various different neurobiological properties, there is currently no universal atlas of brain networks that fits all circumstances. The Organization for Human Brain Mapping has the Workgroup for HArmonized Taxonomy of NETworks (WHATNET) group to work towards a consensus regarding network nomenclature. While the work continues, Uddin, Yeo, and Spreng proposed in 2019 that the following six networks should be defined as core networks based on converging evidences from multiple studies to facilitate communication between researchers.

Default Mode (Medial frontoparietal)

Main article: Default mode network

• The default mode network is active when an individual is awake and at rest. It preferentially activates when individuals focus on internally-oriented tasks such as daydreaming, envisioning the future, retrieving memories, and theory of mind. It is negatively correlated with brain systems that focus on external visual signals. It is the most widely researched network.

Salience (Midcingulo-Insular)

Main article: Salience network

• The salience network consists of several structures, including the anterior (bilateral) insula, dorsal anterior cingulate cortex, and three subcortical structures which are the ventral striatum, substantia nigra/ventral tegmental region. It plays the key role of monitoring the salience of external inputs and internal brain events. Specifically, it aids in directing attention by identifying important biological and cognitive events.
• This network includes the ventral attention network, which primarily includes the temporoparietal junction and the ventral frontal cortex of the right hemisphere. These areas respond when behaviorally relevant stimuli occur unexpectedly. The ventral attention network is inhibited during focused attention in which top-down processing is being used, such as when visually searching for something. This response may prevent goal-driven attention from being distracted by non-relevant stimuli. It becomes active again when the target or relevant information about the target is found.

Attention (Dorsal frontoparietal)

Main article: Dorsal attention network

• This network is involved in the voluntary, top-down deployment of attention. Within the dorsal attention network, the intraparietal sulcus and frontal eye fields influence the visual areas of the brain. These influencing factors allow for the orientation of attention.

Control (Lateral frontoparietal)

Main article: Frontoparietal network

• This network initiates and modulates cognitive control and comprises 18 sub-regions of the brain. There is a strong correlation between fluid intelligence and the involvement of the fronto-parietal network with other networks.

• Versions of this network have also been called the central executive (or executive control) network and the cognitive control network.

Sensorimotor or Somatomotor (Pericentral)

Main article: Sensorimotor network

• This network processes somatosensory information and coordinates motion. The auditory cortex may be included.

Visual (Occipital)

Further information: Visual cortex

• This network handles visual information processing.

Other networks

Different methods and data have identified several other brain networks, many of which greatly overlap or are subsets of more well-characterized core networks.

• Limbic
• Auditory
• Right/left executive
• Cerebellar
• Spatial attention
• Language
• Lateral visual
• Temporal
• Visual perception/imagery

Biology of depression

From Wikipedia, the free encyclopedia

https://en.wikipedia.org/wiki/Biology_of_depression 

Scientific studies have found that different brain areas show altered activity in humans with major depressive disorder (MDD), and this has encouraged advocates of various theories that seek to identify a biochemical origin of the disease, as opposed to theories that emphasize psychological or situational causes. Factors spanning these causative groups include nutritional deficiencies in magnesium, vitamin D, and tryptophan with situational origin but biological impact. Several theories concerning the biologically based cause of depression have been suggested over the years, including theories revolving around monoamine neurotransmitters, neuroplasticity, neurogenesis, inflammation and the circadian rhythm. Physical illnesses, including hypothyroidism and mitochondrial disease, can also trigger depressive symptoms.

Neural circuits implicated in depression include those involved in the generation and regulation of emotion, as well as in reward. Abnormalities are commonly found in the lateral prefrontal cortex whose putative function is generally considered to involve regulation of emotion. Regions involved in the generation of emotion and reward such as the amygdala, anterior cingulate cortex (ACC), orbitofrontal cortex (OFC), and striatum are frequently implicated as well. These regions are innervated by a monoaminergic nuclei, and tentative evidence suggests a potential role for abnormal monoaminergic activity.

Genetic factors

Difficulty of gene studies

Historically, candidate gene studies have been a major focus of study. However, as the number of genes reduces the likelihood of choosing a correct candidate gene, Type I errors (false positives) are highly likely. Candidate genes studies frequently possess a number of flaws, including frequent genotyping errors and being statistically underpowered. These effects are compounded by the usual assessment of genes without regard for gene-gene interactions. These limitations are reflected in the fact that no candidate gene has reached genome-wide significance.

Gene candidates

5-HTTLPR

The 5-HTTLPR, or serotonin transporter promoter gene's short allele, has been associated with increased risk of depression; since the 1990s, however, results have been inconsistent. Other genes that have been linked to a gene–environment interaction include CRHR1, FKBP5 and BDNF, the first two of which are related to the stress reaction of the HPA axis, and the latter of which is involved in neurogenesis. Candidate gene analysis of 5-HTTLPR on depression was inconclusive on its effect, either alone or in combination with life stress.

A 2003 study proposed that a gene-environment interaction (GxE) may explain why life stress is a predictor for depressive episodes in some individuals, but not in others, depending on an allelic variation of the serotonin-transporter-linked promoter region (5-HTTLPR). This hypothesis was widely discussed in both the scientific literature and popular media, where it was dubbed the "Orchid gene", but has conclusively failed to replicate in much larger samples, and the observed effect sizes in earlier work are not consistent with the observed polygenicity of depression.

BDNF

BDNF polymorphisms have also been hypothesized to have a genetic influence, but early findings and research failed to replicate in larger samples, and the effect sizes found by earlier estimates are inconsistent with the observed polygenicity of depression.

SIRT1 and LHPP

A 2015 GWAS study in Han Chinese women positively identified two variants in intronic regions near SIRT1 and LHPP with a genome-wide significant association.

Norepinephrine transporter polymorphisms

Attempts to find a correlation between norepinephrine transporter polymorphisms and depression have yielded negative results.

One review identified multiple frequently studied candidate genes. The genes encoding for the 5-HTT and 5-HT_2A receptor were inconsistently associated with depression and treatment response. Mixed results were found for brain-derived neurotrophic factor (BDNF) Val66Met polymorphisms. Polymorphisms in the tryptophan hydroxylase gene was found to be tentatively associated with suicidal behavior. A meta analysis of 182 case controlled genetic studies published in 2008 found Apolipoprotein E epsilon 2 to be protective, and GNB3 825T, MTHFR 677T, SLC6A4 44bp insertion or deletions, and SLC6A3 40 bpVNTR 9/10 genotype to confer risk.

Circadian rhythm

Depression may be related to the same brain mechanisms that control the cycles of sleep and wakefulness.

Depression may be related to abnormalities in the circadian rhythm, or biological clock.

A well synchronized circadian rhythm is critical for maintaining optimal health. Adverse changes and alterations in the circadian rhythm have been associated various neurological disorders and mood disorders including depression.

Sleep

Sleep disturbance is the most prominent symptom in depressive patients. Studies about sleep electroencephalograms have shown characteristic changes in depression such as reductions in non-rapid eye movement sleep production, disruptions of sleep continuity and disinhibition of rapid eye movement (REM) sleep. Rapid eye movement (REM) sleep—the stage in which dreaming occurs—may be quick to arrive and intense in depressed people. REM sleep depends on decreased serotonin levels in the brain stem, and is impaired by compounds, such as antidepressants, that increase serotonergic tone in brain stem structures. Overall, the serotonergic system is least active during sleep and most active during wakefulness. Prolonged wakefulness due to sleep deprivation activates serotonergic neurons, leading to processes similar to the therapeutic effect of antidepressants, such as the selective serotonin reuptake inhibitors (SSRIs). Depressed individuals can exhibit a significant lift in mood after a night of sleep deprivation. SSRIs may directly depend on the increase of central serotonergic neurotransmission for their therapeutic effect, the same system that impacts cycles of sleep and wakefulness.

Light therapy

Research on the effects of light therapy on seasonal affective disorder suggests that light deprivation is related to decreased activity in the serotonergic system and to abnormalities in the sleep cycle, particularly insomnia. Exposure to light also targets the serotonergic system, providing more support for the important role this system may play in depression. Sleep deprivation and light therapy both target the same brain neurotransmitter system and brain areas as antidepressant drugs, and are now used clinically to treat depression. Light therapy, sleep deprivation and sleep time displacement (sleep phase advance therapy) are being used in combination quickly to interrupt a deep depression in people who are hospitalized for MDD (Major Depressive Disorder).

Increased and decreased sleep length appears to be a risk factor for depression. People with MDD sometimes show diurnal and seasonal variation of symptom severity, even in non-seasonal depression. Diurnal mood improvement was associated with activity of dorsal neural networks. Increased mean core temperature was also observed. One hypothesis proposed that depression was a result of a phase shift.

Daytime light exposure correlates with decreased serotonin transporter activity, which may underlie the seasonality of some depression.

Monoamines

Illustration of the major elements in a prototypical synapse. Synapses are gaps between nerve cells. These cells convert their electrical impulses into bursts of chemical relayers, called neurotransmitters, which travel across the synapses to receptors on adjacent cells, triggering electrical impulses to travel down the latter cells.

Monoamines are neurotransmitters that include serotonin, dopamine, norepinephrine, and epinephrine.

Monoamine hypothesis of depression

Many antidepressant drugs acutely increase synaptic levels of the monoamine neurotransmitter, serotonin, but they may also enhance the levels of norepinephrine and dopamine. The observation of this efficacy led to the monoamine hypothesis of depression, which postulates that the deficit of certain neurotransmitters is responsible for depression, and even that certain neurotransmitters are linked to specific symptoms. Normal serotonin levels have been linked to mood and behaviour regulation, sleep, and digestion; norepinephrine to the fight-or-flight response; and dopamine to movement, pleasure, and motivation. Some have also proposed the relationship between monoamines and phenotypes such as serotonin in sleep and suicide, norepinephrine in dysphoria, fatigue, apathy, cognitive dysfunction, and dopamine in loss of motivation and psychomotor symptoms. The main limitation for the monoamine hypothesis of depression is the therapeutic lag between initiation of antidepressant treatment and perceived improvement of symptoms. One explanation for this therapeutic lag is that the initial increase in synaptic serotonin is only temporary, as firing of serotonergic neurons in the dorsal raphe adapt via the activity of 5-HT_1A autoreceptors. The therapeutic effect of antidepressants is thought to arise from autoreceptor desensitization over a period of time, eventually elevating firing of serotonergic neurons.

Monoamine receptors affect phospholipase C and adenylyl cyclase inside of the cell. Green arrows means stimulation and red arrows inhibition. Serotonin receptors are blue, norepinephrine orange, and dopamine yellow. Phospholipase C and adenylyl cyclase start a signaling cascade which turn on or off genes in the cell. Sufficient ATP from mitochondria is required for these downstream signalling events. The 5HT-3 receptor is associated with gastrointestinal adverse effects and has no relationship to the other monoamine receptors.

Serotonin

The serotonin "chemical imbalance" theory of depression, proposed in the 1960s,  is not supported by the available scientific evidence. SSRIs alter the balance of serotonin inside and outside of neurons: their clinical antidepressant effect (which is robust in severe depression) is likely due to more complex changes in neuronal functioning which occur as a downstream consequence of this.

Initial studies of serotonin in depression examined peripheral measures such as the serotonin metabolite 5-Hydroxyindoleacetic acid (5-HIAA) and platelet binding. The results were generally inconsistent, and may not generalize to the central nervous system. However evidence from receptor binding studies and pharmacological challenges provide some evidence for dysfunction of serotonin neurotransmission in depression. Serotonin may indirectly influence mood by altering emotional processing biases that are seen at both the cognitive/behavioral and neural level. Pharmacologically reducing serotonin synthesis, and pharmacologically enhancing synaptic serotonin can produce and attenuate negative affective biases, respectively. These emotional processing biases may explain the therapeutic gap.

Dopamine

While various abnormalities have been observed in dopaminergic systems, results have been inconsistent. People with MDD have an increased reward response to dextroamphetamine compared to controls, and it has been suggested that this results from hypersensitivity of dopaminergic pathways due to natural hypoactivity. While polymorphisms of the D4 and D3 receptor have been implicated in depression, associations have not been consistently replicated. Similar inconsistency has been found in postmortem studies, but various dopamine receptor agonists show promise in treating MDD. There is some evidence that there is decreased nigrostriatal pathway activity in people with melancholic depression (psychomotor retardation). Further supporting the role of dopamine in depression is the consistent finding of decreased cerebrospinal fluid and jugular metabolites of dopamine, as well as post mortem findings of altered Dopamine receptor D3 and dopamine transporter expression. Studies in rodents have supported a potential mechanism involving stress-induced dysfunction of dopaminergic systems.

Catecholamines

A number of lines of evidence indicative of decreased adrenergic activity in depression have been reported. Findings include the decreased activity of tyrosine hydroxylase, decreased size of the locus coeruleus, increased alpha 2 adrenergic receptor density, and decreased alpha 1 adrenergic receptor density. Furthermore, norepinephrine transporter knockout in mice models increases their tolerance to stress, implicating norepinephrine in depression.

One method used to study the role of monoamines is monoamine depletion. Depletion of tryptophan (the precursor of serotonin), tyrosine and phenylalanine (precursors to dopamine) does result in decreased mood in those with a predisposition to depression, but not in persons lacking the predisposition. On the other hand, inhibition of dopamine and norepinephrine synthesis with alpha-methyl-para-tyrosine does not consistently result in decreased mood.

Monoamine oxidase

An offshoot of the monoamine hypothesis suggests that monoamine oxidase A (MAO-A), an enzyme which metabolizes monoamines, may be overly active in depressed people. This would, in turn, cause the lowered levels of monoamines. This hypothesis received support from a PET study, which found significantly elevated activity of MAO-A in the brain of some depressed people. In genetic studies, the alterations of MAO-A-related genes have not been consistently associated with depression. Contrary to the assumptions of the monoamine hypothesis, lowered but not heightened activity of MAO-A was associated with depressive symptoms in adolescents. This association was observed only in maltreated youth, indicating that both biological (MAO genes) and psychological (maltreatment) factors are important in the development of depressive disorders. In addition, some evidence indicates that disrupted information processing within neural networks, rather than changes in chemical balance, might underlie depression.

Limitations

Since the 1990s, research has uncovered multiple limitations of the monoamine hypothesis, and its inadequacy has been criticized within the psychiatric community. For one thing, serotonin system dysfunction cannot be the sole cause of depression. Not all patients treated with antidepressants show improvements despite the usually rapid increase in synaptic serotonin. If significant mood improvements do occur, this is often not for at least two to four weeks. One possible explanation for this lag is that the neurotransmitter activity enhancement is the result of auto receptor desensitization, which can take weeks. Intensive investigation has failed to find convincing evidence of a primary dysfunction of a specific monoamine system in people with MDD. The antidepressants that do not act through the monoamine system, such as tianeptine and opipramol, have been known for a long time. There have also been inconsistent findings with regard to levels of serum 5-HIAA, a metabolite of serotonin. Experiments with pharmacological agents that cause depletion of monoamines have shown that this depletion does not cause depression in healthy people. Another problem that presents is that drugs that deplete monoamines may actually have antidepressants properties. Further, some have argued that depression may be marked by a hyperserotonergic state. Already limited, the monoamine hypothesis has been further oversimplified when presented to the general public.

Receptor binding

As of 2012, efforts to determine differences in neurotransmitter receptor expression or for function in the brains of people with MDD using positron emission tomography (PET) had shown inconsistent results. Using the PET imaging technology and reagents available as of 2012, it appeared that the D1 receptor may be underexpressed in the striatum of people with MDD. 5-HT_1A receptor binding literature is inconsistent; however, it leans towards a general decrease in the mesiotemporal cortex. 5-HT_2A receptor binding appears to be unregulated in people with MDD. Results from studies on 5-HTT binding are variable, but tend to indicate higher levels in people with MDD. Results with D2/D3 receptor binding studies are too inconsistent to draw any conclusions. Evidence supports increased MAO activity in people with MDD, and it may even be a trait marker (not changed by response to treatment). Muscarinic receptor binding appears to be increased in depression, and, given ligand binding dynamics, suggests increased cholinergic activity.

Four meta analyses on receptor binding in depression have been performed, two on serotonin transporter (5-HTT), one on 5-HT_1A, and another on dopamine transporter (DAT). One meta analysis on 5-HTT reported that binding was reduced in the midbrain and amygdala, with the former correlating with greater age, and the latter correlating with depression severity. Another meta-analysis on 5-HTT including both post-mortem and in vivo receptor binding studies reported that while in vivo studies found reduced 5-HTT in the striatum, amygdala and midbrain, post mortem studies found no significant associations. 5-HT_1A was found to be reduced in the anterior cingulate cortex, mesiotemporal lobe, insula, and hippocampus, but not in the amygdala or occipital lobe. The most commonly used 5-HT_1A ligands are not displaced by endogenous serotonin, indicating that receptor density or affinity is reduced. Dopamine transporter binding is not changed in depression.

Emotional processing and neural circuits

Emotional Bias

People with MDD show a number of biases in emotional processing, such as a tendency to rate happy faces more negatively, and a tendency to allocate more attentional resources to sad expressions. Depressed people also have impaired recognition of happy, angry, disgusted, fearful and surprised, but not sad faces. Functional neuroimaging has demonstrated hyperactivity of various brain regions in response to negative emotional stimuli, and hypoactivity in response to positive stimuli. One meta analysis reported that depressed subjects showed decreased activity in the left dorsolateral prefrontal cortex and increased activity in the amygdala in response to negative stimuli. Another meta analysis reported elevated hippocampus and thalamus activity in a subgroup of depressed subjects who were medication naive, not elderly, and had no comorbidities. The therapeutic lag of antidepressants has been suggested to be a result of antidepressants modifying emotional processing leading to mood changes. This is supported by the observation that both acute and subchronic SSRI administration increases response to positive faces. Antidepressant treatment appears to reverse mood congruent biases in limbic, prefrontal, and fusiform areas. dlPFC response is enhanced and amygdala response is attenuated during processing of negative emotions, the former or which is thought to reflect increased top down regulation. The fusiform gyrus and other visual processing areas respond more strongly to positive stimuli with antidepressant treatment, which is thought to reflect the a positive processing bias. These effects do not appear to be unique to serotonergic or noradrenergic antidepressants, but also occur in other forms of treatment such as deep brain stimulation.

Neural circuits

One meta analysis of functional neuroimaging in depression observed a pattern of abnormal neural activity hypothesized to reflect an emotional processing bias. Relative to controls, people with MDD showed hyperactivity of circuits in the salience network (SN), composed of the pulvinar nuclei, the insula, and the dorsal anterior cingulate cortex (dACC), as well as decreased activity in regulatory circuits composed of the striatum and dlPFC.

Rendition of the Limbic-cortical-striatal-pallidal-thalamic circuit as described by Drevets et al. 2008

A neuroanatomical model called the limbic-cortical model has been proposed to explain early biological findings in depression. The model attempts to relate specific symptoms of depression to neurological abnormalities. Elevated resting amygdala activity was proposed to underlie rumination, as stimulation of the amygdala has been reported to be associated with the intrusive recall of negative memories. The ACC was divided into pregenual (pgACC) and subgenual regions (sgACC), with the former being electrophysiologically associated with fear, and the latter being metabolically implicated in sadness in healthy subjects. Hyperactivity of the lateral orbitofrontal and insular regions, along with abnormalities in lateral prefrontal regions was suggested to underlie maladaptive emotional responses, given the regions roles in reward learning. This model and another termed "the cortical striatal model", which focused more on abnormalities in the cortico-basal ganglia-thalamo-cortical loop, have been supported by recent literature. Reduced striatal activity, elevated OFC activity, and elevated sgACC activity were all findings consistent with the proposed models. However, amygdala activity was reported to be decreased, contrary to the limbic-cortical model. Furthermore, only lateral prefrontal regions were modulated by treatment, indicating that prefrontal areas are state markers (i.e.,. dependent upon mood), while subcortical abnormalities are trait markers (i.e., reflect a susceptibility).

Reward

While depression severity as a whole is not correlated with a blunted neural response to reward, anhedonia is directly correlated to reduced activity in the reward system. The study of reward in depression is limited by heterogeneity in the definition and conceptualizations of reward and anhedonia. Anhedonia is broadly defined as a reduced ability to feel pleasure, but questionnaires and clinical assessments rarely distinguish between motivational "wanting" and consummatory "liking". While a number of studies suggest that depressed subjects rate positive stimuli less positively and as less arousing, a number of studies fail to find a difference. Furthermore, response to natural rewards such as sucrose does not appear to be attenuated. General affective blunting may explain "anhedonic" symptoms in depression, as meta analysis of both positive and negative stimuli reveal reduced rating of intensity. As anhedonia is a prominent symptom of depression, direct comparison of depressed with healthy subjects reveals increased activation of the subgenual anterior cingulate cortex (sgACC), and reduced activation of the ventral striatum, and in particular the nucleus accumbens (NAcc) in response to positive stimuli. Although the finding of reduced NAcc activity during reward paradigms is fairly consistent, the NAcc is made up of a functionally diverse range of neurons, and reduced blood-oxygen-level dependent (BOLD) signal in this region could indicate a variety of things including reduced afferent activity or reduced inhibitory output. Nevertheless, these regions are important in reward processing, and dysfunction of them in depression is thought to underlie anhedonia. Residual anhedonia that is not well targeted by serotonergic antidepressants is hypothesized to result from inhibition of dopamine release by activation of 5-HT2C receptors in the striatum. The response to reward in the medial orbitofrontal cortex (OFC) is attenuated in depression, while lateral OFC response is enhanced to punishment. The lateral OFC shows sustained response to absence of reward or punishment, and it is thought to be necessary for modifying behavior in response to changing contingencies. Hypersensitivity in the lOFC may lead to depression by producing a similar effect to learned helplessness in animals.

Elevated response in the sgACC is a consistent finding in neuroimaging studies using a number of paradigms including reward related tasks. Treatment is also associated with attenuated activity in the sgACC, and inhibition of neurons in the rodent homologue of the sgACC, the infralimbic cortex (IL), produces an antidepressant effect. Hyperactivity of the sgACC has been hypothesized to lead to depression via attenuating the somatic response to reward or positive stimuli. Contrary to studies of functional magnetic resonance imaging response in the sgACC during tasks, resting metabolism is reduced in the sgACC. However, this is only apparent when correcting for the prominent reduction in sgACC volume associated with depression; structural abnormalities are evident at a cellular level, as neuropathological studies report reduced sgACC cell markers. The model of depression proposed from these findings by Drevets et al. suggests that reduced sgACC activity results in enhanced sympathetic nervous system activity and blunted HPA axis feedback. Activity in the sgACC may also not be causal in depression, as the authors of one review that examined neuroimaging in depressed subjects during emotional regulation hypothesized that the pattern of elevated sgACC activity reflected increased need to modulate automatic emotional responses in depression. More extensive sgACC and general prefrontal recruitment during positive emotional processing was associated with blunted subcortical response to positive emotions, and subject anhedonia. This was interpreted by the authors to reflect a downregulation of positive emotions by the excessive recruitment of the prefrontal cortex.

Neuroanatomy

While a number of neuroimaging findings are consistently reported in people with major depressive disorder, the heterogeneity of depressed populations presents difficulties interpreting these findings. For example, averaging across populations may hide certain subgroup related findings; while reduced dlPFC activity is reported in depression, a subgroup may present with elevated dlPFC activity. Averaging may also yield statistically significant findings, such as reduced hippocampal volumes, that are actually present in a subgroup of subjects. Due to these issues and others, including the longitudinal consistency of depression, most neural models are likely inapplicable to all depression.

Structural neuroimaging

GMV reductions in MDD and BD

Meta analyses performed using seed-based d mapping have reported grey matter reductions in a number of frontal regions. One meta analysis of early onset general depression reported grey matter reductions in the bilateral anterior cingulate cortex (ACC) and dorsomedial prefrontal cortex (dmPFC). One meta analysis on first episode depression observed distinct patterns of grey matter reductions in medication free, and combined populations; medication free depression was associated with reductions in the right dorsolateral prefrontal cortex, right amygdala, and right inferior temporal gyrus; analysis on a combination of medication free and medicated depression found reductions in the left insula, right supplementary motor area, and right middle temporal gyrus. Another review distinguishing medicated and medication free populations, albeit not restricted to people with their first episode of MDD, found reductions in the combined population in the bilateral superior, right middle, and left inferior frontal gyrus, along with the bilateral parahippocampus. Increases in thalamic and ACC grey matter was reported in the medication free and medicated populations respectively. A meta analysis performed using "activation likelihood estimate" reported reductions in the paracingulate cortex, dACC and amygdala.

Using statistical parametric mapping, one meta analysis replicated previous findings of reduced grey matter in the ACC, medial prefrontal cortex, inferior frontal gyrus, hippocampus and thalamus; however reductions in the OFC and ventromedial prefrontal cortex grey matter were also reported.

Two studies on depression from the ENIGMA consortium have been published, one on cortical thickness, and the other on subcortical volume. Reduced cortical thickness was reported in the bilateral OFC, ACC, insula, middle temporal gyri, fusiform gyri, and posterior cingulate cortices, while surface area deficits were found in medial occipital, inferior parietal, orbitofrontal and precentral regions. Subcortical abnormalities, including reductions in hippocampus and amygdala volumes, which were especially pronounced in early onset depression.

MDD is associated with reduced FA in the ALIC and genu/body of the CC

Multiple meta analysis have been performed on studies assessing white matter integrity using fractional anisotropy (FA). Reduced FA has been reported in the corpus callosum (CC) in both first episode medication naive, and general major depressive populations. The extent of CC reductions differs from study to study. People with MDD who have not taken antidepressants before have been reported to have reductions only in the body of the CC and only in the genu of the CC. On the other hand, general MDD samples have been reported to have reductions in the body of the CC, the body and genu of the CC, and only the genu of the CC. Reductions of FA have also been reported in the anterior limb of the internal capsule (ALIC) and superior longitudinal fasciculus.

Functional neuroimaging

Studies of resting state activity have utilized a number of indicators of resting state activity, including regional homogeneity (ReHO), amplitude of low frequency fluctuations (ALFF), fractional amplitude of low frequency fluctuations (fALFF), arterial spin labeling (ASL), and positron emission tomography measures of regional cerebral blood flow or metabolism.

Studies using ALFF and fALFF have reported elevations in ACC activity, with the former primarily reporting more ventral findings, and the latter more dorsal findings. A conjunction analysis of ALFF and CBF studies converged on the left insula, with previously untreated people having increased insula activity. Elevated caudate CBF was also reported A meta analysis combining multiple indicators of resting activity reported elevated anterior cingulate, striatal, and thalamic activity and reduced left insula, post-central gyrus and fusiform gyrus activity. An activation likelihood estimate (ALE) meta analysis of PET/SPECT resting state studies reported reduced activity in the left insula, pregenual and dorsal anterior cingulate cortex and elevated activity in the thalamus, caudate, anterior hippocampus and amygdala. Compared to the ALE meta analysis of PET/SPECT studies, a study using multi-kernel density analysis reported hyperactivity only in the pulvinar nuclei of the thalamus.

Brain regions

Research on the brains of people with MDD usually shows disturbed patterns of interaction between multiple parts of the brain. Several areas of the brain are implicated in studies seeking to more fully understand the biology of depression:

Subgenual cingulate

Studies have shown that Brodmann area 25, also known as subgenual cingulate, is metabolically overactive in treatment-resistant depression. This region is extremely rich in serotonin transporters and is considered as a governor for a vast network involving areas like hypothalamus and brain stem, which influences changes in appetite and sleep; the amygdala and insula, which affect the mood and anxiety; the hippocampus, which plays an important role in memory formation; and some parts of the frontal cortex responsible for self-esteem. Thus disturbances in this area or a smaller than normal size of this area contributes to depression. Deep brain stimulation has been targeted to this region in order to reduce its activity in people with treatment resistant depression.

Prefrontal cortex

One review reported hypoactivity in the prefrontal cortex of those with depression compared to controls. The prefrontal cortex is involved in emotional processing and regulation, and dysfunction of this process may be involved in the etiology of depression. One study on antidepressant treatment found an increase in PFC activity in response to administration of antidepressants. One meta analysis published in 2012 found that areas of the prefrontal cortex were hypoactive in response to negative stimuli in people with MDD. One study suggested that areas of the prefrontal cortex are part of a network of regions including dorsal and pregenual cingulate, bilateral middle frontal gyrus, insula and superior temporal gyrus that appear to be hypoactive in people with MDD. However the authors cautioned that the exclusion criteria, lack of consistency and small samples limit results.

Amygdala

The amygdala, a structure involved in emotional processing appears to be hyperactive in those with major depressive disorder. The amygdala in unmedicated depressed persons tended to be smaller than in those that were medicated, however aggregate data shows no difference between depressed and healthy persons. During emotional processing tasks right amygdala is more active than the left, however there is no differences during cognitive tasks, and at rest only the left amygdala appears to be more hyperactive. One study, however, found no difference in amygdala activity during emotional processing tasks.

Hippocampus

Atrophy of the hippocampus has been observed during depression, consistent with animal models of stress and neurogenesis.

Stress can cause depression and depression-like symptoms through monoaminergic changes in several key brain regions as well as suppression in hippocampal neurogenesis. This leads to alteration in emotion and cognition related brain regions as well as HPA axis dysfunction. Through the dysfunction, the effects of stress can be exacerbated including its effects on 5-HT. Furthermore, some of these effects are reversed by antidepressant action, which may act by increasing hippocampal neurogenesis. This leads to a restoration in HPA activity and stress reactivity, thus restoring the deleterious effects induced by stress on 5-HT.

The hypothalamic-pituitary-adrenal axis is a chain of endocrine structures that are activated during the body's response to stressors of various sorts. The HPA axis involves three structure, the hypothalamus which release CRH that stimulates the pituitary gland to release ACTH which stimulates the adrenal glands to release cortisol. Cortisol has a negative feedback effect on the pituitary gland and hypothalamus. In people with MDD this often shows increased activation in depressed people, but the mechanism behind this is not yet known. Increased basal cortisol levels and abnormal response to dexamethasone challenges have been observed in people with MDD. Early life stress has been hypothesized as a potential cause of HPA dysfunction. HPA axis regulation may be examined through a dexamethasone suppression tests, which tests the feedback mechanisms. Non-suppression of dexamethasone is a common finding in depression, but is not consistent enough to be used as a diagnostic tool. HPA axis changes may be responsible for some of the changes such as decreased bone mineral density and increased weight found in people with MDD. One drug, ketoconazole, currently under development has shown promise in treating MDD.

Hippocampal Neurogenesis

Reduced hippocampal neurogenesis leads to a reduction in hippocampal volume. A genetically smaller hippocampus has been linked to a reduced ability to process psychological trauma and external stress, and subsequent predisposition to psychological illness. Depression without familial risk or childhood trauma has been linked to a normal hippocampal volume but localised dysfunction.

Animal Models

Main article: Animal models of depression

A number of animal models exist for depression, but they are limited in that depression involves primarily subjective emotional changes. However, some of these changes are reflected in physiology and behavior, the latter of which is the target of many animal models. These models are generally assessed according to four facets of validity; the reflection of the core symptoms in the model; the predictive validity of the model; the validity of the model with regard to human characteristics of etiology; and the biological plausibility.

Different models for inducing depressive behaviors have been utilized; neuroanatomical manipulations such as olfactory bulbectomy or circuit specific manipulations with optogenetics; genetic models such as 5-HT1A knockout or selectively bred animals; models involving environmental manipulation associated with depression in humans, including chronic mild stress, early life stress and learned helplessness. The validity of these models in producing depressive behaviors may be assessed with a number of behavioral tests. Anhedonia and motivational deficits may, for example, be assessed via examining an animal's level of engagement with rewarding stimuli such as sucrose or intracranial self-stimulation. Anxious and irritable symptoms may be assessed with exploratory behavior in the presence of a stressful or novelty environment, such as the open field test, novelty suppressed feeding, or the elevated plus-maze. Fatigue, psychomotor poverty, and agitation may be assessed with locomotor activity, grooming activity, and open field tests.

Animal models possess a number of limitations due to the nature of depression. Some core symptoms of depression, such as rumination, low self-esteem, guilt, and depressed mood cannot be assessed in animals as they require subjective reporting. From an evolutionary standpoint, the behavior correlates of defeats of loss are thought to be an adaptive response to prevent further loss. Therefore, attempts to model depression that seeks to induce defeat or despair may actually reflect adaption and not disease. Furthermore, while depression and anxiety are frequently comorbid, dissociation of the two in animal models is difficult to achieve. Pharmacological assessment of validity is frequently disconnected from clinical pharmacotherapeutics in that most screening tests assess acute effects, while antidepressants normally take a few weeks to work in humans.

Neurocircuits

Regions involved in reward are common targets of manipulation in animal models of depression, including the nucleus accumbens (NAc), ventral tegmental area (VTA), ventral pallidum (VP), lateral habenula (LHb) and medial prefrontal cortex (mPFC). Tentative fMRI studies in humans demonstrate elevated LHb activity in depression. The lateral habenula projects to the RMTg to drive inhibition of dopamine neurons in the VTA during omission of reward. In animal models of depression, elevated activity has been reported in LHb neurons that project to the ventral tegmental area (ostensibly reducing dopamine release). The LHb also projects to aversion reactive mPFC neurons, which may provide an indirect mechanism for producing depressive behaviors. Learned helplessness induced potentiation of LHb synapses are reversed by antidepressant treatment, providing predictive validity. A number of inputs to the LHb have been implicated in producing depressive behaviors. Silencing GABAergic projections from the NAc to the LHb reduces conditioned place preference induced in social aggression, and activation of these terminals induces CPP. Ventral pallidum firing is also elevated by stress induced depression, an effect that is pharmacologically valid, and silencing of these neurons alleviates behavioral correlates of depression. Tentative in vivo evidence from people with MDD suggests abnormalities in dopamine signalling. This led to early studies investigating VTA activity and manipulations in animal models of depression. Massive destruction of VTA neurons enhances depressive behaviors, while VTA neurons reduce firing in response to chronic stress. However, more recent specific manipulations of the VTA produce varying results, with the specific animal model, duration of VTA manipulation, method of VTA manipulation, and subregion of VTA manipulation all potentially leading to differential outcomes. Stress and social defeat induced depressive symptoms, including anhedonia, are associated with potentiation of excitatory inputs to Dopamine D2 receptor-expressing medium spiny neurons (D2-MSNs) and depression of excitatory inputs to Dopamine D1 receptor-expressing medium spiny neurons (D1-MSNs). Optogenetic excitation of D1-MSNs alleviates depressive symptoms and is rewarding, while the same with D2-MSNs enhances depressive symptoms. Excitation of glutaminergic inputs from the ventral hippocampus reduces social interactions, and enhancing these projections produces susceptibility to stress-induced depression. Manipulations of different regions of the mPFC can produce and attenuate depressive behaviors. For example, inhibiting mPFC neurons specifically in the intralimbic cortex attenuates depressive behaviors. The conflicting findings associated with mPFC stimulation, when compared to the relatively specific findings in the infralimbic cortex, suggest that the prelimbic cortex and infralimbic cortex may mediate opposing effects. mPFC projections to the raphe nuclei are largely GABAergic and inhibit the firing of serotonergic neurons. Specific activation of these regions reduce immobility in the forced swim test but do not affect open field or forced swim behavior. Inhibition of the raphe shifts the behavioral phenotype of uncontrolled stress to a phenotype closer to that of controlled stress.

Altered neuroplasticity

Recent studies have called attention to the role of altered neuroplasticity in depression. A review found a convergence of three phenomena:

1. Chronic stress reduces synaptic and dendritic plasticity
2. Depressed subjects show evidence of impaired neuroplasticity (e.g. shortening and reduced complexity of dendritic trees)
3. Anti-depressant medications may enhance neuroplasticity at both a molecular and dendritic level.

The conclusion is that disrupted neuroplasticity is an underlying feature of depression, and is reversed by antidepressants.

Blood levels of BDNF in people with MDD increase significantly with antidepressant treatment and correlate with decrease in symptoms. Post mortem studies and rat models demonstrate decreased neuronal density in the prefrontal cortex thickness in people with MDD. Rat models demonstrate histological changes consistent with MRI findings in humans, however studies on neurogenesis in humans are limited. Antidepressants appear to reverse the changes in neurogenesis in both animal models and humans.

Inflammation

Various reviews have found that general inflammation may play a role in depression. One meta analysis of cytokines in people with MDD found increased levels of pro-inflammatory IL-6 and TNF-a levels relative to controls. The first theories came about when it was noticed that interferon therapy caused depression in a large number of people receiving it. Meta analysis on cytokine levels in people with MDD have demonstrated increased levels of IL-1, IL-6, C-reactive protein, but not IL-10. Increased numbers of T-Cells presenting activation markers, levels of neopterin, IFN gamma, sTNFR, and IL-2 receptors have been observed in depression. Various sources of inflammation in depressive illness have been hypothesized and include trauma, sleep problems, diet, smoking and obesity. Cytokines, by manipulating neurotransmitters, are involved in the generation of sickness behavior, which shares some overlap with the symptoms of depression. Neurotransmitters hypothesized to be affected include dopamine and serotonin, which are common targets for antidepressant drugs. Induction of indolamine-2,3 dioxygenease by cytokines has been proposed as a mechanism by which immune dysfunction causes depression. One review found normalization of cytokine levels after successful treatment of depression. A meta analysis published in 2014 found the use of anti-inflammatory drugs such as NSAIDs and investigational cytokine inhibitors reduced depressive symptoms. Exercise can act as a stressor, decreasing the levels of IL-6 and TNF-a and increasing those of IL-10, an anti-inflammatory cytokine.

Inflammation is also intimately linked with metabolic processes in humans. For example, low levels of Vitamin D have been associated with greater risk for depression. The role of metabolic biomarkers in depression is an active research area. Recent work has explored the potential relationship between plasma sterols and depressive symptom severity.

Oxidative stress

A marker of DNA oxidation, 8-Oxo-2'-deoxyguanosine, has been found to be increased in both the plasma and urine of people with MDD. This along with the finding of increased F2-isoprostanes levels found in blood, urine and cerebrospinal fluid indicate increased damage to lipids and DNA in people with MDD. Studies with 8-Oxo-2' Deoxyguanosine varied by methods of measurement and type of depression, but F2-Isoprostane level was consistent across depression types. Authors suggested lifestyle factors, dysregulation of the HPA axis, immune system and autonomics nervous system as possible causes. Another meta-analysis found similar results with regards to oxidative damage products as well as decreased oxidative capacity. Oxidative DNA damage may play a role in MDD.

Mitochondrial Dysfunction:

Increased markers of oxidative stress relative to controls have been found in people with MDD. These markers include high levels of RNS and ROS which have been shown to influence chronic inflammation, damaging the electron transport chain and biochemical cascades in mitochondria. This lowers the activity of enzymes in the respiratory chain resulting in mitochondrial dysfunction. The brain is a highly energy-consuming and has little capacity to store glucose as glycogen and so depends greatly on mitochondria. Mitochondrial dysfunction has been linked to the dampened neuroplasticity observed in depressed brains.

Large-scale brain network theory

Instead of studying one brain region, studying large scale brain networks is another approach to understanding psychiatric and neurological disorders, supported by recent research that has shown that multiple brain regions are involved in these disorders. Understanding the disruptions in these networks may provide important insights into interventions for treating these disorders. Recent work suggests that at least three large-scale brain networks are important in psychopathology:

Central executive network

The central executive network is made up of fronto-parietal regions, including dorsolateral prefrontal cortex and lateral posterior parietal cortex. This network is involved in high level cognitive functions such as maintaining and using information in working memory, problem solving, and decision making. Deficiencies in this network are common in most major psychiatric and neurological disorders, including depression. Because this network is crucial for everyday life activities, those who are depressed can show impairment in basic activities like test taking and being decisive.

Default mode network

The default mode network includes hubs in the prefrontal cortex and posterior cingulate, with other prominent regions of the network in the medial temporal lobe and angular gyrus. The default mode network is usually active during mind-wandering and thinking about social situations. In contrast, during specific tasks probed in cognitive science (for example, simple attention tasks), the default network is often deactivated. Research has shown that regions in the default mode network (including medial prefrontal cortex and posterior cingulate) show greater activity when depressed participants ruminate (that is, when they engage in repetitive self-focused thinking) than when typical, healthy participants ruminate. People with MDD also show increased connectivity between the default mode network and the subgenual cingulate and the adjoining ventromedial prefrontal cortex in comparison to healthy individuals, individuals with dementia or with autism. Numerous studies suggest that the subgenual cingulate plays an important role in the dysfunction that characterizes major depression. The increased activation in the default mode network during rumination and the atypical connectivity between core default mode regions and the subgenual cingulate may underlie the tendency for depressed individual to get "stuck" in the negative, self-focused thoughts that often characterize depression. However, further research is needed to gain a precise understanding of how these network interactions map to specific symptoms of depression.

Salience network

The salience network is a cingulate-frontal operculum network that includes core nodes in the anterior cingulate and anterior insula. A salience network is a large-scale brain network involved in detecting and orienting the most pertinent of the external stimuli and internal events being presented. Individuals who have a tendency to experience negative emotional states (scoring high on measures of neuroticism) show an increase in the right anterior insula during decision-making, even if the decision has already been made. This atypically high activity in the right anterior insula is thought to contribute to the experience of negative and worrisome feelings. In major depressive disorder, anxiety is often a part of the emotional state that characterizes depression.