In the context of the proposal put forth here, associations provide the vehicle with which our thoughts advance from one representation to another. Central to the proposal put forward here is the distinction between narrow and broad associative activations. Narrow associative thinking, or rumination, refers to associations that surround a narrow focus (e.g., the context of my bad comment over dinner last night: what I said, what I should have said, the resulting facial expressions and verbal responses to it, possible future implications and so forth). Ruminative, narrow associative processing, as proposed below, might stem from excess prefrontal inhibition (Figure 1a). Broad associative activations, on the other hand, involve more remote associations and, critically, activate associations that make thought processes advance from one context to another smoothly (Figure 1b), thereby minimizing rumination (e.g., “in spite of that miserable comment, dinner was really tasty; never tried the combination of figs with prosciutto before; need to go grocery shopping tomorrow”).

Using human neuroimaging, we recently studied the cortical mechanisms mediating contextual associations in particular [2-4]. The associations that tie items that share the same context consistently activate three main cortical regions: the parahippocampal cortex in the medial temporal lobe (MTL), the medial partial cortex (MPC), and the medial prefrontal cortex (MPFC) [3]. This contextual associations network shows a striking overlap with the cortical network termed the “default network” [5], as we have shown recently [4]. The default network is believed to subserve the mental processes that take place in the brain when we are not engaged in a specific goal-oriented task [6, 7]. This overlap between the default network and the network subserving associative processing of contextually related information is taken as the cortical manifestation of the idea that associative processing is a crucial element of natural thought [4]. In other words, associative activations provide the building blocks that allow our thought processes to proceed from one topic to another. Notably, other tasks, including hedonic processes, reward-related processes and affective processes, also activate parts of this associative network. We have interpreted this common activation by different processes as reflecting the fact that what unites these processes is that they all rely on associative activations at their core [4].

A parallel to the conclusion that positive mood promotes associative thinking is the finding that positive mood is accompanied by a wider attentional “spotlight,” whereas negative mood can be characterized by a more focused type of thinking [14]. For example, when subjects need to accomplish memory or categorization tasks, those in a happier mood utilize available global information whereas the attention of participants in a sad mood is focused on more local properties. Similarly, positive mood and optimism are directly associated with a global processing bias in visual tasks, whereas a local bias is observed for people in negative mood and depression [15]. A negative mood may signal the potential presence of threat and thus might elicit a more focused mode of processing. Indeed, dangerous environments also have the same effect on the scope of attention as that of negative affect [16]. The effects and functional merits of positive emotions, on the other hand, have been framed as expanding thought repertoire (i.e., scope) in a host of dimensions within a comprehensive related framework termed broaden-and-build [17].

Finally, associative processing has been linked with the ability to imagine future events and to predict upcoming information [1]. Therefore, the fact that foresight is severely impaired in depressed individuals [18] provides encouraging support for the idea that associative activation and mood are tightly related.

Rumination seems to be a major vulnerability factor for the negative mood associated with major depression but also with other disorders, such as post-traumatic stress disorder (PTSD) [23], anxiety, obsessive-compulsive disorder (OCD) and more. In fact, given the overlap between rumination and worry, and the prevalence of both [24], the present hypothesis may be relevant across most psychiatric disorders.

One effective therapy in treating stress disorders involves shifting attention from an upsetting stimulus to something relatively neutral [25]. However, merely shifting the focus of attention does not necessarily activate broad associations and thus might lose its efficiency of distracting the ruminating mind as a long-term remedy. Indeed, in accordance with the hypothesis presented here, distraction as a soothing technique is most effective when the identity of the distracter varies frequently [26, 27]. One can therefore consider frequent, broadly associative thinking as providing a higher rate of distraction and thus fewer opportunities to ruminate and dwell on a negative thought.

Interestingly, it has been reported that speed of thought, as manipulated by paced reading, has direct influence on mood [28]. Specifically, these studies suggest that reading faster can make one feel more positively, regardless of whether the content of the text is positive or negative. Reading causes the activation of concepts, and, presumably, faster reading activates more concepts, which could be seen as analogous to the massive activation in broad associative activation (i.e., more concepts activated per time unit). Therefore, although this intriguing finding could be attributed to the mere speed of “mental motion” [22], or perhaps to the fact that a mentally demanding reading activity could serve as a powerful means of distraction from rumination, it could also provide support to the proposal outlined here that increased associative thinking elicits increased positive mood.

Associative processing and the ability to learn associations between previously unrelated items have been attributed primarily to structures of the MTL, particularly the hippocampus [31, 32], and the parahippocampal cortex in the MTL [4, 33]. Several recent studies have demonstrated both functional disturbances and volume reduction of the hippocampal formation in depression [34, 35]. Importantly, the severity of depressive symptoms is directly increased by such hippocampal structural disturbances [36], as well as with a decrease in hippocampal activity [37]. Additionally, successful antidepressant treatment can halt the progression of hippocampal damage [35], and is associated with recovery of hippocampal volume [38]. These results demonstrate the close relationship between depression and the functional and structural integrity of the MTL.

One of the most effective medications for treating depression relies on selective serotonin reuptake inhibitors (SSRI) such as fluoxetine (e.g., Prozac). In recent years it has been observed that SSRIs have a direct influence on adult neurogenesis [39], particularly in the dentate gyrus within the hippocampus [35, 40]. Importantly serotonin has also been shown to play a central role in learning and memory formation [41], which is widely believed to be mediated by the MTL and to rely on associations.

Adult hippocampal neurogenesis is still a new and controversial topic. Nevertheless, the possibilities, especially for our understanding of mental disorder, are profound and thus warrant further attention. There is accumulating evidence that hippocampal neurogenesis is not merely an epiphenomenon of the SSRIs' chemical effect, but actually has a causal effect on the success of SSRI treatment: The newly generated neurons have direct effect on the state of depression, whereby successful antidepressant treatment is conditional upon a successful hippocampal neurogenesis. Indeed, blocking hippocampal neurogenesis substantially reduces the efficiency of antidepressants [29].

Computational modeling work [42] shows that such neurogenesis could be critical for establishing new contexts for behavior in memory, and such contexts, we have argued [4], are based on associative representations and activations. Indeed, associative long-term potentiation (LTP) can be induced in new neurons more readily than in older neurons [43], and the magnitude of LTP can be modulated by serotonin [44]. These converging findings, together, provide a potential mechanism for mediating the increase in associative processing with neurogenesis and SSRI-based therapy (see Box 1).

In a second line of research that is related to the proposal presented here, findings indicate that electroconvulsive therapy (ECT) and transcranial magnetic stimulation (TMS) exhibit anti-depressant effects [45]. How these treatments exert their effects remains an open question. Recently, Mayberg and colleagues [30, 46] demonstrated successfully that electrically stimulating the deep Brodmann area 25 (Figure 2a) can alleviate the symptoms of major depression patients. Area 25, the subgenual cingulate, is situated in the MPFC, which is activated in associative cognitive tasks (Figure 2b). One possible mechanism by which such deep brain stimulation (DBS) in MPFC can affect depression symptoms might be through the regulation of inhibition. Our memory consists of a web of representations that are connected with each other, within several degrees of separation. While this makes for an efficient framework for encoding and retrieval, it is crucial that the activation of one mental representation (e.g., a chair, when we see an image of a chair) would not result in the activation of representations that are irrelevant in the specific context. Inhibition is proposed to play this central role of limiting the extent of related representations that are activated at a given instance (e.g., seeing an image of a chair would activate “chair,” “table” and perhaps “legs,” but should not activate “chicken legs”), thereby giving rise to associations and predictions that are most relevant. In the context of inhibition and mood, an abnormally activated MPFC, as in depression [47], can result in over-inhibition from MPFC to MTL, which might then significantly constrain the scope of associative activation in MTL [48]. Excessive inhibition from MPFC could explain the inability of depressed patients to disengage from debilitating rumination (Figure 1a). Indeed, inhibitory dysfunction can exacerbate depression through rumination [49]. DBS might therefore operate by bringing MPFC back to its normal level of activity (Figure 1b), which brings back inhibition to a level where MTL can resume broader associative activation.

Another advantage of promoting broad associative activations as part of a therapy is that it does not require patients to actively suppress specific thoughts, but rather simply increases the shift from rumination. Explicitly attempting to suppress a specific thought has been shown to be considerably harder than anticipated, and suppression is not effective for alleviating depression, and may even be counter-productive [59]. Fostering a thinking pattern that is more associative in nature instead will divert thought to other themes more readily. Methods such as meditation require trainees to “observe thoughts and let them go.” In a healthy mind with broad associative activations, such process of “letting go” is expected to occur naturally. In this context, it is interesting to note that no claims are made here about whether the pattern of thought can be controlled voluntarily, or whether instead it is automatic. The issue of control is interesting and has already received considerable attention [60], but it is less relevant in the present proposal. The premise of the therapeutic potential of this proposal lies on the idea of restoring the cortical infrastructure that regulates inhibition and allows one to broadly associate and disengage from ruminations in a natural manner. The hypothesis is that when this infrastructure exists, thinking is more associative, independent of voluntary control.

Similar proposals can be made for other conditions that involve mood disorders. For example, patients with PTSD show abnormal activation in the same regions where microstimulation shows benefit for depression symptoms, area BA 25 (as well as BA 24) [61]. PTSD can be seen as a form of rumination, where associative memory retrieval is focused narrowly on a traumatic memory with a limited ability to shift away from this focus. One might therefore suspect that a similar approach for promoting broad associative thinking as a means of disengaging from rumination could be helpful for PTSD patients as well. Similarly, this approach is predicted to prove beneficial also in stress and anxiety disorders, phobias and possibly even in normal aging.

The present hypothesis gives rise to a counterintuitive prediction that attention deficit hyperactivity disorder (ADHD) patients, whose mental processes tend to be doing the exact opposite of ruminating, are expected to have a more positive mood than that of individuals without ADHD. There has not been much research of mood effects in ADHD, and such measurements might be influenced by multiple factors that could counteract each other. Nevertheless, the existing research indicates that, indeed, children with ADHD tend to be relatively more positive thank control subjects, as reflected, for example, by affective valance [62], and positive illusory bias [63]. Furthermore, the hippocampus of ADHD patients has been shown to be of significantly larger volume compared with that of controls [64], and, while acute depression is typically associated with an enlarged and hyperactive amygdala [65], the amygdala of ADHD patients is smaller even than this of controls [64]. With the links made here between broad associative processing, its inhibition, and mood, ADHD could be attributable, at least in part, to deficits in inhibition rather than portrayed exclusively as an attention disorder.

Previous findings show that depression patients are deficient in foresight and future-related thinking [18]. Therefore, one would predict that successful treatment of mood disorders will increase patients' foresight and ability for future-oriented mental operations, such as planning and mental simulation.

Our understanding of mood manifestation in the brain is largely lacking, but a reasonable hypothesis might be that it involves the release and binding of neuroendorphins. If activating certain neural paths results in an increased release of neuroendorphins, one can see how ruminating on the same theme for a long period exhausts the release of such rewarding opioids from the repeatedly active neurons (and why repetition is often boring), whereas activating broad associated representations and pathways that have not been activated recently increases rewarding activations (as well as encourage us to attend novelty in our environment). This aspect of the proposal---that the brain might use the release of neuroendorphins as incentive to activate associations to promote learning and minimize uncertainty---is speculative and will require further interdisciplinary investigation. Nevertheless, recent theoretical and empirical work suggests that a MTL region similar to the one we showed to be activated by contextual associations shows increased fMRI activation for pictures that participants find as pleasant [66, 67], and this area might be at the top of a hierarchy of increased concentrations of opioids. Indeed, the highest levels of opioid binding have been found both in the orbital MPFC and in the MTL [68].

The activation of broad associations, as predicted here, makes mood more positive. Ruminating, on the other hand, makes mood negative. It might be possible that the act of rumination is by itself sufficient for eliciting a negative mood, independent of whether the focus of this rumination is negative (e.g., a bad memory). So why are we not depressed when concentrating on driving, reading, or meditating? One critical difference might be the duration of such attentive focus. Rumination that is associated with depression tends to have a much longer duration than what everyday, emotionally neutral tasks entail. Perhaps more importantly, however, this question helps sharpen the distinction between rumination and a narrow scope of attention: the narrow focus of attention while working towards a goal still advances over time (e.g., during writing), whereas the narrow attention associated with rumination remains focused on the same topic, for a long duration.

In summary, the proposal presented here outlines a cognitive neuroscience framework with which to understand mood. This framework can provide the foundations for future development of non-invasive methods for treating mood disorders, whereby the training and restructuring of the ability for broad associative thinking can elicit improvements that range from structural modifications to mood and behavior (see also Box 2).