“Cold” Cognition in Depression

“Cold” cognition refers to information processing in the absence of any emotional influence (Figure 1). Theoretically, cold cognition is engaged on tests where the stimuli are emotionally neutral and the outcome of the test is not motivationally relevant (though motivational influences could conceivably turn a cold test “hot”; see Might “Cold” Cognition Be Turned “Hot” in Depression? below). Examples of neuropsychological tests usually considered cold include standardized pencil-and-paper assessments commonly used to assess function in neurological patients, for example the California Verbal Learning Test, the Trail-Making Test, and the Wisconsin Card Sort Test. Reliable impairments on such neuropsychological tests were observed from the 1970s onward in depressed patients. Some early studies adopted a classical neuropsychological case series approach,Reference Cavenar, Maltbie and Austin⁶ comparing the performance of individual depressed patients against population norms, identifying deficits of a magnitude judged to be clinically significant in several patients. More frequently, case-control designs were employed, and by the mid-1990s numerous studies comparing specific cognitive measures between groups of depressed patients and comparison subjects had been reported, particularly on memory tests.

Figure 1

Normal “cold” cognition in nondepressed individuals. “Cold” (emotion-independent) cognition is instantiated via a complex set of circuits, including interactions (blue arrows) between the dorsolateral prefrontal cortex (DLPFC), the dorsal anterior cingulate cortex (ACC), and the hippocampus (H). Limbic structures connected to DLPFC, ACC, and H, such as the amygdala (A), the nucleus accumbens (NAcc), and the subgenual portion of the ACC (sgACC) also may also be activated during cold cognition, but are more strongly engaged during “hot” (emotion-laden) cognition (Figure 2). Monoamine neurotransmitter (NT) projections (purple arrows) emanating from the brainstem, including serotonin (5-HT), norepinephrine (NE), and dopamine (DA), may influence cold cognition via modulatory actions in cortical and subcortical regions. Note that most circuit nodes and connections are excluded in this and later figures for clarity, and that some connections may be indirect.

In 1995 Burt etal. Reference Burt, Zembar and Niederehe⁷ performed the first systematic review of this literature, identifying nearly 100 case-control reports examining memory performance in depressed patients. Their meta-analysis identified deficits in patients of standardized effect sizes (Cohen's d) in the range 0.27 (small) to 0.67 (medium-to-large), varying across outcome measures. Surprisingly, patients who were younger exhibited greater deficits. However, several of the studies included subjects with organic neurological illness, and did not match the groups on important demographic variables such as age and educational level, making it difficult to draw firm conclusions. A later meta-analysis by Veiel,Reference Veiel⁸ which utilized more stringent inclusion criteria, identified higher effect sizes for memory, in the range 0.83–0.97 (large), and additionally reported differences in other domains of cognitive function, with only tests in the domain “attention and concentration” apparently relatively spared in depressed patients (however, see next paragraph). This latter result does appear surprising, given Criterion 8 for an MDE: “diminished ability to … concentrate.” Importantly, impairments on paper-and-pencil tests have been observed reliably in unmedicated samples.Reference Porter, Gallagher, Thompson and Young⁹

The advent of theoretically based, computerized cognitive tests in the 1990s provided an important methodological advance in understanding cognition in depression. One example of this approach is in the use of the Cambridge Neuropsychological Test Automated Battery (CANTAB; http://www.cantab.com). Broadly consistent with the results from pencil-and-paper studies described above, impairments were noted on a wide variety of CANTAB tests, including not only memory and executive function,Reference Elliott, Sahakian and McKay¹⁰ but also attentional measures.Reference Swainson, Hodges and Galton¹¹ These later studies employed computerized continuous performance tests (for example the CANTAB Rapid Visual Information Processing test, RVP) to assess sustained attention, in which subjects must detect specific targets presented in a train of hundreds of successively presented stimuli, separated by a sub-second time interval, over several minutes.Reference Swainson, Hodges and Galton¹¹ The marked difference in results from previous studies exemplifies one advantage of utilizing a computerized testing system, which allows stimuli to be presented with greater flexibility and temporal precision than traditional paper-and-pencil assessments, thereby furnishing tests with higher cognitive specificity. Other advantages of computerized testing include automated data collection, resulting in lower inter-administrator variability, as well as standardized recording and scoring of results.

This approach contrasts with earlier non-computerized studies, in which continuous performance paradigms such as the RVP were impractical to administer routinely. As such, the only measures of attention available in the meta-analysis of VeielReference Veiel⁸ were variants of the digit-span test from the Wechsler Adult Intelligence Scale, which does not require a high degree of sustained concentration, and in a recent meta-analyses of executive function in depression was found to be relatively unimpaired.Reference Snyder¹² Several studies identified cognitive impairment during remission,Reference Beats, Sahakian and Levy¹³ though it is possible that some of this continued impairment might be explained in part by residual subclinical symptoms, as meta-analyses have reported small-to-moderate (r values in the range 0.1–0.5) relationships between the degree of cognitive impairment and symptom load.Reference McDermott and Ebmeier¹⁴

Later work confirmed the clinical significance of cognitive deficits during a depressive episode, at least in elderly patients (reviewed in Pimontel etal. Reference Pimontel, Culang-Reinlieb, Morimoto and Sneed¹⁵). For example, Potter and colleaguesReference Potter, Kittinger, Wagner, Steffens and Krishnan^4, Reference Story, Potter, Attix, Welsh-Bohmer and Steffens¹⁶ reported that more cognitively impaired elderly depressed patients improved less following treatment with antidepressant medication. Executive function deficits appear to be particularly reliable predictors of poor treatment response,Reference McLennan and Mathias¹⁷ and some studies have found that severely cognitively impaired depressed patients may benefit from psychological therapy specifically tailored to boosting problem solving.Reference Arean, Raue and Mackin¹⁸ This is consistent with complementary evidence from trials using the cognitive enhancer modafinil as an adjunct to SSRI treatment to improve response,Reference Abolfazli, Hosseini and Ghanizadeh¹⁹ though the cognitive mechanisms underlying this potentially important finding remain to be clarified.

Might “Cold” Cognition Be Turned “Hot” in Depression?

The precise theoretical significance of these reliable group differences on cold cognitive measures has been a matter of debate. Some investigators interpret these effects as reflecting a core feature of depression, likely of central importance in its etiology with the potential to be used as endophenotypes in molecular genetic studies.Reference Glahn, Curran and Winkler²⁰ Others have wondered whether the poor performance observed might reflect a motivational deficit, caused by depressed patients treating task feedback differently to controls.Reference Elliott, Sahakian, Herrod, Robbins and Paykel²¹ This latter interpretation has also received some empirical support. In a study using the CANTAB in depression, Beats etal. Reference Beats, Sahakian and Levy¹³ identified a pattern of responding they termed “catastrophic response to perceived failure.” When depressed patients made an error on a test, they were proportionately more likely than controls to make an error on the subsequent trial. This pattern was confirmed in several studies,Reference Elliott, Sahakian and McKay^10, Reference Murphy, Michael, Robbins and Sahakian²² including one in which comparison patient groups were included, that were matched for overall performance.Reference Elliott, Sahakian, Herrod, Robbins and Paykel²¹

These studies raise the possibility that at least some of the poor performance on neuropsychological tests in depression might be due to altered “hot,” ie, emotion-dependent, cognition (Figures 2 and 3). In other words, ostensibly cold cognitive tasks, especially those featuring explicit feedback, may take on an emotional quality in depressed individuals, and instead of using negative feedback to improve performance, depressed individuals may become discouraged. It is also possible that depressed individuals do not experience positive feedback as intensely as controls, further reducing the motivation to perform well. However, it is important to note that cold cognitive impairments have been observed on tests that do not feature explicit task feedback, and also in individuals who were fully recovered from depression. Therefore both hot and cold cognitive mechanisms are likely to contribute to poor neuropsychological test performance in depressed individuals (Figure 3).

Figure 2

Normal positively biased “hot” (emotion-laden) cognition in nondepressed individuals. Hot cognition is instantiated by circuits including interactions (green arrows) between limbic regions including A, NAcc, and sgACC, the activity in which is profoundly modulated by 5-HT, NE, and DA. These regions share reciprocal connections with DLPFC, ACC, and H, and consequently hot and “cold” (emotion-independent) cognition necessarily interact (for example, motivation alters ostensibly cold cognitive test performance). Nondepressed individuals exhibit positive (green arrows) bottom-up (perceptions/experience) and top-down (expectation) biases, providing resilience to adverse events. Abbreviations and colors as in Figure 1.

Figure 3

Negatively biased hot and impaired cold cognition in depression. Top-down cold cognitive control and monoamine transmission are both compromised (dashed arrows). Depressed individuals exhibit negative (red arrows) bottom-up biases (emotional perceptions and reward experiences) due to disrupted monoamine transmission. They also have negative top-down biases (expectations), resulting in dysfunctional and self-perpetuating negative schemata, which give rise to depressive symptoms. Abbreviations and colors as in Figure 1.

“Hot” Cognition in Depression

In the past decade, several groups have reported that depressed individuals exhibit more negatively biased responses on tests of emotional processing. These are typically variants of cold cognitive tests that were adapted to include emotionally valenced stimuli. A common finding is that never-depressed individuals exhibit a positive bias (Figure 2), possibly reflecting resilience to negative emotional information, and that this is either attenuated or reversed in depressed individuals (Figure 3; see Roiser etal. Reference Roiser, Elliott and Sahakian²³ for a review). Such a negatively biased pattern of responding in depressed individuals, both medicated and unmedicated, has been reported on tests of perception,Reference Joormann and Gotlib²⁴ memory,Reference Matt, Vazquez and Campbell²⁵ attention,Reference Gotlib and Joormann²⁶ and working memory.Reference Joormann and Gotlib²⁷ For example, on the CANTAB Affective Go/No-Go test, on which subjects must respond to one word-type while inhibiting responses to another, Murphy etal. demonstrated that while control individuals responded slightly more quickly to positive than negative target words, the converse was true for depressed patients.Reference Murphy, Sahakian and Rubinsztein²⁸ A similar negative bias was identified using the same test in unmedicated depressed individuals.Reference Erickson, Drevets and Clark²⁹

While studies of emotional bias in depression were first conducted many decades ago, abnormalities in another type of hot cognition, reward and punishment processing, have started to receive attention only relatively recently (see Eshel and RoiserReference Eshel and Roiser³⁰ for a comprehensive review and studies cited therein for further details). This dearth of studies is surprising, given that anhedonia, which is closely related to reward processing, is one of the cardinal symptoms of a depressive episode. Moreover, experimental animal models of depression used for drug discovery frequently focus on behavioral constructs related to reward and punishment processing,Reference Berton, Hahn and Thase³¹ including learned helplessness, behavioral despair, sucrose preference, and intracranial self-stimulation.

Though the literature on reward processing in depression is much smaller than that on emotional biases, some consistent findings have emerged. One is the confirmation of the finding discussed above that depressed patients are hypersensitive to negative feedbackReference Elliott, Sahakian and McKay¹⁰; this finding was determined using tasks that were designed explicitly to assess this process. During a probabilistic reward and punishment reversal learning task featuring two stimuli (one more often associated with positive feedback and the other more often associated with negative feedback), medicated depressed patients showed a pronounced tendency to switch their choice following misleading negative feedbackReference Murphy, Michael, Robbins and Sahakian²²—a result that was later replicated in unmedicated patients.Reference Taylor Tavares, Clark and Furey³² Other studies report hyposensitivity to positive feedback, for example, a failure to liberalize response bias on a difficult task when more points are gained for correct responses than are lost for incorrect responses,Reference Henriques and Davidson³³ or reduced learning from rewarding stimuli.Reference Robinson, Cools, Carlisi, Sahakian and Drevets³⁴

The above tasks rely on the learning of stimulus–outcome associations, while other tests have probed the impact of explicitly providing reward or punishment information about choices on decision-making in depression. One of the first studies to examine this question used the CANTAB Cambridge Gambling Task,Reference Rogers, Everitt and Baldacchino³⁵ which requires participants initially to choose which of two outcomes they think will occur, with the probability of being correct varying, and then to stake points on their decision. A consistent finding across medicated, unmedicated, and even remitted samples is that depressed individuals increase their stake with increasingly better odds (termed “risk adjustment”) to a lower extent than controls,Reference Murphy, Rubinsztein and Michael^36, Reference Rawal, Collishaw, Thapar and Rice³⁷ which possibly reflects ambivalence to winning points. Other studies have used effort-based tasks where subjects must respond quickly in order to achieve rewards, demonstrating reduced motivation in depression.Reference Treadway, Bossaller, Shelton and Zald³⁸ Interestingly, similar findings have been reported in subjects with schizophrenia,Reference Roiser, Stephan and den Ouden^39, Reference Murray, Clark and Corlett⁴⁰ who also experience anhedonia.

An important recent development in this field is the application of formal computational models to understand reward- and punishment-driven behavior. These models can dissect out specific aspects of reward processing behavior (eg, learning, points value, randomness) more precisely than analyses that use measures based on raw data.Reference Montague, Dolan, Friston and Dayan⁴¹ For example, using such a computational approach, Chase etal. demonstrated that reward learning was particularly poor in highly anhedonic depressives.Reference Chase, Frank and Michael⁴² Another recent study used computational modeling to demonstrate more random choices in subjects who scored high on depressive symptom rating scales (though without a categorical diagnosis) when using stimuli of known reward associations to guide decision making.Reference Kunisato, Okamoto and Ueda⁴³

A Cognitive Neuropsychological Model of Depression

What are the theoretical implications of hot and cold cognitive abnormalities in depression? In our view, they mandate a reframing of the classic cognitive model of depression proposed by Beck,Reference Beck⁴⁴ which proposes that depression results from stable, self-reinforcing, dysfunctional negative schemata, and is the inspiration for talking therapy approaches such as CBT. Beck's cognitive model predicts the presence of abnormal hot processing in depression (ie, negative emotional biases) on the basis of “top-down” influences, or what we conceptualize as “negative expectations” (Figure 3). In other words, depressed individuals may exhibit slower responses to happy wordsReference Erickson, Drevets and Clark²⁹ or misinterpret facial expressions as sadReference Joormann and Gotlib²⁴ precisely because they expect to encounter such negative information in the environment. These negative expectations, which include dysfunctional attitudes and negative attributional styles, and give rise to thought processes characteristic of depression, such as negative automatic thoughts and rumination, can be considered a form of “top-down” hot cognition (Figure 3). They are the targets of psychological interventions such as cognitive therapy, which could be conceptualized as training depressed individuals to exert cold cognitive control over their top-down negative biases, for example through working memory, inhibition, and problem solving (Figure 4).

Figure 4

Treating hot and cold top-down cognition in depression. Psychotherapies such as cognitive behavioral therapy (CBT) train patients to challenge their top-down hot biases (negative expectations) and thereby resolve dysfunctional negative schemata. Transcranial magnetic stimulation (TMS) directly activates DLPFC and interconnected structures, improving cold top-down cognitive control and emotional regulation. However, dysfunctional monoamine transmission and bottom-up negative biases (negative perceptions) are not treated directly by CBT or TMS, and may thus persist. Abbreviations and colors as in Figure 1.

However, a complementary view that has gained convincing empirical support over the past decade is that disrupted neurotransmission in systems targeted by antidepressant drugs, such as serotonin, norepinephrine, and dopamine, alters the “bottom-up” processing of emotional stimuli, instantiating “negative perceptions” (Figure 3). Importantly, a large body of evidence suggests that manipulating monoamine transmission experimentally can alter reward and emotional processing biases, in both healthy volunteers and depressed individualsReference Harmer, O'Sullivan and Favaron^45–Reference Roiser, Levy and Fromm⁴⁷ (Figure 5). According to this view, negative biases occur due to compromised monoamine modulation of the neural circuits that process emotional stimuli.Reference Harmer, Goodwin and Cowen⁴⁸

Figure 5

Treating hot bottom-up cognition in depression. Antidepressant medications target the monoamine systems (5-HT, NE, and DA), which may be disrupted in some depressed patients, thereby resolving bottom-up hot biases (negative perceptions). Deep brain stimulation (DBS) in the sgACC may influence circuits instantiating bottom-up negative biases directly, via connections with other regions in limbic circuits such as A. However, top-down negative biases (negative expectations) are not treated directly by antidepressant medications or DBS, and may thus persist. Abbreviations and colors as in Figure 1.

Our cognitive neuropsychological model of depressionReference Roiser, Elliott and Sahakian²³ (Figure 6) proposes an integrated approach, accommodating both the traditional psychological framework (Figure 4) and more recent psychopharmacological findings (Figure 5). We propose that bottom-up biases (negative perceptions), caused by disrupted monoamine transmission, play a causal role in the development of dysfunctional negative schemata, but that the latter can themselves engender top-down biases (negative expectations), thus maintaining negative schemata. This model also proposes a central role for a type of top-down cold cognition, cognitive control, in depression, suggesting that negative perceptions may feed into dysfunctional negative schemata particularly when cognitive control is impaired (Figure 3). Importantly, these different cognitive processes (negative perceptions, negative expectations, and cognitive control) are likely instantiated via the dysfunctional operation of separate, but interacting, neural circuits (Figures 1–3). Being able to identify signals from and manipulate these circuits may enable researchers to better parse the mechanistic heterogeneity of depression, and provide novel approaches to treatment (Figures 4 and 5).

Figure 6

A cognitive neuropsychological model of depression. Red boxes/text indicate factors contributing to the development and maintenance of depressive symptoms. Green boxes/text indicate factors contributing to the treatment of and recovery from depression. 5-HTTLPR: serotonin transporter linked polymorphic region. Reprinted from Roiser etal. Reference Roiser, Elliott and Sahakian²³ with permission.

Understanding Novel Treatments Through Cognition

The findings reviewed in this article, together with associated neuroimaging results, provide an important basis for the understanding of two novel brain circuit-based intervention strategies in treatment-resistant depression. First, repetitive transcranial magnetic stimulation (rTMS) of the dorsolateral prefrontal cortex (DLPFC) is a non-invasive method of stimulation, by which a neural circuit implicated in cognitive control can be manipulated directly. Although rTMS for depression was first attempted in the early 1990s (see George etal. Reference George, Taylor and Short⁴⁹ for a review), the most convincing evidence for efficacy has come from two large double-blind trials,Reference George^50, Reference O'Reardon, Solvason and Janicak⁵¹ both of which reported that in treatment-resistant patients, 2–3 times as many subjects who were administered active stimulation remitted (∼15%) compared to those receiving sham stimulation.

The mechanisms underpinning rTMS in the treatment of depression remain to be completely clarified, but may relate to top-down cold cognition—specifically cognitive control—and its interaction with top-down hot cognitionReference Roiser, Elliott and Sahakian²³ (Figure 4). Several neuroimaging studies have found that depressed subjects exhibit exaggerated prefrontal cortex responses during difficult working memory tasks, interpreted as reflecting prefrontal “inefficiency,” which is not altered by SSRI treatment.Reference Walsh, Williams and Brammer⁵² By contrast, rTMS to the DLPFC has been reported to boost performance on tests requiring cognitive control in depressed subjects,Reference Martis, Alam and Dowd⁵³ and to improve cognitive control over distracting negative information in healthy volunteers.Reference Leyman, De Raedt, Vanderhasselt and Baeken⁵⁴ Some preliminary studies have linked these effects to symptomatic relief directly, reporting that treatment-resistant depressed patients who respond to two weeks of rTMS exhibit improved attentional control over neutral information after a single stimulation session,Reference Vanderhasselt, De Raedt, Leyman and Baeken⁵⁵ when mood effects were not yet apparent, and that treatment response is associated with better inhibition of negative distracting information after 10 days of treatment,Reference Leyman, De Raedt, Vanderhasselt and Baeken⁵⁶ when symptoms had started to remit. Future studies should explore in greater detail whether rTMS exerts its beneficial effects in depression by boosting cognitive control.

Second, for highly treatment-resistant depressed patients, invasive deep brain stimulation (DBS), particularly to the subgenual anterior cingulate cortex (sgACC), has been found to be effective in open-label trials.Reference Mayberg, Lozano and Voon⁵⁷ The rationale for targeting this brain region is the reliable finding from neuroimaging studies that it is over-active in depressed individualsReference Drevets⁵⁸ and increases in metabolism during negative mood.Reference Mayberg, Liotti and Brannan⁵⁹ The sgACC plays an important role in hot cognition and regulates activity in the amygdala,Reference Vidal-Gonzalez, Vidal-Gonzalez, Rauch and Quirk⁶⁰ in which responses to negative stimuli are exaggerated in depression and normalized with SSRI treatmentReference Victor, Furey, Fromm, Ohman and Drevets⁶¹ (Figures 3 and 5). Importantly, responses to negative stimuli in the sgACC are blunted in both depressed patientsReference Grimm, Boesiger and Beck⁶² and healthy individuals at genetic risk for depression,Reference O'Nions, Dolan and Roiser⁶³ thus supporting its role in the instantiation of emotional biases. As with rTMS, the mechanism underlying the beneficial effects of DBS is not completely clear, but may be related to resolving negative bottom-up biases (Figure 5) via altered activity in the sgACC and other interconnected regions, for example the amygdala and orbitofrontal cortex.Reference Mayberg, Lozano and Voon⁵⁷ Future studies should test directly whether stimulation in this region alters bottom-up negative biases in depression.

Disclosures

Jonathan Roiser has the following disclosure: Cambridge Cognition, consultant, consulting fees, retainer. Barbara Sahakian has the following disclosures: Cambridge Cognition, consultant, consulting fees; Janssen Pharmaceuticals, research support, grant; CeNeS Pharmaceuticals, shareholder.