Repeated transcranial magnetic stimulation on dorsolateral prefrontal cortex improves performance in emotional memory retrieval as a function of level of anxiety and stimulus valence


  • Michela Balconi PhD,

    Corresponding author
    1. Research Unit in Neuropsychology of Language, Department of Psychology, Catholic University of the Sacred Heart, Milan, Italy
    • Laboratory of Cognitive Psychology and Neuroscience, Catholic University of the Sacred Heart, Milan, Italy
    Search for more papers by this author
  • Chiara Ferrari MD

    1. Laboratory of Cognitive Psychology and Neuroscience, Catholic University of the Sacred Heart, Milan, Italy
    Search for more papers by this author

Correspondence: Michela Balconi, PhD, Department of Psychology, Catholic University of the Sacred Heart, Milan Largo Gemelli, 1, 20123 Milan, Italy. Email:



Anxiety behavior showed a consistent attentional bias toward negative and aversive memories, induced by a right frontal cortical superiority, based on an unbalance effect between the two hemispheres. The aim of the present study was to explore the role of the left dorsolateral prefrontal cortex (DLPFC) in the memory retrieval process of positive versus negative emotional stimulus, as a function of anxiety level.


A repeated transcranial magnetic stimulation (rTMS) paradigm was used to induce cortical activation of the left DLPFC. Subjects (n = 27; age range, 21–36 years), who were divided into two different groups (high/low anxiety; State–Trait Anxiety Inventory), were required to perform a task consisting of two experimental phases: an encoding phase (lists composed of positive and negative emotional words); and a retrieval phase (old stimuli and new stimuli to be recognized). Moreover, new stimuli (distractors) semantically related or unrelated to the old stimuli were used to test a possible interference effect induced by the semantic association.


rTMS over the left DLPFC affects memory retrieval. High-anxiety subjects benefited in greater measure from frontal left stimulation with a reduced negative bias (increased accuracy and reduced response time for the positive stimuli) and a significant increased performance for the semantically related distractors (reduced interference effect).


Left DLPFC activation favors the memory retrieval of positive emotional information and might limit the unbalance effect induced by right hemispheric superiority in high levels of anxiety.

HEIGHTENED ANXIETY IS associated with reduced recruitment of control mechanisms for emotional cues that are emotionally arousing, potentially aversive, and threat-related.[1] It has also been shown that increased anxiety in healthy subjects induces a reduced involvement of prefrontal control structures (prefrontal cortex, PFC).[2] Models of the processing of anxiety and related behavior have suggested a network of interconnected neuronatomical regions including the amygdala, hippocampus, thalamus, and PFC.[3, 4] The PFC could be crucial in mechanisms underlying the regulation of emotion, such as inhibition.[5, 6] In the case of a pathological level of anxiety in the clinical and pre-clinical field the ability to suppress this automatic processing would be impaired in subjects with anxiety disorder,[7] inducing a sort of attentive bias toward the potentially aversive cues.

A promising and alternative theory called the ‘valence model’, however, explained the relationship between anxiety and emotional information processing by suggesting that withdrawal-related emotions such as anxiety are located in the right hemisphere, whereas approach-related emotions are biased to the left hemisphere.[8-10] Thus, an increased level of anxiety might be associated with dysfunctional right hemispheric activity, with increased activation when aversive conditions are processed.[7, 11] This second model furnished clear evidence of the different behavior induced by positive versus negative emotional stimuli with regard to hemispheric lateralization, suggesting that high anxiety subjects had more right frontal hyperactivation in comparison with the left side. This may induce unbalanced processing of the two stimulus categories, with a consistent bias toward the negative one, meaning that the negative stimuli were selected and processed more quickly than the positive stimuli.[12]

The aim of the present study was therefore to investigate the effect of anxiety on information processing by performing a memory task in which familiar versus novel emotional information was recognized. Specifically, memory retrieval of emotional stimuli was used to verify the effect of anxiety levels in healthy low- or high-anxiety subjects on short-term memory performance.[13, 14] With regards to the brain area contribution to the memory task, it was previously shown that the dorsolateral prefrontal cortex (DLPFC) was implicated in the retrieval of knowledge in both normal and pathological processes.[15] Moreover, some evidence suggests that multiple regions of the PFC have the capacity to perform multiple types of executive control functions (i.e. evaluate, maintain, inhibit, or select). In particular, evidence indicates that the orbitofrontal cortex (OFC) participates in the executive control of information processing and behavioral expression by inhibiting neural activity associated with uncomfortable (e.g. painful) information. Bias effects of anxiety on performance and efficiency have been demonstrated to be greater on tasks imposing demands on the retrieval of more relevant information, that is, potentially threatening information.[2]

Moreover, it has been found that anxiety subjects had a more significantly impaired ability to discard cues that are semantically related to the target stimuli, with specific attentional bias induced by an increased influence of the stimulus-driven attentional system and a decreased influence of the goal-directed attentional system.[16] Thus anxiety could affect the ability to correctly recognize and discard incorrect and irrelevant information, necessary for correct retrieval behavior.

In the present study, anxiety level in healthy subjects was used as an independent group variable to test its effect on the retrieval process in the case of emotional positive and negative cues. Transcranial magnetic stimulation (TMS) was then used to induce increased activation of the left DLPFC. TMS, thought to be the creation of a perturbation, offers a unique opportunity to interact directly with the functioning of a cortical area during the execution of the memory task.[17, 18] Previously it was thought that low-frequency TMS could be effective in anxiety disorder, which might be associated with brain hyperexcitability and related cognitive activation.[19] Also recently, studies in healthy subjects tested the valence hypothesis based on the assumption of increased right hemispheric activity in anxiety disorder. In line with this hypothesis, recent evidence showed that low-frequency TMS of the right DLPFC produced a decreased anxiety-related behavior.[20] Moreover, TMS significantly reduced vigilant attention for fearful faces.[21]

In the present study we took into account the underlying interhemispheric competition by adopting the valence hypothesis.[22] An activation TMS paradigm may be used to increase the cortical excitability of the left hemisphere in order to enhance the response to positive emotional cues. To the best of our knowledge, no previous study has analyzed the effect of the potentiation of the contralateral left hemisphere, which might counterbalance the abnormal hyperactivation of the right hemisphere in anxiety disorder for a retrieval memory task. The main expected effect was a more symmetrical vigilant attention toward both positive and negative memories (unbiased performance).

Moreover, we hypothesized that the potential counterbalancing effect induced by TMS on the left hemisphere would be more effective in the case of higher cognitive demand, especially in high-anxiety subjects. For this reason we verified the impact of the stimulus valence not only related to memory retrieval of the old stimuli, but also linked to the mechanisms for discarding irrelevant information during retrieval[23] when a possible ‘semantic interference’ is produced. To do this we investigated the retrieval process in the presence of some distractors (new stimuli) that were semantically related or unrelated to the old stimuli.[24]



Fifteen women and twelve men (age range, 21–36 years) participated in the experiment. Exclusion criteria were history of psychopathology in the subjects or immediate family. The subjects were all right-handed and with normal or corrected-to-normal visual acuity. They gave informed written consent for participating in the study and the research was approved by the Ethics Committee at the institution where the work was carried out. No payment was provided for their performance.


Subjects sat on a comfortable chair in front of a PC screen. The experimental paradigm consisted of an encoding phase and a retrieval phase. In the encoding phase the participants were asked to memorize some word lists during a specific time window (90 s) for a successive retrieval phase. The retrieval phase was carried out immediately after the encoding phase ended (Fig. 1). Thus each encoding list was followed by a retrieval list. In the retrieval phase, words were randomly presented one by one on the PC screen for 6 s, and subjects were asked to decide whether they had viewed the word before[23] as soon as possible after the presentation of the word on the screen. Response accuracy and response time (RT) were recorded by E-Prime 2.0 Software.

Figure 1.

Experimental procedure for the encoding and retrieval phases. rTMS, repeated transcranial magnetic stimulation.

Two sets of material were used, the first for the encoding phase and the second for the retrieval phase. Words were Italian nouns (from four to seven letters) and of moderate frequency. All the words included were counterbalanced relative to the word length and their abstract versus concrete content.[25] Each word was in Arial 16-point font, in black on a white background.

For the encoding phase nine lists were used, each list composed of 20 words; 10 of them were related to positive emotional content and 10 were related to negative emotional content. For the retrieval phase, stimulus material was composed of a total of 450 stimuli, subdivided into nine lists. Each retrieval list was composed of 50 words grouped in the following categories: old (20 words contained in the encoding lists); and new (30 words not contained in the encoding lists). Each category (old and new) was further divided in two equally distributed subgroups: words with a negative emotional content; and words with a positive emotional content. Moreover, the new words were divided into two equally distributed subcategories, one containing words semantically related (such as gun and pistol) and one containing words not semantically related (such as gun and smile) to the old words.

The familiarity of the word and the emotional valence were assessed for the whole stimulus material before the experimental task, by a group of 14 subjects (7 male, 7 female, mean age, 25.9 ± 2.10 years). Finally, to directly test the semantic link between the old and new word categories, pairs of semantically related and unrelated words were created for each retrieval list.[24, 26] Specifically, the association norms were drawn from the DPSS psycholinguistics database.[27] Prior to the experimental phase some judges evaluated the semantic proximity and relatedness (semantic associates) of each pair on a 9-point Likert scale. They were required to consider the degree of the semantic association (‘how much do you think the two words are semantically related?’).

TMS stimulation

The rTMS was delivered using a Magstim Super Rapid magnetic simulator with a figure-of-eight coil (double wings of 70-mm diameter). We applied rTMS (5 Hz frequency, 1 s, inter-train interval of 5 s) at 100% of the motor threshold on left DLPFC (F3; BA9) immediately upon each retrieval word appearance.

The approximate location of the left DLPFC was automatically identified on the subject's scalp using the SofTaxic navigator system (Brainsight Magstim, SofTaxic Optic 2.0), which uses a set of digitized skull landmarks (nasion, inion, and two preauricular points), and approximately 50 scalp points entered with a FastarackPolhemusdigiter system and an averaged stereotaxic magnetic resonance imaging (MRI) brain atlas in Talairach space.[28] The Talairach coordinates of cortical sites underlying the coil locations were estimated on the basis of an MRI-constructed stereotaxic template (accuracy approx. 1 mm, Talairach space). This scan procedure suggested that TMS was applied over the DLPFC (Talairach −10,40,25 coordinates, medial frontal gyrus; Fig. 2).

Figure 2.

Stimulation area (F3; BA9).

To control the effect of the rTMS stimulation we adopted two control conditions: stimulation of a cortical control site (Cz); and a sham condition (no stimulation). During the sham condition the same intensity and timing of stimulation was used but the coil was held in a way in which no magnetic stimulation reached the brain, because the TMS coil was placed at a 45° angle to the head and the point of maximum activation was superficial compared with active stimulation.[29] Thus, all subjects received 180 trains of rTMS over the left DLPFC, 180 trains of rTMS over the control site (Cz vertex) and 180 trains of rTMS in the sham phase (no real stimulation). The stimulation phase was subdivided into three blocks (total duration of each block, approx. 20 min), with an inter-block interval of approximately 5 min. The stimulation condition order was randomly assigned and counterbalanced.[30] Single-pulse TMS was applied at increasing intensities to determine individual motor threshold by standard procedure.[30]

State–Trait Anxiety Inventory

After the experiment the subjects were asked to answer the State–Trait Anxiety Inventory (STAI).[31] The STAI were scored using norms from the Italian version of the Manual for State Trait Anxiety Inventory (reliability, Cronbach's α = 0.86).[31] Based on the STAI total scores we distinguished two groups with regard to trait anxiety: a high trait-anxiety group (score ≥43) and a low trait-anxiety group (≤39).[32]


Two-factorial mixed-design ANOVAs with four independent factors (Trait anxiety; low/high-anxiety subjects; old/new words, ON; positive/negative emotional content, EC; Condition, F3/control site/sham) were applied on the dependent measures of accuracy (total of correct response/total occurrence for each category) and RT measures. Type I errors associated with inhomogeneity of variance were controlled by decreasing the degrees of freedom using the Greenhouse–Geiser epsilon. Successively, independent ANOVAs were performed for the two distinct subcategories of related and unrelated distractors.

General analysis

A significant main effect was found for ON (F(1,26) = 7.11, P = 0.002; η2 = 0.42), Trait × EC (F(1,52) = 8.78, P ≤ 0.001; η2 = 0.39), Condition × EC (F(2,52) = 9.88, P ≤ 0.001; η2 = 0.40), as well as the for interaction effect Trait × Condition × EC (F(2,52) = 7.76, P = 0.010; η2 = 0.37; Table 1). First, the old stimuli were better recognized than new stimuli (reduced RT). Moreover, as shown by the contrast effects (contrast effect for ANOVA), high-trait anxiety subjects reduced RT in response to negative more than positive stimuli (F(1,26) = 7.79, P = 0.010; η2 = 0.38), whereas low-anxiety subjects did not show significant differences as a function of the stimulus valence. In addition, RT decreased (increased efficiency) in the case of left DLPFC stimulation more for positive stimuli than for negative stimuli (F(1,26) = 4.78, P = 0.020; η2 = 0.35). A consistent reduction of RT was also noted for positive stimulus in left DLPFC stimulation in comparison with control (F(1,26) = 4.51, P = 0.023; η2 = 0.32) and sham (F(1,26) = 4.39, P = 0.026; η2 = 0.31) conditions (Fig. 3). Finally, high-anxiety subjects reduced RT in response to positive stimuli in left DLPFC stimulation in comparison with negative stimuli (F(1,26) = 8.19, P ≤ 0.001; η2 = 0.38; Fig. 4a). In contrast, in the sham and control conditions RT was lower in response to negative stimuli than positive stimuli (F(1,26) = 9.87, P ≤ 0.001; η2 = 0.39). Low-anxiety subjects were more efficient (reduced RT) in response to positive stimuli in the case of TMS applied to the left DLPFC (F(1,26) = 9.65, P ≤ 0.001; η2 = 0.38; Fig. 4b).

Figure 3.

Response time (RT) modulation as a function of stimulation area (F3, left dorsolateral prefrontal cortex; Cz, central control site; sham, no stimulation) and stimulus valence (image, negative vs image, positive).

Figure 4.

Response time (RT) modulation as a function of stimulus valence (image, negative vs image, positive) for (a) high-anxiety and (b) low-anxiety subjects. F3, left dorsolateral prefrontal cortex; Cz, central control site; sham, no stimulation.

Table 1. AI and RT as a function of stimulation condition and emotional valence for old and new stimuli
High-anxiety subjectsOldNew relatedNew unrelated
AIRT (ms)AIRT (ms)AIRT (ms)
  1. AI, accuracy index (total occurrence of correct response/total occurrence); RT, response time.
Cz (control)
Low-anxiety subjects
Cz (control)

Related distractors

For the related distractor category significant effects were found for ON (F(1,26) = 7.90, P ≤ 0.001; η2 = 0.39), Condition × EC (F(1,52) = 8.97, P ≤ 0.001; η2 = 0.46), ON × Condition (F(1,52) = 8.16, P ≤ 0.001; η2 = 0.45) and Trait × EC × Condition (F(1,52) = 6.98, P = 0.010; η2 = 0.36). Specifically, a generally worse performance was identified for new (distractors) than old (target) stimuli, with significant increased RT. For the interaction effects, as shown by contrast analysis, positive stimuli were more efficiently processed (shorter RT) than negative stimuli (F(1,26) = 8.90, P ≤ 0.001; η2 = 0.45) when TMS stimulation was performed on the left DLPFC. Moreover, increased RT was found for distractors than targets in Cz (F(1,52) = 8.13, P ≤ 0.001; η2 = 0.46) and sham (F(1,52) = 7.89, P ≤ 0.001; η2 = 0.42) stimulation.

A significant interaction was found with anxiety level, because high-anxiety subjects had the worst performance for related distractors in comparison with target stimuli (F(1,26) = 8.98, P ≤ 0.001; η2 = 0.45) in terms of higher RT. The negative-valenced distractors were generally better recognized (reduced RT) than positive distractors by high-anxiety subjects. In addition, the negative distractors had reduced RT in Cz (F(1,26) = 10.77, P ≤ 0.001; η2 = 0.44) and sham condition (F(1,26) = 10.98, P ≤ 0.001; η2 = 0.45). In contrast, higher-anxiety subjects were better at discarding the positive distractors as new than the negative distractors (for RT, F(1,26) = 9.02, P ≤ 0.001; η2 = 0.45; for accuracy, F(1,26) = 8.16, P ≤ 0.001; η2 = 0.44) when TMS was applied on the left DLPFC. Moreover, positive distractors were better rejected (shorter RTS) in the case of TMS stimulation than in the Cz (F(1,26) = 10.11, P ≤ 0.001; η2 = 0.45) and sham (F(1,26) = 8.76, P ≤ 0.001; η2 = 0.40) conditions.

Low-anxiety subjects had increased performance (reduced RT) for positive distractors in response to TMS as compared to the Cz (F(1,26) = 9.10, P ≤ 0.001; η2 = 0.43) and sham (F(1,26) = 9.02, P ≤ 0.001; η2 = 0.45) conditions. No other difference was statistically significant.

Unrelated distractors

Similarly, for the unrelated distractors a significant effect was found for Condition × EC (F(1,26) = 9.12, P ≤ 0.001; η2 = 0.46), ON × EC × Condition (F(1,52) = 9.54, P ≤ 0.001; η2 = 0.47) and Trait × EC × Condition (F(1,52) = 6.98, P = 0.010; η2 = 0.43). RT was reduced in response to TMS stimulation on the left DLPFC when subjects recognized more positive than negative stimuli (F(1,26) = 8.70, P ≤ 0.001; η2 = 0.46). The second significant interaction effect was a general increased efficiency in processing old positive stimuli compared to negative stimuli in the case of TMS stimulation on the left DLPFC (F(1,26) = 7.09, P ≤ 0.001; η2 = 0.38). The same trend was observed for the distractors, because the positive distractors had reduced RT compared to the negative distractors (for RT, F(1,26) = 9.50, P ≤ 0.001; η2 = 0.46) when TMS was applied to the left DLPFC. Finally, high-anxiety subjects were better at discarding (shorter RT) the positive distractors as new than the negative distractors (F(1,26) = 9.02, P ≤ 0.001; η2 = 0.45), when TMS was applied on the left DLPFC. Moreover, positive distractors were better rejected in the case of TMS stimulation compared to the Cz (F(1,26) = 10.11, P ≤ 0.001; η2 = 0.45) and sham (F(1,26) = 8.72, P ≤ 0.001; η2 = 0.40) conditions. The negative valenced distractors were generally better recognized (reduced RT) than positive distractors by high-anxiety subjects. In addition, the negative distractors had reduced RT in the Cz (F(1,26) = 10.77, P ≤ 0.001; η2 = 0.44) and sham conditions (F(1,26) = 10.90, P ≤ 0.001; η2 = 0.45). Low-anxiety subjects had increased performance (reduced RT) for positive distractors in response to TMS compared to the Cz (F(1,26) = 9.10, P ≤ 0.001; η2 = 0.43) and sham (F(1,26) = 9.08, P ≤ 0.001; η2 = 0.45) conditions.


In the present study we were concerned primarily with the effects of anxiety on memory performance for emotional cues, with an emphasis on anxiety levels within the normal population. The main focus was on tasks that place significant demands on cognitive resources, such as emotional information retrieval when old (target) and new (distractors) stimuli have to be recognized. Valence effect and semantic interference were also considered to verify the contribution of the left DLPFC in highly complex retrieval processes in relationship with anxiety.

A first general consideration is related to the anxiety level effect. High-anxiety subjects had a generally reduced RT in retrieving emotional information compared to the low-anxiety subjects, as a function of stimulus valence. This result might be represented as a consistent attentional bias toward the potential threatening information by more anxious people. In contrast, low-anxiety subjects were equally responsive to positive and negative categories without significant differences in terms of effectiveness (accuracy) and efficiency (RT). Some support to the assumption that anxiety facilitates the detection of negative aversive information comes from previous studies on attentional bias,[33, 34] in which anxious individuals preferentially attend to or have delayed disengagement from threat-related stimuli. The assumption that anxiety typically reduces attentional control was also related to the existence of two attentional systems, a goal-directed attentional system influenced by knowledge and expectations, and a stimulus-driven attentional system responding to more salient stimuli.[35] The latter should be recruited during the detection of relevant information, and anxiety might intervene to disrupt the balance between these two attentional systems.

A second main effect was related to the cortical correlates underlying the memory retrieval process in the case of emotional stimuli. The contribution of the left DLPFC in modulating the response to emotional cues was elucidated. An increased responsiveness to positive emotional cues was confirmed in the case of induced left DLPFC stimulation. Moreover, this effect was confirmed and accentuated with regard to the anxiety level. High-anxiety subjects had a clear main effect of left DLPFC on performance. The enhancement of the left brain activity produced an increased efficiency (reduced RT). This result would confirm the reduction of the ‘unbalance effect’, which produces an attentional bias toward the negative cues. The valence model of emotional cue processing may explain the subjects' performance.[9, 22] When a functional condition is restored between the two cortical systems deputed to elaborate, respectively, the positive and negative emotional cues, the high-anxiety subjects had a significant reduction of the pre-existing attentional bias. These conclusions are in line with previous results on clinical (such as panic disorder and post-traumatic stress disorder) and subclinical samples[12, 20] that postulated an anxiolytic effect of low-frequency TMS on the right DLPFC. An interesting study that applied low-frequency TMS (cortical de-potentiation) to the frontal left hemisphere identified increased attentional negative bias and a clear anxiety behavior.[36]

Third, a possible interference effect was tested, by comparing semantically related/unrelated distractors. Specifically, we first found a significant increasing of the cognitive costs (worsening performance with higher RT) for the related distractors in the whole sample, with a significant reduced performance for distractors than targets. The related distractor processing may figure as the more complex and effortful cognitive condition, given that it includes both the necessity to reject the irrelevant information and to control the interference induced by the similarity with the target cues.[23] This should have affected the general performance with significantly increased workload for the subjects, especially those with higher levels of anxiety.

The cortical modulation induced by TMS, however, produced a significant reduction of the semantic interference (for the related distractors) and a general improvement of the cognitive efficiency (for the unrelated distractors) in response to a specific stimulus valence. A more significant effect was found as a function of the anxiety level. In fact, for the high-anxiety subjects the TMS stimulation produced more significant and massive improvement in terms of both effectiveness (accuracy) and efficiency (RT) in recognizing the related positive distractors.

As a consequence, the present results provide evidence in favor of the assumption that the potentiation effect induced by TMS on the left frontal side would be more effective in the case of a need for a higher cognitive demand.[15] It is possible that the left DLPFC intervenes to modulate these complex processes, because this is the more cognitively expensive condition, in that it requires specific competencies and a higher workload, with an increased effort for the cognitive system. In parallel, the high-anxiety subjects were more efficient in response to this cognitively effortful condition, benefiting in significant measure from the frontal left stimulation.


This study was supported by Catholic University Research Found D 1.1. 2010. No conflict of interest, financial or other, exists for the present paper.