The time course of ventrolateral prefrontal cortex involvement in memory formation.

Human neuroimaging studies have implicated a number of brain regions in long-term memory formation. Foremost among these is ventrolateral prefrontal cortex. Here, we used double-pulse transcranial magnetic stimulation (TMS) to assess whether the contribution of this part of cortex is crucial for laying down new memories and, if so, to examine the time course of this process. Healthy adult volunteers performed an incidental encoding task (living/nonliving judgments) on sequences of words. In separate series, the task was performed either on its own or while TMS was applied to one of two sites of experimental interest (left/right anterior inferior frontal gyrus) or a control site (vertex). TMS pulses were delivered at 350, 750, or 1,150 ms following word onset. After a delay of 15 min, memory for the items was probed with a recognition memory test including confidence judgments. TMS to all three sites nonspecifically affected the speed and accuracy with which judgments were made during the encoding task. However, only TMS to prefrontal cortex affected later memory performance. Stimulation of left or right inferior frontal gyrus at all three time points reduced the likelihood that a word would later be recognized by a small, but significant, amount (approximately 4%). These findings indicate that bilateral ventrolateral prefrontal cortex plays an essential role in memory formation, exerting its influence between > or = 350 and 1,150 ms after an event is encountered.

INTRODUCTION

The neural bases of human long-term memory have traditionally been studied by assessing the effects of brain damage on memory performance. This work has suggested that encoding information into episodic memory critically depends on a number of interconnected brain regions, in particular the medial temporal lobe and prefrontal cortex (Milner and Petrides 1984; Shimamura 2003; Squire and Knowlton 2000). It is difficult to know from patient data alone, however, whether such regions reflect processing deficits at encoding, retrieval, or both (e.g., Fletcher et al. 1997).

Our understanding of the brain regions specifically associated with memory formation was advanced when techniques were developed to measure neural activity in healthy people. These techniques allow neural activity to be recorded in response to individual events (Dale and Buckner 1997; Josephs et al. 1997; Zarahn et al. 1997) and this activity can be used to predict whether an event is successfully encoded into memory. This is done by contrasting event-related activity as a function of whether the events are remembered or forgotten in a later memory test (the “subsequent memory” approach; Sanquist et al. 1980).

This method has been used with functional magnetic resonance imaging (fMRI) since the late 1990s (Brewer et al. 1998; Wagner et al. 1998b; for reviews, see Paller and Wagner 2002; Wagner et al. 1999). One insight that has been gained is that multiple brain mechanisms are associated with effective episodic encoding. Activity in different sets of brain regions predicts encoding success, depending on the type of material that has to be stored into memory (Kirchhoff et al. 2000; Paller and Wagner 2002; Wagner et al. 1999), the type of processing engaged when that information is encountered (Baker et al. 2001; Davachi et al. 2001; Fletcher et al. 2003; Mitchell et al. 2004; Otten and Rugg 2001a,b), the way in which a memory is later retrieved (Davachi et al. 2003; Otten 2007; Ranganath et al. 2004), and the degree to which study and test processes overlap (Park and Rugg 2008).

A region that has consistently been implicated in the encoding of verbal material during semantic processing tasks is the left prefrontal cortex. Substantial numbers of studies have shown that activity in this region is enhanced for words that are remembered as opposed to forgotten in a later memory test. These effects primarily occur in ventrolateral prefrontal cortex, in particular anterior (Brodmann area [BA] 45/47) and/or posterior (BA 45/44) regions of the left inferior frontal gyrus (IFG; e.g., Baker et al. 2001; Braver et al. 2001; Davachi et al. 2001; Henson et al. 1999; Kirchhoff et al. 2000; Otten and Rugg 2001a; Otten et al. 2001; Wagner et al. 1998b). Effects can extend to more dorsolateral regions (BA 9/46), especially during intentional memorization and associative learning tasks (Blumenfeld and Ranganath 2007; Braver et al. 2001). The effects in left anterior IFG are thought to reflect the influence of semantic elaboration on episodic encoding. This part of cortex has been associated with the processing of a word's semantic attributes (Poldrack et al. 1999) and has also been linked to semantic working memory (Gabrieli et al. 1998). Encoding-related effects in this region may therefore signal the degree to which semantic features are processed and manipulated when information is first experienced (Buckner and Koutstaal 1998; Buckner et al. 2000; Otten et al. 2001; Wagner et al. 1998b, 1999). Events that engage semantic working memory more are likely to result in richer representations with additional details. Such representations are easier to retrieve in a later memory test (Craik and Lockhart 1972; Craik and Tulving 1975).

Although fMRI has been valuable in emphasizing a role for left prefrontal cortex in laying down new memories in healthy adults, it is not possible on the basis of fMRI data alone to determine whether neural activity is necessary for, or merely incidental to, memory formation. fMRI and, indeed, all neuroimaging techniques are inherently correlational. The primary aim of the current study was to establish whether left anterior IFG plays a crucial role in memory encoding. In addition, we sought to examine during what period of time activity in left prefrontal cortex contributes to encoding. The causal role of a brain region and its time course can be assessed with transcranial magnetic stimulation (TMS), which allows the temporary, reversible disruption of neural activity from superficial parts of the brain. It has a spatial resolution precise enough to target functionally specific brain areas and its temporal resolution can parse perceptual and cognitive processes on a millisecond basis (Walsh and Pascual-Leone 2003). The approximate time at which a region exerts its influence on a task can be identified by observing the effects of disrupting activity in this region at different points in time.

There are currently only a handful of studies that have used TMS to address the causal role of prefrontal cortex in long-term memory formation. These studies are listed in Table 1 (for an early review of TMS and memory, see Grafman and Wassermann 1999). The majority of studies (Epstein et al. 2002; Rami et al. 2003; Rossi et al. 2001, 2004, 2006; Sandrini et al. 2003; Škrdlantová et al. 2005) have investigated dorsolateral prefrontal cortex, a region related to executive control processes such as manipulating multiple events in working memory (Blumenfeld and Ranganath 2007; Braver et al. 2001). TMS to this part of cortex leads to decrements in later memory performance with item pairs and complex events. Kahn et al. (2005) focused on a posterior region of ventrolateral prefrontal cortex (involved in phonological processing; Poldrack et al. 1999) and found that stimulation of this region at around 400 ms resulted in a complex pattern of decrements and enhancements of later confident and unconfident recognition of words and pseudowords.

Table 1.

Overview of TMS studies on the causal role of prefrontal cortex in long-term memory formation

Publication	Study Task	Memory Test	Target Area	Time of Stimulation	Control Conditions	Main Findings
Grafman et al. (1994)	Reading of short word lists	Free recall	F7/8, T5/6, P3/4, O1/2	0–500, 250–750 ms (L/R), 500–1,000, 1,000–1,500 ms (L); rTMS at 20 Hz	Sham (noise only)	Impaired recall after L temporoparietal and L/R ventrolateral prefrontal stimulation before 500 ms (relative to sham)
Rossi et al. (2001, 2004)	Indoor/outdoor judgments on complex scenes	Old/new recognition	L/R dorsolateral prefrontal cortex	0–500 ms; rTMS at 20 Hz	Sham (reoriented coil); no TMS	Impaired recognition after L prefrontal stimulation (relative to sham and no TMS)
Epstein et al. (2002)	Intentional memorization of pairs of words and abstract shapes	Matching of shapes to words	L/R dorsolateral prefrontal cortex	680–740 ms; double pulse TMS	Control site TMS (vertex); no TMS	Impaired recall of paired associates after R prefrontal stimulation (relative to L stimulation, sham and no TMS)
Rami et al. (2003)	Logical memory of Rivermead Behavioural Memory Test	Logical memory of Rivermead Behavioural Memory Test	L dorsolateral prefrontal cortex (1 and 5 Hz); R dorsolateral prefrontal cortex and R cerebellum (5 Hz only)	10 s trains of 1 or 5 Hz rTMS with 30 s intervals during study and test	No TMS	Impaired memory performance with 5 Hz R prefrontal stimulation (relative to the other TMS conditions and no TMS)
Sandrini et al. (2003)	Semantic relatedness judgments on word pairs	Cued forced-choice recognition	L/R dorsolateral prefrontal cortex	0–500 ms; rTMS at 20 Hz	Sham (upside-down coil); no TMS	Impaired recognition of unrelated word pairs after L and R prefrontal stimulation (relative to sham and no TMS)
Floel et al. (2004)	Intentional memorization of words and abstract shapes	Old/new recognition with confidence judgments	L/R anterior inferior frontal gyrus (BA 45/47)	0–500 ms; rTMS at 20 Hz	Sham (reoriented coil); no TMS	Impaired confident and overall recognition of words after L prefrontal stimulation and of pictures after R prefrontal stimulation (relative to sham)
Köhler et al. (2004)	Abstract/concrete judgments on words	Old/new recognition with confidence judgments	L/R anterior inferior frontal gyrus (BA 45/47)	200–800 ms; rTMS at 7 Hz	Control site TMS (angular gyrus); no TMS	Enhanced confident and overall recognition after L prefrontal stimulation (relative to control site and no TMS)
Kahn et al. (2005)	Syllable judgments on words and pseudowords	Old/new recognition with confidence judgments	L/R posterior inferior frontal gyrus (BA 44/9)	250, 300, 320, 340, 350, 360, 370, 380, 390, 400, 600 ms; single pulse TMS	No TMS	Enhanced unconfident recognition of words after R prefrontal stimulation at 380 ms, and impaired confident recognition of words after L prefrontal stimulation at 380 ms (relative to no TMS)
Škrdlantová et al. (2005)	Intentional memorization of words and faces	Free recall for words, forced-choice recognition for faces	L dorsolateral prefrontal cortex	0.9 Hz rTMS throughout study task	Sham (reoriented coil)	Impaired free recall of words after L prefrontal stimulation (relative to sham)
Rossi et al. (2006)	Indoor/outdoor judgments on complex scenes	Old/new recognition	L/R dorsolateral prefrontal cortex; L/R intraparietal sulcus	0–500 ms; rTMS at 20 Hz	Sham (reoriented coil); no TMS	Impaired recognition after L prefrontal stimulation (relative to sham and no TMS)

All stimulation took place after study items except where indicated. Rossi et al. (2004) also studied a group of elderly individuals. These results are not considered here. Subjects in the prefrontal TMS conditions in Rossi et al. (2006) were a subset of those reported in Rossi et al. (2001). L = left; R = right.

Three studies have previously examined the contribution of anterior ventrolateral prefrontal cortex, the region typically associated with semantic encoding and the region of interest in the present experiment. Grafman et al. (1994) asked volunteers to read short word lists for later recall. Stimulation of left and right anterior IFG shortly after word onset impaired later recall of a word, relative to a situation where TMS noise was simulated, although no stimulation took place. Floel et al. (2004) studied the intentional encoding of words and abstract shapes. TMS to left anterior IFG over a 500-ms period after item onset impaired later confident and overall recognition of words. The same stimulation to the right hemisphere, by contrast, impaired recognition of shapes. These effects were found relative to TMS to a midfrontal region. Finally, Köhler et al. (2004) also addressed how anterior IFG affects the incidental encoding of words. In an abstract/concrete task, stimulation of this region between 200 and 800 ms after word onset affected later memory performance to the word. This took the form of an increase in later confident and overall recognition after left prefrontal stimulation, relative to TMS to parietal cortex or no TMS.

The above-cited studies have demonstrated the feasibility of using TMS on anterior IFG and also provided preliminary evidence for a causal role of this region in memory formation. However, there remain few studies and a diversity of findings that invite further work. In particular, it is currently unknown what the critical time period is for the contribution of anterior IFG to effective encoding. The previous studies (Floel et al. 2004; Grafman et al. 1994; Köhler et al. 2004) suggest that the anterior IFG is especially important for memory formation immediately after an event is encountered. However, TMS has been applied over a prolonged interval (∼500 ms) only shortly after event onset and it remains an open question whether there is any differentiation within this period or any influence extending beyond it. If activity in anterior IFG reflects the degree to which semantic attributes are processed and manipulated in semantic working memory (Buckner and Koutstaal 1998; Buckner et al. 2000; Otten et al. 2001; Wagner et al. 1998b, 1999), it would be expected that the influence of IFG starts after perceptual processing has taken place and continues well beyond 500 ms.

Here, we used double-pulse TMS to elucidate the time course of anterior IFG. This method maintains temporal specificity while having greater interference effects than those of single-pulse TMS due to summation of the two pulses (e.g., O'Shea et al. 2004; Pitcher et al. 2007). The first pulse was delivered at 350, 750, or 1,150 ms following item onset, with the second pulse given 40 ms later. These times were guided by encoding studies using electrical brain activity (for reviews, see Friedman and Johnson Jr 2000; Paller and Wagner 2002; Rugg 1995; Wagner et al. 1999). The most consistently observed effect is a positive deflection over frontal scalp sites in response to words that are later remembered as opposed to forgotten. The functional interpretation of this deflection is the same as that offered for fMRI activity in left anterior IFG. That is, both effects are thought to reflect the increased processing of a word's semantic attributes, thus increasing the likelihood that the word will later be remembered (e.g., Otten et al. 2007). Although inferring the intracerebral origins of scalp-recorded electrical signals is difficult, a link has been drawn between this positive deflection and activity in left prefrontal cortex (Friedman and Johnson Jr 2000).

The positive deflection related to encoding usually starts around 400 ms and persists for several hundred milliseconds, up to 1 s or even 2 s (e.g., Paller and Wagner 2002; Rugg 1995; Wagner et al. 1999). The semantic processes associated with the modulation are thus expected to be engaged during at least that period of time (Otten and Rugg 2004). The earliest point of TMS application in the present experiment was 350 ms, which was chosen to fall at around the start of encoding-related processes in IFG. The later time points, 750 and 1,150 ms, allowed us to define the longevity of these processes.

Healthy adults completed an incidental encoding task (living/nonliving judgments) on series of words, followed by a recognition memory test incorporating confidence judgments. The encoding task was performed either on its own or while TMS was applied to one of two sites of experimental interest (left or right anterior IFG) or a control site (the vertex). These TMS conditions were administered in different task blocks, with pulses delivered randomly at the three time points within each block. We chose this design to assess both specific and nonspecific effects of TMS. Applying TMS is accompanied by tactile and auditory sensations due to the discharging of the TMS coil and controlling for these is important (Marzi et al. 1998; Walsh and Pascual-Leone 2003). To ensure meaningful inferences about the contribution of anterior IFG, the design included TMS to the vertex, a site known not to be involved in memory formation (cf. Epstein et al. 2002). Task performance can then be contrasted across prefrontal and control site TMS conditions, removing any general changes caused by tactile and auditory stimulation and isolating changes specifically brought about by disrupting activity in anterior IFG. Because it is important to monitor for nonspecific effects (Walsh and Pascual-Leone 2003), task performance was also assessed without TMS.

The question of interest was how stimulation of anterior IFG at 350, 750, and 1,150 ms affects later memory performance, relative to stimulation of the control site. The focus in the present study was on the role of left anterior IFG, but left as well as right IFG were stimulated to discern the contribution of both hemispheres to verbal encoding. It is commonly believed that the left hemisphere is especially important for the encoding of verbal information in healthy younger adults (e.g., Paller and Wagner 2002; Wagner et al. 1999). An earlier TMS study supports this idea (Floel et al. 2004), as do early neuroimaging studies using block designs (Kelley et al. 1998; McDermott et al. 1999; Wagner et al. 1998a). However, few experiments have directly compared the encoding of verbal and nonverbal material with the subsequent memory approach (for an exception, see Kirchhoff et al. 2000) and several experiments that have used verbal material show bilateral activity, albeit more strongly on the left (e.g., Otten et al. 2001). It is therefore not clear to what extent the right PFC contributes to the encoding of verbal material.

METHODS

Participants

The experimental procedures were approved by the joint research ethics committee of the Institute of Neurology and National Hospital for Neurology and Neurosurgery. Fifteen native English-speaking volunteers were recruited via local advertisement and paid £7.50 per hour to take part in the experiment. All gave written informed consent prior to participation. All volunteers were healthy, had no history of neurological or psychiatric illness, had normal or corrected-to-normal vision, and were right-handed (except one ambidextrous volunteer). Their average age was 25 yr (range, 19–32 yr) and eight were women. Four additional volunteers were recruited but not able to complete the experiment because they could not tolerate stimulation at the required minimum 50% threshold of maximum output (see TMS procedures in the following text).

Tasks and procedure

The design of the experiment is illustrated in Fig. 1. The experiment was composed of an incidental study phase, followed by a surprise recognition memory test. At study, participants viewed a sequence of 480 words, presented one word at a time. For each word, the task was to decide whether the word was animate or referred to the property of a living entity. One of two buttons had to be pressed with the left or right index finger (responding hand counterbalanced across subjects). Both speed and accuracy were stressed. A brief practice session familiarized participants with the task.

Fig. 1.

Outline of the study phase of the experiment. Volunteers viewed sequences of words, presented one at a time and separated by 2–3 s. A living/nonliving judgment had to be made on each word. In different blocks, the task was either performed on its own or while transcranial magnetic stimulation (TMS) was administered to one of 3 sites (left anterior inferior frontal gyrus, right anterior inferior frontal gyrus, or vertex). The vertex served as a control site to assess nonspecific effects of TMS. The order of blocks was randomized across volunteers. During the TMS blocks, neural activity was disrupted on each trial with two consecutive TMS pulses separated by 40 ms (double-pulse TMS). Stimulation started at 350, 750, or 1,150 ms, randomly allocated across trials. Following all 4 blocks, volunteers performed a surprise recognition memory test incorporating confidence judgments.

The study sequence was divided into four blocks of 120 words. A different TMS condition was administered in each block. In one of the blocks, the animacy task was performed on its own, without concurrent TMS. The no-stimulation baseline was assessed in a separate task block rather than by interleaving no-TMS and TMS trials (cf. Floel et al. 2004; Kahn et al. 2005; Köhler et al. 2004), given that the mere anticipation of TMS can alter performance and baseline trials can be treated differently depending on the type of TMS with which they are intermixed. In the other three task blocks, decisions were made while TMS was applied to the left anterior IFG, right anterior IFG, or the vertex. The order of the four blocks was randomized across subjects. Left and right anterior IFG were sites of experimental interest. Following Epstein and colleagues (2002), the vertex was used as a control site to assess nonspecific effects of TMS. Large numbers of brain regions have been implicated in successful memory formation, including some used as control sites in previous TMS work (e.g., Köhler et al. 2004). However, there is no evidence to suggest that encoding information into long-term declarative memory relies on the central region beneath the vertex (for reviews, see Blumenfeld and Ranganath 2007; Paller and Wagner 2002; Wagner et al. 1999). During TMS blocks, neural activity was disrupted on each trial with two consecutive TMS pulses separated by 40 ms (double-pulse TMS). As mentioned in the introduction, double-pulse stimulation was used because it maintains some temporal specificity, in this case 40 ms, while having greater interference effects than those of single-pulse TMS due to summation of the two pulses (O'Shea et al. 2004; Pitcher et al. 2007). The first pulse of each pair was delivered at either 350, 750, or 1,150 ms after onset of the study word. The time of stimulation was random on any given trial, with the restriction that stimulation occurred equally often at each time point within a block.

Fifteen minutes after the end of the study phase, volunteers were told that their memory for the words would be tested. TMS was not applied during the test phase. All studied words were presented again, along with words not encountered previously in the experiment. For each word, volunteers were asked to make an old/new recognition judgment, at the same time indicating whether they were confident or unconfident about their decision. One of four buttons had to be pressed with the index and middle fingers of the left and right hands according to the “confident old,” “unconfident old,” “unconfident new,” and “confident new” decision. Confident judgments were always mapped to the middle finger and old/new responses to either the left or right hand (responding hand balanced across participants). Both speed and accuracy were stressed in the instructions. Volunteers first received a short practice list, followed by six blocks of 120 words each. At both study and test, short rest breaks were given in the middle of a block and in between blocks.

Stimuli

The experimental sequences were created from a pool of 843 concrete nouns, selected from the Medical Research Council Psycholinguistics Database (http://www.psy.uwa.edu.au/mrcdatabase/uwa_mrc.htm; Coltheart 1981). The words were between four and nine letters in length and had a written frequency of 1–100 occurrences per million (Kučera and Francis 1967). Six sets of 120 words each were selected at random from this pool, with the restriction that each set had equal numbers of words denoting living and nonliving items and equal distributions of word lengths. Four of these sets were used to create a study list of 480 words, with the remaining two sets added to create a test list of 720 words. The study list was divided into four blocks of 120 trials each, corresponding with the four TMS conditions. The assignment of sets to experimental conditions varied across subjects so that each word occurred as either “new” or “old” and in each TMS condition. Each subject received a different ordering of the words. A further 30 words were selected from the word pool to create practice lists.

All words were presented in central vision on a computer monitor in a black Helvetica font against a gray background. Words were presented one at a time for 300 ms. A neutral warning stimulus (an exclamation mark) was presented 500 ms before each word for 400 ms. There was a 100-ms blank period between the offset of the warning stimulus and the onset of the word. At study, the time in between the onset of a word and the onset of the next warning stimulus varied randomly between 2 and 3 s. At test, the variation was between 2.5 and 3.5 s. A fixation point (a plus sign) was continuously present on the screen other than when stimuli were presented. Words subtended approximate visual angles of 0.7° vertically and 1.3–4° horizontally.

Transcranial magnetic stimulation

TMS pulses were delivered with a Magstim Super Rapid stimulator (Magstim, Whitland, Dyfed, UK) and 70-mm figure-of-eight coils. The prefrontal regions targeted for disruption were localized for each volunteer individually with frameless stereotaxy (Brainsight; Rogue Research, Montreal, Quebec, Canada). To pinpoint the left and right anterior IFG, each volunteer's high-resolution T1-weighted structural MRI scan (acquired on a previous occasion) was spatially transformed into a common coordinate space using the Montreal Neurological Institute normalized canonical brain (Evans et al. 1993) and FSL software (Oxford Centre Functional MRI of the Brain, Univ. of Oxford). These normalized scans were used to transform Talairach coordinates into coordinates in the original structural image. The positions of left and right anterior IFG were estimated by taking the mean Talairach coordinates from Köhler et al. (2004) who used fMRI to identify these regions. Similar coordinates have been observed in other fMRI experiments investigating the encoding of verbal material during animacy tasks (e.g., Morcom et al. 2003; Otten et al. 2001). The Talairach coordinates for the left hemisphere were [x = −48, y = 35, z = 5] and those for the right hemisphere were [x = 45, y = 35, z = 5]. These coordinates roughly correspond with BA 45/47.

The scalp locations for the TMS coils were identified by coregistering each volunteer's head with the structural MRI. A Polaris infrared tracker (NDI, Waterloo, Ontario, Canada) was used to find those locations whose stimulation would direct the maximum field intensity to the two prefrontal locations. These locations were marked on a tight-fitting elasticated cap, which volunteers wore throughout the experiment. The location of the control site, the vertex, was also marked on the cap. This site was measured by taking the midpoints of the distance between the nasion and inion and the left and right preaurical points (Jasper 1958).

The TMS coils were positioned at the designated scalp locations. Coils were replaced and cooled every 60 trials to prevent overheating. TMS intensity was initially set at 120% of motor threshold (the lowest stimulus intensity at which two of three monophasic TMS pulses to the right primary motor hand area produced a minimum finger twitch in the contralateral hand). Intensity was then lowered until stimulation was comfortable enough to complete the experiment. A minimum level of 50% of maximum stimulator output was required to participate in the study. Intensity varied between 50 and 60% of maximum output across volunteers. The TMS coils were held such that the handle pointed downward during stimulation of the prefrontal sites, except for four volunteers, who claimed discomfort with that orientation. For those volunteers, the angle was adjusted until the block could be completed. In the vertex condition, the handle of the coil was placed parallel to the interhemispheric fissure.

During pilot work, we noticed that some participants reported finding TMS to the two prefrontal regions more uncomfortable than TMS to the vertex. The issue of discomfort has not been raised in previous TMS reports on the role of anterior IFG in encoding (e.g., Floel et al. 2004; Köhler et al. 2004). Nonetheless, to ensure that any effects observed in the present study cannot be attributed to discomfort associated with prefrontal TMS, rather than the disruption of prefrontal activity, we asked volunteers to rate their level of discomfort after each TMS condition on a five-point scale. The average rating associated with stimulation of the left prefrontal cortex was 3.6 (range: 2–5) and that with the right cortex 3.5 (range: 1–5). Stimulation of the vertex was always completely comfortable (1). Across-subject correlations between these discomfort ratings and the size of prefrontal TMS-induced effects (performance differences between prefrontal and vertex stimulation) did not reveal any significant correlations for study or test performance. Thus there is no evidence to suggest that discomfort can account for the observed differences in performance (see results).

Analysis procedures

TMS-induced effects were evaluated for performance in the study task and performance in the later recognition memory test. Two types of analysis were carried out: one to assess differences across the four TMS conditions and another to assess differences as a function of time of stimulation. Note that these questions cannot be addressed simultaneously with an omnibus test because TMS condition and time of stimulation were not fully crossed (i.e., the no-TMS condition did not involve stimulation at any point in time).

The first analysis compared performance across the left anterior IFG, right anterior IFG, vertex, and no-TMS conditions to assess specific and nonspecific effects of TMS. Values were collapsed across time of stimulation and entered into repeated-measures ANOVA, incorporating the Greenhouse–Geisser correction for violations against sphericity. Significant differences were pinpointed with three planned comparisons to test the a priori hypotheses. The presence of nonspecific effects was evaluated by contrasting performance across the no-TMS and control site (vertex) conditions. Effects specifically related to disrupting activity in prefrontal cortex were established by comparing performance across the three conditions involving TMS. Given the primary interest in left prefrontal cortex, performance after TMS to this region was contrasted with performance after 1) vertex stimulation to establish a causal role of left prefrontal cortex in memory formation and 2) right prefrontal stimulation to establish the roles of the two hemispheres. Note that TMS to the vertex provides a stricter control than the no-TMS condition to assess effects of prefrontal TMS because it controls for nonspecific effects of the TMS procedure.

The second analysis was directed at the time course of prefrontal cortex. Differences as a function of the instance at which TMS was administered (350, 750, or 1,150 ms) were evaluated with repeated-measures ANOVAs contrasting performance at each time point across the three TMS sites. Reliable differences were localized with post hoc comparisons (Bonferroni-corrected t-test).

Performance in the study task was indexed by how fast and accurately living/nonliving judgments were made. Response times were computed as the mean across all correct responses. For memory performance, two-high-threshold theory (Snodgrass and Corwin 1988) was used to compute measures of recognition accuracy (Pr) and response bias (Br). Pr refers to the difference between the proportions of hits and false alarms and Br to the proportion of false alarms divided by (1 − Pr). Because there was only one false alarm rate, comparing Pr across conditions gives identical results to comparing absolute hit rates. Both Pr and Br were computed irrespective of the confidence with which judgments were made and for confident responses only. Note that these measures are not defined for nonconfident responses in isolation (e.g., Macmillan and Creelman 2005). Recognition performance is reported across all studied words. The same results were obtained when the analyses were restricted to correct study trials.

RESULTS

Study task performance

Table 2 lists the speed and accuracy of living/nonliving judgments during the study task. On average, judgments took 647 ms and were correct on 87% of trials. Collapsed across time of stimulation, speed of responding differed significantly across the four TMS conditions (Fig. 2; see Table 3 for statistics). Judgments were generally made more quickly while TMS was applied relative to the no-stimulation baseline, irrespective of which site was stimulated. The planned comparisons confirmed that TMS nonspecifically affected response speed, as evidenced by a significant difference between the no-TMS and control site conditions. Reaction times did not differ further as a function of specifically disrupting activity in left prefrontal cortex. Neither the comparison between left prefrontal and control site TMS nor that between left prefrontal and right prefrontal TMS was significant.

Table 2.

Study task performance

TMS Condition	Proportion Correct	Response Time, ms
Left anterior IFG
350 ms	0.85 (0.09)	619 (104)
750 ms	0.85 (0.08)	631 (94)
1,150 ms	0.84 (0.09)	618 (90)
Right anterior IFG
350 ms	0.85 (0.08)	645 (118)
750 ms	0.85 (0.06)	631 (87)
1,150 ms	0.85 (0.09)	631 (83)
Vertex
350 ms	0.86 (0.09)	635 (102)
750 ms	0.87 (0.09)	635 (85)
1,150 ms	0.87 (0.08)	641 (102)
No TMS	0.90 (0.05)	691 (86)

Values are across-subject means (SD), n = 15. IFG, inferior frontal gyrus.

Fig. 2.

Study task performance. Response times (bars; left axis) and response accuracy (squares; right axis) of living/nonliving judgments made while the task was performed on its own or while TMS was applied to the left inferior frontal gyrus (IFG), right IFG, or a control site (the vertex). Values are collapsed across time of stimulation. Error bars represent SEs.

Table 3.

Outcome of the statistical analyses

Factor	Study Reaction Time	Study Accuracy	Pr	Br
Overall ANOVA	F(2.00,27.94) = 8.43, P = 0.001	F(2.31,32.28) = 6.89, P = 0.002	F(2.75,38.55) = 4.29, P = 0.012	F(2.55,35.73) = 6.62, P = 0.002
Planned comparisons
Control site vs. no TMS	t14 = −4.04, P = 0.001	t14 = −2.52, P = 0.025	t14 = 2.73, P = 0.016	t14 = 3.46, P = 0.004
Left IFG vs. control site	t14 = −1.23, P = 0.240	t14 = −1.34, P = 0.203	t14 = −2.31, P = 0.037	t14 = −2.69, P = 0.018
Left IFG vs. right IFG	t14 = −1.20, P = 0.250	t14 = 0.04, P = 0.968	t14 = 0.63, P = 0.541	t14 = 0.78, P = 0.451

The overall ANOVAs compared all four TMS conditions (left anterior IFG, right anterior IFG, control site, and no TMS), implementing the Greenhouse–Geisser correction. IFG, inferior frontal gyrus. Control site = vertex.

The same pattern of data emerged for accuracy of responding. Significantly more errors were made during all blocks involving TMS compared with when the task was performed on its own (Fig. 2 and Table 3). Again, the comparison between the control site and no-TMS conditions was statistically reliable, but left IFG did not differ from either vertex or right IFG.

Study task performance was not affected by the time at which TMS was delivered or performance in the later memory test. Repeated-measures ANOVAs using factors of stimulation site (left anterior IFG, right anterior IFG, vertex) and time of stimulation (350, 750, 1,150 ms) did not show significant main or interaction effects for either speed or accuracy of responding (all P > 0.19). Similarly, when study performance was separated according to the type of recognition judgment an item later received, no reliable differences emerged (all P > 0.21). Thus administering TMS generally caused volunteers to respond faster and less accurately during the study task, but this occurred regardless of which site was stimulated, when stimulation took place, and whether a word would later be remembered or forgotten.

Recognition memory performance

Performance in the recognition memory test is shown in Table 4. The speed with which words were recognized was not affected by the TMS condition under which words were encoded or the time at which stimulation occurred (all P > 0.20). However, the accuracy of discriminating between old and new words differed significantly across the four TMS conditions (Fig. 3 and Table 3). Collapsed across confidence, the analysis to establish nonspecific effects of the TMS procedure indicated that words were more likely to be correctly judged as old following stimulation of the control site rather than the no-TMS baseline. In contrast to the study data, recognition accuracy was further affected by the specific disruption of activity in left prefrontal cortex. Pr was significantly lower following stimulation of the left IFG as opposed to the control site, although Pr did not differ between TMS to the left versus right IFG.

Table 4.

Recognition memory performance

Word Type	Recognition Judgment
	Sure “old”	Unsure “old”	Unsure “new”	Sure “new”
A. Proportion of responses
Old
Left anterior IFG
350 ms	0.55 (0.21)	0.20 (0.10)	0.17 (0.12)	0.08 (0.08)
750 ms	0.57 (0.22)	0.20 (0.11)	0.18 (0.12)	0.06 (0.06)
1,150 ms	0.55 (0.22)	0.20 (0.11)	0.17 (0.13)	0.07 (0.07)
Right anterior IFG
350 ms	0.53 (0.24)	0.19 (0.10)	0.19 (0.15)	0.09 (0.09)
750 ms	0.59 (0.22)	0.17 (0.10)	0.18 (0.12)	0.06 (0.07)
1,150 ms	0.55 (0.24)	0.21 (0.12)	0.17 (0.11)	0.07 (0.07)
Vertex
350 ms	0.59 (0.23)	0.19 (0.13)	0.15 (0.13)	0.07 (0.06)
750 ms	0.60 (0.22)	0.20 (0.11)	0.13 (0.11)	0.06 (0.06)
1,150 ms	0.57 (0.22)	0.23 (0.12)	0.13 (0.11)	0.06 (0.08)
No TMS	0.56 (0.20)	0.19 (0.11)	0.17 (0.11)	0.08 (0.07)
New	0.15 (0.14)	0.19 (0.09)	0.39 (0.16)	0.26 (0.14)
B. Mean response time, ms
Old
Left anterior IFG
350 ms	897.0 (127)	1,172.0 (246)	1,260.0 (308)	1,138.0 (290)
750 ms	927.0 (197)	1,260.0 (223)	1,308.0 (339)	1,215.0 (445)
1,150 ms	934.0 (185)	1,217.0 (174)	1,345.0 (297)	1,233.0 (268)
Right anterior IFG
350 ms	926.0 (180)	1,238.0 (196)	1,315.0 (313)	1,224.0 (341)
750 ms	912.0 (129)	1,206.0 (190)	1,259.0 (207)	1,255.0 (247)
1,150 ms	898.0 (152)	1,233.0 (236)	1,308.0 (230)	1,237.0 (296)
Vertex
350 ms	920.0 (198)	1,316.0 (394)	1,298.0 (305)	1,150.0 (267)
750 ms	904.0 (149)	1,192.0 (263)	1,160.0 (257)	1,124.0 (369)
1,150 ms	912.0 (170)	1,158.0 (147)	1,221.0 (279)	1,193.0 (254)
No TMS	920.0 (173)	1,261.0 (214)	1,242.0 (224)	1,160.0 (221)
New	985.0 (184)	1,267.0 (206)	1,246.0 (193)	1,197.0 (129)

Values are across-subject means (SD), n = 15. IFG, inferior frontal gyrus.

Fig. 3.

Memory test performance. Recognition accuracy Pr (bars; left axis) and response bias Br (squares; right axis) for words studied without TMS or while TMS was applied to the left inferior frontal gyrus (IFG), right IFG, or a control site (the vertex). Values are collapsed across time of stimulation. Error bars represent SEs.

Not only recognition accuracy, but also response bias differed across TMS conditions (Fig. 3 and Table 3). Volunteers' willingness to respond “old” to an item showed the same variations as Pr. Again, significant differences occurred between the control site and no-stimulation conditions and the control site and left IFG conditions. The former indicated a nonspecific TMS effect on the likelihood of “old” judgments and the latter that “old” judgments were, in addition, less likely following specific disruption of the left IFG relative to the control site condition. Br did not differ between the left and right IFG conditions. When the analyses were restricted to confident recognition judgments, the same patterns of Pr and Br were observed; however, in that case, they failed to reach significance (P > 0.09).

The time at which TMS was applied during the study phase did not affect Pr or Br, regardless of whether responses were collapsed across levels of confidence. Repeated-measures ANOVAs using factors of stimulation site (left anterior IFG, right anterior IFG, vertex) and time of stimulation (350, 750, 1,150 ms) on the values collapsed across confidence showed reliable main effects [reflecting the decrements associated with left and right IFG stimulation relative to the control site; F(1.89,26.44) = 5.14, P = 0.014 for Pr and F(1.91,26.67) = 6.68, P = 0.005 for Br], but no reliable interactions (all P > 0.87).

DISCUSSION

The findings strongly suggest that the left anterior IFG plays a crucial role in long-term memory formation (cf. Floel et al. 2004; Grafman et al. 1994; Köhler et al. 2004). During an incidental encoding task, any kind of TMS affected the speed and accuracy with which semantic judgments were made on words. However, only TMS to the prefrontal cortex affected later memory performance. Temporarily disrupting activity in anterior IFG reduced the general willingness to respond “old” to words in a recognition memory test performed 15 min later. The ability to discriminate these words from information not encountered in the experiment was also reduced. These effects were observed relative to stimulation of the vertex control site, which provides a stricter control for nonspecific effects of the TMS procedure than the no-TMS baseline.

The decrease in recognition accuracy implies that the representations of words created while prefrontal activity was disrupted were more difficult to access than those of other words. Because the circumstances of retrieval were identical across TMS conditions, these representations must have been weaker or less rich. Language studies have indicated that the left anterior IFG plays a causal role in semantic processing (Gabrieli et al. 1998; Gough et al. 2005; Poldrack et al. 1999); thus it has been suggested that encoding-related activity in this region reflects the influence of semantic processing on effective memory formation (Buckner and Koutstaal 1998; Buckner et al. 2000; Otten et al. 2001a; Wagner et al. 1998b, 1999). The more an item can be semantically elaborated when it is encountered, the more likely it is that a persistent representation will be formed in long-term memory. On this account, disrupting activity in left anterior IFG would be expected to lead to memory representations that are semantically less rich. Consistent with this interpretation, words encoded during TMS to the left prefrontal cortex were more difficult to retrieve in the present experiment (cf. Craik and Lockhart 1972; Craik and Tulving 1975).

Impoverished memory representations may also cause the adoption of a more conservative response strategy. Although response accuracy and response bias are independent (e.g., Macmillan and Creelman 2005; Snodgrass and Corwin 1988), responding “old” may in the present circumstances be less appealing when an item brings to mind only a limited amount of information that can be used to decide whether it occurred earlier. Note that the variations in response bias were observed in a single recognition memory test in which all old items were intermixed with new items. Response bias can thus perhaps counterintuitively change on a trial-by-trial basis. This has been found previously (for review, see Windmann et al. 2002) and demonstrates that response bias not only is under strategic control, but is also influenced by stimulus-related factors.

Importantly, prefrontal TMS at all three time points affected memory performance. Pulses were delivered at 350, 750, and 1,150 ms. Disrupting activity in left anterior IFG at any of these times reduced recognition accuracy and response bias, with no discernible differences across points. Activity in left ventrolateral prefrontal cortex must thus play a crucial role in memory formation between ≥350 and 1,150 ms after word onset. This time course is roughly consistent with encoding-related effects seen in electrical brain activity (Paller and Wagner 2002; Rugg 1995; Wagner et al. 1999). As explained in the introduction, the frontal positive deflection typically observed for items that are later remembered onsets around 400 ms and persists for several hundred milliseconds. This deflection is thought to index the degree to which an item is processed in semantic working memory, similar to what has been proposed for left anterior IFG. The present findings suggest that such semantic elaboration processes are engaged between 350 and 1,150 ms and influence encoding success during at least that period of time. It will be of interest to determine whether these processes can start earlier (cf. Otten et al. 2001b, 2006) or last longer.

Not only stimulation of the left, but also stimulation of the right, anterior IFG affected encoding success. For study as well as test performance, none of the comparisons between TMS to the left versus right prefrontal cortex was reliable. It thus seems that both hemispheres contribute to effective memory formation, at least under the conditions used here. This supplements evidence from studies using fMRI, which often show activity in the right as well as left IFG (e.g., Otten et al. 2001). One possibility is that the living/nonliving task was solved by relying on nonverbal attributes, such as those generated by creating a mental image of the item denoted by a word. Such processing is supported by the right ventrolateral prefrontal cortex (Kelley et al. 1998; Kirchhoff et al. 2000; McDermott et al. 1999; Wagner et al. 1998a). Alternatively, the right hemisphere may always contribute to effective encoding, irrespective of whether nonverbal attributes are engaged (cf. Klingberg and Roland 1998). However, the latter possibility is unlikely given that Floel and colleagues (2004) did not observe a memory decrement following disruption of activity in right ventrolateral prefrontal cortex.

The present data show evidence not only of specific effects of disrupting activity in prefrontal cortex, but also of nonspecific effects associated with the general TMS procedure. Study as well as test performance differed between the no-TMS and control site TMS conditions. General changes brought about by the tactile, auditory, or induced current side effects of TMS are not new to TMS experiments; indeed they are quite common (e.g., Marzi et al. 1998; see Walsh and Pascual-Leone 2003). At study, living/nonliving judgments were made more quickly and less accurately during all task blocks involving TMS compared with when the task was performed on its own. The faster response times, combined with the reports during debriefing, indicate that the prospect of receiving TMS caused volunteers to be more alert and focused on the task, albeit at the expense of committing the occasional error (response speed and response accuracy do not necessarily follow the same pattern during TMS applications; for discussion, see Walsh and Pascual-Leone 2003). Rossi et al. (2001) found a similar pattern of nonspecific TMS effects. Paying more attention at study should lead to better performance in a later memory test (Craik et al. 1996). This was indeed observed following stimulation of the vertex control site. Memory performance following prefrontal stimulation, however, did not differ from that following no TMS. In both cases, performance was lower than that following control site TMS—or, in other words, control site TMS seemingly enhanced later memory performance. This counterintuitive pattern can be accounted for by the combined influence of increased attention during encoding as a result of the general TMS procedure and a selective encoding deficit caused by prefrontal stimulation. For prefrontal TMS, the memory enhancement caused by the increased attention during the study phase was counteracted by the additional detrimental effect of disrupting activity in prefrontal cortex.

Unlike most previous TMS studies on memory formation, we were able to assess both specific and nonspecific effects of TMS because the design included two control conditions: a no-stimulation baseline and stimulation of a control site. Table 1 illustrates the range of control conditions that have been used in previous TMS work (noise-only sham, reoriented-coil sham, separate no-TMS blocks, interleaved TMS, and no-TMS trials). Encouragingly, despite the variations in control conditions, a consistent pattern of findings is emerging to support a causal role of ventrolateral prefrontal cortex in long-term memory formation.

An interesting question is whether distinct prefrontal regions contribute differently to memory encoding. Although there are currently only 12 studies using prefrontal TMS (those reported in Table 1 and our own), it is possible that the effect sizes in these studies suggest a difference related to site of stimulation. A formal meta analysis would be premature in light of the limited number of studies, each of which differs not only with respect to site of stimulation, but also in terms of study task, study material, memory test, stimulation strength, time of stimulation, and experimental protocol. Nonetheless, an initial analysis across those 9 studies that reported sufficient data to compute effect sizes showed that effect size (Cohen's d value) varied between around 0.2 and 1.7, with no discernible pattern related to site of stimulation or any other experimental parameter. Thus evidence for possible differences across prefrontal regions awaits the accumulation of more data.

In the present experiment, the detrimental effect of disrupting activity in ventrolateral prefrontal cortex was, although statistically significant, relatively modest in size. Memory performance was reduced by roughly 4%, which equates to 6 of the 120 total words. The majority of words could thus be successfully encoded and retrieved despite TMS to anterior IFG. It is important to note that there is no reason to assume that the size of an effect has some bearing on its importance. There are many psychological findings that are small but influential (e.g., negative priming and transfer-appropriate processing effects). Nonetheless, there are several reasons why memory decrements following prefrontal TMS may be small. First, activity in anterior IFG may not have been completely disrupted, either because of the strength of stimulation or because stimulation was applied over a duration of only 40 ms. The remaining activity may have enabled a level of processing sufficient to encode most words. Second, although anterior IFG plays a necessary role in memory formation, it is unlikely to be the sole region relevant for encoding. Indeed, isolated lesions to the prefrontal cortex do not typically result in vast memory deficits (Milner and Petrides 1984; Shimamura 2003; Squire and Knowlton 2000). Rather, prefrontal contributions to memory formation are thought to arise from interactions with other regions, notably the medial temporal lobe. These regions may offset any deficits in prefrontal cortex. Third, because encoding is only the initial stage of memory, any disruptions during encoding may be compensated for by later consolidation or retrieval processes.

The final thing to note is that memory decrements were observed across all recognition judgments, irrespective of the confidence with which the judgments were made. This suggests that the semantic processes disrupted by prefrontal TMS are relevant for all encoding-related processes contributing to long-term recognition, not just those supporting recollection. Some fMRI studies find more robust effects in prefrontal cortex for items that are later confidently remembered or recollected (e.g., Wagner et al. 1998b). That pattern was not observed here. It is possible that recollected items carried a large proportion of the TMS effects, even though this was not brought out statistically. It is also possible that, under the present circumstances, disruption to left anterior IFG affects the general strength of a representation in memory. Using a recognition memory test with more levels of confidence, or a source memory paradigm, will help understand the type of long-term memory processes supported by activity in left anterior IFG.

In conclusion, the present findings strengthen the suggestion from only three previous studies that ventrolateral prefrontal cortex plays a critical role in long-term memory formation (cf. Floel et al. 2004; Grafman et al. 1994; Köhler et al. 2004). Importantly, the findings extend these ideas by demonstrating that activity in anterior IFG and the cognitive processes associated with it together exert an influence between ≥350 and 1,150 ms after an event has been encountered. Future work should be directed at assessing whether activity can extend beyond these times and how activity in prefrontal cortex interacts with activity in other brain regions to lay down a new memory.

GRANTS

This work was supported by Wellcome Trust Research Career Development Fellowship WT073147 to L. J. Otten. Stimulus presentation was programmed with the Cogent 2000 software of the physics group of the Wellcome Trust Centre for Neuroimaging.