For many years, it has been known that medial temporal lobe damage in humans produces profound memory impairments, while patients with frontal lobe lesions often perform normally on many standard tests of memory (Lee et al. 2000a; Petrides 1994). The pattern that has emerged from functional neuroimaging studies in healthy volunteers is quite different, with increases in activity reported in medial temporal and frontal lobe areas during many different memory tasks (Buckner et al. 1995 1999; Fletcher and Henson 2001; Lee et al. 2000a). One region that has been consistently activated is the mid-ventrolateral frontal cortex which, in humans, lies below the inferior frontal sulcus and includes Brodmann areas 45 and 47 (Brodmann, 1909). Activity in this region has been reported during spatial, verbal and pattern working memory tasks (Owen et al. 1996a 2000; Stern et al. 2000) but also during episodic memory tests of encoding and retrieval (Fletcher et al. 1998; Lee et al. 2000b; Owen et al. 1996b). The medial temporal lobe structures, including the hippocampus, are also frequently activated during tests of both episodic (Ryan et al. 2001; Schacter et al. 1999) and working memory (Cabeza et al. 2002; Monk et al. 2002; Ranganath and D'Esposito 2001; Stern et al. 1996). In short, while individual studies have reported memory-related activity in the medial temporal lobe system or the mid-ventrolateral frontal cortex, the picture emerging from the imaging literature as a whole is of frequent co-activation across the two regions (Buckner et al., 1999).#

In neuropsychological studies, medial temporal lobe damage frequently impairs simple recognition memory performance, even when the vaguest sense of familiarity should be sufficient to generate a correct response (Owen et al. 1995; Wheeler et al. 1995). Frontal lobe patients are generally unimpaired on such tasks but have difficulties when internally generated intentions or goals are required to generate a response (Lee et al. 2000a; Petrides and Milner 1982; Wheeler et al. 1995). Such evidence suggests that an important factor for understanding the functional relationship between the mid-ventrolateral frontal cortex and the medial temporal lobe system may be the extent to which a volunteer explicitly intends to remember or retrieve a given stimulus and the changes in attentional control that may be consequent upon such an intention. The role of intention has been investigated previously in memory by comparing intentional encoding tasks with so-called ‘incidental’ memory tasks (Buckner et al. 2001; Stark and Okado 2003; Wagner et al. 1998; Otten et al. 2001; Fletcher et al. 2003), although in both cases, activity in the mid-ventrolateral frontal cortex has been reported. However, in many incidental tasks, an intention is still involved, but it is directed away from explicit memorization (e.g., to judging whether words are in uppercase or lowercase letters; Buckner et al. 2000 2001).#

In this study, event-related functional magnetic resonance imaging (fMRI) was used to explicitly test the hypothesis that mid-ventrolateral frontal cortex activity, but not medial temporal lobe activity, will vary according to the degree to which the intention to encode or retrieve information is modulated by a prior instruction cue, all other factors being held constant. Volunteers were instructed on random trials either to just look at pictures of abstract art or to try and remember them for later test. Recognition was examined by asking, on random trials, whether they remembered seeing specific pictures previously. In a fourth condition, volunteers were instructed to passively re-view stimuli shown previously. It was predicted that medial temporal lobe activity would be observed during all experimental conditions when compared to rest, while activity in the mid-ventrolateral frontal cortex would only be observed when volunteers specifically intended to remember or retrieve the stimuli following a specific instruction to do so.#

An initial pilot experiment was conducted to test two aspects of the imaging paradigm: (i) whether explicitly asking volunteers to try and remember stimuli for later recognition lead to improved memory performance compared to when they were asked to simply look at similar stimuli and (ii) whether looking at a single presentation of a stimulus was sufficient to yield significant memory for that stimulus. In both the behavioral pilot study and the functional neuroimaging study, a series of colorful abstract paintings that were neither well known, nor contained easily recognizable objects, were collected and randomly assigned to four experimental conditions (Fig. 1). In the interests of brevity, the two conditions in which the volunteers were asked to simply look at the paintings (without any instruction to remember or recall) will henceforth be referred to as low-intention. The two conditions in which the volunteers were asked explicitly to remember or make a recognition judgement about the paintings will be referred to as high-intention.#

In all conditions, each trial consisted of an instruction presented for 1.5 s. which was followed by a 0.4-s delay and then the presentation of an abstract painting for 3 s. Trials were separated by an inter-trial interval of 0.4 s. In the low-intention encoding condition, the instruction was ‘look at this’. The volunteer's task was just to look at the painting that followed. In the high-intention encoding condition, volunteers received the instruction ‘remember this’. They were told that the task was to remember the painting that followed for possible later recognition. In the high-intention recognition condition, the instruction was ‘have you seen this?’ Here, the volunteer was required to decide whether or not the painting had been seen before. A fourth condition, referred to as low-intention re-viewing was designed to control for the fact that paintings in the high-intention recognition condition were viewed more than once and to examine, as far as is possible, recognition memory in the absence of a specific instruction. In this condition, volunteers were asked to re-view stimuli (with the instruction ‘look at this’) that had been shown previously in the low-intention encoding condition.#

In this study, a novel memory paradigm was used to demonstrate that activity in the mid-ventrolateral frontal cortex, but not the medial temporal lobe system, increases when volunteers have an intention to encode or retrieve information about a stimulus following a specific instruction to do so, all other factors being held constant. In contrast, activity in the parahippocampal gyrus/hippocampus increased in all of the experimental conditions, including those in which no explicit memory instruction was given (see Fig. 4).#

Behavioral data from the pilot experiment confirmed that the condition in which no explicit instruction to remember the stimuli was given, nevertheless, resulted in delayed recognition that was well above chance, even for stimuli that were presented only once during the low-intention encoding condition. Furthermore, repetition of the paintings, without any instruction to remember, significantly improved performance at recognition, again suggesting that the stimuli were attended to and that this alone induced some level of behaviorally relevant mnemonic processing. The results of the functional neuroimaging study clearly demonstrated that the parahippocampal gyrus/hippocampus was significantly active during the low-intention encoding condition, whereas the mid-ventrolateral frontal cortex was not, suggesting that, in the absence of any previously cued intention to remember, memory is mediated primarily by the medial temporal lobe system. Similar results were observed at recognition; that is, re-viewing of stimuli that had been presented previously, with no explicitly cued intention to remember, resulted in significant signal intensity changes in the parahippocampal gyrus/hippocampus, but not in the mid-ventrolateral frontal cortex. These results suggest again, that the conscious experience of familiarity that may lead to successful recognition even in the absence of any explicit instruction to retrieve items from memory, is preferentially mediated by the medial temporal lobe system and not by the mid-ventrolateral prefrontal cortex. This notion concurs closely with the assumptions of relational memory processing accounts of hippocampal system function, which assume that “the hippocampal system provides the critical machinery for binding together the various elements encountered in our interaction with the environment; and it does this binding automatically and obligatorily, in the sense that no strategic intervention is necessary” (Cohen et al., 1999). Similarly, Martin (1999) has noted that the “medial temporal lobe is automatically engaged whenever an event is attended to and volunteers engage in the type of processing that spontaneously occurs in response to a particular type of stimulus.”#

Given such a fundamental role in memory processing, it is notable that relatively few functional neuroimaging studies have reported significant increases in signal intensity in the medial temporal lobe region compared, for example, to the mid-ventrolateral frontal cortex. It has been suggested that the involvement of the medial temporal lobe system in relatively automatic (or ‘bottom–up’) memory processes may effectively ‘cancel out’ activity when experimental and control conditions are compared directly (Martin, 1999). For similar reasons, Stark and Squire (2001), have argued that resting conditions are inappropriate baselines for fMRI studies of memory, particularly with respect to detecting medial temporal lobe activity, as they can reduce, eliminate, or even reverse the sign of activity in this region. The current findings confirm that this is not necessarily the case; in fact, compared to non-events (effectively a resting baseline), significant medial temporal lobe activity was observed during passive viewing of visual stimuli even at the single subject level in 19 of the 20 volunteers scanned.#

What the present results also demonstrate, however, is that when the task instructions (but not the stimuli) at encoding and recognition are changed to encourage additional intentional (or ‘top–down’) control of mnemonic processing, no significant changes are observed in medial temporal lobe activity, despite marked improvements in memory performance. In contrast, significant bilateral increases in signal intensity were observed in the mid-ventrolateral frontal cortex when the volunteers intended to encode and make decisions about the stimuli following an explicit instruction to do so.#

The idea that the frontal cortex is important for intended actions is not a new one, although the precise nature of this involvement has been neither behaviorally, nor anatomically, well specified. As early as 1922, Bianchi (1922) suggested that the frontal lobes are involved in controlling behavior, while the neuropsychological literature throughout the last century is filled with descriptions of frontal lobe patients that imply an impairment of intended actions. For example, deficits in planning (Owen et al. 1990; Shallice 1982 1988), the temporal structuring of behavior (Fuster, 1997), monitoring or manipulation in working memory (Owen 1997; Petrides 1994) and the control of behavior by context (Cohen and Servan-Schreiber, 1992) have been described, all of which imply a disorder of intended actions, yet all of which also lack testable, well-specified components. The current results confirm that even simply changing a task instruction from ‘look’ to ‘look and remember’ is sufficient to elicit a robust change in activity within the mid-ventrolateral region of the prefrontal cortex.#

Although this study was specifically designed to investigate the functions of the mid-ventrolateral frontal cortex by using encoding and recognition within episodic memory as a means by which to increase the demand on intentional processing, previous functional neuroimaging studies in humans have reported activity in this region during many other types of cognitive task. For example, the mid-ventrolateral frontal cortex has been activated in tasks that require the selection, comparison and judgement of stimuli held in short-term and long-term memory (Petrides, 1994), during stimulus selection (Rushworth et al., 1997), when the specification of retrieval cues is required (Dobbins et al., 2002), during the ‘elaboration encoding’ of information into episodic memory (Henson et al. 1999; Wagner et al. 1998) and when judgements of word meaning are required (Kapur et al., 1994). In fact, activity in this region has been reported frequently during tasks that appear to make no direct demands on memory at all, including reversal learning (Cools et al., 2002), inhibition (Konishi et al., 1999), extra-dimensional set-shifting (Nakahara et al. 2002; Hampshire and Owen in press) and task switching (Dove et al., 2000). While each of these tasks has unique requirements that undoubtedly contribute to differences in the overall pattern of activity reported, they all require the self-initiated, conscious (i.e., intentional) selection of appropriate responses, often in the absence of external cues.#

Broadly speaking, the findings from lesion studies in the monkey also suggest that the mid-ventrolateral frontal cortex makes a polymodal contribution to a variety of different tasks that require the initiation and execution of intended actions. In the macaque, the mid-ventrolateral frontal cortex lies below the sulcus principalis on the inferior convexity and comprises areas 12 or 47/12 and 45 (Carmicheal and Price 1994; Petrides and Pandya 1994). Lesions of the ventrolateral frontal cortex, but not the more dorsal cortex surrounding the sulcus principalis, cause impairments in non-spatial delayed-matching-to-sample for single items (Mishkin and Manning 1978; Passingham 1975), spatial and non-spatial delayed alternation (Mishkin et al., 1969), the learning of arbitrary stimulus–response associations (Gaffan 1994; Petrides 1994; Murray and Wise 1997) switching attention to behaviorally relevant aspects of the world (e.g., Dias et al., 1996), and even impair object matching when the sample and the match are simultaneously present and there is no delay component (Rushworth et al., 1997). Thus, once a simultaneous version of the task has been relearned, the imposition of a delay between sample and match poses no more of a problem for a monkey with a ventrolateral frontal lesion than it does prior to surgery (Rushworth et al., 1997). Electrophysiological data from the monkey also support a role for this region in the initiation of a variety of explicit cognitive processes. For example, Sakagami and Niki (1994) trained monkeys to either make or withhold a response depending on which stimulus they were shown. On some blocks of trials, the relevant dimension of the stimulus was its color, on other trials, it was its position or shape. Ventrolateral neurons appeared to encode the stimulus dimension of current interest to the monkey. Similarly, Rao et al. (1997) identified neurons ventral to the principal sulcus, that encoded either, or both, the location and the identity of stimuli presented in a novel delayed response procedure. Remarkably, some neurons adapted flexibly as the emphasis of the task changed during its various stages. Thus, once a target object's identity was no longer relevant many of the ‘what-and-where’ cells no longer coded for object identity but switched to code for object location. This finding suggests that the response of ventrolateral prefrontal ‘memory cells’ is flexible, i.e., they can code different stimulus attributes at different times according to task demands. In other words, they will respond to a stimulus, irrespective of its modality and whenever there is an explicit requirement and an associated intention to do so. Finally, Li et al. (1997) taught monkeys a conditional response task and recorded from ventrolateral cells, while they learned to associate each of learned responses with a new cue. Ventrolateral neurons were particularly modulated during the process of learning the selection rule associated with each of the novel stimuli.#

In summary, the current results bring together much previous literature and suggest that, in both humans and in non-human primates, the mid-ventrolateral frontal cortex plays a crucial role in intended thoughts and actions. What is less clear is whether the precise cognitive and neural mechanisms by which this involvement occurs can be specified any more clearly, beyond this basic operational description. One likely possibility is that the ventrolateral frontal cortex acts by biasing or ‘tuning’ attentional processing between competing representations in modality-specific posterior regions in order to maintain their relevance to current behavioral goals. Such a view is anatomically plausible given the strong bidirectional connections between many posterior cortical association areas and the mid-ventrolateral frontal region, which, in turn, is closely interconnected with the entire lateral prefrontal cortex (Petrides, 1994). Moreover, a frontal module with such properties has been proposed recently (O'Reilly et al. 2002; Frank et al. 2004; see also, Dehaene et al., 1998), although in those computational models, the critical region was defined rather more generally as the ‘lateral prefrontal cortex’. Flexible tuning of task-relevant variables within the mid-ventrolateral frontal cortex is also consistent with accounts of prefrontal function that emphasize its importance in switching (Cools et al. 2002; Konishi et al. 1999; Nakahara et al. 2002; Dove et al. 2000; Hampshire and Owen in press) and the ‘top–down’ modulation of attention (e.g., Owen et al. 1991 1993; Knight 1994; Desimone and Duncan 1995; Dias et al. 1996). Compromising such a function would be expected to affect a wide variety of tasks but particularly any behavior (e.g., an action or a thought), that derives from the subject's plans and intentions (Petrides, 1994).#

In this light, the apparent incongruity between the results of neuropsychological investigations in patients and the findings of many functional neuroimaging studies in healthy volunteers can be more clearly understood. For example, many episodic memory tasks can be performed adequately in a number of different ways; on the basis of judgments of relative familiarity or through the willed (i.e., intentional) recollection of encoded information (Jacoby and Dallas 1981; Mandler 1980). Information about stimulus familiarity may be sufficient to allow many patients with frontal lobe damage, or more specifically, damage to the mid-ventrolateral frontal cortex, to generate correct responses in the absence of more specific information about the actual content of the remembered information. The common observation that patients with frontal lobe lesions can perform perfectly well on certain tasks which undoubtedly tap episodic and working memory processes (e.g., Owen et al., 1995; for review, see Wheeler et al., 1995), is entirely consistent with this suggestion. In contrast, an extensive neuropsychological literature from patients with damage to the medial temporal lobe structures confirms that such patients are often impaired at relatively simple memory tasks that can be solved on the basis of stimulus familiarity, including those that require only single item recognition (e.g., Kimura 1963; Milner 1968; Owen et al. 1995; Piggott and Milner 1993). The fact that the mid-ventrolateral frontal cortex and the medial temporal lobe system are commonly co-activated in functional neuroimaging studies suggests that, in healthy volunteers performing at optimal levels, both regions will be recruited routinely for most memory tests, whether or not they are both absolutely necessary for successful performance.#