The cingulate cortex and limbic systems for emotion, action, and memory

The anterior cingulate cortex (ACC), itself a limbic structure, has connections with a set of other limbic and related areas including the amygdala and orbitofrontal cortex (OFC) involved in emotion and reward-related processing (see Fig. 1, and for details “Anterior cingulate cortex: anatomy and connections”) (Rolls 2014a, 2018a). This set of limbic and related structures related to the ventral or ‘what’ processing streams (Ungerleider and Haxby 1994), and these ventral processing streams themselves, provide a major source of ‘what’ and ‘reward’ input into the hippocampal memory system via the perirhinal and entorhinal cortex (blue in Fig. 1) (Rolls 2018b; Rolls and Wirth 2018).

The posterior cingulate cortex has connections from parietal structures such as the precuneus and lateral parietal areas and is involved in spatio-topographical and related memory functions (Cavanna and Trimble 2006; Rolls 2015; Rolls and Wirth 2018; Leech and Sharp 2014; Kircher et al. 2002; Vogt 2009; Vogt and Pandya 1987; Vogt and Laureys 2009). This limbic region related to the dorsal or ‘where’ processing systems (Ungerleider and Haxby 1994) provides a second major source of input into the hippocampal memory system, via the parahippocampal gyrus (areas TF and TH) and the entorhinal cortex (red in Fig. 1) (Rolls 2018b; Rolls and Wirth 2018). Because the ACC and its related limbic areas, and the posterior cingulate cortex and its related areas, have such different connections and functions, it has been argued that we should no longer think of a single limbic system, but instead of two (or more) limbic processing systems (Rolls 2015).

However, a key concept is that the orbitofrontal/anterior cingulate/amygdala set of limbic areas related to ventral steam processing, and the posterior cingulate cortex related to dorsal stream processing, enable the ventral and dorsal processing streams to be brought together in the hippocampus, so that we can form memories of ‘what’ happened ‘where’, which is prototypical of episodic memory (Rolls 2016a, 2018b; Kesner and Rolls 2015). In an interesting twist, there is in fact a connection from the orbitofrontal cortex to the posterior cingulate cortex (Vogt and Laureys 2009), which provides a path for reward- and punishment-related information to enter the hippocampus via the dorsal route as well as by the ventral route (Rolls 2018b; Rolls and Wirth 2018).

This conceptual framework is developed a little more next, with the more detailed evidence provided later. It should be noted that this framework applies to primates including humans, with the principles of operation being considerably different in rodents, due to the much less well-developed orbitofrontal cortex, and visual and even taste cortical processing areas (Rolls 2016a, 2018a, 2019a).

The orbitofrontal cortex represents the reward value of stimuli (Rolls 2000, 2014a, 2018a, 2019a, b; Tremblay and Schultz 1999; Rolls et al. 1989; Small et al. 2001). It is in a sense an output region for all the sensory systems, including taste, olfaction, visual, auditory, and somatosensory, that represents ‘what’ a stimulus is, and uses that information to build what are frequently multimodal representations but in value space rather than in ‘what’ or stimulus identity space. Orbitofrontal cortex neurons focus on reward value representations for stimuli and know little about actions.

The orbitofrontal cortex sends inputs to the ACC about the value of stimuli, that is, about goals including the value of outcomes (the reward received) and the expected value. The ACC in combination with the midcingulate motor area, which contains representations of actions, interfaces actions to outcomes (rewards or punishers received) using action–outcome learning, and also takes into account the cost of actions to obtain the goal when selecting actions (Rushworth et al. 2012; Kolling et al. 2016; Rolls 2019a). The anterior and midcingulate cortical areas are thus relevant to emotion, for they implement the instrumental goal-directed actions that the instrumental reinforcers involved in emotion produce (Rolls 2014a, 2019a, b). In the context of its representations of value, damage to the anterior cingulate areas does influence emotion (Rolls 2014a, 2019a).

The ACC operates as a system that performs goal-directed actions to obtain rewards and avoid punishers, and takes into account the outcomes received after actions, in that it is sensitive to devaluation of the goal, and will not select an action if the goal has been devalued. This is in contrast to the basal ganglia, which implement a stimulus–motor response mapping which becomes automated as a habit after much learning, and is not sensitive to devaluation of the goal (Rolls 2014a, 2019a, b).

The ACC areas occupy approximately the anterior third of the cingulate cortex (Fig. 2). They are separate from a midcingulate/cingulate motor area (Vogt 2009; Vogt et al. 2003) that may be involved in action selection (Rushworth et al. 2004, 2011; Noonan et al. 2011). The ACC includes area 32, the pregenual cingulate cortex (where genu refers to the knee of the corpus callosum); area 25, the subgenual cingulate cortex; and a part of area 24 (Fig. 2) (Price 2006a, b; Öngür et al. 2003; Ongür and Price 2000). [It is however noted that the midcingulate cortex has been described as having an anterior and a posterior part, with the criteria including cytoarchitecture (Vogt 2016) (see Fig. S1), and that the anterior part may overlap with or be similar to what is described as the supracallosal part of the ACC in “Anterior cingulate cortex: functional neuroimaging and neuronal activity”.]

As shown in Fig. 2, the ACC receives strong inputs from the orbitofrontal cortex (Carmichael and Price 1995; Morecraft and Tanji 2009; Vogt 2009; Carmichael and Price 1996; Devinsky et al. 1995; Vogt and Pandya 1987; Vogt 2019). The ACC is also characterized by connections with the amygdala (Carmichael and Price 1995; Morecraft and Tanji 2009; Vogt 2009). The ACC also has connections with some temporal cortical areas involved in memory including the parahippocampal gyrus (which provides via the entorhinal cortex a bridge to the hippocampus); and with the rostral superior temporal gyrus, the auditory superior temporal gyrus, and the dorsal bank of the superior temporal sulcus (Vogt 2009; Rolls 2016a, 2018b; Saleem et al. 2008; Insausti et al. 1987) (see Fig. 2). [The cortex in the superior temporal sulcus contains visual neurons that respond to face expression, gesture, and head movement (Hasselmo et al. 1989a, b).]

In more detail, a ‘medial prefrontal network’ (mainly ACC) selectively involves medial areas 14r, 14c, 24, 25, 32, and 10m, rostral orbital areas 10o and llm, and agranular insular area Iai in the posterior orbital cortex in macaques (Carmichael and Price 1996). An ‘orbital’ prefrontal network links most of the areas within the orbital cortex, including areas Iam, Iapm, Ial, 121, 12m, and 12r in the caudal and lateral parts of the orbital cortex, with areas 131, 13m, and 13b in the central orbital cortex, which have further onward connections to the rostral orbital area 11l (Carmichael and Price 1996; Price 2006b; Ongür and Price 2000). Several orbital areas (including 13a, 12o, and 11m) have connections with both the medial and orbital networks. Many of these areas are shown in Fig. 2. It is very interesting that this medial prefrontal network has connections with the posterior cingulate/retrosplenial cortex and parahippocampal cortex (Saleem et al. 2008), and has access to the hippocampus in this way, whereas the orbitofrontal cortex has projections to the perirhinal cortex (Saleem et al. 2008), and thus has access to the hippocampus via a more ventral route (Fig. 1).

A very interesting new finding about ACC connectivity in relation to what follows is that the medial orbitofrontal cortex has strong functional connectivity with the pregenual cingulate cortex, in both of which rewards are represented; and that the lateral orbitofrontal cortex (and inferior frontal gyrus) has strong functional connectivity with the supracallosal, more dorsal, ACC area, both of which are activated by unpleasant aversive stimuli (Rolls et al. 2018). This was shown in a resting state fMRI investigation with 254 healthy participants (Rolls et al. 2018). [Functional connectivity refers to correlations between the fMRI BOLD signal in different brain regions, and may include trans-synaptic effects as well as reflecting direct connections (Van Essen et al. 2019). The concept is that brain areas with high correlations in their signals are likely to have direct or indirect connections. I aim to include the term ‘functional’ when this type of connectivity is referred to.] Parcellation (to identify different clusters of voxels with similar functional connectivity) was performed based on the functional connectivity of individual ACC voxels in the control participants (Fig. 3). (The functional connectivity was measured by the correlation of the resting state fMRI signals between pairs of voxels.) A pregenual and subcallosal subdivision (1, green) has strong functional connectivity with the medial orbitofrontal cortex and connected areas (Fig. 3), which are implicated in reward (Fig. 4). The supracallosal subdivision (2, red), which is activated by unpleasant stimuli and non-reward, has high functional connectivity with the lateral orbitofrontal cortex and adjacent inferior frontal gyrus areas (Fig. 3), also activated by unpleasant stimuli (Fig. 4) (Rolls et al. 2018). These functional connectivities provide support for the hypothesis that the reward-related medial orbitofrontal cortex provides inputs to the pregenual cingulate cortex, also activated by rewards; and that the lateral orbitofrontal cortex, implicated in effects of non-reward and punishers, provides inputs to the supracallosal part of the ACC, which is also activated by unpleasant stimuli (Rolls et al. 2018; Rolls 2019a).

It should be noted that the medial prefrontal network includes area 14 (gyrus rectus) in the most medial part of the orbital cortex, and that area 14 should not be included in Price and colleagues’ ‘orbital prefrontal network’ (Carmichael and Price 1996). Their ‘orbital prefrontal network’ includes areas on the posterior, central and lateral, orbital surface (agranular insular areas Ial, Iam, Iapl and Iapm, and orbital areas 13b, 13l, 13m, 11l, 12r, 12m and 12l, see Fig. 2), which is described as the orbitofrontal cortex (Rolls 2014a, 2018a, 2019a, b). The term ‘ventromedial prefrontal cortex’ (vmPFC) is not well defined anatomically, and refers generally to a medial and ventral region of the prefrontal cortex which probably can be taken to include the ventral parts of the ‘medial prefrontal network’ of Price and colleagues, and also probably medial parts of the orbitofrontal cortex and medial area 10 (Rolls 2019a). Indeed, different parts of the ventromedial prefrontal cortex have different functional connectivity with other brain areas (Rolls et al. 2019a). The ventromedial prefrontal cortex may play a particular role in decision-making, especially in the choice between different rewards (Grabenhorst and Rolls 2011; Rolls et al. 2010a, b; Glascher et al. 2012; Rolls 2019a).

The outputs of the ACC reach further back in the cingulate cortex towards the midcingulate cortex, which includes the cingulate motor area (Vogt 2009; Vogt et al. 1996; Morecraft and Tanji 2009; Vogt 2016). The ACC also projects forward to the medial prefrontal cortex area 10 (Price 2006b; Ongür and Price 2000). Another route for output is via the projections to the striatum/basal ganglia system. The connections with the temporal lobe have been described above (Saleem et al. 2008). The ACC, including the subgenual cingulate cortex area 25, has outputs that can influence autonomic/visceral function via the hypothalamus, midbrain periaqueductal grey, and insula, as does the orbitofrontal cortex (Rempel-Clower and Barbas 1998; Critchley and Harrison 2013; Price 2006b; Ongür and Price 2000).

Lesion studies in humans (Camille et al. 2011) and macaques (Rudebeck et al. 2008) have provided evidence for a dissociation in the role of the ACC in action–outcome associations to guide behaviour; and of the orbitofrontal cortex in stimulus–outcome associations to update the expected value (Rushworth et al. 2012). Lesions of the ACC in rats impair the ability to take account of the costs of actions, and there is a complementary human neuroimaging study (Croxson et al. 2009).

An investigation more closely related to the understanding of emotion showed that patients with selective surgical lesions of the anterior ventral part of the ACC and/or medial prefrontal cortex area BA9 were impaired on voice and face expression identification, had changed social behaviour, such as inappropriateness, and had changes in their subjectively experienced emotions (Hornak et al. 2003). In line with these results in humans, reduced social vocalization and social and emotional changes were found in monkeys with anterior cingulate lesions (Hadland et al. 2003).

Complementing the human anterior cingulate damage effects on voice and in some cases face expression identification (Hornak et al. 2003), neuroimaging studies of vocal expression identification have reported orbital and medial prefrontal/cingulate activation. For example, non-verbal sounds expressing fear, sadness, and happiness activated orbital cortex BA11 and medial BA9 (Morris et al. 1996). Further, fearful sounds activated the ACC (BA32 and BA24) (Phillips et al. 1998). Facial expression identification produces activation in orbital and medial regions that include BA32/24 (anterior cingulate) and medial BA9 (Blair et al. 1999; Dolan et al. 1996; Nakamura et al. 1999).

Complementing the human anterior cingulate damage effects on the subjective experience of emotion (Hornak et al. 2003), activations were found in the ventral ACC and medial BA9 during self-generated emotional experience when participants recalled emotions of sadness or happiness (Lane et al. 1997a, b, 1998).

Based on this evidence, the hypothesis is that the ACC receives inputs from the orbitofrontal cortex and amygdala about expected rewards and punishers, and about the outcomes, the rewards and punishers (which include face and voice expression) received, and that these are used as the value representations in action–outcome learning performed in the cingulate cortex.

The anterior cingulate cortex receives information from its topologically nearby neocortical area, the orbitofrontal cortex (most of which is neocortical in primates, and which receives from ventral stream ‘what’ areas), and also the amygdala (also a recipient of ventral stream projections), and projects this information to a number of areas, including autonomic areas in the brainstem as well as in the insula, to the midcingulate cortex, and to the striatum. Because reward value is important in producing emotions, the anterior cingulate cortex becomes involved in emotion (Rolls 2014a, 2018a, 2019a), and damage to it in humans can impair emotions (Hornak et al. 2003). This provides for an emotion-related function of the cingulate cortex.

Another important function of the cingulate cortex is, via the midcingulate premotor area with its connections to neocortical motor areas, to associate actions with outcomes, as indicated by the connections shown in green in Fig. 1. My proposal is that convergence of reward or outcome information from the ACC, and of information about actions from the posterior cingulate cortex, occurs in the cingulate cortex leading to outputs via the midcingulate motor area, which projects to premotor areas including the premotor cortex area 6 and the supplementary motor area (see green arrows in Fig. 1). This provides connectivity for action–outcome learning (Rolls 2019a). The ACC receives reward and punishment outcome information from the orbitofrontal cortex (OFC). The posterior cingulate cortex receives information about actions from the parietal cortex. Then these two types of information are brought together towards the mid-part of the cingulate cortex including the cingulate premotor area (Vogt 2016), which with its connections to premotor neocortical areas can select the action that is most likely, given the action–outcome learning performed within this cingulate system, to obtain the goal, the desired outcome (Rolls 2019a).

In addition, the parietal areas have projections to medial frontal areas connected with the dorsal parts of the ACC (Vogt 2009), and these projections may also provide a route for action-related information to reach the cingulate action–outcome learning system (Rolls 2019a).

A third important function of the cingulate cortex is related to the hippocampal memory system, as shown in Fig. 1. This function is introduced by some anatomical considerations.

The cingulate cortex is largely agranular, that is, it does not have a well-developed layer 4 with granule cells (although a layer 4 is present in areas 23, 31, and 32). However, the cingulate cortex is similar to the neocortex (or isocortex) in other ways, with clear layers 2 and 3, and 5 and 6. Indeed, the cingulate cortex has a high cell density in layers 5 and 6. A term used to describe this cortex is proisocortex (Pandya et al. 2015). Another region of proisocortex is the posterior-most part of the primate orbitofrontal cortex which is agranular and includes the most posterior part of area 13. [The more anterior part of area 13 is granular (Rolls 2019a, b).] Other regions of proisocortex include the agranular anterior insula, the temporal pole, the parahippocampal cortex, and the rostral perirhinal cortex (Pandya et al. 2015). It has been hypothesized that neocortical areas develop in evolution from these proisocortical areas (Pandya et al. 2015).

The connections of these proisocortical areas include forming connectional bridges between neocortical areas and areas such as the hippocampus (which is termed allocortex). Indeed, this is evident in Fig. 1, which shows how perirhinal cortex forms a bridge from the ventral stream neocortical areas (including temporal lobe cortical areas) to the hippocampus (and back), and how the posterior cingulate cortex and parahippocampal cortex form a bridge from the dorsal stream neocortical areas (including the parietal cortex) to the hippocampus (and back) (via the entorhinal cortex in both cases). It is notable that the orbitofrontal cortex and ACC project information, probably about rewards, via both the ventral and dorsal routes, to the hippocampal system (via the perirhinal cortex in the ventral route; and via the posterior cingulate cortex in the dorsal stream to parahippocampal cortex connections).

This is consistent with the framework I propose that the cingulate cortex provides a bridge linking neocortical areas with the hippocampal memory system (Rolls 2018b; Rolls and Wirth 2018), with spatial information reaching the hippocampal system via the posterior cingulate cortex, and reward-related information reaching the hippocampus via the anterior cingulate and orbitofrontal cortex (Rolls 2019a).

This cingulate connectivity is further compared with that of the hippocampus, which receives information from the ventral ‘what’ processing stream (blue) and the dorsal ‘where’ or ‘action’ processing stream (red) in Fig. 1. The entorhinal cortex area 28 is the main entry for cortical connections to and from the hippocampus. The forward projections to the hippocampus are shown with large arrowheads, and the backprojections with small arrowheads. The main ventral stream connections to the hippocampus which convey information about objects, faces, etc. are in blue, and the main dorsal stream connections which convey ‘where’ information about space and movements are in red. The ventral ‘what’ visual pathways project from the primary visual cortex V1 toV2, then V4, then posterior inferior temporal visual cortex (PIT), then anterior inferior temporal visual cortex (AIT), then perirhinal cortex (areas 35/36), and thus to entorhinal cortex (Rolls 2016a). The dorsal ‘where’ visual pathways project from V1 to V2, then MT (middle temporal), then LIP (lateral intraparietal), then parietal area 7 (lateral) and medial (including the precuneus), then to posterior cingulate cortex areas 23/32) including the retrosplenial cortex (areas 29/30) and thus to the parahippocampal gyrus (areas TF and TH), and then entorhinal cortex. Area 22 is the superior temporal auditory association cortex. Reward information reaches the hippocampus from the orbitofrontal cortex (OFC), ACC (areas 32 and 25), and amygdala; but also via the orbitofrontal cortex to posterior cingulate cortex connections. The lateral prefrontal cortex areas 9 and 46 involved in working memory connect via the posterior cingulate cortex. The hippocampus enables all the high-order cortical regions to converge into a single network in the hippocampal CA3 region, which is involved in episodic memory (Rolls 2015, 2016a; Kesner and Rolls 2015).

A difference of hippocampal connectivity therefore from that of the cingulate cortex is that the inputs to the hippocampus from the ventral stream (blue in Fig. 1) are about objects and faces, etc. (from the temporal cortical visual areas), as well as reward information. This fits with the concept that the hippocampus is able to associate together ‘what’ or reward information (blue in Fig. 1) with spatial or temporal information (red in Fig. 1) to enable the formation of episodic memories about objects and faces, and where they were seen on a particular occasion (Rolls 2016a, 2018b; Rolls and Wirth 2018; Kesner and Rolls 2015; Rolls and Kesner 2006). In contrast, it is argued that the cingulate cortex performs the different computational function of associating actions with outcomes, so needs reward and spatial/action information, but less object information, and performs computations that may be based on several recent occasions in which outcomes resulted from actions (Rolls 2019a).

It should be noted that the anatomical connectivity shown in Fig. 1 provides two routes for reward-related information to reach the hippocampal memory system. The first, more direct, route is from the orbitofrontal and amygdala to the perirhinal cortex with information derived from the ventral visual and auditory ‘what’ processing streams (blue in Fig. 1). The second route for reward-related information is via the ACC (which receives its reward-related information from the orbitofrontal cortex) which projects to the posterior cingulate cortex and also to the parahippocampal gyrus, providing access to the hippocampal memory system via the dorsal route. There are also direct projections from the orbitofrontal cortex to the posterior cingulate cortex (green in Fig. 1). The presence of reward-related information in the posterior cingulate cortex and its connected parietal areas is likely to relate to the roles of these areas in some types of decision-making (Hayden et al. 2008; Platt and Glimcher 1999).