Conserved Sequence Processing in Primate Frontal Cortex

Abstract

An important aspect of animal perception and cognition is learning to recognize relationships between environmental events that predict others in time, a form of relational knowledge that can be assessed using sequence-learning paradigms. Humans are exquisitely sensitive to sequencing relationships, and their combinatorial capacities, most saliently in the domain of language, are unparalleled. Recent comparative research in human and nonhuman primates has obtained behavioral and neuroimaging evidence for evolutionarily conserved substrates involved in sequence processing. The findings carry implications for the origins of domain-general capacities underlying core language functions in humans. Here, we synthesize this research into a 'ventrodorsal gradient' model, where frontal cortex engagement along this axis depends on sequencing complexity, mapping onto the sequencing capacities of different species.

Keywords: cognition; evolution; human; language; monkey; neurobiology.

Figure 1

Taxonomy of Sequencing Complexity and Primate Abilities. Relationships between events in a sequence can vary in complexity on multiple dimensions (Box 1, main text). (A) One dimension of complexity (Y-axis) that is relevant for language begins with the capacity to evaluate adjacent relationships (or dependencies) between items in a sequence. This requires an incoming element to be at least temporarily held in memory, compared and associated with the next element in the sequence. (B) Another level of complexity, which human infants learn during the first year of life , is processing nonadjacent relationships. In this case, a relationship must be learned between two elements separated in time. For example, nonhuman primates have been shown to learn that the first element in a three element-long sequence predicts the final element, while ignoring an intervening element that is uninformative about the nonadjacent relationship between the first and third element . Moreover, squirrel monkeys were able to learn that the pitch of the first tone in a sequence predicted the final tone, separated by a variable number of tones of a different pitch . This requires holding a stimulus in working memory and comparing it to another element over one or more intervening elements so that the nonadjacent dependency can be established, assessed, or monitored. (C) Hierarchical relationships ​, prominent in language, reflect even higher levels of complexity, such as one phrase (e.g., ‘AB’) being nested inside of another ‘AB’ phrase, requiring multiple ‘A-B’ associations to be simultaneously held in memory. Relative to this taxonomy of sequencing complexity, some currently known sequence processing capabilities of nonhuman primates are illustrated (see the section ‘Sequence Learning: A Candidate Language Precursor’ in the main text). Question marks denote experiments in which species showed no evidence of sensitivity to certain types of sequencing violations (see for details), indicating uncertainty as to whether these species are able to process sequences at the given level of complexity (see Outstanding Questions, main text).

Figure 2

Evolutionarily Conserved Brain Areas for Processing Adjacent Sequence Relationships in Human and Monkey Frontal Cortex. Humans and rhesus macaques were first exposed to exemplary sequences from an artificial grammar that generates variable sequences, based on several adjacent relationships (bottom right of Figure 1, main text). The participants were then presented with ‘consistent’ and ‘violation’ testing sequences during fMRI scanning. Group results from 12 human participants are displayed alongside representative results in an individual macaque, from the three that were studied (for details, see [9]). The results showed voxels in the brain that responded more strongly to violation sequences than to consistent sequences, shown here on rendered lateral surface representations of the human and monkey brain. The effects are illustrated for the right hemisphere, but are not significantly lateralized . A key result was the strong engagement of ventral frontal and opercular cortex (vFOC) in both monkeys and humans, including the frontal operculum and anterior insula. These findings highlight the role of these regions in processing adjacent sequencing relationships. Areas 44 and 45 were not strongly engaged in either species, although the effect was statistically more pronounced in monkeys than in humans. These general impressions are supported by independent neuroimaging evidence in monkeys and humans (see the section Sequence Processing in the Primate Brain: Testing Neuroevolutionary Hypotheses’ in the main text). Parietal activation was observed in both species, including area 39 in humans and area 7 in macaques. These regions form part of the dorsal processing pathway and, in humans, are involved in a range of language tasks, including sentence comprehension , , , , but are not thought to form part of the core perisylvian circuit for hierarchical syntax . Involvement of parietal regions is less evident than the involvement of frontal cortex during sequence processing in humans (e.g., 39, 61) and is not consistently observed in monkeys , .

Figure 3

Key Figure: A Gradient of Frontal Network Engagement as a Function of Sequencing Complexity Sequence processing, similar to many behaviorally complex operations, is supported by an interacting network of brain areas. Increasing sequence complexity modulates the network and the involvement of particular regions in distinct ways, as illustrated. When predictions are relatively simple (e.g., adjacent relationships ​where one element directly predicts the next element in the sequence), they engage a network that includes the ventral frontal and opercular cortex (vFOC) in both humans and monkeys. In humans, when sequence relationships ​become more complex (either nonadjacent relationships ​or nested relationships, where several associations must be processed), the combinatorial codes required for integrating and mediating the respective dependencies involve interaction between regions of frontal cortex and other parts of the spatially distributed network, as illustrated. Presenting any given stimulus may elicit a distinct pattern of neuronal (unit) responses in sensory cortex (illustrated by the dot patterns). Through associative learning, these neuronal response patterns can become representationally more similar for elements that predictably co-occur across sequences , . This process can occur several times for multiple or nested associations, but would require interactions between more dorsal cortical areas, as illustrated, that together can mediate and reinstate the multiple associations available throughout the network (see the section ‘Ventrodorsal Gradient Model of Frontal Cortex Function for Sequencing Operations’ in the main text). The human brain has specialized, functionally and anatomically, to support abilities more complex than those of nonhuman animals, with language a salient example. Therefore, alongside evolutionarily conserved, functional homologies, important divergences are also likely to exist (see Outstanding Questions, main text). Some evidence for a divergence of mechanisms comes from studies showing that areas 44 and 45 have greater variability in sequence processing across species , .