Richard E. Passingham

Many suggestions have been made as to the functions of the prefrontal (PF) cortex. However, they involve labelling areas using psychological terminology. But what we need to know is how the PF cortex operates. We argue that understanding this must start with describing the flow of information. We illustrate this argument by considering three PF areas. Each has a unique pattern of inputs and outputs, and we suggest that the implication is that each performs a unique transformation from the inputs to the outputs. The caudal PF cortex transforms input that is maintained by attention or short-term memory into the target of the appropriate eye movement. The mid-dorsal PF cortex transforms input concerning the order of objects or actions into the target of the appropriate eye and hand movements, thus supporting sequences of action. The ventral PF cortex transforms input concerning an object or sound into prospective activity that encodes the associated object or sound. However, it is important to appreciate that the mid-dorsal and ventral PF cortex are specialized for encoding abstract transformations, irrespective of the specific actions or objects. The advantage is that this enables generalization to novel problems that have the same underlying logic. We account for the difference between fast learning and slow learning in this way. The human brain has co-opted these mechanisms so as to support intelligence. Non-verbal tests of IQ typically use sequences of letters, numbers or designs. These test the ability to understand the abstract rules that apply. Here the activations lie in the mid-dorsal PF cortex. Verbal tests typically assess the ability to understand semantic associations. These can be presented either in pictorial or verbal form. Here the activations lie in the ventral PF cortex.

The motivation for an imaging study should be better than simply “little is known about the neural basis of X.” As in other areas of science, interest should be in phenomena that are in need of explanation. These could concern normal cognition, for example that it is difficult to carry out two tasks at the same time; or abnormal cognition, for example that there are patients who hear voices. Answering questions such as these can require a combination of methods, which can be best achieved by collaboration between scientists who are experts in different methods. With the maturation of brain imaging, users should think of themselves not as “imagers” but as neuroscientists.

Functional imaging is primarily a correlational technique. Although there are methods for analyzing the causal structure of systems, proof of causation requires methods that intervene in the workings of the system. If area A influences area B, preventing activity in A should diminish activity in B. The interference with activity in area A can be permanent, as after a brain lesion, or temporary, as with the application of transcranial magnetic stimulation. However, to understand the workings of the system we need to know not only that area A influences area B, but also how it does so. This involves the synchronization of activity in the two areas. To study this one needs methods such as electro-encephalography and magneto-encephalography that provide evidence of synchronized and dynamic oscillations at different frequencies.

The different areas of the brain do not work in isolation. Instead they operate as integrated systems. These systems can be identified because areas within the same system have a similar, though not identical, pattern of connections. Because the areas are also interconnected, the activations of areas within a system tend to covary over time, whether the subject is at rest or engaged in a task. Thus the different systems can be identified on the basis of the degree to which they covary or are activated independently. There are two aims in studying functional systems. One is to identify the nature of information that flows from one area to another. The other is to characterize the causal structure of the system. Several methods are available to do this, and these have been particularly useful for studying top-down effects, as in the voluntary control of attention.

A critical step in finding out how the human brain supports cognition is to identify the operations that are performed by the individual parts. In the case of the cortex the parts are the different cytoarchitectonic areas. Just as connectional fingerprints can be established for each area, so their functions can be discerned by functional fingerprints. These show the degree to which an area is engaged across a wide range of cognitive tasks. The aim of comparing these plots for different areas is to work out the transformation that the different areas perform, from input to output. Given the complexity of the brain it can help to produce computational models of how it works and to relate the operations that are carried out by different areas to different components or parameters of the model.

Once the statistical analysis has been performed, the location of the activations has to be established. The importance of this step lies in the fact that the location provides a link to the anatomical connections, since it is these that constrain the functions of the area. Each cortical area has a unique overall pattern of extrinsic connections, and it is this connectional fingerprint that provides the anatomical basis for functional localization. Thus, it is critical that an activation is assigned to the correct area, and that the localization is not described in terms of a general region, if within that region there are subareas with different connections. One can account for the function in terms of the anatomical connections. The identification of the correct cytoarchitectonic area usually depends on the warping of the image so that it fits a standardized template and the use of a probabilistic atlas to identify the most likely area for the activation.

Having recorded a signal, it is necessary to interpret its functional significance. The way in which this is done is to relate the signal to a psychological condition. As in other branches of science, an experimental condition is contrasted with a control condition. The interpretation is clearest when these differ in just one respect, though this can be difficult to achieve. Standard statistics are used to evaluate the significance of the difference. However, the analysis of imaging data can be onerous and many methods have been developed to avoid false-positive and false-negative results. These include robust correction for the number of statistical comparisons that are made, as the image is made up of thousands of voxels across many regions. Researchers also use targeted region-of-interest analysis; in this case the region must be specified beforehand. One must also study enough subjects: if small groups are used, the study may be underpowered.

Imaging is one of the major tools for neuroscience but it should be seen against the background of neuroscience in general. The brain can be studied by charting its connections, by recording its activity, or by intervening in its workings. In animals the connections can be demonstrated by tracer techniques, recordings can be taken of the activity of single cells, and lesions can be placed in selective cytoarchitectonic areas. Brain imaging provides methods for studying the human brain. Connections can be inferred using diffusion weighted imaging, activations can be recorded using functional magnetic resonance imaging, and transcranial magnetic brain stimulation can be used to intervene. The development of these methods has provided the means for studying the neural basis of cognitive abilities, including those that are unique to the human brain.

There has been no agreement as to whether the prefrontal cortex is especially enlarged in the human brain. To answer this question, we analyzed the only two datasets that provide information on total prefrontal cortex volume based on cytoarchitectonic criteria. One delineated the prefrontal cortex proper on the basis of cytoarchitectonic criteria; the other used a proxy of the prefrontal cortex based on a cytoarchitectonic delineation of the frontal lobe. To investigate whether all cortical association areas, including the prefrontal cortex, are enlarged in the human brain, we scaled the different areas to a common reference, the primary visual cortex. To investigate whether the prefrontal cortex is more enlarged than other association areas, we scaled it relative to its inputs from and outputs to other nonprimary areas. We carried out separate regression analyses using different data samples as a predictive baseline group: data for monkeys alone informs us on whether great apes are different from monkeys; data for all non-human anthropoids, including great apes, informs us on whether humans are different from all other primates. The analyses show that the value for the human prefrontal cortex is greater than expected, and that this is true even when data for the great apes are included in the analysis. They also show that the chimpanzee prefrontal cortex is greater than expected for a monkey with a similar sized cortex. We discuss possible functional consequences.