6 | JANUARY 2004 | VOLUME 5 www.nature.com/reviews/neuro

H I G H L I G H T S

Although it is clear that the prefrontal cortex isresponsible for cognitive control — ensuring that ouractions are appropriate not just in light of our sensoryinput but also for a given context or event — theorganization of the prefrontal cortex is controversial.Koechlin et al. propose a new model for the functioningof the lateral prefrontal cortex (LPFC) in whichdifferent areas form a hierarchy of cognitive control,from the premotor cortex to more anterior regions.

The model is best understood by considering a well-known example of cognitive control. If a phone rings, adefault response would be to answer it. However, if youare at a friend’s house (a different context), you wouldnot usually answer the phone. But this contextualcontrol can be overridden if, for example, your friendhas gone out and has asked you to take any calls.

In the model proposed by Koechlin and colleagues,there are three levels of control. The premotor cortex isresponsible for ‘sensory control’ — answering a ringingphone. More anteriorly, the caudal LPFC carries out‘contextual control’ — using cues that accompany thestimulus to tell you that it is inappropriate to answeryour friend’s phone. And the rostral LPFC isresponsible for ‘episodic control’, in which earlier cues(a conversation with your friend) tell you how youshould respond during a given episode. In thishierarchy, signals flow from the most anterior parts ofthe LPFC to the more posterior parts and the premotorcortex, conveying top–down information to controlour actions.

The authors tested their model using behaviouralstudies and functional imaging — with rather moreabstract stimuli — as well as information theory forquantifying cognitive control. Subjects were given tasksin which either sensory or contextual informationcould be varied, either alone or with episodic

information. As predicted by the model, reaction timesincreased with stimulus, context and episode factors,and both stimulus and context effects were additivewith the episode effect.

Functional magnetic resonance imaging (fMRI)allowed the authors to see whether different types ofcognitive control were reflected in increased activity indifferent parts of the LPFC. Consistent with thecascading, top–down nature of the model, activity inthe premotor cortex showed effects of stimulus,context and episode; caudal LPFC showed effects ofcontext and episode, but not stimulus; and rostralLPFC only showed effects of episode. To test theprediction that the effects of context and episode inmore posterior areas result from top–down controlfrom the most anterior areas, the authors used astructural equation model to investigate the effectiveconnectivity of the LPFC. The results also supportedthe three-tiered model of cognitive control.

In this model, representations are distributed in theLPFC depending on their temporal structure, ratherthan their content or internal complexity. It joins anumber of other models of prefrontal function and isunlikely to be the last, but it can also explain thepattern of results seen in a number of other studies inwhich rostral and caudal LPFC showed activity thatdepended on whether cognitive control was contextualor episodic. The story is far from over, but this studycould be an important chapter.

Rachel JonesReferences and links

ORIGINAL RESEARCH PAPER Koechlin, E. et al. The architecture ofcognitive control in the human prefrontal cortex. Science 302, 1181–1185(2003)FURTHER READING Wood, J. N. & Grafman, J. Human prefrontal cortex:processing and representational perspectives. Nature Rev. Neurosci. 4,139–147 (2003) | Duncan, J. An adaptive coding model of neural function inprefrontal cortex. Nature Rev. Neurosci. 2, 820–829 (2001)

Cascading cognitive control

C O G N I T I V E N E U R O S C I E N C E

NeuromancyIt was definitely the headlinethat got me reading. “Devicesthat read human thought nowpossible” reported the San Francisco Herald(10 November 2003),referring to studies presentedat the past Society forNeuroscience meeting.Happily, my initial dismay athaving missed the postersubsided when I discoveredthat the story was actually onbrain–machine interfaces thatharness neuronal activity toguide mechanical limbs.Such interfaces had alreadybeen used in monkeys, andthe newspaper, prompted bypreliminary results on patientswith Parkinson’s disease,discussed the possibility ofusing them in people.

Relieved by the fact that mysecret thoughts will remainprivate for now, my attentionturned to the financial twist ofthe story. On the one hand,Miguel Nicolelis, a leader inthis field, is quoted as saying“I have no interests in anybusiness … I want to havefun; I don’t want to makemoney. What I am very afraidof is that people who reallywant to make a buck out ofthis will be rushing into theclinical thing.” On the other,John Donaghue, anotherexpert on this area, pointedout that he and hiscolleagues “realized the onlyway to fully exploit thetechnology was to form acompany capable of raisingthe money needed to carryout very expensive clinicalstudies”. Cyberkinetics Inc. isthe result of their realization.

Regardless of whose sideyou are on, a growingnumber of neuroscientistsare becoming aware of thefinancial implications of whatthey do in the lab, and spendmore time with theircolleagues over at theTechnology Transfer Office.Although many of them aresurely driven by their desire tofoster technologicalprogress, I suspect we wouldnot need a mind-readingdevice to spot the phrase‘lucrative patent’ in theirthoughts.

Juan Carlos López

IN THE NEWS

© 2004 Nature Publishing Group