- Email: [email protected]
Visual Pursuit: An Instructive Area of Cortex
Current Biology Vol 15 No 16 R638
transcription can. Given this result, it is more likely that a loss of tRNA tertiary structure triggers adenylation of the hypomethylated tRNAiMet in vivo and not the lack of m1A58 [13,19]. This is consistent with the finding that TRAMP efficiently adenylated a tRNAala containing two point mutations predicted to disrupt its structure while the wild-type tRNAala was inefficiently adenylated in vitro [2]. This result also implies that all the components needed to recognize aberrantly structured RNAs are present within purified TRAMP. The same structural perturbations may be required for the adenylation of snRNA, snoRNA and 5s rRNAs, but since most of these RNAs have been characterized as full-length or their 5′′ and 3′′ ends remain uncharacterized this seems unlikely for this set of RNAs. These new findings promise to provide insights into how nuclear RNA surveillance plays an important role in regulating eukaryotic gene expression.
3.
4.
5.
6.
7.
8.
9.
10.
References 1. LaCava, J., Houseley, J., Saveanu, C., Petfalski, E., Thompson, E., Jacquier, A., and Tollervey, D. (2005). RNA degradation by the exosome is promoted by a nuclear polyadenylation complex. Cell 121, 713–724. 2. Vanacova, S., Wolf, J., Martin, G., Blank,
11.
12.
D., Dettwiler, S., Friedlein, A., Langen, H., Keith, G., and Keller, W. (2005). A new yeast poly(A) polymerase complex involved in RNA quality control. PLoS Biol. 3, e189. Wyers, F., Rougemaille, M., Badis, G., Rousselle, J.C., Dufour, M.E., Boulay, J., Regnault, B., Devaux, F., Namane, A., Seraphin, B., et al. (2005). Cryptic pol II transcripts are degraded by a nuclear quality control pathway involving a new poly(A) polymerase. Cell 121, 725–737. Kushner, S.R. (2002). mRNA decay in Escherichia coli comes of age. J. Bacteriol. 184, 4658–4665. Li, Z., Reimers, S., Pandit S and Deutscher MP (2002). RNA quality control: degradation of defective transfer RNA. EMBO J. 21, 1132–1138. Piper, P.W., Bellatin, J.A., and Lockheart, A. (1983). Altered maturation of sequences at the 3¢ terminus of 5S gene transcripts in a Saccharomyces cerevisiae mutant that lacks a RNA processing endonuclease. EMBO J. 2, 353–359. Kuai, L., Fang, F., Butler, J.S. and Sherman, F. (2004). Polyadenylation of rRNA in Saccharomyces cerevisiae. Proc. Natl. Acad. Sci. USA 101, 8581–8586. Allmang, C., and Kufel, J. (1999). Functions of the exosome in rRNA, snoRNA and snRNA synthesis. EMBO J. 18, 5399–5410. van Hoof, A., Lennertz, P., and Parker, R. (2000). Yeast exosome mutants accumulate 3′′-extended polyadenylated forms of U4 small nuclear RNA and small nucleolar RNAs. Mol. Cell. Biol. 20, 441–452. Mitchell, P., Petfalski, E., Shevchenko, A,. Mann, M., and Tollervey, D. (1997). The exosome: a conserved eukaryotic RNA processing complex containing multiple 3′′→5′′ exoribonucleases. Cell 91, 457–466. Aravind, L., and Koonin, E.V. (1999). DNA polymerase beta-like nucleotidyltransferase superfamily: identification of three new families, classification and evolutionary history. Nucleic Acids Res. 27, 1609–1618. Saitoh, S., and Chabes, A. (2002). Cid13
Visual Pursuit: An Instructive Area of Cortex Recent experiments have revealed an area of visual cortex that provides a velocity error signal which enables the eye to learn to pursue targets when they move in a predictable way. R.H.S. Carpenter Eye movements exist to make up for our visual defects. The most debilitating is that our retinal receptors are very slow, so that we cannot see properly when the retinal image is moving. Usually, this is because of movement of the head, and the resulting slippage of the entire visual scene generates a powerful reflex, the optokinetic response, which moves the eye in such a way as to
reduce the retinal slip: a simple negative feedback system, in which retinal slip velocity is in effect an error signal. The neural circuits for this response are relatively simple, located for the most part in the brainstem. Here, neurons coding for large-scale retinal slip velocity in different directions send this information to neurons in the vestibular nuclei whose function — with help from the semicircular canals — is to estimate head velocity, and thus
is a cytoplasmic poly(A) polymerase that regulates ribonucleotide reductase mRNA. Cell 109, 563–573. 13. Kadaba, S., and Krueger, A. (2004). Nuclear surveillance and degradation of hypomodified initiator tRNAMet in S. cerevisiae. Genes Dev. 18, 1227–1240. 14. de la Cruz, J., Kressler, D., Tollervey, D., and Linder, P. (1998). Dob1p (Mtr4p) is a putative ATP-dependent RNA helicase required for the 3′′ end formation of 5.8S rRNA in Saccharomyces cerevisiae. EMBO J. 17, 1128–1140. 15. Inoue, K., Mizuno, T., Wada, K. and Hagiwara, M. (2000). Novel RING finger proteins, Air1p and Air2p, interact with Hmt1p and inhibit the arginine methylation of Npl3p. J. Biol. Chem. 275, 32793–32799. 16. Ho, Y., Gruhler, A., Heilbut, A., Bader, G.D., Moore, L., Adams, S.L., Millar, A., Taylor, P., Bennett, K., Boutilier, K., et al. (2002). Systematic identification of protein complexes in Saccharomyces cerevisiae by mass spectrometry. Nature 415, 180–183. 17. Huh, W.K., Falvo, J.V., Gerke, L.C., Carroll, A.S., Howson, R.W., Weissman, J.S., and O'Shea, E.K. (2003). Global analysis of protein localization in budding yeast. Nature 425, 686–691. 18. Takano, A., Endo, T., and Yoshihisa, T. (2005). tRNA actively shuttles between the nucleus and cytosol in yeast. Science 309, 140-142. 19. Anderson, J., Phan, L., Cuesta R, Carlson, B.A., Pak, M., Asano, K., Bjork, G.R., Tamame, M., and Hinnebusch, A.G. (1998). The essential Gcd10p-Gcd14p nuclear complex is required for 1methyladenosine modification and maturation of initiator methionyl-tRNA. Genes Dev. 12, 3650–3662.
Department of Biological Sciences, Marquette University, PO Box 1881, Milwaukee, Wisconsin 53201, USA. DOI: 10.1016/j.cub.2005.08.002
in turn to generate equal and opposite compensatory eye movements [1]. There is, however, another way that retinal slip can arise, which poses more of a computational problem. A cat intent on a mouse running through undergrowth needs the retinal image of its prey to be stationary, but if it achieves this there will be a powerful signal from the optokinetic mechanism generated by the backwards retinal slip of the undergrowth itself, which will tend to hold the eye firmly stationary. So what is needed is a system that can selectively inhibit optokinesis except for a defined target region, and one that can also continue the eye’s tracking even when the mouse is briefly obscured by leaves and branches [2,3]. This in
Dispatch R639
turn implies prediction of the mouse’s path and velocity, much as an anti-aircraft gun predicts the future position of a plane. But the task is made hugely more difficult by the fact mentioned earlier, that retinal information is so very slow: it takes some 40 milliseconds or so to reach even the lowest levels of the brain, and simple visual reaction times are of the order of 180 milliseconds or more. All of this demands a control system of some sophistication, which is both flexibly selective and capable of learning and therefore of prediction. As might be expected, the oculomotor subsystem that generates these kinds of movements — the smooth pursuit system — has evolved only relatively recently, mostly in predator species whose retina contains a fovea, a central region specialised for high-definition vision [4–6]. The learning and prediction are easy to demonstrate (Figure 1) [7]. Asked to follow a target moving repetitively and predictably — a pendulum, for example — at first one’s eye movements are relatively poorly matched to target velocity; but within a very few repetitions of the pattern of movement there is a dramatic improvement in performance. We can show that this is because the system is actively predicting the motion by suddenly interposing a mask that covers part of the target’s movement: despite the lack of visual input the eye continues to track the invisible movement, albeit not quite as accurately (below, Figure 1). So it is clear that smooth pursuit is not merely driven directly by the error signal: it must contain some kind of predictive model. This in turn implies that errors must be able to tweak the model in addition to driving the response in the first place – the system has to use regulatory parametric feedback as well as the much more familiar direct negative feedback. A recent paper by Megan Carey and colleagues [8], working in Steve Lisberger’s lab, has provided some welcome insight into the neural mechanisms that may underlie these processes of instruction and learning, linking
Figure 1. Learning to pursue. Top, a record of a human subject starting to track a target moving sinusoidally in the horizontal plane at 0.5 Hz, showing the rapid improvement as the movement of the target is learnt. Bottom, tracking the same sinusoidal target, but now it is obscured during the portion shown in grey: the velocity continues even though there is no stimulus. (My unpublished data.)
Eye
Target
Mask
2s
them to an area of cortex (medial temporal or MT) where neurons have long been known to carry signals related to retinal slip, of a kind that would make them good candidates for providing error information for regulating smooth pursuit [9,10]. Macaques were trained to follow a repetitive visual target moving horizontally at a constant velocity [8]. This task provides a good background on which to study how the smooth pursuit system learns to predict target movement. A previous study from the lab [11] had shown that, if a vertical perturbation is introduced at a particular point on each
Current Biology
sweep, the pursuit system soon learns to anticipate it. The learning is revealed by occasionally presenting a sweep in which the perturbation does not occur: in these ‘probe’ trials, the eyes nevertheless persist in making the expected vertical deviation, despite the absence of an error. In this more recent paper, instead of a real error signal, a vertical perturbation was induced by microstimulation of an appropriate part of MT. The time-course of the resultant vertical velocity of the eye is now more complex, because the evoked upward eye movement evokes actual retinal slip that feeds back into the Response to stimulus Velocity (deg/s) 0.5
Stimulate
0
MT
Stimulate (300 ms)
Velocity error
Teach SPS
Motor command
‘Probe’ - no stimulus 0.5
0
Drive Curent Biology
Figure 2. Learning to predict a perturbation. Left, simplified representation of the neural pathways relevant to these experiments. When an unexpected perturbation of the target movement occurs, a visual velocity error signal is generated that goes to the medial temporal cortical area (MT). This in turn generates an ‘instructive’ command to the oculomotor smooth pursuit sub-system (SPS), which has the dual effect of driving the eye to give immediate correction of the error (direct feedback: arrow) and modifying stored programs so that the perturbation is anticipated in future (parametric feedback: circle). In the first experiment, MT is briefly stimulated, causing both a vertical perturbation of the eye’s velocity, representing the system’s attempt to compensate for the presumed displacement of the target (upper trace, right), and modification of the stored program which — after many trials with MT stimulation — is revealed in a ‘probe’ trial in which MT is not stimulated and yet a compensatory response still occurs.
Current Biology Vol 15 No 16 R640
Response to stimulus Velocity (deg/s) 0.5
Stimulate No visual feedback Vision
3.
0
MT
4.
Stimulate (300 ms)
5.
Teach
Motor command
‘Probe’ - no stimulus 0.5
6.
SPS
0
Drive Curent Biology
7.
Figure 3. Demonstrating that MT is genuinely instructional. Left, the underlying pathways, as before: but now visual feedback from the eye is prevented, so that the initial perturbation caused by MT stimulation is no longer modified by the actual visual feedback that would otherwise result from it, and the perturbation of vertical velocity is enhanced (top right). Below right, in probe trials in the same monkey the system demonstrates that it has modified its stored program, even though no real visual errors occurred.
system to generate a corrective downward response to counteract it. In probe trials, with no stimulation of MT, a learned component of the response is revealed (Figure 2), just as in the previous study. This in itself is not particularly informative about the role of MT in smooth pursuit learning. Stimulation almost anywhere in the oculomotor system would be expected to do something of the sort, because of the error signal that is bound to arise through visual feedback whenever the eye is artificially perturbed. To control for this, a second experiment was performed: this time, visual feedback was prevented by stabilising the target, moving it vertically by an amount exactly equal to the vertical eye movement at every moment (Figure 3). In this way, vertical retinal slip is eliminated, and as a result of this lack of feedback, the perturbation is more sustained; and once again, in probe trials the eye moves vertically even though MT has not been stimulated. Another difference is that because of the absence of corrective feedback, the response is no
longer complicated by a second, compensatory phase in opposition to the first. This elegant and convincing experiment therefore suggests strongly that MT is part of a route by which tracking errors both initiate immediate correction, and also cause information to be stored that results in anticipation of the perturbation in future. What it does not tell us, unfortunately, is just how the learning itself is implemented. This has been an area of active speculation for several decades [12–17], but as yet we are not much nearer understanding even the type of learning process that is going on, let alone the neuronal details of its implementation. But now that we have a way of injecting instructional signals into the system, without visual feedback coming up from behind and complicating things, there is perhaps some hope of making progress. References 1. Carpenter, R.H.S. (1988). Movements of the Eyes, 2nd edn. (London: Pion). 2. Collewijn, H., and Tamminga, E.P. (1984). Human smooth pursuit and saccadic eye movements during voluntary pursuit of
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
different target motions on different backgrounds. J. Physiol. 351, 217–250. Becker, W., and Fuchs, A.F. (1985). Prediction in the oculomotor system: smooth pursuit during transient disappearance of a visual target. Exp. Brain Res. 57, 562–575. Walls, G.L. (1962). The evolutionary history of eye movements. Vision Res. 2, 69–80. Pola, J., and Wyatt, H.J. (1992). Smooth movement: response characteristics, stimuli and mechanisms. In: Eye Movements. Edited by Carpenter, R.H.S. (London: MacMillan), pp. 138–156. Missal, M., Lefèvre, P., Crommelinck, M., and Roucoux, A. (1995). Evidence for high-velocity smooth pursuit in the trained cat. Exp. Brain Res. 106, 509–512. Travis, R.C., and Dodge, R. (1930). Ocular pursuit of objects which temporarily disappear. J. Exp. Psychol. 13, 98–112. Carey, M.R., Medina, J.F., and Lisberger, S.G. (2005). Instructive signals for motor learning from visual cortical area MT. Nat. Neurosci. 8, 813–816. Maunsell, J.H., and van Essen, D.C. (1983). Functional properties of neurons in middle temporal visual area of the macaque monkey. 1. Selectivity for stimulus direction, speed and orientation. J. Neurophysiol. 49, 1127–1147. Ferrera, V.P., and Lisberger, S.G. (1997). Neuronal responses in visual areas MT and MST during smooth pursuit target selection. J. Neurophysiol. 78, 1433–1446. Medina, J.F., Carey, M.R., and Lisberger, S.G. (2005). The representation of time for motor learning. Neuron 45, 157–167. Bahill, A.T., and McDonald, J.D. (1983). Model emulates human smooth pursuit system producing zero-latency target tracking. Biol. Cybern. 48, 213- 222. Robinson, D.A., Gordon, J.L., and Gordon, S.E. (1986). A model of the smooth pursuit eye movement system. Biol. Cybern. 55, 43–57. Krauzlis, R.J., and Lisberger, S.G. (1989). A control systems model of smooth pursuit eye movements with realistic emergent properties. Neural Comput. 1, 116–122. Barnes, G.R., and Asselman, P.T. (1991). The mechanism of prediction in human smooth pursuit eye movements. J. Physiol. 439, 439–461. Barnes, G., Grealy, M., and Collins, S. (1997). Volitional control of anticipatory ocular smooth pursuit after viewing, but not pursuing, a moving target: evidence for a reafferent velocity store. Exp. Brain Res. 116, 445–455. Churchland, M.M., and Lisberger, S.G. (2001). Experimental and computational analysis of monkey smooth-pursuit eye movements. J. Neurophysiol. 86, 741–759.
The Physiological Laboratory, University of Cambridge, Cambridge CB2 3EG, UK. DOI: 10.1016/j.cub.2005.08.004
transcription can. Given this result, it is more likely that a loss of tRNA tertiary structure triggers adenylation of the hypomethylated tRNAiMet in vivo and not the lack of m1A58 [13,19]. This is consistent with the finding that TRAMP efficiently adenylated a tRNAala containing two point mutations predicted to disrupt its structure while the wild-type tRNAala was inefficiently adenylated in vitro [2]. This result also implies that all the components needed to recognize aberrantly structured RNAs are present within purified TRAMP. The same structural perturbations may be required for the adenylation of snRNA, snoRNA and 5s rRNAs, but since most of these RNAs have been characterized as full-length or their 5′′ and 3′′ ends remain uncharacterized this seems unlikely for this set of RNAs. These new findings promise to provide insights into how nuclear RNA surveillance plays an important role in regulating eukaryotic gene expression.
3.
4.
5.
6.
7.
8.
9.
10.
References 1. LaCava, J., Houseley, J., Saveanu, C., Petfalski, E., Thompson, E., Jacquier, A., and Tollervey, D. (2005). RNA degradation by the exosome is promoted by a nuclear polyadenylation complex. Cell 121, 713–724. 2. Vanacova, S., Wolf, J., Martin, G., Blank,
11.
12.
D., Dettwiler, S., Friedlein, A., Langen, H., Keith, G., and Keller, W. (2005). A new yeast poly(A) polymerase complex involved in RNA quality control. PLoS Biol. 3, e189. Wyers, F., Rougemaille, M., Badis, G., Rousselle, J.C., Dufour, M.E., Boulay, J., Regnault, B., Devaux, F., Namane, A., Seraphin, B., et al. (2005). Cryptic pol II transcripts are degraded by a nuclear quality control pathway involving a new poly(A) polymerase. Cell 121, 725–737. Kushner, S.R. (2002). mRNA decay in Escherichia coli comes of age. J. Bacteriol. 184, 4658–4665. Li, Z., Reimers, S., Pandit S and Deutscher MP (2002). RNA quality control: degradation of defective transfer RNA. EMBO J. 21, 1132–1138. Piper, P.W., Bellatin, J.A., and Lockheart, A. (1983). Altered maturation of sequences at the 3¢ terminus of 5S gene transcripts in a Saccharomyces cerevisiae mutant that lacks a RNA processing endonuclease. EMBO J. 2, 353–359. Kuai, L., Fang, F., Butler, J.S. and Sherman, F. (2004). Polyadenylation of rRNA in Saccharomyces cerevisiae. Proc. Natl. Acad. Sci. USA 101, 8581–8586. Allmang, C., and Kufel, J. (1999). Functions of the exosome in rRNA, snoRNA and snRNA synthesis. EMBO J. 18, 5399–5410. van Hoof, A., Lennertz, P., and Parker, R. (2000). Yeast exosome mutants accumulate 3′′-extended polyadenylated forms of U4 small nuclear RNA and small nucleolar RNAs. Mol. Cell. Biol. 20, 441–452. Mitchell, P., Petfalski, E., Shevchenko, A,. Mann, M., and Tollervey, D. (1997). The exosome: a conserved eukaryotic RNA processing complex containing multiple 3′′→5′′ exoribonucleases. Cell 91, 457–466. Aravind, L., and Koonin, E.V. (1999). DNA polymerase beta-like nucleotidyltransferase superfamily: identification of three new families, classification and evolutionary history. Nucleic Acids Res. 27, 1609–1618. Saitoh, S., and Chabes, A. (2002). Cid13
Visual Pursuit: An Instructive Area of Cortex Recent experiments have revealed an area of visual cortex that provides a velocity error signal which enables the eye to learn to pursue targets when they move in a predictable way. R.H.S. Carpenter Eye movements exist to make up for our visual defects. The most debilitating is that our retinal receptors are very slow, so that we cannot see properly when the retinal image is moving. Usually, this is because of movement of the head, and the resulting slippage of the entire visual scene generates a powerful reflex, the optokinetic response, which moves the eye in such a way as to
reduce the retinal slip: a simple negative feedback system, in which retinal slip velocity is in effect an error signal. The neural circuits for this response are relatively simple, located for the most part in the brainstem. Here, neurons coding for large-scale retinal slip velocity in different directions send this information to neurons in the vestibular nuclei whose function — with help from the semicircular canals — is to estimate head velocity, and thus
is a cytoplasmic poly(A) polymerase that regulates ribonucleotide reductase mRNA. Cell 109, 563–573. 13. Kadaba, S., and Krueger, A. (2004). Nuclear surveillance and degradation of hypomodified initiator tRNAMet in S. cerevisiae. Genes Dev. 18, 1227–1240. 14. de la Cruz, J., Kressler, D., Tollervey, D., and Linder, P. (1998). Dob1p (Mtr4p) is a putative ATP-dependent RNA helicase required for the 3′′ end formation of 5.8S rRNA in Saccharomyces cerevisiae. EMBO J. 17, 1128–1140. 15. Inoue, K., Mizuno, T., Wada, K. and Hagiwara, M. (2000). Novel RING finger proteins, Air1p and Air2p, interact with Hmt1p and inhibit the arginine methylation of Npl3p. J. Biol. Chem. 275, 32793–32799. 16. Ho, Y., Gruhler, A., Heilbut, A., Bader, G.D., Moore, L., Adams, S.L., Millar, A., Taylor, P., Bennett, K., Boutilier, K., et al. (2002). Systematic identification of protein complexes in Saccharomyces cerevisiae by mass spectrometry. Nature 415, 180–183. 17. Huh, W.K., Falvo, J.V., Gerke, L.C., Carroll, A.S., Howson, R.W., Weissman, J.S., and O'Shea, E.K. (2003). Global analysis of protein localization in budding yeast. Nature 425, 686–691. 18. Takano, A., Endo, T., and Yoshihisa, T. (2005). tRNA actively shuttles between the nucleus and cytosol in yeast. Science 309, 140-142. 19. Anderson, J., Phan, L., Cuesta R, Carlson, B.A., Pak, M., Asano, K., Bjork, G.R., Tamame, M., and Hinnebusch, A.G. (1998). The essential Gcd10p-Gcd14p nuclear complex is required for 1methyladenosine modification and maturation of initiator methionyl-tRNA. Genes Dev. 12, 3650–3662.
Department of Biological Sciences, Marquette University, PO Box 1881, Milwaukee, Wisconsin 53201, USA. DOI: 10.1016/j.cub.2005.08.002
in turn to generate equal and opposite compensatory eye movements [1]. There is, however, another way that retinal slip can arise, which poses more of a computational problem. A cat intent on a mouse running through undergrowth needs the retinal image of its prey to be stationary, but if it achieves this there will be a powerful signal from the optokinetic mechanism generated by the backwards retinal slip of the undergrowth itself, which will tend to hold the eye firmly stationary. So what is needed is a system that can selectively inhibit optokinesis except for a defined target region, and one that can also continue the eye’s tracking even when the mouse is briefly obscured by leaves and branches [2,3]. This in
Dispatch R639
turn implies prediction of the mouse’s path and velocity, much as an anti-aircraft gun predicts the future position of a plane. But the task is made hugely more difficult by the fact mentioned earlier, that retinal information is so very slow: it takes some 40 milliseconds or so to reach even the lowest levels of the brain, and simple visual reaction times are of the order of 180 milliseconds or more. All of this demands a control system of some sophistication, which is both flexibly selective and capable of learning and therefore of prediction. As might be expected, the oculomotor subsystem that generates these kinds of movements — the smooth pursuit system — has evolved only relatively recently, mostly in predator species whose retina contains a fovea, a central region specialised for high-definition vision [4–6]. The learning and prediction are easy to demonstrate (Figure 1) [7]. Asked to follow a target moving repetitively and predictably — a pendulum, for example — at first one’s eye movements are relatively poorly matched to target velocity; but within a very few repetitions of the pattern of movement there is a dramatic improvement in performance. We can show that this is because the system is actively predicting the motion by suddenly interposing a mask that covers part of the target’s movement: despite the lack of visual input the eye continues to track the invisible movement, albeit not quite as accurately (below, Figure 1). So it is clear that smooth pursuit is not merely driven directly by the error signal: it must contain some kind of predictive model. This in turn implies that errors must be able to tweak the model in addition to driving the response in the first place – the system has to use regulatory parametric feedback as well as the much more familiar direct negative feedback. A recent paper by Megan Carey and colleagues [8], working in Steve Lisberger’s lab, has provided some welcome insight into the neural mechanisms that may underlie these processes of instruction and learning, linking
Figure 1. Learning to pursue. Top, a record of a human subject starting to track a target moving sinusoidally in the horizontal plane at 0.5 Hz, showing the rapid improvement as the movement of the target is learnt. Bottom, tracking the same sinusoidal target, but now it is obscured during the portion shown in grey: the velocity continues even though there is no stimulus. (My unpublished data.)
Eye
Target
Mask
2s
them to an area of cortex (medial temporal or MT) where neurons have long been known to carry signals related to retinal slip, of a kind that would make them good candidates for providing error information for regulating smooth pursuit [9,10]. Macaques were trained to follow a repetitive visual target moving horizontally at a constant velocity [8]. This task provides a good background on which to study how the smooth pursuit system learns to predict target movement. A previous study from the lab [11] had shown that, if a vertical perturbation is introduced at a particular point on each
Current Biology
sweep, the pursuit system soon learns to anticipate it. The learning is revealed by occasionally presenting a sweep in which the perturbation does not occur: in these ‘probe’ trials, the eyes nevertheless persist in making the expected vertical deviation, despite the absence of an error. In this more recent paper, instead of a real error signal, a vertical perturbation was induced by microstimulation of an appropriate part of MT. The time-course of the resultant vertical velocity of the eye is now more complex, because the evoked upward eye movement evokes actual retinal slip that feeds back into the Response to stimulus Velocity (deg/s) 0.5
Stimulate
0
MT
Stimulate (300 ms)
Velocity error
Teach SPS
Motor command
‘Probe’ - no stimulus 0.5
0
Drive Curent Biology
Figure 2. Learning to predict a perturbation. Left, simplified representation of the neural pathways relevant to these experiments. When an unexpected perturbation of the target movement occurs, a visual velocity error signal is generated that goes to the medial temporal cortical area (MT). This in turn generates an ‘instructive’ command to the oculomotor smooth pursuit sub-system (SPS), which has the dual effect of driving the eye to give immediate correction of the error (direct feedback: arrow) and modifying stored programs so that the perturbation is anticipated in future (parametric feedback: circle). In the first experiment, MT is briefly stimulated, causing both a vertical perturbation of the eye’s velocity, representing the system’s attempt to compensate for the presumed displacement of the target (upper trace, right), and modification of the stored program which — after many trials with MT stimulation — is revealed in a ‘probe’ trial in which MT is not stimulated and yet a compensatory response still occurs.
Current Biology Vol 15 No 16 R640
Response to stimulus Velocity (deg/s) 0.5
Stimulate No visual feedback Vision
3.
0
MT
4.
Stimulate (300 ms)
5.
Teach
Motor command
‘Probe’ - no stimulus 0.5
6.
SPS
0
Drive Curent Biology
7.
Figure 3. Demonstrating that MT is genuinely instructional. Left, the underlying pathways, as before: but now visual feedback from the eye is prevented, so that the initial perturbation caused by MT stimulation is no longer modified by the actual visual feedback that would otherwise result from it, and the perturbation of vertical velocity is enhanced (top right). Below right, in probe trials in the same monkey the system demonstrates that it has modified its stored program, even though no real visual errors occurred.
system to generate a corrective downward response to counteract it. In probe trials, with no stimulation of MT, a learned component of the response is revealed (Figure 2), just as in the previous study. This in itself is not particularly informative about the role of MT in smooth pursuit learning. Stimulation almost anywhere in the oculomotor system would be expected to do something of the sort, because of the error signal that is bound to arise through visual feedback whenever the eye is artificially perturbed. To control for this, a second experiment was performed: this time, visual feedback was prevented by stabilising the target, moving it vertically by an amount exactly equal to the vertical eye movement at every moment (Figure 3). In this way, vertical retinal slip is eliminated, and as a result of this lack of feedback, the perturbation is more sustained; and once again, in probe trials the eye moves vertically even though MT has not been stimulated. Another difference is that because of the absence of corrective feedback, the response is no
longer complicated by a second, compensatory phase in opposition to the first. This elegant and convincing experiment therefore suggests strongly that MT is part of a route by which tracking errors both initiate immediate correction, and also cause information to be stored that results in anticipation of the perturbation in future. What it does not tell us, unfortunately, is just how the learning itself is implemented. This has been an area of active speculation for several decades [12–17], but as yet we are not much nearer understanding even the type of learning process that is going on, let alone the neuronal details of its implementation. But now that we have a way of injecting instructional signals into the system, without visual feedback coming up from behind and complicating things, there is perhaps some hope of making progress. References 1. Carpenter, R.H.S. (1988). Movements of the Eyes, 2nd edn. (London: Pion). 2. Collewijn, H., and Tamminga, E.P. (1984). Human smooth pursuit and saccadic eye movements during voluntary pursuit of
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
different target motions on different backgrounds. J. Physiol. 351, 217–250. Becker, W., and Fuchs, A.F. (1985). Prediction in the oculomotor system: smooth pursuit during transient disappearance of a visual target. Exp. Brain Res. 57, 562–575. Walls, G.L. (1962). The evolutionary history of eye movements. Vision Res. 2, 69–80. Pola, J., and Wyatt, H.J. (1992). Smooth movement: response characteristics, stimuli and mechanisms. In: Eye Movements. Edited by Carpenter, R.H.S. (London: MacMillan), pp. 138–156. Missal, M., Lefèvre, P., Crommelinck, M., and Roucoux, A. (1995). Evidence for high-velocity smooth pursuit in the trained cat. Exp. Brain Res. 106, 509–512. Travis, R.C., and Dodge, R. (1930). Ocular pursuit of objects which temporarily disappear. J. Exp. Psychol. 13, 98–112. Carey, M.R., Medina, J.F., and Lisberger, S.G. (2005). Instructive signals for motor learning from visual cortical area MT. Nat. Neurosci. 8, 813–816. Maunsell, J.H., and van Essen, D.C. (1983). Functional properties of neurons in middle temporal visual area of the macaque monkey. 1. Selectivity for stimulus direction, speed and orientation. J. Neurophysiol. 49, 1127–1147. Ferrera, V.P., and Lisberger, S.G. (1997). Neuronal responses in visual areas MT and MST during smooth pursuit target selection. J. Neurophysiol. 78, 1433–1446. Medina, J.F., Carey, M.R., and Lisberger, S.G. (2005). The representation of time for motor learning. Neuron 45, 157–167. Bahill, A.T., and McDonald, J.D. (1983). Model emulates human smooth pursuit system producing zero-latency target tracking. Biol. Cybern. 48, 213- 222. Robinson, D.A., Gordon, J.L., and Gordon, S.E. (1986). A model of the smooth pursuit eye movement system. Biol. Cybern. 55, 43–57. Krauzlis, R.J., and Lisberger, S.G. (1989). A control systems model of smooth pursuit eye movements with realistic emergent properties. Neural Comput. 1, 116–122. Barnes, G.R., and Asselman, P.T. (1991). The mechanism of prediction in human smooth pursuit eye movements. J. Physiol. 439, 439–461. Barnes, G., Grealy, M., and Collins, S. (1997). Volitional control of anticipatory ocular smooth pursuit after viewing, but not pursuing, a moving target: evidence for a reafferent velocity store. Exp. Brain Res. 116, 445–455. Churchland, M.M., and Lisberger, S.G. (2001). Experimental and computational analysis of monkey smooth-pursuit eye movements. J. Neurophysiol. 86, 741–759.
The Physiological Laboratory, University of Cambridge, Cambridge CB2 3EG, UK. DOI: 10.1016/j.cub.2005.08.004