CE: AC ED: Susan Koshy Op: NS WNR: lww_wnr_3822 NEUROREPORT

VISION, CENTRAL

Processing speed in recurrent visual networks correlates with general intelligence Jacob Jolija,d, Danielle Huismana, Steven Scholtea, Ronald Hamela, Chantal Kemnerc and Victor A.F. Lammea,b a

Department of Psychology, University of Amsterdam, bVision and Cognition Lab, Netherlands Ophthalmic Research Institute, Amsterdam, cSection Developmental Disorders, Department of Psychiatry, University Medical Center, Utrecht University, Utrecht, The Netherlands and dLaboratory of Psychophysics, Brain Mind Institute, Swiss Federal Institute of Technology Lausanne, Switzerland

Correspondence and requests for reprints to Dr Jacob Jolij, University of Exeter, School of Psychology, Washington Singer Laboratories, Perry Road, Exeter EX4 4QG, United Kingdom Tel: + 44 (0) 1392 264626; fax: + 44 (0) 1392 264623; e-mail: [email protected] Sponsorship: J.J. is supported by a grant of the Social Sciences Council of the Netherlands Organization for Scienti¢c Research. Received 6 June 2006; accepted 18 July 2006

Studies on the neural basis of general £uid intelligence strongly suggest that a smarter brain processes information faster. Di¡erent brain areas, however, are interconnected by both feedforward and feedback projections. Whether both types of connections or only one of both types are faster in smarter brains remains unclear. Here we show, by measuring visual evoked potentials during a texture discrimination task, that general £uid intelligence shows a

strong correlation with processing speed in recurrent visual networks, while there is no correlation with speed of feedforward connections. The hypothesis that a smarter brain runs faster may need to be re¢ned: a smarter brain’s feedback connections run fasc 2006 Lippincott Williams & ter. NeuroReport 00:000^ 000 ! Wilkins.

Keywords: intelligence, recurrent processing, scene segmentation, visual processing

Introduction People differ in mental abilities. If one assumes that ‘mind is what the brain does’, these differences should be attributed to individual differences in information-processing capacities of the central nervous system. One of the hypotheses that have been put forward recently is that smarter people have faster brains: that is, brighter persons process information more rapidly and hence have higher mental abilities. This view is supported by both psychophysical and neurophysiological data [1,2]. General fluid intelligence correlates strongly with a measure called ‘inspection time’. Inspection time is a measure of how fast the visual system extracts information from a given stimulus. It is measured by means of the socalled ‘pi-paradigm’. In a typical pi-paradigm, a study participant has to decide which of two legs of a P-figure is longer. The P-figure is substituted by a mask shortly after it is presented, thus rendering it invisible if the stimulus-mask interval is set at an appropriate length. The stimulus-mask interval at which a participant can perform the task accurately (e.g., performing at 80%) is taken as the inspection time. Correlations between intelligence measures and inspection time are about "0.5 [1], suggesting that individual differences in inspection time explain a large portion of variance in intelligence. Information processing speed in the central nervous system can also be measured using event related potentials.

c Lippincott Williams & Wilkins 0959- 4965 !

Latency of the P300 component, a component thought to reflect cognitive processing of a stimulus, shows a correlation of "0.3 with performance IQ scores, as does the steepness of the so-called N1–P2 curve: the first negative (N1) and second positive (P2) peaks in the evoked potential are closer to each other in more intelligent persons, suggesting that information processing is faster in intelligent individuals [2]. It is difficult, however, to link these event related potential components to specific neural processes. Therefore, what goes faster in smarter brains remains unclear. To gain a better understanding of the neural basis of individual differences in mental abilities, it would be helpful to identify neural circuits or modes of neural processing contributing to general fluid intelligence. Cortical areas are interconnected by an abundance of corticocortical connections. These connections can be roughly divided into feedforward and feedback connections. Feedforward connections signal information from lower sensory areas to higher-tier areas, while feedback connections carry information from higher-tier areas back to lower areas. These connections have been studied extensively in the visual system, both on an anatomical and on a functional level [3,4]. In the primate visual system, feedforward connections carry information from the primary visual cortex (V1) to higher areas. During this feedforward sweep of activation,

Vol 00 No 00 & & 2006

1

NEUROREPORT information is extracted from a visual scene. Within 100 ms, neurons throughout the entire brain respond to visual stimuli. Fairly complex decisions can be made on the basis of this feedforward activity, such as visual categorization [5], although these decision-making capabilities appear to be ‘hard-wired’ in the system. More ‘cognitive’ functions of the visual system, such as scene segmentation, visual attention and visual awareness, seem to rely on recurrent signals from higher visual areas to lower visual areas, such as V1 [4,6,7]. Scene segmentation, a process in which the visual system divides a visual scene into parts, has been studied using visual evoked potentials (VEPs). Scene segmentation is an important step in figure–ground segregation and in forming object representations. Texture discrimination tasks are often used in laboratory settings to study scene segmentation, and a considerable number of VEP studies have been done using texture discrimination tasks, showing that negative VEP components around 200 ms reflect scene segmentation [8–10]. Single unit recordings in macaque monkeys and imaging studies in humans have provided good evidence that these VEP components are generated by re-entrant signals from higher cortical areas back to V1: deactivating or lesioning circumstriate areas does not affect single-unit activity up to 100 ms, while scene segmentationrelated components are selectively suppressed [11,12]. Earlier components in the VEP, such as the P100, do not seem to rely on re-entrant processing, but more on the feedforward sweep of neural information transfer [13,14]. Therefore, scene segmentation-related activity in the VEP may provide a direct measure of recurrent processing in the visual system. In this study, we investigated the relationship between speed of recurrent and feedforward processing and general intelligence. We used a slightly adapted version of the checkerboard stimuli used by Lamme et al. [8]. We presented homogeneous and checkerboard texture stimuli, made up out of short line segments. These stimuli allow for a strong segmentation of texture squares against a background. To isolate components related to scene segmentation, and thus recurrent processing, homogenous VEPs are subtracted from the checkerboard VEPs. The peak latency of the scene segmentation-related components in this subtraction VEP provide an index of the speed of recurrent processing. To index feedforward activation, we determined P100 peak latencies in the raw checkerboard VEPs. Both these indices of neural processing speed were correlated with scores on the advanced version of Raven’s Progressive Matrices Test [15]. Our results show that processing speed in recurrent, but not feedforward visual networks explain a large portion of the variance of differences in general intelligence.

Methods Study participants Twenty-seven freshman psychology students (10 males, age 18–26 years, mean 20.4 years) participated in the experiment. All participants had normal or corrected-to-normal vision. Written informed consent was obtained from all of them before the experiment. Participants were rewarded with study credits for their participation. All participants were naive towards the purpose of the experiment, and had never participated in a texture discrimination experiment before.

2

Vol 00 No 00 & & 2006

JOLIJ ETAL.

Intelligence measures The intelligence of participants was measured using the 20min timed version of Raven’s Advanced Progressive Matrices Test [15]. In this test, participants are required to fill out as many items out of the untimed Advanced Progressive Matrices Test as possible in 20 min. This shortened version of the test is a reliable predictor of the actual Raven score (Hamel and Schmittmann, in press). The test was administered during the test week, a compulsory part of the BSc Psychology course at the University of Amsterdam, ca 6 months before the electroencephalogram (EEG) recordings. Visual stimulation Stimuli were generated on a PC and displayed on a 19-inch Iiyama monitor, with a refresh rate of 100 Hz and a screen resolution of 1024 by 768 pixels. Participants were seated 100 cm from the computer monitor, and were instructed to fixate on a red fixation dot that was present throughout each trial. Each trial started with a homogeneous texture that was displayed for 300–800 ms. Then, the homogeneous texture was replaced by another homogeneous texture, a 901 orientation contrast checkerboard or a 201 orientation contrast checkerboard, each with a 33% ratio. The background orientation of this second texture was always chosen in such a way that it was different from the background orientation of the initial texture. Neural activity was measured in two conditions: a passive condition, in which participants had to passively observe the stimuli, and an active condition, in which participants had to respond to checkerboards by pressing a button with their dominant hand as quickly as possible. The second texture remained on screen for 500 ms in the passive condition, or until the participant responded in the active condition, after which a new trial started. The size of the individual squares of the checkerboard was 2.41 by 2.21 visual arc. Electroencephalogram recording and analysis EEG was recorded using a 48-channel ActiveOne EEG system (Biosemi Inc., Amsterdam, The Netherlands) with active electrodes. Horizontal and vertical electro-oculograms were measured to control for eye blinks. The EEG signal was digitized and sampled at 256 Hz on a separate acquisition computer. Markers were sent with the EEG signal by the stimulation computer to allow for offline segmentation. Per condition, three blocks of 10 min were recorded; each block contained 450 trials. EEG data were analyzed using Analyzer (BrainVision Products GmbH, Gilching, Germany). The raw signal was filtered between 1 and 15 Hz and then segmented in epochs of 1200 ms (200 ms before stimulus–1000 ms after stimulus) on the basis of markers sent with visual stimulation. Segments were excluded from analysis when maximum amplitude in a segment exceeded 100 mV. A baseline correction (baseline 20 ms) was performed for each segment. VEPs were computed for homogeneous and checkerboard trials. To compute the scene segmentation specific signal, homogeneous trials were subtracted from checkerboard trials. The resulting trials were averaged, thus resulting in a VEP representing neural activity related to scene segmentation (see Fig. 1). For both conditions, and per participant, peak latencies in this subtraction signals were computed by

AQ1

NEUROREPORT

RECURRENT PROCESSING AND INTELLIGENCE

(a)

Homogeneous textures

500 −800 ms

500 ms or until response

different from that of a P300 [8–10,13,16]. Peak latencies for the P100 component were computed by finding maximum amplitude in the interval 90–130 ms in the checkerboard VEPs for the 901 orientation contrast checkerboards and the 201 orientation contrast checkerboards, thus resulting in two P100 latencies per participant. We used SPSS version 10.0 for Windows (SPSS Inc., Chicago, Illinois, USA) to compute Pearson correlations between Raven scores and peak latencies.

90°checkerboards

Results

500 −800 ms

500 ms or until response 20°checkerboards

500 −800 ms

(b)

(µV) 8

500 ms or until response

P100

6 Homogeneous textures

4 2 −2 −4 −6 −8

One participant was excluded from analyses as she forgot to fill out one page of the Raven test. Raven scores of the remaining participants ranged from 9 to 27. The mean score was 20.7, with a standard deviation of 4.5. The median score was 22. Latencies and amplitudes of the VEP components of interest are summarized in Table 1. In both the passive and the active conditions, we found significant correlations between peak latencies of scene segmentation VEP-related components and Raven scores: "0.33 (Po0.05) (passive condition) and "0.36 (Po0.05) (active condition) for the 201 checkerboards and "0.37 (Po0.05) (passive condition) and "0.41 (Po0.05) (active condition) for the 901 checkerboards (Fig. 2). No significant correlations were observed between peak amplitudes and Raven score. Correlations between Raven scores and P100 latencies and amplitudes were not significant in either condition. Reaction times did not correlate with Raven scores (see Table 2).

100

200

300

400

Checkerboard textures (90° shown here) Scene segmentation related component

Fig. 1 Typical trial run and visual evoked potential (VEP) analysis. (a) Example of a typical trial run. After a homogeneous texture, presented for 300^ 800 ms, a 201 orientation contrast, a 901 orientation contrast or a homogeneous texture could appear. Background and foreground elements were always replaced with new elements. In the passive condition, the texture was presented for 500 ms; in the active condition, it was presented until the participant pressed a button. (b) Rationale of theVEP analysis. By averaging all homogeneous orientations (green), and subtracting this signal from the average of all checkerboard presentations of a given orientation contrast (blue), we could isolate the segregation speci¢c signal (red). Latencies of the P100 and the segregation speci¢c components give an index of the speed of feed-forward and feedback activations in the visual system. Here the activity in channel POz for a 901 orientation contrast texture is shown.

finding the minimum amplitude in the interval 200–400 ms in channel POz according to the international 10/20 system, resulting in two latencies for recurrent processing per participant. Please note that this is not a P300: here we report a negative peak, while the P300 is a positive peak. In addition, the topographical distribution of this peak is

Discussion Using VEPs in a simple texture discrimination task, we were able to index the latency of both feedforward and feedback activations in the visual system. We found that processing speed in recurrent but not in feedforward visual networks correlates with general intelligence. These results extend earlier work reporting a correlation between the P300 and intelligence [2]. While the neural processes underlying the P300 are still under discussion [17], the neural generators of the VEP components we describe here have been characterized using single cell measurements in monkeys and functional magnetic resonance imaging and transcranial magnetic stimulation in humans [8–14,16]. This makes it possible to further specify the relationship between intelligence and neural processing. Our results corroborate earlier work with inspection time. Inspection time is measured using a visual masking paradigm. Recent work on visual masking shows that masking a stimulus interferes with recurrent processing of the masked stimulus: if a new stimulus is presented before re-entrant signals from higher visual areas reach early visual areas, masking occurs [18,19]. Feedforward signals remain unaffected by a visual mask. Our work suggests that inspection time in brighter participants is smaller because re-entrant signals reach early visual areas faster. Is there a causal link, however, between processing speed in recurrent visual networks and intelligence? A recent study by Luciano et al. [20] suggests that there is no causal relationship between perceptual speed and intelligence. Hence, a link between recurrent connections in the visual system and intelligence seems to be no more than a covariation. One hypothesis might be that brighter indivi-

Vol 00 No 00 & & 2006

3

NEUROREPORT

JOLIJ ETAL.

Table 1 Mean amplitudes (in mV) and latencies (in ms) of visual evoked potential components re£ecting feedforward and re-entrant processing 201 Passive condition

Feedforward Re-entrant

901 Passive condition

201 Active condition

901 Active condition

Amplitude

Latency

Amplitude

Latency

Amplitude

Latency

Amplitude

Latency

5.6 (2.9) "3.1 (1.8)

119 (5) 316 (52)

5.9 (3.6) "5.8 (3.0)

117 (5) 302 (33)

5.7 (2.6) "9.6 (4.3)

125 (6) 323 (33)

6.2 (2.9) "14 (5.1)

124 (6) 293 (28)

Numbers in brackets indicate standard deviations.

25

25

20

20

Raven score

Raven score

(a)

15

15 10

10 5 200

250

350 300 Latency (ms)

5 250

400

300 350 Latency (ms)

400

(b) 25 Raven score

Raven score

25 20 15 10

20 15 10

5

5 300

250

350

400

250

275

Latency (ms)

300

325

350

Latency (ms)

Fig. 2 Scatterplots of Raven scores and visual evoked potential (VEP) latencies. Latency of scene segmentation-related VEP components is plotted against Raven score for all conditions in the experiment. Circles show individual data points, lines show best linear ¢t. (a) Checkerboards of 201: passive condition (left panel) and active condition (right panel). (b) Checkerboards of 901: passive condition (left panel) and active condition (right panel). Table 2 Correlations between Raven scores and visual evoked potential components re£ecting feedforward and re-entrant processing 201 Passive condition

Feedforward Re-entrant

901 Passive condition

201 Active condition

901 Active condition

Amplitude

Latency

Amplitude

Latency

Amplitude

Latency

Amplitude

Latency

0.22 0.17

"0.13 "0.33*

0.17 "0.19

"0.11 "0.37*

0.31 "0.13

"0.20 "0.36*

0.17 "0.10

"0.11 "0.41*

*Po0.05 (one tailed), indicates signi¢cant correlation.

duals are able to direct their attention faster towards the texture stimuli, and thus have faster recurrent interactions within their visuals systems. Texture segregation, however, as well as the segregation-specific VEP signals we recorded here, occurs in complete absence of attention [21,22]. Furthermore, if more intelligent participants are able to direct their attention faster, or in a more efficient way, this

4

Vol 00 No 00 & & 2006

should show up in the latency or amplitude of the P100, as the P100 is known to be modulated by attention [23]. No correlation is observed between P100 latency and amplitude with Raven scores. Therefore, it seems unlikely that individual differences in attention for the texture stimuli may fully explain our results.

RECURRENT PROCESSING AND INTELLIGENCE

As genetic factors explain a large portion of variance both in inspection time and in general intelligence, it has been proposed that myelination of axons may be the mediating factor [24]. Our results show that such a hypothesis needs refinement. Not all neuronal processing is faster in brighter individuals: it seems that faster processing is limited to recurrent networks. Hence, in the search for the biological basis of intelligence, one may need to focus on myelination of exclusively cortico-cortical connections, explaining why earlier studies focusing on myelination in the peripheral nervous system did yield little result [2]. Finally, we need to consider the possibility that the VEP components we recorded do not necessarily reflect purely feedforward and feedback processing. Although we may assume, on the basis of single-cell work, and on human psychophysiology, that the bulk of neural activity underlying the P100 and scene segmentation-related VEP components is related to feedforward and feedback processing, respectively [8–14,16,23], it is very well possible that some feedback processing underlies the P100, and some feedforward processing underlies scene segmentation-related components. Even if this may the case, our results still show that general intelligence only correlates with the speed of specific neural processes: the speed of initial encoding of sensory input does not correlate with general intelligence, but the speed of subsequent analysis leading to scene segmentation does. Conclusion Do smarter brains run faster? Here we demonstrate that this hypothesis needs to be refined: not all processes in smarter brains run faster. We show evidence that mainly a smarter brain’s feedback connections run faster.

Acknowledgements We thank Jelte Wicherts and Suzanne Kats for their help with acquisition of the Raven scores, and Frouke Hermens for technical assistance with data analysis.

References 1. Kranzler JH, Jensen AR. Inspection time and intelligence: a meta-analysis. Intelligence 1989; 13:329–347. 2. Deary IJ, Caryl PG. Neuroscience and human intelligence differences. Trends Neurosci 1997; 20:365–371. 3. Salin PA, Bullier J. Corticocortical connections in the visual system: structure and function. Physiol Rev 1995; 75:105–154.

NEUROREPORT 4. Lamme VAF, Roelfsema PR. The different modes of vision offered by feedforward and feedback processing. Trends Neurosci 2000; 23:571–579. 5. VanRullen R, Koch C. Visual selective behavior can be triggered by a feedforward process. J Cogn Neurosci 2003; 15:209–217. 6. Lamme VAF, Supe`r H, Landman R, Roelfsema PR, Spekreijse H. The role of primary visual cortex (V1) in visual awareness. Vis Res 2000; 40:1507– 1521. 7. Supe`r H. Cognitive processing in the primary visual cortex: from perception to memory. Rev Neurosci 2002; 13:287–298. 8. Lamme VAF, Van Dijk BW, Spekreijse H. Texture segregation is processed by primary visual cortex in man and monkey. Evidence from VEP experiments. Vis Res 1992; 32:797–807. 9. Bach M, Meigen T. Electrophysiological correlates of texture segregation in the human visual evoked potential. Vis Res 1992; 32:417–424. 10. Caputo G, Casco C. A visual evoked potential correlate of global figure– ground segmentation. Vis Res 1999; 39:1597–1610. 11. Lamme VAF, Supe`r H, Spekreijse H. Feedforward, horizontal, and feedback processing in the visual cortex. Curr Opin Neurobiol 1998; 8:529–535. 12. Hupe´ JM, James AC, Payne BR, Lomber SG, Girard P, Bullier J. Cortical feedback improves discrimination between figure and background by V1, V2, and V3 neurons. Nature 1998; 394:784–787. 13. Scholte HS, Jolij J, Lamme VAF. The cortical processing dynamics of edge detection and scene segmentation. In: Breitmeyer B, Ogmen H, editors. The first half second: the microgenesis and temporal dynamics of unconscious and conscious visual processes. Cambridge, Massachusetts: MIT Press; 2006. pp. 73–89. 14. Heinen K, Jolij J, Lamme VAF. Figure–ground segregation requires two distinct periods of activity in V1: a transcranial magnetic stimulation study. Neuroreport 2005; 16:1483–1487. 15. Raven J, Raven JC, Court JH. Manual for Raven’s progressive matrices and vocabulary scales. Oxford, UK: Oxford Psychologists Press; 1998. 16. Scholte HS. Scene segmentation. PhD thesis. Amsterdam: University of Amsterdam; 2003. 17. Nieuwenhuis S, Aston-Jones G, Cohen JD. Decision making, the P3, and the locus coeruleus-norepinephrine system. Psychol Bull 2005; 131:510– 532. 18. Enns JT, Di Lollo V. What’s new in visual masking? Trends Cogn Sci 2000; 4:345–352. 19. Lamme VAF, Zipser K, Spekreijse H. Masking interrupts figure–ground signals in V1. J Cogn Neurosci 2002; 14:1044–1053. 20. Luciano M, Posthuma D, Wright MJ, de Geus EJ, Smith GA, Geffen GM, et al. Perceptual speed does not cause intelligence, and intelligence does not cause perceptual speed. Biol Psychol 2005; 70:1–8. 21. Scholte HS, Witteveen SC, Spekreijse H, Lamme VAF. The influence of inattention of neural correlates of scene segmentation. Brain Res 2006; 1076:106–115. 22. Schubo A, Meinecke C, Schroger E. Automaticity and attention: investigating automatic processing in texture segmentation with eventrelated brain potentials. Brain Res Cogn Brain Res 2001; 11:341–361. 23. Coull JT. Neural correlates of attention and arousal: insights from electrophysiology, functional neuroimaging and psychopharmacology. Prog Neurobiol 1998; 55:343–361. 24. Posthuma D, De Geus EJ, Boomsma DI. Perceptual speed and IQ are associated through common genetic factors. Behav Genet 2001; 31:593–602.

Vol 00 No 00 & & 2006

5

AUTHOR QUERY FORM

LIPPINCOTT WILLIAMS and WILKINS JOURNAL NAME: WNR ARTICLE NO: 3822 QUERIES AND / OR REMARKS QUERY NO

Q1

Details Required Please provide all details of Ref. 'Hamel and Schmittmann (in press) to be included in the Reference list.

Author's Response