Representation of Numerical and Sequential Patterns in ...

Article

Representation of Numeri

Highlights

d The monkey brain is capable of representing numerical and

sequence patterns

d fMRI responses to number and to sequence are segregated in

the monkey

d The human inferior frontal gyrus responds to both types of

patterns

d Humans and monkeys differ even in a simple sequence

learning paradigm

Wang et al., 2015, Current Biology 25, 1966–1974August 3, 2015 ª2015 Elsevier Ltd All rights reservedhttp://dx.doi.org/10.1016/j.cub.2015.06.035

Authors

Liping Wang, Lynn Uhrig, Bechir

Jarraya, Stanislas Dehaene

[email protected] (L.W.),[email protected] (S.D.)

In Brief

Wang et al. used fMRI in untrained

macaques and humans to investigate the

brain areas involved in representing the

abstract patterns underlying a series of

tones. While number and sequence

patterns are available to macaques, a

unique integrated response to both

patterns was observed in human inferior

frontal and superior temporal cortex.

Current Biology

Article

Representation of Numerical and SequentialPatterns in Macaque and Human BrainsLiping Wang,1,2,4,5,* Lynn Uhrig,1,2,3 Bechir Jarraya,1,2,3,6,7 and Stanislas Dehaene1,2,8,9,*1Cognitive Neuroimaging Unit, INSERM, 91191 Gif-sur-Yvette, France2CEA, DSV, I2BM, NeuroSpin Center, 91191 Gif-sur-Yvette, France3Inserm Avenir-Bettencourt-Schueller Team, 91191 Gif-sur-Yvette, France4Key Laboratory of Brain Functional Genomics (MOE and STCSM), Institute of Cognitive Neuroscience, School of Psychology and Cognitive

Science, East China Normal University, 200062 Shanghai, China5NYU-ECNU Institute of Brain and Cognitive Science, NYU Shanghai, 200062 Shanghai, China6Universite Versailles Saint-Quentin-en-Yvelines, 78000 Versailles, France7Neuromodulation Unit, Department of Neurosurgery, Foch Hospital, 92150 Suresnes, France8College de France, 75005 Paris, France9Universite Paris-Sud, 91400 Orsay, France

*Correspondence: [email protected] (L.W.), [email protected] (S.D.)http://dx.doi.org/10.1016/j.cub.2015.06.035

SUMMARY

The ability to extract deep structures from auditorysequences is a fundamental prerequisite of languageacquisition. Using fMRI in untrained macaques andhumans, we investigated the brain areas involved inrepresenting two abstract properties of a series oftones: total number of items and tone-repetitionpattern. Both species represented the number oftones in intraparietal and dorsal premotor areas andthe tone-repetition pattern in ventral prefrontal cortexand basal ganglia. However, we observed a jointsensitivity to both parameters only in humans, withinbilateral inferior frontal and superior temporal re-gions. In the left hemisphere, those sites coincidedwith areas involved in language processing. Thus,while someabstractpropertiesof auditorysequencesare available to non-human primates, a recentlyevolved circuit may endow humans with a uniqueability for representing linguistic and non-linguisticsequences in a unified manner.

INTRODUCTION

A major issue for cognitive neuroscience is to determine how hu-

man representational abilities differ from those of other species.

Language acquisition is a prime example of fast learning that

seems unique to humans. It is often proposed that the faculty of

language reflects a broader human-specific ability to acquire

and represent recursive structures [1] or regular combinations

of symbols [2]. Sensitivity to abstract patterns and regularities is

essential to mathematics and music, two other faculties uniquely

developed in humans [3, 4]. Searching for comparative evidence

on the neural representation of rules and symbols may therefore

shed a unique light on the evolutionary origins of human cognition.

Previous studies have shown that the discovery of numerical

or logical patterns called ‘‘algebraic rules’’ [5] is a powerful

1966 Current Biology 25, 1966–1974, August 3, 2015 ª2015 Elsevier

mechanism, already available to human infants, and may play

an important role in the acquisition of language. In a seminal

study [5], 7-month-old infants were exposed for only 2 min to se-

quences respecting a regularity such as AAB (all sequences

contain two identical sounds followed by a different one). Infants

later attended longer to stimuli that violated the rule than to novel

stimuli that respected it. Such evidence suggests that infants

could detect and memorize at least some aspects of the regular

pattern governing the stimuli (e.g., the initial repetition of two

sounds, or the change in the last item). It has been claimed

that monkeys and some birds may possess the rudiments of

this ability [6–8], but current evidence remains inconclusive

[9–12]. Although non-human primates can learn patterns based

on number [13] or artificial grammars [8], we still do not know

whether and how the neural networks underlying such abstract

features differ in monkey and human brains.

At the brain level, electrophysiological recordings in mon-

keys have shown that single neurons in prefrontal and parietal

cortical regions can encode motor patterns such as AABB

or ABAB, where A and B are unspecified gestures [14–16].

However, these results raise several issues. First, those brain

representations were only studied after extensive training;

demonstrations of numerical or symbolic coding in untrained

animals, such as the presence of number-tuned neurons in pa-

rietal and prefrontal cortex [17], are quite scarce. Second,

electrophysiological studies, unlike brain-imaging studies

[18–20], do not allow for a direct comparison of the neural cir-

cuits for sequence representation in monkeys and humans at

the whole-brain level.

fMRI allows exploring whole-brain activity in monkeys and hu-

mans. In a recent fMRI study [20], we demonstrated that, once

monkeys repeatedly heard a specific auditory melody aaaab

(where a and b are two fixed tones), hearing a deviant sequence

aaaaa led to widespread activation in temporal, parietal, and

prefrontal cortices, at sites similar to humans [21]. This novelty

response, however, is ambiguous. It might simply indicate that

monkeys can memorize melodies and detect a novel one. Alter-

natively, it could arise from a sensitivity to the violation of

abstract properties such as number (‘‘four sounds plus another

Ltd All rights reserved

Figure 1. Experimental Design

(A–C) Experimental design (A and B) and main effects of number and sequence changes in monkeys and humans (C).

(A) In different runs, subjects habituated to auditory stimuli respecting a fixed sequence pattern: AAAA or AAAB.

(B) They were then presented with rare test stimuli forming a 23 2 design, respecting the existing pattern, changing the number of items (from four to two or six),

changing the repetition pattern (e.g., going from AAAB to AAAA or vice versa), or changing both (e.g., going from AAAA to AAAAAB). Variability in temporal

spacing and pitch ensured that only the abstract numerical or repetition pattern was predictable.

(C) Main effects of number and sequence change were identified at a whole-brain level using fMRI (contrasts of N+ > N� and of S+ > S�).

Abbreviations are as follows: 6VR, ventral premotor; ACC, anterior cingulate cortex; VIP, ventral inferior parietal; IFG, inferior frontal gurus; IPS, intraparietal

sulcus; SMA, supplementary motor area.

one’’) or tone-repetition pattern (e.g., ‘‘one sound is different’’ or

‘‘the last sound is different’’).

The present paradigm was therefore designed to probe the

sensitivity of monkeys and humans to such abstract auditory

properties and to identify whether the two species use similar

brain areas for this task. We used fMRI to visualize whole-brain

activity while awake monkeys and humans were passively

exposed to auditory sequences. During an initial habituation

phase, subjects heard sequences with a fixed pattern (AAAB

or AAAA). Critically, A and B could be any of several sounds,

and duration and temporal spacing were constantly varied

such that only the pattern itself could be learned (Figures 1 and

S1). Using fMRI, we then tested for brain responses to novel

sequences that either respected the original pattern or violated

it. The violations consisted in changing the total number of items

(e.g., going from AAAB to AB or to AAAAAB), changing the tone-

repetition pattern (going from AAAA to AAAB or vice versa), or

changing both (e.g., going from AAAA to AAAAAB). Again, con-

trols ensured that discrimination could not be based on other

non-numerical parameters (see Figure S1). This design resulted

in four test conditions: N�S�, new exemplars of same rule;

N+S�, isolated number deviants; N�S+, isolated sequence de-

viants; and N+S+, double deviants (Figure 1). Importantly, both

species were naive to the auditory sequences, had not been

actively trained to discriminate them, and simply performed an

unrelated eye-fixation task while the auditory stimuli were

presented.

Current Biology 25, 1966

RESULTS

The first question is, can monkeys identify the invariable pattern

underlying the variable sequences? If monkey brains extract the

pattern, then they should generalize to novel exemplars and

respondonly topattern-violating items. If, on theotherhand,mon-

key brains only store specific melodies in memory, then even the

N�S� test sequences should elicit a novelty reaction. Thus, we

first examined brain regions responsive to the N�S� test stimuli

with only frequency changes. ContrastingN�S�with habituation

showed no significant brain responses (voxel-wise, p < 0.005,

corrected by false discovery rate [FDR], p < 0.05), suggesting

that monkeys extracted the pattern and generalized it to novel

items.

We then examined the monkey brain responses to pattern-

violating sequences. To identify brain regions responsive to

number change, we first compared sequences with an isolated

violation in number (N+S�) to those without any violation

(N�S�) (Figure 2A). Despite a certain variability in the activations

observed in individual monkeys (see Figure S2), 14 significant

activation peaks were identified in the group analysis (Table

S1; p < 0.005, FDR p < 0.05). In parietal areas, the strongest acti-

vation was found bilaterally in the ventral part of the intraparietal

sulcus (VIP), a site previously found to contain number-sensitive

neurons [22, 23] (Figure 2A). Both two- and six-tone number

deviants elicited significantly higher VIP activations than the

four-tone test stimulus (t test, p < 0.05; Figure S4A). In addition,

–1974, August 3, 2015 ª2015 Elsevier Ltd All rights reserved 1967

Figure 2. Brain Activation to Isolated Num-

ber Changes

(A) Brain activation to isolated number deviants

(N+S�) versus control stimuli (N�S�) projected on

lateral and top views of the brain (monkeys, t > 2.7,

p < 0.005, FDR p < 0.05 corrected; humans, t > 3.0,

p < 0.001, FDR p < 0.05 corrected).

(B) Brain activation (betas) at specific peaks in the

four test conditions (N+S+, N+S�, N�S+, and

N�S�; brain activation was identified as the mean

beta weight of the regression of the fMRI response

onto the corresponding predictor). An asterisk

denotes a significant main effect of number

[(N+S+) + (N+S�) > (N�S�) + (N�S+)]. Error bars

indicate 1 SE.

Abbreviations are as follows: 6DR, dorsal pre-

motor; 8A, area 8A; aINS, anterior insular cortex;

pSTS, posterior superior temporal sulcus; ACC,

anterior cingulate cortex; VIP, ventral inferior pari-

etal; IFG, inferior frontal gyrus; IPS, intraparietal

sulcus; SMA, supplementary motor area. See also

Table S1.

the posterior bank of the inferior arcuate sulcus (area 6DR, i.e.,

dorsal premotor F2/F4) was also activated bilaterally, as were

the posterior superior temporal sulcus (pSTS), anterior insula

(aINS), supplementary motor area (SMA), and anterior cingulate

(ACC). Similar results were found when we examined a second,

more stringent contrast for a response to number, namely the

main effect of number change, regardless of the presence of

a concomitant change in sequence [(N+S+) + (N+S�) >

(N�S�) + (N�S+)] (Figure 1C): responses were found in areas

VIP, 6DR, pSTS, SMA, and ACC, but not aINS (p < 0.005,

pFDR < 0.05 corrected). Finally, we computed a third, even

more stringent statistic consisting in a conjunction analysis for

N+S+ > N�S+ and N+S� > N�S� (Figure 4A), which therefore

searched for a replicable response to number change, whether

or not there also was a concomitant change in sequence pattern.

Again, this conjunction criterion identified areas VIP, pSTS, SMA,

and ACC (conjunction null, p < 0.01, pFDR < 0.05). As indicated by

the histograms in Figure 2B, those areas showed a positive acti-

vation (relative to the mean of the habituation stimuli in this run)

whenever the total number of tones suddenly changed and a

null or, in some cases, a negative activation (suggesting further

habituation) whenever this number remained equal to its habitu-

ation value. The main brain regions (VIP and ACC) showing the

1968 Current Biology 25, 1966–1974, August 3, 2015 ª2015 Elsevier Ltd All rights reserved

number effect were also observed in indi-

vidual monkeys (Figures S2 and S3).

Thus, monkeys possess a set of regions

responsive to number change, irrespec-

tive of concomitant changes in sound fre-

quency, timing, and sequence pattern.

When we ran the same paradigm in

humans, we observed activations to nu-

merical deviance (N+S� > N�S�) in bilat-

eral intraparietal sulcus (IPS) (p < 0.001,

pFDR < 0.05 corrected), at a site plausibly

homologous to the monkey site (Fig-

ure 2A). As in monkeys, additional activa-

tions were observed bilaterally in the

pSTS, SMA, medial prefrontal (mPFC), and aINS (Table S1).

However, different from monkeys, humans showed an intense

activation in bilateral inferior frontal gyri (IFG) (Figure 2A, Human).

There was a significant main effect of number in IPS, IFG,

pSTS, SMA, medial prefrontal (mPFC), and aINS areas (see plots

of activation in Figure 2B). These areas also remained in the

conjunction analysis (p < 0.01, pFDR < 0.05 corrected), indicating

a context-independent numerical response (Figure 4A). The IPS,

left IFG, and SMA closely overlapped with areas active during

mental calculation in the same subjects (Figure S4B).

In summary, the numerical feature of auditory sequences

engaged highly similar networks in both species. The areas

involved primarily belonged to a dorsal auditory pathway [24].

The intraparietal cortex, previously involved in number represen-

tation [3, 22, 23], was a dominant node in both species, yet we

also observed additional activation of the IFG only in humans.

We next probed the sensitivity to sequence changes in both

species. In the monkey group analysis, nine cortical regions

were activated by isolated sequence changes, i.e., whenever

the sequence pattern suddenly changed from AAAB to AAAA

or vice versa (N�S+ > N�S�; p < 0.005, pFDR < 0.05 cor-

rected; Figures 1C and 3A; Table S2). Unlike for number, a

ventral auditory pathway was seen. In the frontal cortex, the

Figure 3. Brain Activation to Isolated

Sequence Changes

Brain activations to isolated sequence changes

(violations of the tone-repetition pattern) in mon-

keys and humans.

(A) Brain activation to isolated sequence deviants

(N�S+) versus control stimuli (N�S�) in monkeys

(t > 2.7, p < 0.005, FDR p < 0.05 corrected) and

humans (t > 3.0, p < 0.001, FDRp < 0.05 corrected).

(B) Brain activations in the four test conditions

(same format as Figure 2).

Abbreviations are as follows: aINS, anterior insular

cortex; pSTS, posterior superior temporal sulcus;

IFG, inferior frontal gyrus; TP, temporal pole. See

also Table S2.

strongest activations were found bilaterally in area 6VR

(i.e., ventral premotor area F5) and the ventral part of dorsolat-

eral prefrontal cortex (VLPFC, area 46v), extending to the

anterior insula. In the temporal lobe, the bilateral pSTS and

left anterior STS (aSTS) were significantly activated. A subcor-

tical region in the basal ganglia, previously engaged in

sequence chunking [25], was also activated (left caudate,

x = �2, y = 5, z = 4; see Figure S4D). Our second statistical

criterion, the main effect of sequence violation [(N+S+) +

(N�S+) > (N�S�) + (N+S�)], confirmed the contribution to

sequence processing of areas 6VR/VLPFC, pSTS, aSTS, and

caudate, but not aINS (Figure 3B). Finally, the most stringent

conjunction analysis, searching for areas activated in both

contrasts N+S+ > N+S� and N�S+ > N�S� established

that areas 6VR/VLPFC, pSTS, and left aSTS reacted to

sequence change even when there was a concomitant change

in number (conjunction null, p < 0.01, pFDR<0.05 corrected;

Figure 4A). The main brain regions showing the sequence

Current Biology 25, 1966–1974, August 3, 2015

effect (6VR and caudate) were also

observed in individual monkeys (Figures

S2 and S3).

Similar analyses in humans revealed

that, as in monkeys, isolated sequence

deviants activated the bilateral pSTS and

temporal polar (TP) cortices (p < 0.001,

pFDR < 0.05 corrected; Figure 3A). Also

consistent with the monkey results was a

bilateral activation in the basal ganglia

(putamen, x = 21, y = 9, z = 7, t = 3.62;

and x = �23, y = 7, z = 14, t = 3.15; Fig-

ure S4D). Furthermore, humans again

showed an additional intense frontal

activation in the bilateral IFG. All of these

areas were confirmed using two addi-

tional criteria for sequence responses,

namely the main effect of sequence viola-

tion (Figure 3B) and the conjunction anal-

ysis (Figure 4A).

In summary, a striking difference

between species was that the IFG was

activated by both sequence and number

changes in humans, while number and

sequence violations seemed to involve

distinct networks in monkeys. To quantify this observation, we

explored which regions showed intersecting statistical maps

for the main effects of number and sequence change, thus

possibly operating as integrative structures or ‘‘hubs.’’ In mon-

keys, no significant regions showed joint effects, either additively

or with an interaction of both factors (both p < 0.005, corrected

by FDR). Humans were different: conjunction analysis identified

additive joint effects of number and sequence in bilateral pSTS

and IFG (Figure 4A; Table S3).

To confirm this finding with highest statistical sensitivity, we

applied a leave-one-out cross-validation approach: we first

used number-change (N+ > N�) or sequence-change (S+ >

S�) contrasts as functional localizers to define the voxels of

interest, in all but one of fMRI runs, then examined the responses

to either the N+ > N� or S+ > S� contrasts in the left-out run.

In monkeys, the results confirmed that the regions of IPS and

ACC/SMA were selectively activated to the number deviants,

and regions of aSTS, pSTS, and 6V were activated to the

ª2015 Elsevier Ltd All rights reserved 1969

Figure 4. Evidence for a Uniquely Human Joint Sensitivity to Number

and Sequence Patterns

(A) Conjunction analyses identified the areas for number change detection,

regardless of a concomitant change in sequence pattern (conjunction of

N+S+ >N�S+ andN+S� >N�S�, shown in red) and, conversely, the areas for

sequence change detection, regardless of a concomitant change in number

(conjunction of N+S+ > N+S� and N�S+ > N�S�, shown in green). Brain

activations in monkeys and humans are projected on lateral and top views of

the brains. Maps are thresholded at t > 2.4 (monkey) and at t > 3.0 (human),

which corresponds to the conjunction null, p < 0.01, FDR p < 0.05 corrected.

Arrows indicate the uniquely human peaks common to both number and

sequence conjunction analyses (Table S3).

(B) Conjunction map of the two main effects of number and sequence change

in humans (contrasts (N+S+) + (N+S�) > (N�S�) + (N�S+) and (N+S+) +

(N�S+) > (N�S�) + (N+S�); t > 2.5, conjunction null, p < 0.01, corrected by

FDR) superimposed to human cytoarchitectonically defined areas 44 (blue

box) and BA45 (green box). Note that the joint activations in IFG are confined to

BA44.

(C) Activation in subject-specific language-responsive voxels within seven

regions of interest (ROIs) in humans. Within each subject, voxels responsive to

sentences processing (p < 0.01, uncorrected) in each localizer were identified.

Brain activation within those voxels is plotted for the four test conditions.

1970 Current Biology 25, 1966–1974, August 3, 2015 ª2015 Elsevier

sequence deviants (Figure S3). Crucially, even with this sensitive

analysis, no activations were observed in the generalization

across conditions (S+ > S� contrast in voxels isolated by the

N+ > N� contrast, or vice versa), confirming that the number-

change and sequence-change networks involve non-overlap-

ping voxels. Human results, by contrast, not only replicated the

brain areas showing the number effect in IPS, SMA, and IFG

and the sequence effect in TP, putamen, pSTS, and IFG but

also showed cross-condition generalization, with a joint activa-

tion of both effects in bilateral pSTS and IFG (Figure S3).

To specifically test inferior frontal regions in both monkey and

human, we then performed a three-way ANOVA with factors of

region (i.e., voxels identified either by the N+/N� contrast or by

the S +/S� contrast) 3 number change 3 sequence change,

using as dependent variable the mean fMRI activation of the

cross-validated voxels within human IFG and monkey area F5,

respectively. In humans, there was a significant main effect of

number (F(1,16) = 8.47, p = 0.01) and sequence (F(1,16) = 4.65,

p = 0.03), but no significant interactions (region 3 number,

F(1,16) = 1.29, p = 0.27; region 3 sequence, F(1,16) = 0.02, p =

0.96; number3 sequence, F(1,16) = 0.56, p = 0.46; region3 num-

ber 3 sequence, F(1,16) = 0.99, p = 0.33). Hence, in humans, IFG

voxels showed joint effects of number and sequence, irre-

spective of the contrast used to identify them. In monkeys, by

contrast, the results showed a significant main effect of

sequence (F(1,29) = 29.8, p = 0.0002) and a significant effect

in region 3 sequence interaction (F(1,29) = 4.95, p = 0.03), but

no number effect (F(1,29) = 0.10, p = 0.75) or any significant

region 3 number interaction (F(1,29) = 0.72, p = 0.41). This is

consistent with monkey IFG responding only to sequence

change and only in voxels identified using the sequence change

contrast.

In the human left hemisphere, the IFG and pSTS regions coin-

cided with those previously identified as forming a core network

for language syntax [26]. Indeed, probabilistic cytoarchitectonic

maps located the human IFG conjunction effect to Brodmann’s

area 44 (Figure 4B), previously associated with hierarchical

linguistic and non-linguistic sequences [27, 28]. The fact that

our paradigm involves a minimal form of ‘‘syntax,’’ yet with

non-verbal stimuli (tones), may explain why the human IFG was

recruited in both left and right hemispheres, as also found during

artificial language processing [27], musical syntax processing

[4], mathematical calculation [3], and hierarchically organized

behavior [29].

To clarify whether the human left-hemispheric areas in our

study corresponded precisely with language-processing re-

gions, we compared the human sequence-change responses

to the areas identified in the same group of subjects using an

independent language localizer [30]. The results showed over-

lapping areas for non-verbal sequence change and for verbal

sentence processing in TP and putamen. The left IFG showing

a sequence effect was just posterior to the activations in sen-

tence processing, with a slight overlap (Figure S4B). A further

Activations in IFGoper, IFGorb, and pSTS showed significant main effects

of number and sequence change (N, number effect; S, sequence effect;

**p < 0.05 corrected; *p < 0.05 uncorrected; see Table S4 for p values). Error

bars represent ±1 SEM.

Ltd All rights reserved

Figure 5. Representational Similarity Anal-

ysis of Activations to Number and Sequence

Changes

Left: ROIs in human prefrontal cortex and inferior

frontal gyrus (IFG) and monkey frontal cortex (PFC)

and F5. ROIs were created by SPM WFU_Pick-

Atlas Toolbox (http://fmri.wfubmc.edu/software/

pickatlas). Monkey F5 ROIs were defined as

spheres of 8-mm radius centered on the peak

sequence effect (left: [�16, �4, 7]; right: [15, 4, 7]).

Within these large regions, we first searched for

subject-specific voxels activated in the contrast of

all sound sequences relative to rest. We then

treated this list of voxels as a vector and examined

the correlation of the brain responses to number

change and to sequence change. Right: mean

correlation coefficients (±SEM) between the brain

responses to number change and to sequence

change in these voxels. The mean correlations in

humans (with both PFC and IFG/F5 ROIs) are

significantly higher than those in the monkeys

(ANOVA, main effect of species, **p < 0.001).

comparison between our auditory sequence-responsive areas

and other sequence-processing studies of the human hier-

archical organization of motor actions [29] and the structure

complexity of human language [27] showed highly consistent

activations in the inferior frontal regions (Figure S5).

Because overlapping fMRI activations may arise spuriously at

the group level, we sought to confirm our findings in individual

subjects. We used an independent functional localizer for

language [30] to identify subject-specific voxels activated during

sentence processing. All subjects showed significant voxels in

each of seven left-hemispheric regions of interest (ROIs) from a

previous study of language constituent structure [26]. Within

those voxels, we then evaluated the contrasts for non-verbal

number and sequence violations (Figure 4C). The results

confirmed joint effects of number and sequence in left inferior

frontal language area IFGoper (mean coordinates �41, 8, 23;

number: F(1, 16) = 29.86, p < 0.001, Bonferroni-corrected for

seven ROIs; sequence: F(1, 16) = 35.79, p < 0.001, corrected;

and interaction effect: F(1, 16) = 10.70, p < 0.05, corrected). At a

less significant level, joint effects were also found in IFGorb

(�45, 32, �6; number: F(1, 16) = 9.83, p < 0.05 corrected;

sequence: F(1, 16) = 7.77, p < 0.05 uncorrected) and in pSTS

(�50, �42, 4; number: F(1, 16) = 7.27, p < 0.05 uncorrected;

sequence: F(1, 16) = 8.24, p < 0.05 uncorrected).

Current Biology 25, 1966–1974, August 3, 2015

The human IFG areas showing these

effects are thought to be putative cy-

toarchitectonic homologs of ventrolateral

PFC areas 6VR, 44, and 45 in the ma-

caque monkey [31]. Indeed, in those

regions, monkeys showed a sequence

effect, but unlike humans, they exhibited

no number effect. It could, however, be

argued that the latter is merely a null

result. In a final analysis, we therefore

asked whether the functional conver-

gence of number and sequence re-

sponses in humans and their dissociation

in monkeys could be directly demonstrated. We used represen-

tational similarity analysis to quantify the within-subject topo-

graphic resemblance of IFG activations to number and sequence

in the frontal cortex of both species (Figure 5). We first identified

subject-specific voxels activated by all auditory sounds

(voxel, p < 0.005, uncorrected), then computed the correlation

of number- and sequence-change contrasts over those voxels.

An ANOVA on the correlation coefficients, with factors of hemi-

sphere (left or right) and species (human or monkey), revealed

a highly significant difference between species (both PFC and

IFG/F5 ROIs, p < 0.001; Figure 5). Humans showed significantly

positive correlations in bilateral IFG (two-tailed t test, p < 0.05) as

well as in the entire prefrontal cortex (p < 0.05), indicating a func-

tional convergence of number and sequence responses to the

same voxels. Monkeys, on the other hand, tended toward a

negative correlation, both in area F5 proper (p = 0.07) and in fron-

tal cortex (p = 0.08), indicating a functional segregation.

DISCUSSION

Our results indicate that monkeys, like humans, are sensitive to

the numerical and sequential structures of auditory sequences.

In line with previous reports, we found that the macaque and hu-

man brain possess homologous networks that spontaneously

ª2015 Elsevier Ltd All rights reserved 1971

encode the abstract feature of number. The intraparietal sulcus

has long been related to the coding of cardinal and ordinal num-

ber in bothmonkeys [22, 23] and humans [3, 32]. Area VIP, where

number-change responses were found in the present and past

work [22], contains neurons tuned to specific numbers of serially

presented tones, exactly as required to encode the present stim-

uli [22, 23]. Our whole-brain results, using auditory sequences in

naive monkeys, converge with a recent monkey electrophysi-

ology study demonstrating numerical responses to visual sets

of dots in naivemonkeys in area VIP [17]. Behavioral experiments

confirm that untrained monkeys [13] and humans [32] are able to

spontaneously discriminate numbers in various auditory or visual

formats.

The present study further demonstrates that the monkey brain

is also able to spontaneously represent the algebraic pattern

(AAAB or AAAA) underlying otherwise variable auditory se-

quences. Algebraic capacities of this sort were previously

explored behaviorally in 7-month infants [5] and in tamarin mon-

keys [33]. It is the first time, however, that the brain mechanisms

of this ability are identified. In monkeys, the cortical areas

showing the sequence effect involve the ventral auditory

pathway (pSTS, aSTS/TP, and VLPFC/IFG), confirming its

important role in auditory pattern discrimination [34]. Monkey

VLPFC was previously involved in auditory object identification

and macaque vocalizations [35]. Single neurons in the rhesus

VLPFC are tuned to categories of auditory vocalizations [36].

The VLPFC has thus been proposed to be a component of neural

circuit involved in the categorization of socially meaningful sig-

nals. Macaque prefrontal neurons were also found to represent

action sequence boundaries [37]. In this respect, PFC likely

operates in tight connection with the basal ganglia, where

we observed sequence-change responses in both species

and where single-unit recordings in monkey have revealed

learning-related changes in firing patterns related to habit forma-

tion and sequence chunking [25]. Studies of human patients with

lesions or neurodegenerative diseases also suggest a role for the

basal ganglia in syntactic rule processing [38].

Beyond this inter-species similarity, our data demonstrate that

human bilateral inferior frontal areas (mainly BA44) and posterior

superior temporal sulcus possess a unique ability for multi-

feature integration, as they show correlated effects of number

and sequence change that are absent in monkeys. In the left

hemisphere, these regions overlap with those involved in the for-

mation of syntactic constituent structures [26]. The human left

BA44 is specifically engaged in the representation of nested hi-

erarchical structures, both with linguistic [28] and non-linguistic

stimuli [27, 29]. It responds not only to simple rules, as tested

here, but also to complex compound rules [27, 28, 39, 40].

Comparative fMRI studies of resting-state networks indicate

that IFG and pSTS are uniquely expanded [41, 42] and function-

ally interconnected in humans compared to monkeys [19].

Compared to macaques, humans exhibit a much stronger

coupling between auditory association areas and IFG, including

the specific pSTS/BA44 network identified here [43]. Anatomi-

cally, the human IFG is expanded and more asymmetrical, its

mini columns are more spaced [44], and it connects to the tem-

poral lobe via a prominent fiber tract, the arcuate fasciculus,

which is much smaller or absent in other non-human primates

[45]. All of these observations concur to suggest an important

1972 Current Biology 25, 1966–1974, August 3, 2015 ª2015 Elsevier

anatomical, functional, and, possibly, representational change

in those regions.

A methodological difficulty that confronts any comparative

brain-imaging research is that the human adult subjects always

receive considerably more education and training than the

monkeys. Our results may therefore indicate that education to

linguistic, mathematical, ormusical structures affects the encod-

ing of auditory sequences in IFG and that similar results would

have been obtained after equally extensive training in monkeys.

While this is a possibility, behavioral studies indicate that infants

already possess a capacity to quickly grasp algebraic [5] and

numerical patterns [46] in the first days of life, prior to education,

whereas apes still exhibit striking deficiencies in acquiring count-

ing symbols after months of training [47]. Infant imaging studies

demonstrate that an organized language network is already

detectable in 2-month-old infants [48], encompassing the hu-

man-specific left-hemisphere network identified here, which in

macaque monkeys is remarkably underdeveloped anatomically

and functionally [33, 41–43, 45]. Our comparative observations

are therefore most compatible with the hypothesis that evolution

endowed the human brain with novel inferior frontal cortical

and superior temporal circuits capable of representing regular

sequence patterns. At the very least, the present results indicate

that, given similar stimulus exposure, human adults achieve a

more integrated representation of auditory sequences thanmon-

keys and that this enhanced learning ability relates to IFG-pSTS

circuitry.

If our hypothesis is correct, while monkeys can represent

abstract properties such as ‘‘four sounds’’ or ‘‘one item is

different,’’ evolution granted humans with the specific ability to

quickly unify these representations into a single nested structure

such as ‘‘three identical items, then a different one.’’ The recent

evolution of such brain representations in the Homo lineage, if

confirmed, would provide a tantalizing explanation for the joint

emergence of human-specific linguistic, mathematical, and

musical skills in recent paleontological times.

EXPERIMENTAL PROCEDURES

Subjects

We tested three adult rhesus macaques (two females and one male, 4–6 kg,

8–9 years of age). All procedures were conducted in accordance with the

European Convention for Animal Care (86-406) and the NIH’s Guide for the

Care and Use of Laboratory Animals, and they were approved by the Institu-

tional Ethical Committee (CETEA 10-003). We also tested 19 healthy human

subjects with no known neurological or psychiatric pathology. Two human

subjects were excluded for being unable to maintain fixation on the white

cross on screen during the auditory sequence. Human subjects gave written

informed consent to participate in this study, which was approved by the local

Ethics Committee.

Auditory Paradigm and Stimuli

Auditory stimuli were presented usingMATLAB (MathWorks), via MR-compat-

ible headphones (MRConfon) with an intensity of 70 dB. The onset asynchrony

from one sequence to the next was fixed to 2,400 ms. The auditory sequences

in the paradigm included habituation (standard) and rare test (deviant) stimuli

(Figure 1). The habituation series comprised four identical tones (denoted

AAAA) or three identical tones followed by a different tone (denoted AAAB).

In each fMRI run, only one type of habituation stimulus (AAAA or AAAB) was

presented. Each run started with a first period of rest (14.4 s, 6 TRs, TR =

2.4 s), followed by 5 blocks of 30 sequences each (72 s, 30 TRs), separated

by a silent period (14.4 s, 6 TRs). The onset asynchrony from one sequence

Ltd All rights reserved

to the next was fixed to 2,400 ms. The total duration for one run was 446.4 s

(186 TRs). The first block of each run always contained 30 habituation tokens,

in order to establish the global regularity (AAAA or AAAB). In the following four

blocks, 24 tokens (80%) were of the habituation type and 6 (20%, never ap-

pearing as the first 6 tokens) were test stimuli deviating from the current rule

in one of four possible ways. The test stimuli followed a 23 2 factorial design,

containing two levels (same or different) for each of two factors (number or

sequence). Hence, there were four conditions in the test stimuli: N�S�:

non-number/non-sequence deviants (stimuli respecting the same rule as the

habituation trials and only deviating in frequency and timing); N+S�: number

deviants (two or six tones, e.g., AA or AAAAAA when habituation is AAAA);

N�S+: sequence deviants (e.g., AAAB when habituation is AAAA or vice

versa); and N+S+: number and sequence (double) deviants (e.g., AB or

AAAAAB when habituation is AAAA). Deviant stimuli were always followed

by at least one standard stimulus. In each fMRI run, each of four testing blocks

only contained one type of deviants. The order of the four testing blocks was

randomized.

To test for generalization, all test stimuliwerecomposedentirely of novel tones

not heard during habituation. The frequency of tones in the habituation stimuli

was selected from 500, 800, 1,280, and 2,048 Hz, while that in the test stimuli

was selected from 700, 1,120, and 1,792 Hz (indicated by different line colors

inFigure1).Toensure thatonly numberchangewas responsible for theobserved

response in the number deviant trials, we performed control over non-numerical

parametersaspreviously described [32, 46] (FigureS1). Total sequenceduration

(TSD) varied among350, 950 and 1,550ms in the habituation stimuli (always four

items) by controlling individual tone duration (ITD, among 25, 50, and 75ms) and

inter-stimulus-interval (ISI, among 50, 83.3, 216, 283.3, 416, and 450 ms). The

test stimuli were always a series of tones (two, four, or six items) of 50 ms ITD

and 250 ms ISI, which ensured that the TSD (varied among 350, 950 and

1,550 ms) and total tone duration (TTD, varied among 100, 200, and 300 ms)

were all familiar and presented equally often in the habituation and test stimulus

sets. Thus, only number mattered: ITD and ISI could not be used for discrimina-

tion because theywere equalized for two, four, and six item test sequences, and

TSD and TTD could not be used either because these parameters had all been

presented equally often during habituation.

To ensure the novelty of auditory stimuli, the subjects (both monkeys and

humans) received no training with the stimuli or the paradigm outside of scan-

ner. During the whole scanning period, the monkeys were trained to fixate on a

red dot (0.35� 3 0.35�) on a screen in front of the monkey’s head. The human

subjects were asked to fixate on the central cross on the screen while paying

attention to the auditory stimuli. This instruction was adopted because prior

research showed that novelty responses disappear under inattention [21].

In order to assess the awareness of regularities, the following list of ques-

tions was given to human subjects after the scanning session:

(1) Did you notice that most stimuli comprised four items?

(2) Did you notice infrequent stimuli with two or six items?

(3) Did you notice infrequent stimuli in the form of AAAB when the global

rule was AAAA?

(4) Did you notice infrequent stimuli in the form of AAAA when the global

rule was AAAB?

(5) Did you notice infrequent stimuli in the form of AB or AAAAAB when the

global rule was AAAA?

(6) Did you notice infrequent stimuli in the form of AA or AAAAAA when the

global rule was AAAB?

Once a subject completed the scanning sessions, she or he was asked to

comment on the auditory sounds. The 17 subjects scored 5.6/6 on average,

indicating nearly full awareness of the number and sequence regularities.

Data Acquisition and Analysis

See the Supplemental Experimental Procedures for details.

SUPPLEMENTAL INFORMATION

Supplemental Information includes Supplemental Experimental Procedures,

five figures, and four tables and can be found with this article online at

http://dx.doi.org/10.1016/j.cub.2015.06.035.

Current Biology 25, 1966

ACKNOWLEDGMENTS

The research leading to these results has received funding from the European

Commission Seventh Framework Programme (FP7/2007-2013) under grant

agreement 604102 (Human Brain Project), from the National Key Fundamental

Research (973) Program of China Grant 2013CB329501, from the program

‘‘Investissements d’Avenir’’ ANR-10-IAIHU-06, and it was supported by

INSERM, CEA, College de France, Universite Paris-Sud, Universite de Ver-

sailles Saint-Quentin en Yvelines, the French Ministry of Research (ANR-

1403-01 Construct), the Bettencourt-Schueller foundation, the Fondation

Roger de Spoelberch, the Fondation pour la Recherche Medicale, and the

Franco-Chinese Foundation for Science and its Applications (FFCSA). We

thank Morgan Dupont and the NeuroSpin facility for superior technical

assistance.

Received: March 2, 2015

Revised: June 6, 2015

Accepted: June 16, 2015

Published: July 23, 2015

REFERENCES

1. Hauser, M.D., Chomsky, N., and Fitch, W.T. (2002). The faculty of lan-

guage: what is it, who has it, and how did it evolve? Science 298, 1569–

1579.

2. Deacon, T.W. (1997). The Symbolic Species: The Co-Evolution of

Language and the Brain (New York: W. W. Norton & Co).

3. Dehaene, S., Spelke, E., Pinel, P., Stanescu, R., and Tsivkin, S. (1999).

Sources ofmathematical thinking: behavioral and brain-imaging evidence.

Science 284, 970–974.

4. Koelsch, S., and Siebel, W.A. (2005). Towards a neural basis of music

perception. Trends Cogn. Sci. 9, 578–584.

5. Marcus, G.F., Vijayan, S., Bandi Rao, S., and Vishton, P.M. (1999). Rule

learning by seven-month-old infants. Science 283, 77–80.

6. Gentner, T.Q., Fenn, K.M., Margoliash, D., and Nusbaum, H.C. (2006).

Recursive syntactic pattern learning by songbirds. Nature 440, 1204–

1207.

7. Abe, K., and Watanabe, D. (2011). Songbirds possess the spontaneous

ability to discriminate syntactic rules. Nat. Neurosci. 14, 1067–1074.

8. Wilson, B., Slater, H., Kikuchi, Y., Milne, A.E., Marslen-Wilson, W.D.,

Smith, K., and Petkov, C.I. (2013). Auditory artificial grammar learning in

macaque and marmoset monkeys. J. Neurosci. 33, 18825–18835.

9. Beckers, G.J., Bolhuis, J.J., Okanoya, K., and Berwick, R.C. (2012).

Birdsong neurolinguistics: songbird context-free grammar claim is prema-

ture. Neuroreport 23, 139–145.

10. Fitch, W.T., and Friederici, A.D. (2012). Artificial grammar learning meets

formal language theory: an overview. Philos. Trans. R. Soc. Lond. B Biol.

Sci. 367, 1933–1955.

11. ten Cate, C., and Okanoya, K. (2012). Revisiting the syntactic abilities of

non-human animals: natural vocalizations and artificial grammar learning.

Philos. Trans. R. Soc. Lond. B Biol. Sci. 367, 1984–1994.

12. van Heijningen, C.A., de Visser, J., Zuidema, W., and ten Cate, C. (2009).

Simple rules can explain discrimination of putative recursive syntactic

structures by a songbird species. Proc. Natl. Acad. Sci. USA 106,

20538–20543.

13. Jordan, K.E., Maclean, E.L., and Brannon, E.M. (2008). Monkeys match

and tally quantities across senses. Cognition 108, 617–625.

14. Wallis, J.D., Anderson, K.C., and Miller, E.K. (2001). Single neurons in pre-

frontal cortex encode abstract rules. Nature 411, 953–956.

15. Shima, K., Isoda, M., Mushiake, H., and Tanji, J. (2007). Categorization of

behavioural sequences in the prefrontal cortex. Nature 445, 315–318.

16. Eiselt, A.K., and Nieder, A. (2013). Representation of abstract quantitative

rules applied to spatial and numerical magnitudes in primate prefrontal

cortex. J. Neurosci. 33, 7526–7534.

–1974, August 3, 2015 ª2015 Elsevier Ltd All rights reserved 1973

17. Viswanathan, P., and Nieder, A. (2013). Neuronal correlates of a visual

‘‘sense of number’’ in primate parietal and prefrontal cortices. Proc.

Natl. Acad. Sci. USA 110, 11187–11192.

18. Gil-da-Costa, R., Stoner, G.R., Fung, R., and Albright, T.D. (2013).

Nonhuman primate model of schizophrenia using a noninvasive EEG

method. Proc. Natl. Acad. Sci. USA 110, 15425–15430.

19. Mantini, D., Corbetta, M., Romani, G.L., Orban, G.A., and Vanduffel, W.

(2013). Evolutionarily novel functional networks in the human brain?

J. Neurosci. 33, 3259–3275.

20. Uhrig, L., Dehaene, S., and Jarraya, B. (2014). A hierarchy of responses to

auditory regularities in the macaque brain. J. Neurosci. 34, 1127–1132.

21. Bekinschtein, T.A., Dehaene, S., Rohaut, B., Tadel, F., Cohen, L., and

Naccache, L. (2009). Neural signature of the conscious processing of audi-

tory regularities. Proc. Natl. Acad. Sci. USA 106, 1672–1677.

22. Nieder, A. (2012). Supramodal numerosity selectivity of neurons in primate

prefrontal and posterior parietal cortices. Proc. Natl. Acad. Sci. USA 109,

11860–11865.

23. Nieder, A., Diester, I., and Tudusciuc, O. (2006). Temporal and spatial

enumeration processes in the primate parietal cortex. Science 313,

1431–1435.

24. Bizley, J.K., and Cohen, Y.E. (2013). The what, where and how of auditory-

object perception. Nat. Rev. Neurosci. 14, 693–707.

25. Graybiel, A.M. (1998). The basal ganglia and chunking of action reper-

toires. Neurobiol. Learn. Mem. 70, 119–136.

26. Pallier, C., Devauchelle, A.D., and Dehaene, S. (2011). Cortical represen-

tation of the constituent structure of sentences. Proc. Natl. Acad. Sci.

USA 108, 2522–2527.

27. Bahlmann, J., Schubotz, R.I., and Friederici, A.D. (2008). Hierarchical arti-

ficial grammar processing engages Broca’s area. Neuroimage 42,

525–534.

28. Makuuchi, M., Bahlmann, J., Anwander, A., and Friederici, A.D. (2009).

Segregating the core computational faculty of human language fromwork-

ing memory. Proc. Natl. Acad. Sci. USA 106, 8362–8367.

29. Koechlin, E., and Jubault, T. (2006). Broca’s area and the hierarchical

organization of human behavior. Neuron 50, 963–974.

30. Pinel, P., Thirion, B., Meriaux, S., Jobert, A., Serres, J., Le Bihan, D.,

Poline, J.B., and Dehaene, S. (2007). Fast reproducible identification

and large-scale databasing of individual functional cognitive networks.

BMC Neurosci. 8, 91.

31. Petrides, M., and Pandya, D.N. (2002). Comparative cytoarchitectonic

analysis of the human and the macaque ventrolateral prefrontal cortex

and corticocortical connection patterns in the monkey. Eur. J. Neurosci.

16, 291–310.

32. Piazza,M., Izard, V., Pinel, P., Le Bihan, D., and Dehaene, S. (2004). Tuning

curves for approximate numerosity in the human intraparietal sulcus.

Neuron 44, 547–555.

33. Saffran, J., Hauser, M., Seibel, R., Kapfhamer, J., Tsao, F., and Cushman,

F. (2008). Grammatical pattern learning by human infants and cotton-top

tamarin monkeys. Cognition 107, 479–500.

1974 Current Biology 25, 1966–1974, August 3, 2015 ª2015 Elsevier

34. Leaver, A.M., Van Lare, J., Zielinski, B., Halpern, A.R., and Rauschecker,

J.P. (2009). Brain activation during anticipation of sound sequences.

J. Neurosci. 29, 2477–2485.

35. Averbeck, B.B., and Romanski, L.M. (2006). Probabilistic encoding of vo-

calizations in macaque ventral lateral prefrontal cortex. J. Neurosci. 26,

11023–11033.

36. Gifford, G.W., 3rd, MacLean, K.A., Hauser, M.D., and Cohen, Y.E. (2005).

The neurophysiology of functionally meaningful categories: macaque

ventrolateral prefrontal cortex plays a critical role in spontaneous catego-

rization of species-specific vocalizations. J. Cogn. Neurosci. 17, 1471–

1482.

37. Fujii, N., and Graybiel, A.M. (2003). Representation of action sequence

boundaries by macaque prefrontal cortical neurons. Science 301, 1246–

1249.

38. Teichmann, M., Dupoux, E., Kouider, S., Brugieres, P., Boisse, M.F.,

Baudic, S., Cesaro, P., Peschanski, M., and Bachoud-Levi, A.C. (2005).

The role of the striatum in rule application: the model of Huntington’s

disease at early stage. Brain 128, 1155–1167.

39. Badre, D., Kayser, A.S., and D’Esposito, M. (2010). Frontal cortex and the

discovery of abstract action rules. Neuron 66, 315–326.

40. Reverberi, C., Gorgen, K., and Haynes, J.D. (2012). Compositionality

of rule representations in human prefrontal cortex. Cereb. Cortex 22,

1237–1246.

41. Chaplin, T.A., Yu, H.H., Soares, J.G., Gattass, R., and Rosa, M.G. (2013). A

conserved pattern of differential expansion of cortical areas in simian

primates. J. Neurosci. 33, 15120–15125.

42. Hill, J., Inder, T., Neil, J., Dierker, D., Harwell, J., and Van Essen, D. (2010).

Similar patterns of cortical expansion during human development and

evolution. Proc. Natl. Acad. Sci. USA 107, 13135–13140.

43. Neubert, F.X., Mars, R.B., Thomas, A.G., Sallet, J., and Rushworth, M.F.

(2014). Comparison of human ventral frontal cortex areas for cognitive

control and language with areas in monkey frontal cortex. Neuron 81,

700–713.

44. Schenker, N.M., Buxhoeveden, D.P., Blackmon, W.L., Amunts, K., Zilles,

K., and Semendeferi, K. (2008). A comparative quantitative analysis of

cytoarchitecture andminicolumnar organization in Broca’s area in humans

and great apes. J. Comp. Neurol. 510, 117–128.

45. Rilling, J.K., Glasser, M.F., Preuss, T.M., Ma, X., Zhao, T., Hu, X., and

Behrens, T.E. (2008). The evolution of the arcuate fasciculus revealed

with comparative DTI. Nat. Neurosci. 11, 426–428.

46. Izard, V., Dehaene-Lambertz, G., and Dehaene, S. (2008). Distinct cerebral

pathways for object identity and number in human infants. PLoS Biol. 6,

e11.

47. Matsuzawa, T. (2009). Symbolic representation of number in chimpan-

zees. Curr. Opin. Neurobiol. 19, 92–98.

48. Dehaene-Lambertz, G., Dehaene, S., and Hertz-Pannier, L. (2002).

Functional neuroimaging of speech perception in infants. Science 298,

2013–2015.

Ltd All rights reserved

Current Biology

Supplemental Information

Representation of Numerical and Sequential

Patterns in Macaque and Human Brains

Liping Wang, Lynn Uhrig, Bechir Jarraya, and Stanislas Dehaene

Supplemental Information

Supplemental Figures and Legends

Figure S1. (Supplements Figure 1)

Fig. S1. Controlling for non-numerical parameters Following the logic of our previous papers *S1-3+, several methodological precautions were taken to ensure that the subjects’ novelty responses were solely based on number. First, the habituation stimuli (top) varied in pitch (500, 800, 1280, or 2048 Hz), total sequence duration (TSD; 350, 950 or 1550 ms), and individual tone duration (ITD; 25, 50 or 75 ms). As a consequence of these choices, the stimuli also varied in stimulus onset asynchrony (SOA) and total sound energy (TSE). Thus, subjects were incited to focus on the constant parameter of number (always 4 items). Second, the test stimuli (bottom) varied in total number (2, 4 or 6 items) while controlling for other differences. All test stimuli used a novel set of pitch values (700, 1120, or 1792 Hz). The numerical deviants (2 or 6 items) and the numerical standard (4 items) were matched in individual tone duration (ITD) and tempo (equal SOA). Once averaging over the two deviants (2 and 6 items), as was done in the fMRI analysis, they also shared the same average total sequence duration (TSD) and total sound energy (TSE). We also ensured that all of these test parameter values appeared equally often during habituation and were therefore equally familiar. Thus, none of these parameters could explain the novelty response observed selectively for number deviants compared to the numerical standard sequence.

Figure S2. (Supplements Figure 2&3)

VIP in Monkey A (Number effect):

VIP in Monkey K (Number effect):

IPS in Monkey J (Number effect):

ACC in Monkey J (Number effect):

SMA and ACC in Monkey K (Number effect):

6VR and Caudate in Monkey J (Sequence effect):

6VR and Caudate in Monkey K (Sequence effect):

Fig. S2. Individual activation patterns in the three monkeys for the main effects of number change ((N+S+) + (N+S-) > (N-S-) + (N-S+) contrast) and of sequence change ((N+S+) + (N-S+) > (N-S-) + (N+S-) contrast) (t>2.6, p<0.005, uncorrected). Main findings are also shown with multiple brain images in individual monkeys. Overall, our findings were consistent across monkeys, with some individual differences. For the main effect of number, all three monkeys showed activations in intraparietal sulcus (IPS), and two out of three monkeys showed activations in the region of pSTS and ACC. However, the locations of pSTS and ACC varied among three monkeys, indicating inter-individual variability. Activation in SMA was observed in Monkey K. Although the monkeys performed the fixation task well (>85%), monkey J and K showed the activations in several sites in visual cortex. For the main effect of sequence, two out of three monkeys displayed consistent activations in 6VR and caudate. The activations in pSTS and aSTS were also observed in at least two monkeys, but with certain inter-individual variability in locations. Multiple slices of brain images were displayed in individual monkeys. The inter-slice-space was 2 mm. Brain activations (betas) are shown in the four test conditions (N+S+, N+S-, N-S+ and N-S-) at specific peaks (same format as figure 2B). The coordinates refer to the monkey Montreal Neurological Institute template *S4+. Error bars indicate one standard error. Abbreviations: 6VR, ventral lateral prefrontal cortex; pSTS, posterior

superior temporal sulcus; ACC, anterior cingulate cortex; IPS, intraparietal sulcus; SMA, supplementary motor area.

Figure S3. (Supplements Figure 2&3) Test-restest analysis of Number and Sequence Effect across the three monkeys Number effect:

Sequence effect:

Test-retest analysis of individual monkeys, separately for the Number and Sequence effect Monkey A - Number effect

Monkey K - Number effect

Monkey J - Number effect

Monkey A - Sequence effect

Monkey K - Sequence effect

Monkey J - Sequence effect

Cross-validation analysis

Fig. S3. Whole brain maps after test-retest analysis (p<0.005, uncorrected, threshold for intra-class correlation *ICC+>0.64) across monkeys and runs showing the number and sequence effect. The test-retest analysis was performed with the ICC-Toolbox (Intra-class correlation coefficient) (http://www.kcl.ac.uk/ioppn/depts/neuroimaging/research/imaginganalysis/Software/ICC-Toolbox.aspx) *S5+. The ICC map was defined by *S6+: 𝐼𝐶𝐶 = (𝐵𝑀𝑆 − 𝐸𝑀𝑆)/(𝐵𝑀𝑆 + (k-1)EMS). Basically, the method estimates the correlation of subject signal intensities between runs, modeled by a two-way ANOVA, with random subject effects and fixed run effects, where BMS is between-subjects mean square, EMS is error mean square, and k is the number of repeated runs. The results largely confirmed the group analysis data, and showed VIP (slices -22 to -18) and ACC (slices 10 to 14) activations for the number-change effect, and 6VR (slices 2 to 8) and caudate (slices 2 to 6) activations for the sequence-change effect. Although the activations in pSTS were weak, they could be found at slightly different locations in both effects (slices -16 and -12). Likewise, the test-retest results in individual monkeys, which estimated the correlation between even and odd runs within each subject using the ICC-Toolbox (p<0.005, uncorrected, threshold ICC=0.64), confirmed the activation results from individual monkeys. In general, monkey J showed a different activation pattern from other two monkeys for both effects. For number effect, all three monkeys showed IPS activations. Monkey K and J showed activations in ACC. Monkey K and J showed weak activation in SMA. For sequence effect, all three monkeys showed 6VR and caudate activations. For both effects, pSTS and aSTS activations showed a high degree of variability across monkeys.

Cross-validation analysis. Upper: Monkey brain maps after cross-validation (p<0.005, uncorrected) showing selective responses in IPS and ACC/SMA to number change and 6V and pSTS to sequence change. Crucially, there is no activation in the S+/S- contrast in voxels localized using the N+/N- contrast, or conversely in the N+/N- contrast in voxels localized using the S+/S- contrast, indicating distinct neural representations for those two types of changes in monkeys. Lower: Human brain maps after cross-validation (p<0.005, uncorrected) showing the selective regions in IPS, pSTS, SMA and IFG to number change, and pSTS, TP, putamen and IFG to sequence change. Joint activations are found in pSTS and IFG areas in both of the cross-condition generalization tests, i.e. the S+/S- contrast in voxels localized using the N+/N- contrast, and the N+/N- contrast in voxels localized using the S+/S- contrast, suggesting an integration of those two types of changes in humans. Method: The SPM_SS (Subject-specific analysis toolbox, http://www.nitrc.org/projects/spm_ss/) was used to perform voxel-based multi-block analyses by using either N+>N- or S+>S- contrast as functional localizers. In this leave-one-out cross-validation analysis, for each human subject, we loop over each fMRI run, keeping it separate for the rest of the data. These data are then used to identify the regions of interest responsive to either number-change (N+>N- contrast) or sequence-change (S+>S- contrast). Finally, we examine the responses to both N+>N- and S+>S- conditions in those ROIs in the left-out of data. Upper = ROIs defined by number-change contrast; Lower=ROIs defined by sequence-change contrast; p<0.005, uncorrected). The monkey data were analyzed in a similar way as the human data, except that the loop over subjects, which was used in humans, was replaced by a loop over each experimental day and monkey.

Figure S4. (Supplements Figure 4)

A

B

C

Fig. S4. A. Left: Ventral intraparietal (VIP) activations showing a main effect of number change in monkeys. Activation maps ((N+S+) + (N+S-) > (N-S-) + (N-S+) contrast) superimposed on coronal section (X=6, t >2.7, p < 0.005, corrected by FDR). Right: The beta values at the activation peaks of the bilateral VIP areas were averaged and plotted in the three test conditions: sequences with 2 tones, 4 tones and 6 tones. The activations elicited by 2 tones and by 6 tones were both significantly higher than those evoked by 4 tones (t-test, p<0.05). This response pattern replicates previous observations in humans *S1+ and indicates a consistent response to numerical novelty for both numbers smaller (2) and larger (6) than the habituation value (4). There were no activations in VIP with other contrasts for conditions 2<4<6, nor for 6<4<2, suggesting that activation in this region was dominated by a numerical adaptation effect. Error bars indicate one standard error. B: Spatial relation of the number-change and sequence-change effects to the fMRI localizer for calculation and language in humans. (Left). Relation of the number change effect to the fMRI activation during mental calculation. The bilateral intraparietal, SMA and left IFG activations to auditory sequences deviating in number (number effect, red cluster) show a large degree of overlap with the activation during mental calculation (blue cluster), particularly in the left IFG (overlap is represented in pink). (Right). Relation of the sequence change effect to the fMRI activation during sentences processing. The activations in the posterior temporal lobe and temporal pole (TP) showing a sequence change effect (pink cluster) overlap with the language-related activations in temporal pole and middle temporal regions during sentence reading and sentence listening (yellow cluster; overlap shown in light pink). Overlap is quite reduced in the left IFG, because most of the activation to the sequence effect falls just posterior to the group-level language activation. Images are thresholded at voxel p<0.001 (uncorrected). Conventions are the same as in Fig. 2 and 3. C. Subject-specific search for overlapping regions showing a main effect of number change and of calculation in humans. Within each subject, we searched bilateral IPS areas (ROIs from ref. *S7+) for voxels responsive during mental calculation (p<0.001, PFDR <0.05 corrected). Then the

D

average fMRI activations from the main sequence paradigm within these voxels were submitted to a 2 by 2 ANOVA with factors of number change and sequence change. Conventions are the same as in Fig. 2 and 3. A significant main effect of number change is indicated by asterisks in the bilateral IPS areas (ANOVA, Left: F (1,16)=6.12, p<0.05; Right: F (1,16)= 4.70, p<0.05). D. Effect of sequence change in the basal ganglia in monkeys and humans. Activation maps ( (N+S+) + (N-S+) > (N-S-) + (N+S-) contrast) are superimposed on transverse sections (left) and coronal sections (right) in monkey (upper panels) (t >2.7, p < 0.005, corrected by FDR) and human (lower panels) (t>3.0, p < 0.001, corrected by FDR). See Table S2 for the coordinates.

Figure S5. (Supplements Figure 5)

Figure S5. Comparison of IFG activations evoked by sequence change (C & D) with activations in human studies of the hierarchical organization of motor actions (A, from ref. *S8+) and the structural complexity of human language (B, from ref. *S9+). In panels C and D, activations showing the sequence effect are superimposed on the maximum probability maps of cytoarchitectonic areas using the SPM toolbox *S10+. A, orange and yellow indicate voxels activated in transitions between simple motor chunks. Blue lines indicate the boundaries of Brodmann’s area 44. B, orange and yellow indicate voxels showing a main effect of the structural complexity of sentences. L: left; R: right.

Tables

Table S1

Regions showing brain activation in monkeys and humans with numerical deviants (N+S- >

N-S-)

Monkeys

x y z t Value Brain Regions

-7 -23 10 4.60 Left parietal

-7 -24 17 4.65

-10 -27 8 4.82

-11 -12 10 4.43

9 -25 11 3.67 Right parietal

3 -23 11 4.12

8 -12 10 4.32

-9 10 11 4.35 Left dorsal premotor

13 -3 14 3.87 Right dorsal premotor

-1 3 18 3.21 Supplementary motor

-14 -2 -1 4.55 Left anterior insular

-16 -11 -3 4.21 Left posterior superior temporal

18 -11 -4 3.63 Right posterior superior temporal

-2 9 12 4.43 Anterior cingulate

Significant peaks at a voxel level of p<0.005 corrected by FDR.

Humans

x y z t Value Brain Regions

-45 -43 40 4.30 Left parietal

-30 -60 46 4.72

-27 -58 43 4.99

42 -43 49 4.52 Right parietal

45 -34 49 5.06

42 -55 55 3.78

-38 11 28 4.38 Left inferior frontal

-47 29 19 3.52

48 11 22 6.64 Right inferior frontal

51 11 19 4.79

42 35 16 4.19

39 38 16 6.70 Right middle frontal

-24 41 28 3.81 Left middle frontal

0 14 52 7.57 Supplementary motor

-30 23 -2 6.99 Left anterior insular

33 26 -2 5.58 Right anterior insular

-63 -37 4 5.54 Left posterior superior temporal

54 -37 4 5.76 Right posterior superior temporal

-51 -7 19 4.32 Left postcentral

63 -7 28 4.92 Right postcentral

48 2 -17 3.65 Right temporal pole

Significant peaks at a voxel level of p<0.001 corrected by FDR.

Table S2

Regions showing brain activation in monkeys and humans with sequential deviants (N-S+ >

N-S-)

Monkeys

x Y z t Value Brain Regions

-16 -4 7 3.74 Left ventral lateral prefrontal

15 4 7 3.95 Right ventral lateral prefrontal

-2 5 4 4.45 Left Caudate

-12 -4 4 4.71 Left anterior insular

9 -5 8 3.26 Right anterior insular

-16 -12 -4 3.94 Left posterior superior temporal

-16 -6 -9 3.84 Left anterior superior temporal

11 -11 -2 4.28 Right posterior superior temporal

-9 -11 9 4.38 Right precentral

Significant peaks at a voxel level of p<0.005 corrected by FDR.

Humans

x Y z t Value Brain Regions

-42 9 25 3.94 Left inferior frontal

-42 2 23 4.07

60 14 25 5.17 Right inferior frontal

-23 7 14

3.65 Left Putamen

21 9 7 3.62 Right Putamen

-63 31 1 4.11 Left posterior superior temporal

51 -40 7 4.10 Right posterior superior temporal

-45 5 -14 3.65 Left temporal pole

51 11 -17 3.55 Right temporal pole

63 -16 37 4.54 Right postcentral

-51 -1 49 5.15 Left precentral

48 -19 22 3.63 Right primary auditory

Significant peaks at a voxel level of p<0.001 corrected by FDR.

Table S3

Regions showing joint activation of number and sequence effects in humans

x y z t Value Brain Regions

-45 8 22 4.09 Left inferior frontal gyrus

-60 8 22 4.67

-39 23 25 2.85

60 11 22 4.26 Right inferior frontal gyrus

39 2 22 3.48

36 23 1 2.73 Right Insular

-51 -46 4 3.28 Left posterior superior temporal

51 -43 7 2.90 Right posterior superior temporal

-51 -1 49 4.08 Left precentral

57 -19 49 2.79 Right postcentral

57 11 -14 4.96 Right temporal pole

Significant peaks at a voxel level of p<0.01, conjunction null, corrected by FDR.

Table S4

p-values from ANOVA test of main effects in the 7 main ROIs*S11+.

Region Number Effect Sequence Effect

corrected uncorrected corrected uncorrected

TP ns ns 0.002 0.0003

aSTS ns ns ns ns

pSTS ns 0.02 ns 0.01

TPJ ns ns ns ns

IFGorb 0.04 0.006 0.09 0.01

IFGtri <10-3 <10-4 ns 0.06

BA44 <10-3 <10-4 <10-4 <10-5

Supplemental Experimental Procedures

Data Acquisition

Monkey: Functional images were acquired in a 3T scanner (Siemens, Tim Trio), using

a T2*-weighted gradient echo-planar imaging (EPI) sequence (TR= 2.4 s, TE = 20 ms, 1.5 mm3

voxel size) with a single transmit-receiver surface coil. The monkeys were trained to sit in a

sphinx position in a primate chair with their head fixed. MION (monocrystalline iron oxide

nanoparticle, Sinerem or Feraheme, AMAG Pharmaceuticals, MA) contrast agent (10 mg/kg,

i.v.) was injected to monkeys before scanning*S12+. Eye movements were monitored and

recorded by an eye tracking system (ISCAN ETL-200, ISCAN Inc. MA, USA). A total of 282 runs

(52452 volumes, 30 sessions) were collected from 3 monkeys: monkey A: 97 runs (18042

volumes, 11 sessions); monkey J: 112 runs (20832 volumes, 10 sessions), and monkey K: 73

runs (13578 volumes, 9 sessions).

Human: The same 3T scanner equipped with a twelve-channel coil was used to

collect human functional imaging data, with T2*EPI sequence (TR = 2.4 s, TE = 30 ms, Matrix

= 64 x 64, 3 mm3 voxel size). Eye movements were collected using an EyeLink 1000 eye-

tracker (SR Research, ON, Canada). A total of 120 runs from 17 human participants were

acquired and analyzed.

Monkey functional images were realigned, corrected for slice timing, spatially

normalized to the monkey MNI anatomical template*S4+ using Python and FSL

(http://www.fmrib.ox.ac.uk/fsl/). Images were resampled to 1 mm isotropic and smoothed

with an isotropic Gaussian kernel (3 mm full width at half maximum)*S13+.

Human functional volumes were corrected for slice timing differences, realigned to

correct for motion correction, spatially normalized using the parameter obtained from the

normalization of the anatomy.

After imaging preprocessing, active brain regions were identified by performing

voxel-wise GLM analyses implemented in SPM8 (http://www.fil.ion.ucl.ac.uk/spm) in both

monkeys and humans. Only runs with a fixation rate higher than 85% were included in the

database. In a first SPM model, the nine predictors included: *1-3+ the onsets of the

habituation stimuli with TSD 350, 950 and 1550 ms, and *4-9+ the onsets of the test stimuli:

N-S-, N+S- (two items), N+S- (six items), N-S+, N+S+ (two items) and N+S+ (six items). These

nine events were modeled as delta functions convolved with the canonical hemodynamic

response function for MION and its temporal and dispersion derivatives *S14+. Parameters

of head motion derived from realignment were also included in the model as covariates of

no interest.

Two second-level SPM group analyses were performed, separately for monkeys

(pooling over the three monkeys) and humans (pooling over the 17 subjects). These models

were full factorial two-way ANOVA model with two main factors: Number change and

Sequence change. Each model included one image from each of the relevant conditions (N-

S-, N+S-, N-S+ and N+S+) and each run. These images were obtained, for each run, from the

first-level SPM model, using the following contrasts: N-S- : predictor 4 > mean of 1+2+3,

N+S- : mean of predictor 5+6 > mean of 1+2+3, N-S+ : predictor 7 > mean of 1+2+3; and

N+S+ : mean of predictor 8+9 > mean of 1+2+3.

The number deviant activation regions were identified as the group analysis map of

the comparison of BOLD signals between the N+S- and N-S- conditions (Fig. 2A). The

sequence deviant activation regions were acquired as the comparison of BOLD signals

between the N-S+ and N-S- conditions (Fig. 3A). The main effect of number was identified

with the contrast of (N+S+) + (N+S-) > (N-S-) + (N-S+). The main effect of sequence was

defined with the contrast of (N+S+) + (N-S+) > (N-S-) + (N+S-).

The coordinates (referring to the atlas of*S4+) of the activation peaks and their t

values were included in Table S1 and S2.

A classical threshold of voxel p < 0.001, corrected for multiple comparisons using the

False Detection Rate correction (PFDR<0.05) was applied to human fMRI data. For monkeys,

owing to the smaller number of subjects and signal-to-noise level, a lower voxel-wise

threshold was applied (p < 0.005), yet still with PFDR<0.05 correction for multiple

comparisons.

The plots (Fig. 2B & 3B) at the activation peaks were generated by extracting the β-

weight of SPM regressions of individual sessions’ data with the hemodynamic functions of

the four conditions, and then plotting the mean and SE of these β-weights. The anatomical

registration of the peaks (Fig. 2 & 3, Table S1&S2) was confirmed using Caret software *S15-

16+. The coordinates were obtained from the SPM software using the monkey Montreal

Neurological Institute template *S4+.

To examine the generalization of the number and sequence rules, conjunction maps

were generated (Fig. 4A). The conjunction analysis of number effect consisted in detecting

the voxels significantly activated in both contrast images (N+S+ > N-S+) and (N+S- > N-S-).

Similarly, the conjunction analysis for the sequence effect consisted in identifying the voxels

appearing in both contrast images (N+S+ > N+S-) and (N-S+ > N-S-). Owing to the fact that

these analyses bore on two simultaneous subtests (conjunction null), the statistical

threshold was lowered to p<0.01, again with false detection rate correction PFDR<0.05.

The conjunction analysis of both number and sequence effects was also performed

to detect voxels significantly activated in both contrast images *(N+S+) + (N+S-) > (N-S-) + (N-

S+)+ and *(N+S+) + (N-S+) > (N-S-) + (N+S-)+).

Spatial relation of the number and sequence effects to the fMRI localizer in humans

Regions showing the effects of number and sequence were compared with those

involved in calculation and language (sentence reading), as localized by a previously

published event-related functional localizer (for details, see*S7+). Briefly, this 5-minute

functional magnetic resonance imaging (fMRI) captures brain responses during auditory and

visual perception, left- and right-hand motor/actions, reading, language comprehension and

mental calculation at an individual level. In this sequence, ten different types of stimuli were

presented with rest periods (black screen) serving as null events baseline. Functional images

for the localization were acquired in the 17 human subjects with a TR of 2.4s and a voxel size

of 3×3×3 mm. In the present study, we only used two contrasts: sentence processing

(relative to rest) in order to localize regions involved in language processing; and auditory

mental calculation (relative to the processing of non-calculation related sentences), which

functionally identified the bilateral number-related IPS areas. In both cases, the stimuli

were spoken and written sentences that varied in length from 6 to 8 words. Written words

were flashed successively in four successive screens (250 ms on, 100 ms interstimulus

interval). Subjects just had to passively read or hear sentences (e.g. “Rain made the road

dangerous”, and to respond when a probe sentence was displayed (e.g. “press three times

on the right button”), or to calculate when they read or heard a calculation problem (e.g.

“compute 3 plus 6”). The auditory and visual sentences activated bilateral superior and

middle temporal gyrus, left middle frontal and left inferior frontal gyrus. The contrast of

sentence processing was obtained by the conjunction analysis (p<0.01, uncorrected)

between the auditory and visual sentence contrasts.

As reported previously *S7+, calculation activated the bilateral intraparietal sulcus

(Fig. S4B, top left panel, denoted in blue, p<0.001, PFDR <0.05 corrected), SMA and IFG

regions (Fig. S4B, bottom left panel, denoted by blue). The calculation-related regions mainly

overlapped with the region showing the number effect (p<0.001, PFDR <0.05 corrected,

denoted by red) in our primary auditory sequence test. In addition, the activations in the

bilateral putamen and TP showing the sequence effect (Fig. S4B, right panel, denoted in

yellow pink cluster, p<0.001, PFDR <0.05 corrected) largely overlapped with the activations

during sentence reading (yellow cluster, p<0.001, PFDR <0.05 corrected). The activations

showing the sequence effect in left IFG (peak at *-60,8,22+) fell just posterior to those in the

functional localizer (peak at *-54,20,10+).

Analysis of language-related ROIs

We performed an analysis of individual region of interests (ROIs) using the subject-

specific analysis toolbox (SPM_SS)*S17+ for the 17 human subjects that had been tested with

the functional language localizer*S7+. For each subject, within each of seven language-

related ROIs reported by Pallier et.al. (2011)*S11+, we first used our functional localizer to

identify subject-specific voxels activated by reading sentences relative to rest (voxel p<0.01,

uncorrected). The SPM_SS then identified these voxels and extracted, within each ROI, the

subject-specific contrast values (four contrast images: N-S-, N+S-, N-S+ and N+S+), and these

values were further tested for the main effects of number and sequence (Fig. 4C). The

results appear in table S3, with or without Bonferoni correction for multiple comparisons

over 7 regions.

Overlap between anatomy structure and functional SPM images

The SPM Anatomy Toolbox of Eickhoff et al. *S10+, containing human probabilistic

cytoarchitectony maps, was used to identify the overlap our functional images and cyto-

architectonically defined anatomical regions. As explained in the main text, two main

activations were identified in human left and right inferior frontal gyri (IFG) areas showing

both number and sequence effects (conjunction analysis, Fig. 4A, Human, denoted by

yellow). The overlap analysis showed that the left IFG activation covered 20.2% of BA44

(54.4 voxels across 17 subjects), and the corresponding local maximum showed a probability

of 56% of lying in BA44, located at x=-60, y=8, z=22 (T=6.2). The other cluster of activations

in the right IFG covered 16.5% of BA44 (32.7 voxels across all subjects) and 3.3% of BA45

(11.4 voxels). The corresponding local maximum showed a probability of 54% of lying in

BA44 (at x=60, y=11, z=22, T=5.4) (Fig. 4B). Comparison with another published map of

receptor density profiles*S18+ tentatively suggests a localization within bilateral Brodmann’s

area 44d.

Supplemental References

S1. Piazza, M., Izard, V., Pinel, P., Le Bihan, D., and Dehaene, S. (2004). Tuning curves for

approximate numerosity in the human intraparietal sulcus. Neuron 44, 547-555.

S2. Hauser, M.D., Dehaene, S., Dehaene-Lambertz, G., and Patalano, A.L. (2002).

Spontaneous number discrimination of multi-format auditory stimuli in cotton-top

tamarins (Saguinus oedipus). Cognition 86, B23-32.

S3. Dehaene, S., Izard, V., Piazza, M. (2005). Control over non-numerical parameters in

numerosity experiments. Unpbulished Manuscript (available on

http://www.unicog.org).

S4. Frey, S., Pandya, D.N., Chakravarty, M.M., Bailey, L., Petrides, M., and Collins, D.L.

(2011). An MRI based average macaque monkey stereotaxic atlas and space (MNI

monkey space). Neuroimage 55, 1435-1442.

S5. Caceres, A., Hall, D.L., Zelaya, F.O., Williams, S.C., and Mehta, M.A. (2009). Measuring

fMRI reliability with the intra-class correlation coefficient. Neuroimage 45, 758-768.

S6. Shrout, P.E., and Fleiss, J.L. (1979). Intraclass correlations: uses in assessing rater

reliability. Psychological bulletin 86, 420-428.

S7. Pinel, P., Thirion, B., Meriaux, S., Jobert, A., Serres, J., Le Bihan, D., Poline, J.B., and

Dehaene, S. (2007). Fast reproducible identification and large-scale databasing of

individual functional cognitive networks. BMC Neurosci 8, 91.

S8. Koechlin, E., and Jubault, T. (2006). Broca's area and the hierarchical organization of

human behavior. Neuron 50, 963-974.

S9. Makuuchi, M., Bahlmann, J., Anwander, A., and Friederici, A.D. (2009). Segregating

the core computational faculty of human language from working memory. Proc Natl

Acad Sci U S A 106, 8362-8367.

S10. Eickhoff, S.B., Stephan, K.E., Mohlberg, H., Grefkes, C., Fink, G.R., Amunts, K., and

Zilles, K. (2005). A new SPM toolbox for combining probabilistic cytoarchitectonic

maps and functional imaging data. Neuroimage 25, 1325-1335.

S11. Pallier, C., Devauchelle, A.D., and Dehaene, S. (2011). Cortical representation of the

constituent structure of sentences. Proc Natl Acad Sci U S A 108, 2522-2527.

S12. Vanduffel, W., Fize, D., Mandeville, J.B., Nelissen, K., Van Hecke, P., Rosen, B.R.,

Tootell, R.B., and Orban, G.A. (2001). Visual motion processing investigated using

contrast agent-enhanced fMRI in awake behaving monkeys. Neuron 32, 565-577.

S13. Uhrig, L., Dehaene, S., and Jarraya, B. (2014). A hierarchy of responses to auditory

regularities in the macaque brain. J Neurosci 34, 1127-1132.

S14. Leite, F.P., Tsao, D., Vanduffel, W., Fize, D., Sasaki, Y., Wald, L.L., Dale, A.M., Kwong,

K.K., Orban, G.A., Rosen, B.R., et al. (2002). Repeated fMRI using iron oxide contrast

agent in awake, behaving macaques at 3 Tesla. Neuroimage 16, 283-294.

S15. Peeters, R., Simone, L., Nelissen, K., Fabbri-Destro, M., Vanduffel, W., Rizzolatti, G.,

and Orban, G.A. (2009). The representation of tool use in humans and monkeys:

common and uniquely human features. J Neurosci 29, 11523-11539.

S16. Nelissen, K., Borra, E., Gerbella, M., Rozzi, S., Luppino, G., Vanduffel, W., Rizzolatti, G.,

and Orban, G.A. (2011). Action observation circuits in the macaque monkey cortex. J

Neurosci 31, 3743-3756.

S17. Fedorenko, E., Hsieh, P.J., Nieto-Castanon, A., Whitfield-Gabrieli, S., and Kanwisher,

N. (2010). New method for fMRI investigations of language: defining ROIs

functionally in individual subjects. J Neurophysiol 104, 1177-1194.

S18. Amunts, K., Lenzen, M., Friederici, A.D., Schleicher, A., Morosan, P., Palomero-

Gallagher, N., and Zilles, K. (2010). Broca's region: novel organizational principles and

multiple receptor mapping. PLoS biology 8.