Conceived and designed the experiments: Y. Li Y. Liu CY TJ. Analyzed the data: Y. Li Y. Liu JL CY. Contributed reagents/materials/analysis tools: WQ KL. Wrote the paper: Y. Li Y. Liu CY TJ. Data collection: Y. Liu JL WQ KL CY.
The authors have declared that no competing interests exist.
Intuitively, higher intelligence might be assumed to correspond to more efficient information transfer in the brain, but no direct evidence has been reported from the perspective of brain networks. In this study, we performed extensive analyses to test the hypothesis that individual differences in intelligence are associated with brain structural organization, and in particular that higher scores on intelligence tests are related to greater global efficiency of the brain anatomical network. We constructed binary and weighted brain anatomical networks in each of 79 healthy young adults utilizing diffusion tensor tractography and calculated topological properties of the networks using a graph theoretical method. Based on their IQ test scores, all subjects were divided into general and high intelligence groups and significantly higher global efficiencies were found in the networks of the latter group. Moreover, we showed significant correlations between IQ scores and network properties across all subjects while controlling for age and gender. Specifically, higher intelligence scores corresponded to a shorter characteristic path length and a higher global efficiency of the networks, indicating a more efficient parallel information transfer in the brain. The results were consistently observed not only in the binary but also in the weighted networks, which together provide convergent evidence for our hypothesis. Our findings suggest that the efficiency of brain structural organization may be an important biological basis for intelligence.
Networks of interconnected brain regions coordinate brain activities. Information is processed in the grey matter (cortex and subcortical structures) and passed along the network via whitish, fatty-coated fiber bundles, the white matter. Using maps of these white matter tracks, we provided evidence that higher intelligence may result from more efficient information transfer. Specifically, we hypothesized that higher IQ derives from higher global efficiency of the brain anatomical network. Seventy-nine healthy young adults were divided into general and high IQ groups. We used diffusion tensor tractography, which maps brain white matter fibers, to construct anatomical brain networks for each subject and calculated the network properties using both binary and weighted networks. We consistently found that the high intelligence group's brain network was significantly more efficient than was the general intelligence group's. Moreover, IQ scores were significantly correlated with network properties, such as shorter path lengths and higher overall efficiency, indicating that the information transfer in the brain was more efficient. These converging evidences support the hypothesis that the efficiency of the organization of the brain structure may be an important biological basis for intelligence.
Researchers have long studied the biological basis for intelligence and have found increasing evidence relating high performance on intelligence quotient (IQ) tests to the coordination of multiple brain regions, utilizing both structural and functional brain imaging techniques
It is well accepted that the human brain, which can be viewed as a large, interacting and complex network with nontrivial topological properties
In the present study, we tested the hypothesis that individual intelligence is associated with the individual's brain structural organization. Specifically, higher intelligence test scores correspond to a higher global efficiency of the individual's brain anatomical network. We performed our study on 79 healthy young adults, basically using the DTT method proposed by Gong et al.
We successfully constructed binary and weighted anatomical networks for each of the 79 subjects in the form of symmetric connectivity matrixes using our method (see
Both the T1 template and the AAL template showed in the right column are in the MNI space, with image dimensions of 181 mm×217 mm×181 mm and voxel dimensions of 1 mm×1 mm×1 mm. Both the rT1 image and the transformed AAL template overlaid on it showed in the left column are in the DTI native space of one randomly selected individual, with image dimensions of 256 mm×256 mm×45 mm and voxel dimensions of 1 mm×1 mm×3 mm. The homologous brain regions in AAL template were coded in different colors because the areas in the left and right hemispheres were considered separately.
A 90×90 symmetric matrix in which the
(A), (B) and (C): A 3D presentation of the network in anatomical space, in which the green points correspond to the 90 AAL regions defined in
Please note that the fiber bundles showed here may be only parts of a specific major white matter tract, rather than the entire tract.
Region name | Abbreviation | Region name | Abbreviation |
Precentral | PreCG | Lingual | LING |
Frontal_Sup | SFG | Occipital_Sup | SOG |
Frontal_Sup_Orb | SFGorb | Occipital_Mid | MOG |
Frontal_Mid | MFG | Occipital_Inf | IOG |
Frontal_Mid_Orb | MFGorb | Fusiform | FG |
Frontal_Inf_Oper | IFGoper | Postcentral | PoCG |
Frontal_Inf_Tri | IFGtri | Parietal_Sup | SPG |
Frontal_Inf_Orb | IFGorb | Parietal_Inf | IPG |
Rolandic_Oper | ROL | SupraMarginal | SMG |
Supp_Motor_Area | SMA | Angular | ANG |
Olfactory | OLF | Precuneus | PCUN |
Frontal_Sup_Medial | SFGmed | Paracentral_Lobule | PCL |
Frontal_Mid_Orb | FGMedOrb | Caudate | CAU |
Rectus | RECT | Putamen | PUT |
Insula | INS | Pallidum | PAL |
Cingulum_Ant | ACC | Thalamus | THA |
Cingulum_Mid | MCG | Heschl | HES |
Cingulum_Post | PCC | Temporal_Sup | STG |
Hippocampus | HIP | Temporal_Pole_Sup | STGp |
ParaHippocampal | PHIP | Temporal_Mid | MTG |
Amygdala | AMYG | Temporal_Pole_Mid | MTGp |
Calcarine | CAL | Temporal_Inf | ITG |
Cuneus | CUN |
Abbreviations: AAL, Automated Anatomical Labeling.
Threshold value | SOBCC group mean (SD) | E group mean (SD) | Cp group mean (SD) | Lp group mean (SD) | γ group mean (SD) | λ group mean (SD) | E_glob group mean (SD) |
1 | 90 (0.16) | 1185 (±101) | 0.52 (±0.01) | 2.32 (±0.09) | 1.83 (±0.15) | 1.11 (±0.02) | 0.50 (±0.02) |
2 | 90 (0.35) | 921 (±85) | 0.50 (±0.02) | 2.60 (±0.11) | 1.82 (±0.12) | 1.13 (±0.02) | 0.45 (±0.02) |
4 | 89 (0.66) | 694 (±74) | 0.48 (±0.02) | 2.99 (±0.17) | 2.07 (±0.21) | 1.17 (±0.03) | 0.40 (±0.02) |
5 | 89 (0.75) | 625 (±68) | 0.47 (±0.02) | 3.15 (±0.19) | 2.13 (±0.23) | 1.18 (±0.04) | 0.39 (±0.02) |
Abbreviations: SOBCC, Size of Biggest Connected Component; SD, Standard deviation.
Notes: E, Cp, Lp denote the number of edges, average clustering coefficient and mean shortest path length of the network respectively. γ and λ denote the small-world properties of the network. E_glob denotes the absolute global efficiency of the network. Detailed definitions can be found in the
Network measures included the total number of edges
Brain networks | N | Cp | Lp | γ | λ | E_glob |
Anatomical network (Gong |
78 | 0.49 | 2.32 | 4.07 | 1.15 | Not reported |
Anatomical network (Iturria-Medina |
90 | Not reported | Not reported | 1.85 | 1.12 | Not reported |
Morphological network (He |
54 | 0.30 | 3.05 | 2.36 | 1.15 | Not reported |
Functional network (Achard |
90 | 0.53 | 2.49 | 2.37 | 1.09 | Not reported |
Functional network (Salvador |
45 | 0.25 | 2.82 | 2.08 | 1.09 | Not reported |
Notes: N, Cp, Lp denote the number of nodes, average clustering coefficient and mean shortest path length of the network respectively. γ and λ denote the small-world properties of the network. E_glob denotes the absolute global efficiency of the network. Detailed definitions can be found in the
As shown in
Threshold value | Topological properties | Value, group mean (SD) | |||
GI (n = 42) | HI (n = 37) | GI |
|||
1 | E | 1160.95 (95.51) | 1211.51 (101.19) | 0.025 | |
Cp | Binary | 0.52 (0.01) | 0.52 (0.01) | 0 .213 | |
Weighted | 0.60 (0.02) | 0.61 (0.01) | 0.063 | ||
Lp | Binary | 2.34(0.09) | 2.30(0.07) | 0.019 | |
Weighted | 0.16 (0.03) | 0.14 (0.02) | <0.001 * | ||
E_glob | Binary | 0.50(0.02) | 0.51(0.02) | 0.019 | |
Weighted | 10.28 (1.74) | 11.79 (2.20) | 0.001 * | ||
2 | E | 899.05 (78.61) | 946.11 (85.65) | 0.013 | |
Cp | Binary | 0.50 (0.02) | 0.50 (0.02) | 0 .217 | |
Weighted | 0.57 (0.02) | 0.57 (0.02) | 0.162 | ||
Lp | Binary | 2.63 (0.11) | 2.57 (0.10) | 0.015 | |
Weighted | 0.16 (0.03) | 0.14 (0.02) | <0.001 * | ||
E_glob | Binary | 0.45 (0.02) | 0.46 (0.02) | 0.012 | |
Weighted | 10.31 (1.74) | 11.83 (2.23) | 0.001 * | ||
4 | E | 670.14 (66.01) | 721.08 (74.22) | 0.002 * | |
Cp | Binary | 0.47 (0.02) | 0.49 (0.02) | 0.018 | |
Weighted | 0.52 (0.02) | 0.54 (0.02) | 0.006 * | ||
Lp | Binary | 3.04 (0.17) | 2.94 (0.15) | 0.006 * | |
Weighted | 0.15 (0.02) | 0.13 (0.02) | 0.001 * | ||
E_glob | Binary | 0.40 (0.02) | 0.41 (0.02) | 0.004 * | |
Weighted | 10.40 (1.73) | 11.91 (2.27) | 0.001 * | ||
5 | E | 601.86 (60.05) | 650.86 (68.36) | 0.001 * | |
Cp | Binary | 0.47 (0.02) | 0.47 (0.02) | 0.183 | |
Weighted | 0.51 (0.02) | 0.52 (0.02) | 0.034 | ||
Lp | Binary | 3.21 (0.19) | 3.01 (0.17) | 0.004 * | |
Weighted | 0.15 (0.02) | 0.13 (0.02) | <0.001 * | ||
E_glob | Binary | 0.38 (0.02) | 0.39 (0.02) | 0.003 * | |
Weighted | 10.44 (1.72) | 11.97 (2.30) | 0.001 * |
Significance was set at
Abbreviations: GI, General Intelligence; HI, High Intelligence.
Intelligence test scores included full scale IQ (FSIQ), performance IQ (PIQ) and verbal IQ (VIQ) (see
In the case of binary networks,
Threshold value | Topological properties | FSIQ | PIQ | VIQ | ||||
PCC | PCC | PCC | ||||||
1 | E | 0.173 | 0.132 | 0.159 | 0.167 | 0.167 | 0.146 | |
Cp | Binary | 0.040 | 0.730 | 0.006 | 0.960 | 0.068 | 0.559 | |
Weighted | 0.114 | 0.322 | 0.071 | 0.537 | 0.146 | 0.206 | ||
Lp | Binary | −0.192 | 0.094 | −0.179 | 0.120 | −0.184 | 0.108 | |
Weighted | −0.312 | 0.006 * | −0.289 | 0.011 * | −0.297 | 0.009 * | ||
E_glob | Binary | 0.189 | 0.099 | 0.176 | 0.125 | 0.182 | 0.114 | |
Weighted | 0.302 | 0.008 * | 0.273 | 0.016 * | 0.291 | 0.010 * | ||
2 | E | 0.187 | 0.104 | 0.186 | 0.106 | 0.166 | 0.149 | |
Cp | Binary | 0.123 | 0.286 | 0.110 | 0.341 | 0.121 | 0.294 | |
Weighted | 0.131 | 0.256 | 0.111 | 0.336 | 0.134 | 0.245 | ||
Lp | Binary | −0.198 | 0.085 | −0.206 | 0.073 | −0.172 | 0.135 | |
Weighted | −0.338 | 0.003 * | −0.325 | 0.004 * | −0.314 | 0.005 * | ||
E_glob | Binary | 0.200 | 0.081 | 0.206 | 0.073 | 0.176 | 0.125 | |
Weighted | 0.304 | 0.007 * | 0.277 | 0.015 * | 0.293 | 0.010 * | ||
4 | E | 0.254 | 0.026 * | 0.264 | 0.020 * | 0.217 | 0.058 | |
Cp | Binary | 0.175 | 0.127 | 0.133 | 0.250 | 0.197 | 0.085 | |
Weighted | 0.218 | 0.057 | 0.179 | 0.119 | 0.233 | 0.042 * | ||
Lp | Binary | −0.270 | 0.017 * | −0.285 | 0.012 * | −0.230 | 0.044 * | |
Weighted | −0.357 | 0.001 * | −0.330 | 0.003 * | −0.338 | 0.003 * | ||
E_glob | Binary | 0.272 | 0.017 * | 0.284 | 0.012 * | 0.233 | 0.041 * | |
Weighted | 0.306 | 0.007 * | 0.275 | 0.015 * | 0.297 | 0.009 * | ||
5 | E | 0.265 | 0.020 * | 0.255 | 0.025 * | 0.243 | 0.033 * | |
Cp | Binary | 0.017 | 0.884 | −0.015 | 0.898 | 0.040 | 0.727 | |
Weighted | 0.111 | 0.336 | 0.076 | 0.513 | 0.129 | 0.264 | ||
Lp | Binary | −0.275 | 0.015 * | −0.287 | 0.011 * | −0.237 | 0.038 * | |
Weighted | −0.365 | 0.001 * | −0.339 | 0.003 * | −0.347 | 0.002 * | ||
E_glob | Binary | 0.277 | 0.015 * | 0.279 | 0.014 * | 0.246 | 0.031 * | |
Weighted | 0.309 | 0.006 * | 0.277 | 0.015 * | 0.300 | 0.008 * |
Significance was set at
Abbreviations: FSIQ, Full Scale IQ; PIQ, Performance IQ; VIQ, Verbal IQ; PCC, Partial Correlation Coefficient.
To further localize the association with intellectual performance, the local efficiency (
Brain regions (Abbreviation) | FSIQ | PIQ | VIQ | |||
PCC | PCC | PCC | ||||
PoCG _L | 0.289 | 0.011 * | 0.242 | 0.034 * | 0.303 | 0.007 * |
STGp _R | −0.265 | 0.020 * | −0.255 | 0.025 * | −0.248 | 0.030 * |
MCG _R | 0.256 | 0.025 * | 0.190 | 0.098 | 0.286 | 0.012 * |
AMYG _R | 0.249 | 0.029 * | 0.262 | 0.021 * | 0.232 | 0.042 * |
MOG _L | 0.215 | 0.061 | 0.282 | 0.013 * | 0.130 | 0.260 |
MTGp _R | 0.211 | 0.066 | 0.290 | 0.010 * | 0.106 | 0.357 |
MFG _R | 0.180 | 0.116 | 0.249 | 0.029 * | 0.091 | 0.433 |
The threshold value was set at
Abbreviations: FSIQ, Full Scale IQ; PIQ, Performance IQ; VIQ, Verbal IQ; PCC, Partial Correlation Coefficient.
Brain regions (Abbreviation) | FSIQ | PIQ | VIQ | |||
PCC | PCC | PCC | ||||
PoCG _L | 0.349 | 0.002 * | 0.345 | 0.002 * | 0.313 | 0.006 * |
IFGoper _R | 0.333 | 0.003 * | 0.274 | 0.016 * | 0.343 | 0.002 * |
CUN _L | 0.303 | 0.007 * | 0.253 | 0.026 * | 0.310 | 0.006 * |
PCUN _R | 0.283 | 0.013 * | 0.266 | 0.019 * | 0.262 | 0.022 * |
PCC _R | 0.275 | 0.016 * | 0.245 | 0.032 * | 0.270 | 0.017 * |
IFGorb _L | 0.271 | 0.017 * | 0.194 | 0.092 | 0.324 | 0.004 * |
MOG _L | 0.258 | 0.023 * | 0.320 | 0.005 * | 0.168 | 0.143 |
SOG _R | 0.257 | 0.024 * | 0.272 | 0.017 * | 0.216 | 0.059 |
MOG _R | 0.255 | 0.025 * | 0.267 | 0.019 * | 0.208 | 0.070 |
PreCG _L | 0.252 | 0.027 * | 0.223 | 0.051 | 0.262 | 0.021 * |
MTG _L | 0.247 | 0.030 * | 0.212 | 0.064 | 0.251 | 0.028 * |
THA _R | 0.244 | 0.033 * | 0.168 | 0.145 | 0.278 | 0.014 |
PAL _R | 0.240 | 0.036 * | 0.216 | 0.059 | 0.251 | 0.028 * |
SFGorb _L | 0.240 | 0.036 * | 0.158 | 0.171 | 0.288 | 0.011 * |
MFGorb _L | 0.237 | 0.038 * | 0.174 | 0.129 | 0.266 | 0.019 * |
PAL _L | 0.234 | 0.040 * | 0.166 | 0.148 | 0.269 | 0.018 * |
CUN _R | 0.222 | 0.052 | 0.225 | 0.049 * | 0.179 | 0.119 |
RECT _L | 0.220 | 0.054 | 0.143 | 0.213 | 0.268 | 0.018 * |
HIP _R | 0.208 | 0.069 | 0.267 | 0.019 * | 0.137 | 0.236 |
The threshold value was set at
Abbreviations: FSIQ, Full Scale IQ; PIQ, Performance IQ; VIQ, Verbal IQ; PCC, Partial Correlation Coefficient.
We reproduced our investigations utilizing different brain parcellation schemes for network construction (see
In this study, we successfully constructed binary and weighted anatomical networks for individual brains from 79 healthy young adults using a DTT method. Network topological properties were analyzed and prominent small-world attributes were found. These findings are in accordance with the findings of previous human brain network studies that were done at a macro scale level
After its introduction by Watts and Strogatz
We identified hub regions, which have been identified by previous studies of functional or anatomical networks
In general, the topological properties of the brain anatomical network constructed in our current study are compatible with the findings of previous human brain network studies. However, some discrepancies exist between our results and previous findings, such as the exact values of the topological properties including the small-world indices. These discrepancies may be due to differences in data types and analytical methods.
In this study, global efficiency of the brain anatomical network was higher in the HI groups than in the GI groups, and positive correlations between intelligence tests scores and the global efficiency of the networks were found in all the healthy young adults while controlling for age and gender. These findings were consistently observed in the different situations we tested, including the binary and the weighted networks we constructed, the different brain parcellation schemes we employed (see
Many previous studies have related intelligence to different structural and functional properties of the brain. Positive correlations between IQ and total brain volume have been reported by several research teams who used structural imaging techniques on different populations with different scan protocols and different intelligence measures
In a recent study performed by our group
Subjects with higher IQ scores consistently showed more edges (
As reviewed by Roth and Dicke
In conclusion, we successfully constructed binary and weighted anatomical networks of the individual brains of 79 healthy adults. These networks showed topological properties that included a prominent small-world attribute that was quite comparable with the findings of previous human brain network studies. More importantly, extensive analysis consistently revealed significant correlations between intelligence test scores and brain anatomical network properties across all subjects, providing convergent evidence for our hypothesis that a more efficient brain structural organization may be an important biological basis for higher intelligence. Our study may provide new clues for understanding the mechanism of intelligence.
It should be noted that the healthy adults included in this current work have been used in previous studies performed by our group for different purposes
Seventy-nine normal subjects (44 males and 35 females, mean age = 23.8 years, range = 17–33 years) were recruited by advertisement. Each subject was examined using the Chinese Revised Wechsler Adult Intelligence Scale (WAIS-RC)
After a full explanation, all subjects gave voluntary written informed consent according to the standards set by the Ethical Committee of Xuanwu Hospital of Capital Medical University.
Diffusion tensor images of all the subjects were obtained on a 3.0-T Siemens MRI scanner. A single shot echo planar imaging sequence (TR = 6000 ms, TE = 87 ms) was employed. Diffusion sensitizing gradients were applied along 12 non-collinear directions (b = 1000 s/mm2), together with a non-diffusion-weighted acquisition (b = 0 s/mm2). An integrated parallel acquisition technique was used with an acceleration factor of 2, which can reduce the acquisition time with less image distortion from susceptibility artifacts. From each subject, 45 axial slices were collected. The field of view was 256 mm×256 mm; the acquisition matrix was 128×128 and zero filled into 256×256; the number of excitations was 3; and the slice thickness was 3 mm with no gap, which resulted in a voxel-dimension of 1 mm×1 mm×3 mm. A 3D T1-weighted image for each subject was obtained using a magnetization prepared rapid gradient echo sequence. The imaging parameters were a field of view of 220 mm×220 mm, TE of 2 s, TR of 2.6 ms, flip angle of 9°, and a voxel-dimension of 1 mm×1 mm×1 mm.
Both the DTI data and T1-weighted data were visually inspected by two radiologists for apparent artifacts arising from subject motion and instrument malfunction. Distortions in the diffusion tensor images caused by eddy currents and simple head motions were then corrected by FMRIB's Diffusion Toolbox (FSL 4.0;
First, we employed the AAL template
To provide more support for our current investigation, we also employed the parcellation scheme used by Gong et al.
Subsequently, DTT was performed on every subject. Seed points were selected as voxels with an FA value greater than 0.3 in each node region
Two AAL node regions
We further developed our investigation into weighted anatomical networks by assigning a weighted index to each entry of the binary network constructed in the previous section. In the human brain network study by Hagmann et al.
To investigate other possible weighted indices, we also employed the average FA value of all the reconstructed fiber bundles between two connecting regions and implemented the subsequent inspection of anatomical network properties and statistical analyses on the resulting networks as well. Details can be found in
A complex network can be represented as a graph in which nodes correspond to the elements of the system and arcs to the interactions between them
We used
We used
The subgraph
For the binary network, the absolute clustering coefficient of a node
The mean shortest absolute path length of a node is defined as:
The global efficiency of the network
The local efficiency of the
The concept of “small-world”, originally proposed by Watts and Strogatz
To further explore the configuration of the brain network, we examined the hub regions and degree distribution of the binary anatomical networks we constructed. Extended details can be found in
A two-sample
Partial correlations between intelligence test scores and global brain network properties (E,
There are several methodological issues in our present study that need to be addressed.
First, a deterministic tractography method was utilized for network construction. We realize that this kind of fiber tracking method has a limited capacity for resolving crossing fiber bundles
Second, in contrast to a population-based network analysis, which may tend to exclude false-negative connections
Third, we developed our investigations from a binary to a weighted anatomical network by introducing different weighted indices. Although no existing studies can directly validate our method, our results showed that either using the number or the average FA value of the existing fiber bundles between two regions can lead to a network topology similar to that found in previous human brain network studies (see
Finally, a risk of this study is that some of the fiber tracts reconstructed by our method may not belong to the specific AAL region. This could happen if the white matter voxels included in the fiber tracking procedure were not truly adjacent to the cortex. Additionally, the choice of the relatively high FA threshold of 0.3 for the seed voxel in our current study might increase this possibility, since it may exclude low FA sub-cortical white matter areas as seed regions. To address this issue Gong et al.
Investigation of different brain parcellation schemes
(0.12 MB DOC)
Investigation of different weighted indices
(0.06 MB DOC)
Investigations of hubs and degree distribution in the brain network
(0.12 MB DOC)
Significant partial correlation between E and IQ scores under the scheme with 78 nodes.
(0.11 MB TIF)
Significant partial correlation between Lp and IQ scores under the scheme with 78 nodes.
(0.40 MB TIF)
Significant partial correlation between E_glob and IQ scores under the scheme with 78 nodes.
(0.39 MB TIF)
Significant partial correlation between Lp and IQ scores using average FA as weighted index.
(0.11 MB TIF)
Significant partial correlation between E_glob and IQ scores using average FA as weighted index.
(0.11 MB TIF)
Degree distributions of the group-based network and the binary networks of three randomly selected subjects.
(0.12 MB TIF)
The authors thank Drs. Rhoda E. and Edmund F. Perozzi for English language editing. The authors thank Dr. Ming Song for constructive suggestions. The authors also thank Mr. Jiefeng Jiang for helping out with 3D presentation.