Article Figures & Data
Figures
- Figure 1.
A flow chart of the neural genetic investigation analysis for language. A, The definition of language activation regions and language genetic clusters. The top-right panel displays the language activation map, which was derived from the HCP language fMRI task. In this task, subjects listened to stories adapted from Aesop's fables and answered questions related to them (as shown in the top-left panel). The language map was further subdivided into 40 regions using the Brainnetome Atlas (Fan et al., 2016). For each region, genetic modeling was performed, and the regions with a genetic factor (A or D) were identified as genetically influenced language regions. Then, 25 genetically influenced language regions were merged based on neural activation, resulting in six language genetic clusters (as shown in the bottom-right panel). Slice views and projected brain images were prepared in MRIcron (https://www.nitrc.org/projects/mricron) and BrainNet Viewer (Xia et al., 2013), respectively. B, Gene expression analyses of language clusters. The left panel displays the region × gene expression matrix, which was generated through multiple preprocessing steps and used for subsequent analyses. Two types of decoding were conducted: decoding between genetically influenced language regions and nonlanguage regions to test whether the language genetic clusters exhibited systematically different genetic profiles compared with the rest of the brain and decoding among different language genetic clusters to examine differences among language genetic clusters. The genes that contributed to these two classifications were identified accordingly. C, The investigation of genetic factors for multiple object domains. The genetic impact of each language cluster on various object domains, including tools, body parts, and faces, was evaluated using the HCP working memory task. In this task, subjects viewed pictures from four object domains (i.e., tools, faces, body parts, and places; Barch et al., 2013) over two runs (shown in the left panel). To avoid the influence of working memory and focus on our interests, tools, body parts, and faces, whole-brain contrast images of these three object domains versus places were calculated and measured at the individual level after model estimation. Multiple lines of analyses (shown in the right panel) were conducted to further test the genetic relationship between language and other object domains for each language genetic cluster. The first validation analysis tested whether each language cluster exhibited a genetic effect on specific object domains when the analyses were restricted to the “language genetic voxels.” The second validation analysis investigated whether there were common factors influenced by genetics between language and specific object domains with the existing genetic influence through the CP model.
- Figure 2.
Genetic influence found in multiple cortical and subcortical language regions. A, Multiple types of language regions. After genetic modeling, multiple types of language regions were defined. The left panel shows the genetically influenced language regions, which were identified by the presence of a significant genetic factor (A or D). Out of 25 language regions, 22 regions were best modeled by model DE, and 3 regions were best modeled by model AE (p < 0.05, compared with the control model E). The middle panel shows the environmentally influenced language regions, which were identified by having a significant common environmental factor (C) in the model CE. Six language regions showed significant common environment effects (p < 0.05, compared with the control model E). The right panel shows the other language regions, which did not have any significant genetic or environmental factors in the models. A, D, and E denote the additive genetic factor, dominance genetic factor, and unique environment factor, respectively. See language genetic influence result without a language mask in Extended Data Figure 2-1. B, Genetic influence found in the six language clusters. For each language cluster, the left top and bottom panels show the ROI and best model for each language cluster, respectively. A, D, and E denote the additive genetic factor, dominance genetic factor, and unique environment factor, respectively, and their values are indicated below each factor. The scatterplots show the ICC across the subject pairs in the MZ and DZ groups. Each dot represents a twin pair in each group.
- Figure 3.
Gene expression patterns across language genetic clusters. A, Common genetic expression profiles revealed by decoding between genetically influenced language regions and other nonlanguage regions. Bar graphs represent the average accuracy of the classification between genetically influenced language regions and nonlanguage regions over 10,000 iterations. Error bars indicate the standard error. The left panel displays the classification result based on the two hemispheres. The right panel shows the classification result based on the left hemisphere. The gray line represents the chance level. See functional labels based on Neurosynth for genes contributing to commonality of language clusters in Extended Data Table 3-1. B, Different genetic expression profiles revealed by decoding between different language genetic clusters. The red line represents the actual accuracy of the classification among regions of different language genetic clusters. The distribution represents the accuracy of classification after disrupting labels and repeating the process 10,000 times. See functional labels based on Neurosynth for genes contributing to differences of different language clusters in Extended Data Table 3-2.
- Figure 4.
Differential genetic influence on responses to tools and body parts in different language clusters. A, Genetic model results of different language clusters on responses to tools, body parts, and faces. For each language cluster of interest, the neural location is shown in the left panel. The best genetic models for face, tool, and body part processing are displayed from left to right. A, D, and E represent the additive genetic factor, dominance genetic factor, and unique environment factor, respectively, and their values are shown below each factor. The model with a significant genetic effect (AE model or DE model, p < 0.05, FDR-corrected, compared with the control model E) is framed with a red rectangle. See genetic model results for all the language genetic clusters after excluding specific object domain-responsive voxels in Extended Data Table 4-1. B, Genetic model results at the voxel level. The cluster-level results were validated at the voxel level. For each language cluster showing a genetic effect on any object domain, genetic model analyses were conducted for the object domain of interest (with a genetic effect at the cluster level) on the “language genetic voxels.” The “language genetic voxels” were voxels with a best model that included a genetic factor (A or D factor, compared with the E model, at a threshold of p < 0.05), and they are displayed in the left panel. The best model for the object domain of interest is shown. A, D, and E represent the additive genetic factor, dominance genetic factor, and unique environmental factor, respectively, and their values are indicated below each factor. The scatterplots illustrate the ICC across the twin pairs in the MZ and DZ groups. Each dot represents a twin pair in each group.
- Figure 5.
CP model results. For each language cluster of interest, the neural location is shown in the left panel. The activation indexes of the language (indicated by the headphone icon) and the object domain with a genetic effect are fitted into the CP model to determine whether the common factor has a genetic influence. The resulting model is shown in the right panel. A, D, and E denote the additive genetic factor, dominance genetic factor, and unique environment factor, respectively. The factor value is shown above it. The value between cognitive ability (i.e., language and object domains) and common factors represent the loading value of each cognitive capability on the common factor.
Tables
Region number Mask label MZ twin DZ twin rMZ-rDZ 2*rDZ-rMZ Initial model Best model ΔAIC p Cluster r p r p 1 SFG_L_7_1 0.28 0.001 0.11 0.194 0.17 −0.06 ADE DE 7.63 0.002 5 2 MFG_L_7_2 0.25 0.004 0.41 0.001 −0.16 – ACE CE 13.86 *** – 3 IFG_L_6_1 0.18 0.029 0.21 0.051 −0.03 – ACE CE 4.03 0.014 – 4 IFG_L_6_2 0.34 0.000 0.15 0.134 0.19 −0.04 ADE DE 12.90 *** 5 5 IFG_L_6_3 0.23 0.007 0.06 0.325 0.17 −0.11 ADE DE 4.43 0.011 5 6 IFG_L_6_4 0.27 0.002 0.10 0.227 0.17 −0.07 ADE DE 7.19 0.002 5 7 IFG_L_6_5 0.26 0.002 0.18 0.084 0.08 0.10 ACE E – – – 8 ORG_L_6_2 0.26 0.003 0.23 0.040 0.03 0.20 ACE CE 8.59 0.001 – 9 ORG_L_6_6 0.26 0.003 0.26 0.021 0.00 – ACE CE 9.60 *** – 10 PrG_L_6_1 0.09 0.163 0.04 0.370 0.05 −0.01 ADE E – – – 11 PrG_L_6_3 0.16 0.047 0.30 0.010 −0.14 – ACE CE 5.58 0.006 – 12 STG_L_6_1 0.21 0.013 0.17 0.099 0.04 0.13 ACE CE 5.15 0.008 – 13 STG_L_6_2 0.30 0.001 −0.03 0.586 0.33 −0.36 ADE DE 7.82 0.002 4 14 STG_R_6_2 0.26 0.002 −0.13 0.833 0.39 −0.52 ADE DE 4.68 0.01 4 15 STG_L_6_3 0.23 0.008 −0.09 0.752 0.32 −0.41 ADE DE 3.14 0.023 4 16 STG_R_6_3 0.32 0.000 −0.09 0.747 0.41 −0.50 ADE DE 9.39 *** 4 17 STG_L_6_4 0.27 0.002 0.06 0.335 0.21 −0.15 ADE DE 6.74 0.003 2 18 STG_R_6_4 0.30 0.001 −0.13 0.842 0.43 −0.56 ADE DE 7.04 0.003 4 19 STG_L_6_5 0.30 0.001 0.10 0.229 0.20 −0.10 ADE AE 8.64 0.001 1 20 STG_R_6_5 0.33 0.000 0.02 0.444 0.31 −0.29 ADE DE 11.10 *** 3 21 STG_L_6_6 0.32 0.000 0.14 0.145 0.18 −0.04 ADE AE 11.14 *** 2 22 STG_R_6_6 0.41 0.000 0.00 0.513 0.41 −0.41 ADE DE 17.21 *** 3 23 MTG_L_4_2 0.37 0.000 0.09 0.252 0.28 −0.19 ADE DE 15.37 *** 1 24 MTG_R_4_2 0.37 0.000 0.09 0.243 0.28 −0.19 ADE DE 15.82 *** 1 25 MTG_L_4_4 0.41 0.000 0.15 0.130 0.26 −0.11 ADE DE 19.96 *** 1 26 MTG_R_4_4 0.38 0.000 0.01 0.475 0.37 −0.36 ADE DE 14.64 *** 3 27 ITG_L_7_3 0.36 0.000 0.04 0.383 0.32 −0.28 ADE DE 13.22 *** 1 28 FuG_L_3_1 0.06 0.247 0.33 0.005 −0.27 – ACE AE 0.15 0.14 – 29 pSTS_L_2_1 0.38 0.000 −0.03 0.588 0.41 −0.44 ADE DE 14.72 *** 2 30 pSTS_R_2_1 0.31 0.000 −0.10 0.772 0.41 −0.51 ADE DE 7.49 0.002 3 31 pSTS_L_2_2 0.30 0.001 −0.13 0.832 0.43 −0.56 ADE DE 6.83 0.003 2 32 IPL_L_6_5 0.26 0.002 0.12 0.180 0.14 −0.02 ADE DE 6.95 0.003 5 33 IPL_L_6_6 0.15 0.058 −0.03 0.600 0.18 −0.21 ADE DE 0.11 0.15 – 34 PoG_L_4_2 0.14 0.062 0.05 0.343 0.09 −0.04 ADE DE 0.35 0.13 – 35 PoG_R_4_2 0.08 0.201 0.00 0.508 0.08 −0.08 ADE E – – – 36 Hipp_L_2_1 0.00 0.486 0.09 0.243 −0.09 – ACE E – – – 37 BG_L_6_1 0.04 0.341 −0.04 0.611 0.08 −0.12 ADE E – – – 38 BG_L_6_3 −0.08 0.806 −0.13 0.832 0.05 – ACE E – – – 39 BG_L_6_5 0.28 0.001 0.10 0.215 0.18 −0.08 ADE AE 8.10 0.001 6 40 BG_R_6_5 0.24 0.005 −0.14 0.847 0.38 −0.52 ADE DE 3.79 0.016 6 Note: *** denotes p < 0.001.
- Table 2.
Genetic model results for all the language genetic clusters at the cluster-overlapping level
Cluster Language Body Face Tool Best model Compared with model E Best model Compared with model E Best model Compared with model E Best model Compared with model E ΔAIC p ΔAIC p ΔAIC p ΔAIC p LRATL DE 19.38 <0.001 DE 2.00 0.044 E – – E – – LSTG DE 8.24 0.001 DE 1.79 0.044 E – – DE 1.59 0.045 RSTG DE 11.05 <0.001 E – – E – – DE 3.80 0.016 LRPAC DE 5.95 0.003 E – – E – – DE 0.60 0.069 LFPC AE 8.56 <0.001 E – – E – – E – – LRDC DE 6.46 0.002 DE 3.05 0.044 E – – DE 8.39 0.004 Note: ΔAIC denotes the degree to which the best model is better than the control model. The p values were FDR-corrected.
- Table 3.
Genetic model results for all the language genetic clusters at the voxel-overlapping level
Cluster Language Body Face Tool Best model Compared with model E Best model Compared with model E Best model Compared with model E Best model Compared with model E ΔAIC p ΔAIC p ΔAIC p ΔAIC p LRATL DE 34.53 <0.001 DE 2.03 0.046 E – – E – – LSTG DE 17.32 <0.001 DE 1.53 0.046 E – – DE 0.54 0.077 RSTG DE 16.54 <0.001 E – – E – – DE 4.34 0.019 LRPAC DE 13.24 <0.001 E – – E – – DE 0.45 0.077 LFPC DE 14.12 <0.001 E – – E – – E – – LRDC DE 22.88 <0.001 DE 4.36 0.028 E – – DE 2.91 0.040 Note: ΔAIC denotes the degree to which the best model is better than the control model. The p values were FDR-corrected.
Figure 2-1
Language genetic influence result without a language mask A. Multiple types of whole-brain regions. After genetic modeling without restricting the language activation map, multiple types of regions were identified. B. The clustering result of the genetic regions without a language activation mask. Download figure 2-1, DOCX file.
Table 3-1
Functional labels based on Neurosynth for genes contributing to commonality of language clusters Note: For each gene, the functional terms from Neurosynth represent the terms with the most similar meta-analysis whole-brain activation map to the gene’s whole-brain map. The correlation values indicate the correlation of the gene’s whole-brain expression with the term’s meta-analysis result. Download Table 3-1, DOC file.
Table 3-2
Functional labels based on Neurosynth for genes contributing to differences of different language clusters Note: For each gene, the functional terms from Neurosynth represent the terms with the most similar meta-analysis whole-brain activation map to the gene’s whole-brain map. The correlation values indicate the correlation of the gene’s whole-brain expression with the term’s meta-analysis result. Download Table 3-2, DOC file.
Table 4-1
Genetic model results for all the language genetic clusters after excluding specific object domain responsive voxels Note: For each language genetic cluster, the cognitive abilities with existing genetic effects (AE model or DE model, p values < 0.05, compared with the control model E, uncorrected) are marked with yellow. ΔAIC denotes the degree to which the best model is better than the control model. Download Table 4-1, DOC file.
In this issue
