Genes Expressed in Specific Areas of the Human Fetal Cerebral Cortex Display Distinct Patterns of Evolution

The developmental mechanisms through which the cerebral cortex increased in size and complexity during primate evolution are essentially unknown. To uncover genetic networks active in the developing cerebral cortex, we combined three-dimensional reconstruction of human fetal brains at midgestation and whole genome expression profiling. This novel approach enabled transcriptional characterization of neurons from accurately defined cortical regions containing presumptive Broca and Wernicke language areas, as well as surrounding associative areas. We identified hundreds of genes displaying differential expression between the two regions, but no significant difference in gene expression between left and right hemispheres. Validation by qRTPCR and in situ hybridization confirmed the robustness of our approach and revealed novel patterns of area- and layer-specific expression throughout the developing cortex. Genes differentially expressed between cortical areas were significantly associated with fast-evolving non-coding sequences harboring human-specific substitutions that could lead to divergence in their repertoires of transcription factor binding sites. Strikingly, while some of these sequences were accelerated in the human lineage only, many others were accelerated in chimpanzee and/or mouse lineages, indicating that genes important for cortical development may be particularly prone to changes in transcriptional regulation across mammals. Genes differentially expressed between cortical regions were also enriched for transcriptional targets of FoxP2, a key gene for the acquisition of language abilities in humans. Our findings point to a subset of genes with a unique combination of cortical areal expression and evolutionary patterns, suggesting that they play important roles in the transcriptional network underlying human-specific neural traits.

NA: "Not available". For genes lists: the gene is not in the considered sub-list. For qvalues: q-value could not be estimated. List of genes next to an HAR containing at least one TFBS with turnover at FDR<0.05; C: TFs for these TFBS with FDR<0.05 and their expression values; qvalue and fold-change (PFO vs. PT and 17GW vs. 19 GW) for these genes. In green genes with significant (q<0.01) PFO vs. PT differential expression, in red TFs related to TFBS with PFO vs. PT (q<0.01) Second table:

Supplementary
Same list, but only with genes or TF with PFO vs. PT q<0.01, Red TF related to TFBS which are themselves differentially expressed in PT vs. PFO at q<0.01; location of the considered gene and the HAR in which the TFBS are predicted. Some genes appear more than one time if the TFBS are in two different HARs.

Supplementary Table 4:
Genes differentially expressed in PFO vs. PT (q<0,01), that are nearby a HAR containing at least one transcription factor (TF) with a significant loss or gain of predicted binding sites in the human lineage (FDR<0.05). The corresponding TFs are in the 2 nd columm, TFs differentially expressed in PFO vs PT are in red. Genes differentially expressed (between PFO vs. PT) among the genes differentially or similarly regulated by human and chimpanzee FOXP2 (represented in Figure 7C).

3D reconstruction
Assembling the picture stack. The pictures of the frozen block face were manually aligned one at a time with Adobe Photoshop CS (Adobe Systems Inc.®). The picture of the first section was used as background onto which a picture of the adjacent section was superimposed, as a second layer. This "layer 2" was made semitransparent and aligned using the transform-rotate and transform-scale commands to correct misalignments due to camera instability or block removal for the night. This allowed a reliable overlaying of the consecutive pictures. When the alignment was optimal, "layer 2" was made opaque again and the same procedure was applied to the next picture, and so on for the whole set of pictures. Each aligned layer was saved afterwards in an individual jpeg file numbered according to its position along the zaxis. This series of jpeg images corresponds to the image stack needed to perform the 3D reconstruction of the brain. The first step of the model construction was to outline the outer limits of the brain using free hand drawing to get its general shape. As 3D-doctor allowed adding slices to an existing stack, it was possible to start drawing from a restricted number of pictures, add the next slices progressively and check while drawing the global quality of the alignment for any drifting occurring in the stack due to human alignment errors.
Transition from one section to the next was smooth. Only a very limited number of sections had to be realigned before the stack was judged optimal.
Once the brain boundaries were traced on all images, these boundaries were smoothened using the "boundary process-smooth boundary" command, reducing the distance between nodes to 10, to reduce computer processor time when generating the surface model. The available surface-rendering modes were tested; "complex rendering" gave the best results and was therefore used in all subsequent steps using the most detailed parameters (x=1, y=1, z=1). The generated surface model was smoothened three times using the "tools-smooth surface" command to reduce the "steps" generated by this mode of reconstruction and to give a more natural-looking appearance to the brain structures.
Selection of the target areas on the 17GW model. We wanted to compare the gene expression pattern of two cortical areas -Broca and Wernicke areas -which in the adult brain would be involved in speech to determine if some cortical asymmetry was already present in our foetal brains. Since most major cortical grooves start to develop during the third trimester, we decided to draw on our 17GW model a wide frontal area that in adult brain would encompass without doubts PFO and PT, encompassing Broca and Wernicke areas. These areas were drawn first on the left side of the model.
Once the choice of limits of these areas was firm, we reported them exactly on the right side of the 17GW brain.
In order to project exactly the same areas onto the right hemisphere, we generated two half models of the 17GW brain using the edit-boundary process-split object command. The left side of the brain including its chosen target areas was then flipped left to right in the model viewer using the tools-mirror model command and moved to merge as exactly as possible with the right side model. The target areas were the drawn on the right side model to match as closely as possible their left counterparts.
Transferring the target areas to the 19GW model. We first tried to match the 17GW model inside the 19GW model to transfer the target area for one model to the second, as we had done to transfer the left areas to the right hemisphere in the 17GW model.
We observed that the size and shape of the hemispheres was too different between these models to allow unambiguous areas transfer between them on such a visual-only basis. We therefore used the 3D doctor registration function to transform the shape of our 17GW model into the shape and size of our 19GW model, creating a 17GW p19 model, which then could truly be matched to the actual 19GW model.
The registration command applies only to the pictures of the stack of origin. It will stretch the voxels according to the demand imposed by the new reference points, modifies the angle of slicing and reslices to match the angle and number of slices of the stack of destination. This command does not apply to the hand drawn boundaries.
Yet we needed our chosen areas to go though the same transformation as the whole brain. To solve this issue, we exported for each image of our origin stack the pictures of the window showing the frozen block image and its hand-drawn boundaries, creating a new stack, identical to the first one but with the hand drawn boundaries integrated into the frozen block image. In Photoshop® the selected cortical area boundaries were then filled with colour so they would be easier to recognize after registration reslicing. Registration requires the user to define at least 4 "identical" reference points on the two models. We tried different reference points and various numbers of them. The best result was obtained using 6 reference points placed respectively at the tip of the frontal, temporal and occipital lobes of the two hemispheres.
Once registration was done, the boundaries of the brain and of the four selected cortical areas were drawn on the new 17GW p19 stack. These boundaries were then saved and imported to their corresponding frozen block picture in the 19WG stack.
Since transformation from the 17GW model to the 19GW model was not a perfect point to point morphing, some area boundaries had to be slightly rotated or shifted to follow more closely the actual curves of the 19GW brain. Once all areas boundaries were drawn on the 19GW model, the 17GW p19 model was inserted inside the 19GW model to check the area match.

Dissection of the 3D-selected cortical areas on the frozen sections
Pictures of the window showing the frozen block image corresponding to the areas to be actually dissected and its hand drawn boundaries were exported to Photoshop. The area boundaries were filled with bright printer friendly colour for easier visual identification. The images were flipped horizontally so they could be placed under their respective frozen section. The left side of the brain was marked on each picture to avoid left-right confusion during dissection. The images were printed in colour to the exact size of their frozen section counterpart. This was checked with the cresyl violet staining of adjacent sections. The colour pictures were put inside a small RNAse-out treated plastic cover and placed under their respective frozen section for dissection. The slides were dissected on a bed of carboice to avoid RNA degradation.