Broad Integration of Expression Maps and Co-Expression Networks Compassing Novel Gene Functions in the Brain

Transcription

1 Supplementary Information Broad Integration of Expression Maps and Co-Expression Networks Compassing Novel Gene Functions in the Brain Yuko Okamura-Oho a, b, *, Kazuro Shimokawa c, Masaomi Nishimura b, Satoko Takemoto b, Akira Sato d, Teiichi Furuichi d, Hideo Yokota b, a Brain Research Network (BReNt) Sakurayama, Zushi-shi, Kanagawa, , Japan b Image Processing Research Team, Extreme Photonics Research Group, RIKEN Center for Advanced Photonics 2-1 Hirosawa Wako-shi Saitama, , Japan c Department of Health Record Informatics, Tohoku Medical Megabank Organization, Tohoku University 4-1 Seiryo-chou Aoba-ku Sendai-shi Miyagi, , Japan d Department of Applied Biological Science, Faculty of Science and Technology, Tokyo University of Science 2641 Yamazaki, Noda-shi, Chiba, , Japan * : Correspondence for Yuko Okamura-Oho, M.D., Ph.D. Brain Research Network (BReNt) Sakurayama, Zushi-shi, Kanagawa, , JAPAN Phone : yoho@brent-research.org website: 1

2 List of Supplementary Materials Video S1: Introduction to the integrated analysis of gene expression maps and co-expression networks using a ViBrism-DB dataset Table S1: Biological interactions and GO terms rich in co-expression hub genes (2 spreadsheets) Table S2: GO terms accumulated in SETs of genes (21 spreadsheets) Table S3: List of groups of genes composing co-expression networks Table S4: probe information and QC of RNA and microarray 2

3 Figure S1. Transcriptome Tomography: a technique for high-throughput 3D reconstruction of gene expression maps (a sample map of Slitrk6 in the present dataset) As described in the previous paper 10, at least three genetically identical specimens, such as whole mouse brains of C57BL6/J, are sectioned in each of three orthogonal planes to produce fractions. In the study, first, a total of 28 1-mm (5 m 200 sections per batch)-thick fractions (13 in coronal, 9 in sagittal and 6 in horizontal sectioning series) were produced using three mm 3 -sized mouse brains, as shown in pseudo colors in the left panel. The gene expression densities in the fractions (fraction data) are measured with 36,558 probes on a microarray. Second, three more brains were processed as biological replicates. 61 fractions in total were subjected to the analysis (see Note). In the process of material sectioning, a block-face image is obtained before each 5- m section is cut. The block face images indicate the outlines of the fractions of the brains in 3D space, as shown in the left panel. The outlines of the whole brains are used for registration to align the brains in the same coordinates. The measured expression densities are assigned to the voxels of the registered fractions. Six values of the densities in each voxel obtained from 6 series of the fractions are averaged (3 series are shown in the left panel). The average values represent the expression density of the voxel, as shown in images of pseudo-colored expression densities of the fractions and the 3D expression maps, with or without the 80% cut-off filter. 3

4 Note: The sectioning process cannot produce brain fractions treated as exact biological replicates. Therefore, slightly different 28 biological replicate pairs are defined between fractions of the two replicate series using r of expression densities in the fractions as a similarity measure 10 ; that is, 5 fractions in the second series are not used for the ANOVA. 4

5 number of target probes having k (Log 10 ) number of target probes having k (Log 10 ) number of target probes having k (Log 10 ) r>0.85 y= -1.16x+3.87 R: SE: y= -2.16x+3.60 R: SE: probe number in SET (k) (Log 10 ) r>0.75 y= -0.75x+3.28 R: SE: y= -1.56x+3.82 R: SE: probe number in SET (k) (Log 10 ) r>0.65 y= -0.55x+2.72 R: SE: y= -1.38x+4.07 R: SE: probe number in SET (k) (Log 10 ) 5

6 Figure S2. The distribution of probe number in SETs ( ) calculated in the computationally randomized and non-randomized datasets Using total probes, was calculated for each probe with a threshold of r > 0.85, 0.75 and 0.65 in the non-randomized dataset, which was the fraction data used in Figure 1A, and in a computationally randomized dataset as described in the Methods: then, the distributions of were plotted in the same way as in Figure 1A and shown in Log 10 scales. Closed circles and gray cross marks represent the results of non-randomized and randomized datasets, respectively. Dashed lines and dotted lines with formulas indicate the power law fit, respectively, with the correlation coefficient (R) and the standard error (SE): 95% confidence intervals are shown with chain lines (Log 10 < 2 for r > 0.85, otherwise Log 10 < 2.5). With a threshold of r > 0.85, the distributions in the two datasets are significantly distinguished throughout the range of Log 10 > 0 with a larger number of probes exhibiting in the non-randomized dataset (p < 0.05). We concluded that the SETs with r > 0.85 were non-randomly formed with statistical significances in the fraction data. Whereas with thresholds of r > 0.75 and 0.65, the distributions in the two datasets are not distinguished at the range of approximately Log 10 1 and 2, respectively; that is, a SET formed at these ranges of cannot be distinguished from a randomly formed probe combination with the r thresholds. Thus, we decided to perform SET analysis with a threshold of r > 0.85 in this study. Intriguingly, all the distributions of for the randomized datasets as well as the non-randomized ones follow a power law feature. However, the probes accompanied by Log 10 > 2 are not present in the randomized dataset with the threshold of r > Then we decided to set a threshold of for co-expression hub genes in this study. 6

7 Figure S3. Distribution of cells expressing Nkx2-1 in figures of the open resource, GENSAT, which corroborates our mapping results GENSAT employs a BAC 7

8 transgenic technique to express a fluorescent marker protein that is under the control of the promoter and regulatory elements of the gene of interest 26. This technique elicits the cells expressing the gene of interest. Selected figures of the GENSAT database showing distribution of Nkx2-1 expression cells in the adult mouse brain are displayed. The cells are present in areas caudal to the olfactory bulb (cob, cells indicated with arrows), in the ventral forebrain (vfb, cells with arrowheads) and regionally in the hypothalamus (Hy, regions with asterisks). Original images are seen in the database of The Gene Expression Nervous System Atlas (GENSAT) Project, NINDS Contracts N01NS02331 & HHSN C to The Rockefeller University (New York, NY): image ID=38514, and

9 Figure S4. Co-expression gene groups of Wnt3/Fzd1/Pax3 (A) and Wnt7a/Fzd7/Pax6 (B) displaying different expression patterns in ISH Figures selected from ISH datasets of ABA show distribution of gene expression in the cerebellum (wm: white matter, gl: granular layer, ml: molecular layer). The Purkinje cell layer is between the ml and the gl: and Purkinje cells (with arrows) and Bergmann glia cells (with closed arrowheads) are located in this layer. Granular cells are located in the gl (with open arrowheads). The cells are stained with probes of genes, symbols of which are given at the top left of the figure panels and representative cells are indicated with marks. Wnt Genes of the two gene groups are expressed in the Purkinje cells but Fzd1/Pax3 and Fzd7/Pax6 are present in different types of cells, the Bergmann glia cells and the Granular cells, respectively. This histological evidence supports the distinct local function of each gene group, which was assumed in the network analysis. 9

10 Figure S5. Co-expression gene groups of Lhx2/Foxg1/Nr2e1/Tbr1/Chrm1 and Lhx6/Foxo6/Fezf2/Tbr1/Chrna5 displaying similar but distinctive expression patterns in ISH Figures selected from ISH datasets of ABA corroborate distribution of gene expression shown in our analysis. Arrows and arrowheads indicate the upper layer 10

11 and the lower layer of the cerebral cortex, respectively. CPu: caudate putamen, vfb: ventral forebrain. Lhx2/Foxg1/Nr2e1/ Chrm1 are expressed similarly with high density in the upper layer of the cortex. Tbr1 functioning in regulating the layer identity in the cortex belongs to both groups and is expressed in the upper and lower layers. Fezf2/Chrna5, regulating GABAergic/glutamatergic transition, are expressed in the lower layer cortex and subcortical regions (CPu and vfb) with a few in the upper layer cortex. Lhx6 is present in known areas of GABAergic neurons such as the subcortical regions and the cortical interneurons. 11

12 Figure S6. Lhx9 expressing cells present in neuron groups of visual and auditory input system and of their integration system in the adult mouse brain Figures are obtained from the database of GENSAT (gene.id=184&mouseage.id=1). Lhx9 expressing cells are present in the neurons of auditory input system (dark blue), visual input system (black) and their integration system (dark green) including the lateral entorhinal cortex 52 and the paraventricular thalamic nucleus

13 Figure S7. Co-expression of Pax6/Wnt7a in the olfactory bulb areas (OB) and the cerebellum (CB) ISH images of the OB, the whole brain and the CB are shown in the left, center and right panels, respectively. Expression of Pax6 and Wnt7a genes, symbols of which are given at the top of the center panels, is shown at the postnatal stages of 7 days (P7) and 21 days (P21) as indicated. Arrows in the OB panels at the stages indicate gene expression in areas of the rostral migratory stream (rms) from the subventricular zone (svz) to the olfactory bulb (ob). In the CB panels at P7 both genes are present in the outer external granular layer (oegl) and the granular layer (gl), but rarely in the inner 13

14 external granular layer (iegl). At P21 they are present in the gl. The two streams from the svz to the ob along the rms in the OB and from the oegl to the gl across the iegl in the CB are architectures for persistent neurogenesis through to the adult stage and during three weeks after birth, respectively. The prominent expression of Pax6/Wnt7a seen in the streams suggests their co-function in the neurogenesis at the neonatal stage. In the adult stage both genes are expressed in the stream of the OB (see ABA images) but Wnt7a expression is restricted to the Purkinje cell layer in the CB (see Supplementary Fig. S4). This may be compatible to the persistent adult neurogenesis in the OB. Other abbreviations cc: corpus callosum, wm: white matter, ml: molecular layer. 14