Temporal dynamics and genetic control of transcription in the human prefrontal cortex.

A global view of the PFC transcriptomea, Histogram of subject ages in the brain collection. Birth is indicated by the dashed white line. This colour scale serves as the legend for all panels in Fig. 1 and Fig. 2b. b, Box plot of absolute rates of expression change within each stage of life. Because the rates of expression change are so high early in life, the y-axis scale is different for fetal and infant stages than for all other stages. The two horizontal dotted lines in the left panel show the entire extent of the y axis in the right panel. Only microarray probes showing systematic variation with age (R2 > 0.5) were included (n = 8,704 probes). Age ranges: fetal, 14–20 gestational weeks; infant, 0–6 months; child, 1–10 years; decades as labelled. Open points depict the mean expression correlation across subjects within each age stage (Pearson’s r calculated across all expression measures; y axis scale at far right). Filled points depict the mean expression correlation across subjects between adjacent age stages. The grey histogram displays the distribution of ages that marked a change in the trajectory of expression for genes across the postnatal lifespan (see Methods). c, A global view of dynamic PFC transcription across the lifespan, using MDS (distance = 1 − r, stress = 20.5) and expression correlation. Each mRNA sample is represented as a single point coloured by the age of the subject. Pairwise distances are derived from correlation across all expression measures between subjects, such that proximity indicates similarity, whereas distance indicates dissimilarity. d, PCA of transcription in PFC across the lifespan. PCA was performed on data from all gene expression probes to represent each mRNA sample as a single point coloured by the age of the subject.

Reversal of fetal expression changes in infancy and ageinga, Scatter plot of fetal and infant gene expression change with age. Each gene is represented as a single point. Only genes with slopes at P < 0.05 in both stages were included in this analysis (n = 1,502 genes measured by 1,819 probes: Supplementary Table 2). The number of genes in each quadrant is indicated in black. The P-values listed were derived from Pearson’s χ2 tests comparing the proportion of genes in each quadrant compared to an expected proportion of 0.25. Key functional gene groups are highlighted and listed in the quadrants where they are over-represented. b, Depiction of individual genes’ expression across the human lifespan, illustrating the four patterns of expression across fetal and infant development shown in a. The gene depicted in each panel is an example drawn from gene groups highlighted in the corresponding quadrants in a. Expression levels (normalized log2(sample/reference)) are on the y axis, with age on the x axis. The linear age scale is variable within each of the individual stages of life as labelled. Subjects are coloured by age as in Fig. 1a. GW, gestational week. Clockwise from top left: NADH dehydrogenase (ubiquinone) 1α subcomplex 7, 14.5 kDa (NDUFA7; ATP synthesis); γ-aminobutyric acid A receptor, α1 (GABRA1; synapse); stathmin-like 2 (STMN2; axon); and cyclin B2 (CCNB2; cell cycle). c, Visualization of ageing (50s) and infant expression trajectories among genes showing reversal between fetal development and infancy (same genes depicted in a, lower right and upper left panels, n = 1,105). The red line depicts a linear fit to the data.

Genetic control of PFC gene expressiona, Position and strength of associations of SNP genotype with gene expression levels (distance of the SNP from the associated TSS is plotted on the x axis). Only gene–SNP pairs lying on the same chromosome are displayed here. Negative log10(P value) is on the y axis. Only P values <0.0001 are included in this analysis. The genome-wide Bonferroni-corrected P = 0.05 is shown as a horizontal dashed grey line: P = 0.05/1.89 × 1010 associations = 2.6 × 10−12 (11.6 on this y-axis scale). The numbers of associations listed in the plot refer to those passing this genome-wide alpha level. Both solid blue curves depict a local nonlinear regression fit (loess, span = 0.5) of association strength across distance from the TSS. The blue fit lying close to the y axis is a fit plotted in the same scale as the plotted P values (y axis at left). The second blue fit is the same fit, plotted on an expanded y axis (y axis at right). The minimum value of this fit both upstream and downstream from the TSS (that is, approximately where the local estimate of enrichment for greater association reaches zero) is marked with a vertical blue dashed line. b, The most significant observed association of a single SNP with the expression of one gene across the lifespan (highlighted in a in green): the rs1045599 SNP lies within the ZSWIM7 gene on chromosome 17 (ZSWIM7, zinc finger, SWIM-type containing 7). Age scales are defined as in Fig. 2b. ZSWIM7 expression level (normalized log2(sample/reference)) is on the y axis and is coloured by rs1045599 genotype.

The genome produces a consistent molecular architecture in PFCGlobal comparison of genetic and transcriptional differences between subjects. Each point represents a comparison of two subjects in the collection. Genetic distance between subjects is depicted on the x axis as the number of differing alleles over the portion of the genome interrogated. Transcriptional distance is shown on the y axis as 1 minus the correlation across all gene expression values from the subjects (as used in Fig. 1c). Each subject comparison is coloured to indicate the races (AA, African American; Cauc., Caucasian) of the two individuals involved in the comparison. The thick black curve is an estimate of the local mean (loess, span = 0.25) of transcriptional distance as it varies across genetic distance. The thin black curves depict fits to the residuals around this mean. Only African American and Caucasian sample comparisons are visualized here (>96% of the collection).