The temporal dynamics of genome expression throughout the body and its genetic and epigenetic control are central to a synthetic understanding of how a relatively small number of DNA molecules can give rise to an entire human. Similarly, temporal expression patterns in neural tissue and their regulation across the lifespan will elucidate molecular mechanisms involved in the formation, mature function and degeneration of the human brain.
Previous studies have combined transcriptome and genetic analyses to investigate the genetic control of gene expression in human cell lines1-3
. Few studies have applied these genomic techniques to human neural tissue4-7
or human brain disease8
. Others have focused on the transcriptome in human fetal brain tissue9
, temporal patterns of gene expression in postnatal life10
, and gene co-expression patterns in the brain11,12
. Here we describe the combination of genome-wide DNA and RNA analyses in a large collection of meticulously curated human brain specimens to produce a comprehensive view of how the expression of the human genome in the prefrontal cortex (PFC) progresses from fetal development through ageing and how sequence variation in the genome impacts on these expression patterns.
The post-mortem brain tissue collection (n = 269 subjects without neuropathological or neuropsychiatric diagnosis) spans the majority of the human lifespan (). From each subject in the brain collection, RNA from PFC grey matter was analysed using spotted oligonucleotide microarrays yielding data from 30,176 gene expression probes. DNA from cerebellar tissue was studied with Illumina BeadChips producing 625,439 single nucleotide polymorphism (SNP) genotypes for each subject.
A global view of the PFC transcriptome
The absolute rate of expression change within each life stage was quantified for all genes using linear models (, box plot). The rate of expression change during fetal development is much faster than at any other stage in human life. Changes during infancy are much slower, yet still more rapid than at any later time in life. After the first half year of postnatal life, rates of expression change slow markedly, and continue to slow during the childhood and teenage years, subsequently maintaining a low rate of change through the 20s, 30s and 40s. After this period, rates of expression change begin to rise again through several decades, and in the aged human brain, change reaches and then exceeds rates observed during teenage years.
The distribution of expression trajectory turning points was investigated across postnatal life (, grey histogram). Rates of expression change decrease from childhood through the teenage years (blue boxes) as many genes redirect expression trajectories (peak in grey histogram near 20 years). In contrast, in ageing, expression change accelerates (yellow–orange boxes) as more genes enter turning points (the minor peak in the grey histogram near 60 years).
The correlation of expression measures across subjects was explored within each age stage and between adjacent stages (, points). Transcription in PFC appears most similar across individuals at the beginning of life and then again to a lesser extent nearer its end, demonstrating the most diversity during the years of mature brain functioning, when age-dependent rates of expression change are lowest (this observation is also clear in ). The separation of mean within- and between-age stage correlations observed early in life indicates the occurrence of fundamentally distinct transcriptional programs within fetal, infant and childhood development, followed by a smoother more continuous progression of change throughout the rest of the lifetime.
To obtain a global perspective on transcription in PFC across the human lifetime, expression profile correlations were combined with multidimensional scaling (MDS) to reduce the complexity of the expression data and produce an intuitive visualization of global patterns (). The spatial progression of the colour scale in this plot is a reflection of age-dependent change in human PFC transcription. Even within the brief 6-week range of fetal development examined, there is clearly observable systematic expression change with time (along the vertical axis). Following fetal development, the path of global transcriptional change alters markedly, progressing steadily away from the fetal state through the neonatal, infant and childhood ages, each of which has a relatively distinct identity compared with other periods (across the horizontal axis). A second redirection of global transcriptional change occurs at the end of the teenage years (also observed from a different perspective in , grey histogram), followed by a more linear progression through adulthood and into ageing. This global view was also used to inspect the effects of covariates (Supplementary Fig. 1
In another global view of prefrontal transcription, the age effect within the fetal samples is effectively illustrated using principal components analysis (PCA, ). The first principal component (PC1) separates the fetal from postnatal samples, whereas the second (PC2) appears to align with age effects within both the fetal and postnatal samples. The directions of the fetal and postnatal age effects along PC2 appear to be in opposition. Additionally, fetal expression changes are negatively correlated with those in other stages of early life: infancy r
= −0.45, P
= 1.3 × 10−90
; childhood r
= −0.48, P
= 1.5 × 10−47
; and teenage years r
= −0.18, P
= 2.3 × 10−8
(including only probes with slopes at P
< 0.05 in both stages, Supplementary Table 1
). This might indicate that select fetal expression changes are reversed at different times across the lifespan, beginning immediately after birth.
To investigate further this observation of reversing trajectories, genes showing significant expression change across age in both fetal and infant development were compared directly (). Approximately three-quarters of genes showing significant change in both stages reverse their direction of expression change between fetal and early postnatal life, with most changing from an increase in utero to a decrease in the months after birth.
Reversal of fetal expression changes in infancy and ageing
To gain functional insight into these changing expression patterns, the genes within each of the quadrants in were interrogated for the over-representation of functional gene groups. Detailed functional group lists for each of the quadrants are contained in Supplementary Table 3
. This examination of gene expression trajectories in early life may give a global genomic perspective on mechanisms in neural development that have been well studied at the individual cell and gene level: genes involved in cell division are over-represented among genes for which expression decreases during both fetal development and infancy. Inversely, genes related to the synapse are over-represented among genes showing expression increases during both stages. This pair of findings is a genomic reflection of the well-characterized decrease in cell proliferation with opponent increase in neuronal differentiation through both late fetal and early infant development.
In contrast to synaptic components, genes with axonal function are highly enriched among genes showing increasing expression during fetal development followed by decreases after birth. This coordinated reversal of expression trajectories among axonal genes while many synaptic genes continue to increase in infancy is probably a genomic view of the process of pruning exuberant axons while synapse development and maturation at appropriate target sites advance13
. Specific gene expression changes in synaptic and axonal genes during fetal and infant life are listed in Supplementary Table 4
Genes in ATP synthesis also show a reversal of expression patterns, but in this case, decreasing during fetal development and rising after birth. In fetal development, energy metabolism seems to be slowing along with the decrease in cellular proliferation, consistent with cell division as the primary energy consuming process during fetal development. However, after birth, proliferation in the PFC continues to slow while expression of energy metabolism genes increases markedly. Other functional gene groups with increasing expression during these first postnatal months include genes involved in Ca2+
transport, gated ion channels, voltage-gated K+
channels and active ion transport (Supplementary Table 3
), indicating that neuronal maturation and activity now drive energy production.
This functional analysis of expression trajectories also reveals potentially novel mechanisms in early cortical development: in the heavily populated quadrant showing increasing expression in the fetus and decreasing expression in infancy, 22 of the top 49 over-represented gene groups are microRNA (miRNA) target gene groups (Supplementary Table 3
= 6.5 × 10−5
and below). Together, these miRNA target groups account for 266 of the 673 genes in this quadrant (40%). miR-9 targets are the most highly enriched of these miRNA target gene groups. miR-9 is brain-specific14
and is used reiteratively in diverse processes in neural development, including patterning, neurogenesis and differentiation15,16
, as well as cell migration17
The reversal of fetal expression trajectories is also seen much later in life. Fetal expression trajectories show a strong negative correlation with changes observed in the sixth decade of life (50s) (r
= −0.46, P
= 2.4 × 10−21
; Supplementary Table 1
). This finding is consistent with the age-dependent repression of neuronal genes observed previously18
. Whereas fetal expression trajectories show negative correlation with both infant and 50s trajectories, expression trajectories in infancy do not correlate with those observed in the 50s (Supplementary Table 1
). However, within the set of genes showing trajectory reversal between fetal and infant ages, expression change in infancy and in the 50s share a striking amount of similarity (). Therefore, although infant expression changes do not globally resemble those happening later in life, the specific reversal of fetal expression trajectories seen in infancy is mirrored within changes in ageing.
These fetal reversals in ageing can also be demonstrated by comparing our observations in fetal development with recent findings in ageing. Genes with significant increases during fetal development are enriched for genes shown to decrease in the ageing cortex19
, whereas genes decreasing during fetal development are enriched for genes known to increase in ageing (P
= 1.0 × 10−6
= 4.6 × 10−11
, respectively; see Supplementary Table 5
). Similar reversals are also seen in genes reported to change in Alzheimer’s disease20
: fetal increases are enriched for genes downregulated in Alzheimer’s disease and fetal decreases are enriched for genes upregulated in Alzheimer’s disease (P
= 2.2 × 10−21
= 7.1 × 10−7
, respectively; see Supplementary Table 5
). Hence, in the PFC, the reversal of specific expression patterns from in utero
development occurs in infancy and then again much later in normal ageing and in the neuropathological processes of Alzheimer’s disease.
To explore the genetic control of prefrontal expression patterns, DNA from the sample collection was interrogated with high-density SNP microarrays to catalogue common genomic polymorphisms. All possible associations of SNP genotypes with gene expression levels were examined (expression quantitative trait loci, or eQTL): n
= 30,176 expression probes × 625,439 SNP genotypes = 1.89 × 1010
(~19 billion) possible associations. Consistent with previous observations, we see that individual SNPs can profoundly affect the expression level of individual genes. When considering data across all subjects, 1,628 individual associations surpass genome-wide Bonferroni correction. Association analysis was also conducted within the African American and Caucasian samples separately (significant associations for all analyses are in Supplementary Table 6
The strength and location of associations relative to transcriptional start sites (TSS) are explored in . Consistent with past eQTL studies across many organisms, we find that effects proximal to TSSs are of greater average strength than associations across greater distances or across chromosomes (). There are considerably more strong associations downstream (3′) from the TSS than upstream. This is consistent with previous observations8
, and demonstrates that downstream polymorphisms (often within gene sequences) that impact on expression are stronger and/or more numerous than alterations at equal distances upstream (potentially in promoter or enhancer sequences). Additionally, expression-associated SNPs are biased towards positions within genes (fold enrichment = 1.61, P
= 2.9 × 10−76
). Within this gene bias, both exonic and intronic locations are over-represented, but to vastly different degrees (fold enrichment = 4.3 and 1.4, P
= 5.0 × 10−94
and 1.2 × 10−32
Genetic control of PFC gene expression
The single strongest association observed was between the expression of the ZSWIM7
gene and SNP rs1045599, located within this same gene (). This association of genotype with expression level is observed across all ages and races studied. Similar to this analysis, the freely available interactive stand-alone application that we have developed enables the visualization of expression data across the lifespan and the exploration of genetic associations for individual gene queries (http://www.libd/braincloud
). We invite the research community to explore this resource with their own interests.
To explore the relationship between the genome as a whole and the PFC transcriptome as a whole, we compared genetic distance and transcriptional distance in all possible pairwise subject comparisons (). Although individual SNPs clearly have an impact on the expression of individual genes ( and Supplementary Table 6
) globally, there is no association of genetic distance between individual humans with the similarity of their prefrontal transcriptional profiles (, R2
The genome produces a consistent molecular architecture in PFC
This dramatic lack of association between genetic distance and transcriptome distance across our sample is a surprising result that requires further interrogation. It is possible that no association is found in because most of the genetic polymorphisms measured do not impact on gene expression. Therefore, we repeated this search for association by investigating global transcriptional distance across a focused subset of the genetic data: only SNPs involved in genome-wide significant SNP–expression associations were considered. This analysis also revealed no association between focused genetic distance and global transcriptional distance (Supplementary Fig. 2
). In addition, these same analyses performed within individual races showed no association between global transcriptional distance and genetic distance when either global or focused genetic distance was used.
We conclude that despite the many genetic polymorphisms that individually can affect the expression of single genes, the human genome produces a consistent molecular architecture in the human prefrontal cortex across the lifespan. This is true across (the human) race. It is possible that individual genetic traits and complex combinations of traits that disrupt this architecture are selected against in the general population and would not appear in studies of normal human brain development. The clear observation of associations of individual genetic polymorphisms with gene expression () in the absence of a relationship between global genetic and transcriptome profiles () demonstrates our ability to analyse microscale genetic effects while macroscale interactions remain elusive. It is perhaps useful to consider each individual complete genome as a grand combination of variants which is acted upon (in evolution and in environment) and which acts (in development, biological function and disease) as a whole, rather than individual genetic traits in isolation. Characterization of the higher-order interactions within this whole is a great challenge facing biologists today.