Coordinated Expression of Phosphoinositide Metabolic Genes during Development and Aging of Human Dorsolateral Prefrontal Cortex

Background Phosphoinositides, lipid-signaling molecules, participate in diverse brain processes within a wide metabolic cascade. Hypothesis Gene transcriptional networks coordinately regulate the phosphoinositide cascade during human brain Development and Aging. Methods We used the public BrainCloud database for human dorsolateral prefrontal cortex to examine age-related expression levels of 49 phosphoinositide metabolic genes during Development (0 to 20+ years) and Aging (21+ years). Results We identified three groups of partially overlapping genes in each of the two intervals, with similar intergroup correlations despite marked phenotypic differences between Aging and Development. In each interval, ITPKB, PLCD1, PIK3R3, ISYNA1, IMPA2, INPPL1, PI4KB, and AKT1 are in Group 1, PIK3CB, PTEN, PIK3CA, and IMPA1 in Group 2, and SACM1L, PI3KR4, INPP5A, SYNJ1, and PLCB1 in Group 3. Ten of the genes change expression nonlinearly during Development, suggesting involvement in rapidly changing neuronal, glial and myelination events. Correlated transcription for some gene pairs likely is facilitated by colocalization on the same chromosome band. Conclusions Stable coordinated gene transcriptional networks regulate brain phosphoinositide metabolic pathways during human Development and Aging.

The complexity of brain phosphoinositide metabolism limits our understanding the roles of phosphoinositides in Development and Aging and our ability to design therapeutic interventions in disease states [10,[13][14][15][16][17]. One way to address these limitations may be to analyze agerelated transcription of phosphoinositide genes in brain over the lifespan. During Development (0 to~20 years), the human brain undergoes marked nonlinear changes in synaptic and dendritic growth and pruning, neuronal loss, glial elaboration and myelination, in arachidonic and docosahexaenoic acid concentrations, and it shifts from ketone body to glucose consumption for ATP synthesis [18][19][20][21][22][23][24]. During later Aging (21+ years), brain function and metabolism are maintained in a more homeostatic range, although risk for neurodegeneration increases [25].
In the present study, we used BrainCloud to compare age-related expression in human dorsolateral prefrontal cortex of 49 genes involved in phosphoinositide synthesis, degradation, and signaling [1,2]. Based on our prior studies [29,30], we hypothesized that we could identify coordinated expression of these genes during the Development and Aging intervals. Such changes might correspond to changes in biochemical reactions involving the gene products and be facilitated by colocalization on a chromosomal band [29][30][31][32][33][34].

Methods
We selected 49 genes involved in phosphoinositide metabolism, based on canonical pathways reported in Ingenuity Pathway Analysis (IPA) (Ingenuity Systems, Redwood City, CA, http:// www.ingenuity.com) and other sources [1,2]. Expression data for each gene were exported from the BrainCloud database from 231 males and females ranging in age from birth to 78 years [26]. No subject had a history of significant psychiatric, neurological disorder, or drug abuse, or postmortem evidence of neuropathology.
As described in our prior studies, we separated the samples into Development (0 to 20.95 years, 87 subjects) and Aging (21 to 78.2 years, 144 subjects) intervals [29,30]. Gender and race breakdowns, as well as a description of the data in BrainCloud, have been reported earlier [29,30].
Twenty-two of the 49 genes chosen were detected by more than one probe in the Brain-Cloud database. When possible (18 of these 22 genes), the probe covering all possible alternate transcripts of the gene was chosen using the Gene View tab on BrainCloud. The probe that covered all transcripts also was the highest intensity probe for all but one gene (PIP5K1A), for which we used the probe covering all the transcripts. If one probe did not cover all possible transcripts, we chose the highest intensity probe (SLC5A3, PTGS2, and ITPK1).
Statistical tests were performed using Partek Genomics Suite (Version 6.6, Partek, St. Louis, MO, USA) and GraphPad Prism 5 (Version 5.02, GraphPad Software, La Jolla, CA, USA). First, a t-test was performed in Partek to determine whether mean expression levels differed significantly between the Aging and Development intervals for each gene. Pearson's r correlations were performed in GraphPad Prism 5 (Graph Pad Software, La Jolla, CA) for each gene, to determine correlation with age in each interval. Visual observation during Development suggested nonlinear expression for some genes. We therefore compared goodness of fit of the data with a nonlinear equation, Y = (Y0 -Plateau) Ã exp(-K Ã A) + Plateau (where Y = expression level at age A, and Y0 expression level at A = 0 years), to goodness of fit with a linear regression between 0 and 20 years [29,30].
Data also were analyzed with Cluster 3.0 software [35], without filtering or adjustment. As described earlier [29,30], we calculated distance between probes using the Euclidean distance calculation and clustering using the centroid linkage method expression [36]. The output.cdt file was loaded into TreeView program [37] to generate figures showing relatedness among genes of interest.
Similarity matrices using Pearson's r correlations and hierarchical clustering also were created using Partek, to make correlation (heat) maps showing correlation coefficients and gene clusters in expression between pairs of genes during Development and Aging [29,30]. Correlation data from the heat maps were used to construct corresponding statistical significance matrices. GraphPad Prism was used to calculate Pearson's r correlations for pairs of genes closely located on the same chromosome band.
Ethics Statement. This research was supported entirely by the Intramural Programs of the National Institute on Aging, the National Institute of Alcohol Abuse and Alcoholism, and the National Institute of Mental Health, National Institutes of Health. No author has a conflict of interest. Samples were collected under NIH protocol number NCT00001260, 900142, which include written informed consent from next-of-kin including consent for clinical records to be used. Every brain is consented.
The other second messenger released by PLC mediated hydrolysis is membrane-bound DAG, which binds to and activates protein kinase C (PKC) and other kinases that phosphorylate intracellular proteins [2]. DAG can be lost by phosphorylation by diacylglycerol kinase (DGKE) and recycled into phosphatidylinositol (PI), effectively terminating activation of PKC, or it can be hydrolyzed by diacylglycerol lipase (DAGLA) to arachidonic acid (AA). Arachidonic acid in turn can be oxidized by COX-2 (PTGS2) to produce pro-inflammatory metabolites like prostaglandin E 2 (PGE 2 ) (Fig 1, lower right) [29,48]. Arachidonic acid also may be released from PI by calcium-dependent cPLA 2 type IVA (PLA2G4A), then converted to PGE 2 by COX-2 (Fig 1, top) [29,31]. Table 1 lists the 49 genes in this analysis, their corresponding protein names, chromosomal locations, probes used, and results of t-tests comparing mean expression levels between Aging and Development. All fold changes are less than |2|, suggesting relatively stable expression throughout the lifespan. SYNJ2 has the highest fold increase (1.81, p = 10 −15 ) and SLC2A13 has the largest fold decrease (-1.67, p = 10 −9 ) in Aging compared with Development. Correcting for 49 comparisons, at p < 0.001, 14 genes are higher and of 6 genes are lower in mean expression during Aging than Development. Table 2 shows statistically significant correlations between gene expression levels and age in the two intervals. Correcting for 98 comparisons, at p < 0.0005 fewer age correlations are evident during Aging than Development. Only expression of ITPKB (r = 0.34) and of GRASP (r = -0.38) correlates significantly with age during the Aging interval. During Development, on the other hand, expression of PIP42K2A, SYNJ2, SACM1L, IMPA1, PLA2G4A, SYNJ1, PIK3CB, PIK3R4, INPP1, PRKCD, PIK3CD, PIK3C3, and ITPKB increase significantly, while expression of AKT1, PIK3R2, IPMK, CYTH3, PIP5K1C, and PIK3C2B decrease significantly.

Age correlations in expression of genes and gene pairs
Visual observation of expression levels during Development suggested nonlinear changes for some genes. We tested this by comparing nonlinear to linear goodness of fits for each of the 49 genes. For 10 of them, as illustrated in Fig 2, expression of ITBKB, PIP4K2A, SYNJ2, PRKCD, GRASP, PLA2G4A and PTGS2 increase non-linearly in the first years of life before reaching a plateau, whereas expression of PIK3C2B, CYTH3, and DGKE decline before reaching a plateau.
Colocalization on a chromosome may facilitate transcription of genes whose protein products participate in tightly connected metabolic pathways [29,32]. To consider this mechanism for our genes, we list in Table 3 statistically significant (p < 0.0001) correlations between genes on the same chromosomal band. On 1q25, expression of PLA2G4A (cPLA 2 Type IVA) correlates with expression of PTGS2 (COX-2) during both Development and Aging. Genes located on 3p21-23, PLCD1, SACMIL, IP6K1, and PRKCD, are significantly correlated with each other during Development and/or Aging. Significant correlations during both intervals occur between PIK3R4 and PIK3CB on 3q22 and between ITPK1 and AKT1 on 14q32. During Development DGKE and PRKCA are significantly correlated on 17q22. Cooperative clustered transcription correlations within extended groups TreeView dendrograms can identify genes whose transcription is coordinated or clustered in a hierarchical cascade, indicating relatedness and common cellular processes [29,30,49]. For example, the Development dendrogram ( Fig 3A) shows hierarchical interactions of IMPA2 and ISYNA1 (involved in myo-inositol synthesis); of PLCD1, IPMK, PIK3R3, and PI4K2A; and of SACMIL, PIK3CA, IMPA1, PIK3C3, and INPP1. PLA2G4A, PTGS2 GRASP, and SLC2A13 are distant (more dissimilar expression patterns) from the other genes. The Aging dendrogram (Fig 3B) shows that IMPA2 is closely tied to PLCD1 and that PLA2G4A also is distant from the other genes. Pearson's correlation matrices (heat maps) of pairwise correlations among the 49 genes were created using unsupervised hierarchical clustering within the Development ( Fig 4A) and Aging (Fig 4B) intervals. Hierarchical clustering row and column titles are not conserved between heat maps in the two intervals, as they represent the highest probability of correctly clustering genes based on Pearson's r correlation in each interval. In the figures, genes that are positively intercorrelated within a cluster are highlighted in red, while those that are negatively intercorrelated are shown in blue.
Three distinct clusters of genes with highly positively intercorrelated expression levels are identified in both the Development and Aging intervals (green outlined boxes on x-axis). During Development (Fig 4A)   The heat map for Aging (Fig 4B) also identifies three distinct groups of intercorrelated genes, with many similarities (bolded) to the respective Development groups in Fig

Discussion
Based on the literature, we constructed a phenotypic cascade containing known pathways of brain phosphoinositide metabolism and identified 49 genes whose protein products participate in this cascade (Fig 1). Among the 49 genes, we identified three groups in both Development and Aging and showed that the groups have similar intercorrelations and partially overlapping composition in the two intervals. In both intervals, ITPKB, PLCD1, PIK3R3, ISYNA1, IMPA2, INPPL1, PI4KB, and AKT1 are Group 1, PIK3CB, PTEN, PIK3CA, and IMPA1 are in Group 2, and SACM1L, PI3KR4, INPP5A, SYNJ1, and PLCB1 are in Group 3. Genes in Group 1 are negatively correlated with genes in Groups 2 and 3. Genes in Groups 2 and 3 are positively correlated across groups.
The similar gene groups and their interrelations in Development and Aging suggest that stable transcriptional networks underlie brain phosphoinositide metabolism throughout the entire lifespan, despite marked phenotypic differences between the two intervals [18][19][20][21][22][23][24]. Such stable networks likely arose through evolutionary constraints that maximized functional efficiency and minimized energy requirements of the metabolic pathways that were regulated [3,50,51]. A number of mechanisms may have contributed to their elaboration, such required coupling of reaction products and enzymes, enzyme colocalization at a cellular cite, organization of the nucleosome to bring genes and promoter regions together for co-transcription, and gene colocalization on common chromosomal bands [29,32,52].
We considered gene colocalization on the same chromosomal band in Table 3. Thus, expression levels correlate during Development and Aging for PLA2G4A and PTGS2 Age Transcription of Brain Phosphoinositide Enzymes colocalized on 1q25, consistent with evidence that arachidonic acid may have to be liberated from phospholipid by PLA 2 before it can be oxidized to PGE 2 by COX-2 [53,54]. Levels for PIK3R4 and PIK3CB on 3q22 and for ITPK1 and AKT1 on 14q32 also correlate significantly in the two age intervals. Levels for PLCD1, SACMIL, IP6K1, and for PRKCD on 3p21-23 correlate during Development and/or Aging, and levels for DGKE and PRKCA on 17q22 correlated during Development.
The dendrograms of Fig 4 identify certain gene expression hierarchies in the two intervals. For example, the Development dendrogram (Fig 4A) shows hierarchical relations among IMPA2, ISYNA1, PLCD1, IPMK, PIK3R3, and PI43K, which are distant from PLA2G4A and PTGS2 within the arachidonic acid cascade [29]. PLA2G4A also is separated from the phosphoinositide hierarchy in the Aging dendrogram (Fig 3B).
Our separating Aging and Development in this study is consistent with our prior studies using BrainCloud [29,30], and with phenotypic differences between the two intervals [18][19][20][21][22][23][24]. Of the 49 genes studied, 15 have a higher and 5 genes a lower mean expression level during Aging than Development (Table 1). Significant age correlations also are fewer during Aging than Development (Table 2), but lesser variation during Aging may have reduced our power to determine statistically significant correlations [55].
Nonlinear increases in ITBKB, PIP4K2A, SYNJ2, PRKCD, GRASP, PLA2G4A and PTGS2 expression and decreases in PIK3C2B, CYTH3, and DGKE expression during Development likely correspond to the many nonlinear phenotypic changes that have ben described, suggesting a role for phosphoinositides in them. There is rapid neuronal loss in the first year of life, dendritic growth followed by pruning over a 15 year period, declining myelination, changing arachidonic and docosahexaenoic acid concentrations throughout the entire interval, and conversion from ketone body to glucose use for oxidative metabolism in the first year [18][19][20][21][22][23][24].
During Development, the positive age correlations of SYNJ1 and SYNJ2 expression ( Table 2), whose protein products modify clathrin-mediated synaptic endocytosis [2,39], correspond to a reported increase in SYNJ immunoreactivity and dendritic spine density in brain [19,62]. SYNJ2 expression is 1.8 fold higher in Aging than Development, suggesting late stage synaptic changes [63]. Finally, SYNJ1, SYNJ2, and INPP1 are in Group 3 with PTEN in both intervals. It is reasonable that genes whose proteins degrade PI(3,4,5)P 3 should be positively correlated with PTEN, because if PTEN negatively regulates PI(3,4,5)P 3 signaling it could do so by interacting with other enzymes that degrade PI(3,4,5)P 3 .
The 1.67 fold decrement in SLC2A13 (HMIT) expression in Aging compared with Development does not correspond to a decreased brain myo-inositol concentration [64,65]. However, the negative age correlation in GRASP expression (Table 2) during Aging suggests glutamatergic alterations, since GRASP links Group 1 metabotropic glutamate receptors to neuronal proteins [66]. The highly positive correlation of ITPKB expression during Aging differs from a report that ITPKB mRNA was not increased with age in postmortem brain [67].
Class IA PI3K dimers evolved from a single enzyme in unicellular eukaryotes [68]. They consist of a p110 catalytic and a p85 regulatory subunit, each of which has three isoforms, PIK3CA, PIK3CB, and PIK3CD, and PIK3R1, PIK3R2 and PIK3R3, respectively [68]. Expression of PIK3R2 decreases while that of PIK3R4 increases during Development (Table 2), demonstrating the principle of divergent expression after gene duplication [69]. The regulatory subunit PIK3R3 is in Group 1 while the catalytic subunits PIK3CA and PIK3CB are in Group 3 in both the Development and Aging heat maps and the genes in the two groups are negatively correlated.
Although BrainCloud is a powerful tool for examining gene expression changes with age, it has limitations. It only contains data for the dorsolateral prefrontal cortex [26], but expression patterns differ between brain regions [70]. It also does not distinguish between cell types, which can also have distinct transcriptional trajectories [27,28,70]. However, BrainCloud does have a large number of samples, which increases its statistical power. The Allen Human Brain Atlas contains data only from 3 men, while the Loerch et al. study contains 28 brain samples [26][27][28].
In the future, it would be of interest to investigate mechanisms underlying coordinated transcription in relation to changing levels of the transcribed proteins. Methylation of gene promoters, histone acetylation and methylation state, transcription factors, miRNAs, DNA sequences of cis-elements (transcription factor binding sites), and feedback regulation by metabolites can influence expression [71][72][73]. Genes whose expression decreases with age appear to have higher promoter GC content than the other genes [71], suggesting differences in methylation state [74].
In summary, we hade described coordinated changes during Development and Aging in transcription of genes coding for multiple aspects of brain phosphoinositide metabolism, suggesting important roles for these genes. Three somewhat similar groups of genes with distinct expression intercorrelations were identified in each of the two intervals, and some pairwise correlations could be related to colocalization on the same chromosomal band. Nonlinear changes during Development likely participate in concurrent nonlinear phenotypic changes within this period. Mechanisms of coordinated transcription in normal as well as pathological human brain deserve to be explored further.