SEARCH

SEARCH BY CITATION

Keywords:

  • Autism-spectrum disorder;
  • brain development;
  • coherent-gene groups;
  • cortex;
  • hierarchical coherent-gene group model;
  • hierarchical transcription-factor network;
  • informational entropy;
  • neurogenesis;
  • neurotransmission;
  • schizophrenia;
  • self-organizing maps;
  • signaling pathway;
  • synaptic epigenesis;
  • synaptogenesis

Abstract

  1. Top of page
  2. Abstract
  3. Materials and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. References
  8. Supporting Information

We have described a strategy to analyze the data available on brain genes expression, using the concept of coherent-gene groups controlled by transcription factors (TFs). A hierarchical model of gene-expression patterns during brain development was established that identified the genes assumed to behave as functionally coding. Analysis of the concerned signaling pathways and processes showed distinct temporal gene-expression patterns in relation with neurogenesis/synaptogenesis. We identified the hierarchical tree of TF networks that determined the patterns of genes expressed during brain development. Some ‘master TFs’ at the top level of the hierarchy regulated the expression of gene groups. Enhanced/decreased activity of a few master TFs may explain paradoxes raised by the genetic determination of autism-spectrum disorders and schizophrenia. Our analysis showed gene–TF networks, common or related, to these disorders that exhibited two maxima of expression, one in the prenatal and the other at early postnatal period of development, consistent with the view that these disorders originate in the prenatal period, develop in the postnatal period, and reach the ultimate neural and behavioral phenotype with different sets of genes regulating each of these periods. We proposed a strategy for drug design based upon the temporal patterns of expression of the concerned TFs. Ligands targeting specific TFs can be designed to specifically affect the pathological evolution of the mutated gene(s) in genetically predisposed patients when administered at relevant stages of brain development.

Several difficulties impede our understanding of the relationships between the organization of the genome and the neural phenotype of the brain: (1) The human genome has only 22 000 structural genes and (2) the differences in sequences of mouse, rat, monkey, chimpanzee and human genomes are so small that genome evolution, as opposed to brain evolution, appears strikingly non-linear (Changeux 1983, 2004; Ebersberger et al. 2002; Jiang Z et al. 2007; Konopka et al. 2012; Liu et al. 2012). Plausible explanations for such non-linear evolution include (1) the combinatorial co-expression of genetic determinants creating networks of interacting genes accounting for normal brain development (Johnson et al. 2011; Kang et al. 2011) and its pathologies (Ben-David and Shifman 2012; Torkamani et al. 2010), (2) the length of postnatal development (13–15 years in Homo sapiens), accompanied by major increase of brain weight (about fivefold before adulthood) (Lagercrantz et al. 2010), (3) the selective stabilization and elimination (pruning) of synapses under the control of evoked or spontaneous activity elicited by interactions with the physical, biological, social and cultural environment (Benoit & Changeux 1975, 1978; Changeux 2004; Changeux & Danchin 1976; Changeux et al. 1973; Luo & O'Leary 2005; Purves & Lichtman 1980; Stretavan et al. 1988); and eventually, (4) chromatin epigenesis (Lagercrantz 2012). Concomitantly, the development of brain transcriptome follows definite sequential patterns in the course of pre- and postnatal development (Bayès et al. 2011; Fu et al. 2011; Kang et al. 2011; Liu et al. 2012; Somel et al. 2009; Stead et al. 2006) that need further study.

Attempts to model the relationship between genotype and phenotype in multicellular organisms are fundamentally based on extension of the bacterial operon scheme (Jacob & Monod 1961) to eukaryotic cell differentiation and embryonic development (Britten & Davidson 1969; Davidson 2010; Driever & Nüsslein-Volhard 1998; Jiang & Levine 1993; Kerszberg & Changeux 1994a,b, 1998; Li E & Davidson 2009; Monod & Jacob 1961; Wolpert 1969). According to this scheme, transcription factors (TFs) serve as diffusible signaling allosteric proteins, binding to specific DNA elements in the promoter regions and triggering (or inhibiting) in cis the transcription of adjacent genes by DNA–RNA polymerase (Mannervik et al. 1999), depending on the organization of their complex with RNA polymerase and DNA (Liu et al. 2010), as with other TFs. Moreover, ligands that allosterically regulate TF conformations (or prevent their proper organization) can affect their function. Transcription factors may, in addition, control the transcription of their own structural gene (Noll 1993; Thayer et al. 1989; Walther & Grüss 1991), generating autocatalytic feedback loops, which may serve as gene switches in embryonic development (Kerszberg and Changeux 1994a,b, 1998). TFs also control other genes, including TFs genes, thus generating hierarchical trees of gene-expression patterns in the course of development (Davidson 2010; Kerszberg and Changeux 1998; Larson et al. 2010).

Yet applying this paradigm to brain genes and relevant pathological phenotypes is not straightforward. The major difficulty resides in the hundreds of genetic determinants that predispose to brain disorders, such as autism-spectrum disorders (ASDs) (Girirajan et al. 2011; Sebat et al. 2009) and schizophrenia (Bayès et al. 2011; Kirov et al. 2012). The lack of a simple relationship between gene sets and how brain connectivity becomes established requires, in our opinion, a deeper analysis of overall gene expression in brain development.

We explored the hypothesis that genomic subsystems of interacting genes are formed during brain development, thus extending the standard analysis previously based on single genes. We looked for correlations between gene-expression patterns in the developing brain, assuming that they show underlying common molecular processes of gene regulation. Using time series of gene-expression data from rat (Stead et al. 2006) and, to some extent, human brain (Somel et al. 2010), we (1) looked for occurrence of correlated groups of co-expressed genes – coherent gene groups (CGGs) – that differ in their developmental evolution in rat cerebral cortex (CTX), hippocampus (HPC) and hypothalamus (HYP), (2) distinguished successive steps of gene-expression patterns in CTX evolution that correlate with its anatomical development, postnatal synaptogenesis being considerably prolonged in humans, (3) elucidated the signaling pathways and TFs controlling CGGs assumed to behave as functionally coding units with distinct patterns, in particular those that occur at birth and postnatally and are controlled by the current activity of the neural network and thus (4) uncovered nested hierarchical interactions between TFs during brain development that possibly account for the distinct temporal evolution of CGGs in major brain disorders such as ASDs and schizophrenia. Building on these findings, we aim to develop a novel strategy of drug design to treat these disorders based on a hierarchical network of TFs, paving the way to new pharmacological tools: TF orthosteric and allosteric ligands administered at relevant sensitive stages of brain development in genetically predisposed patients.

Materials and methods

  1. Top of page
  2. Abstract
  3. Materials and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. References
  8. Supporting Information

The strategy used applies the microarray data analysis and clustering with self-organized maps (SOMs) consisted of the following steps: (A) initial processing of microarray data and first-level SOM clustering (which continues by the three branches), (B) determination of gene zones associated with significant developmental changes and brain disorders, (C) analysis of stages of development and quick comparison between rat and human brain development and (D) pathway analysis, (E) functional and hierarchical network analysis, (F) elimination of genes associated with brain disorders – schizophrenia and ASD; and (G) directions to the future drug design. A flowchart of methods used is shown in Fig. S1, Supporting Information.

Step A included (1) calculation of mean from replicates, (2) generation of time series, (3) selecting set of genes by significance and (4) clustering genes into the set of SOMs, which we call the first-level SOMs to distinguish them from the two following SOM clusterings using Gene Expression Dynamic Inspector (gedi) program (Eichler et al. 2003; Guo et al. 2006). In Step B we selected the top of differentially expressed genes and mapped them back into the first-level SOMs, forming the gene zones that showed expression-profile similarities. Step C contained two independent parallel processes: SOM informational entropy calculation and orthogonal second-level SOM generation that resulted in obtaining maps each representing clusters of developmental days, or ‘coherent-day groups’(CDG), for the each of 650 CGGs. Both of these parameters – informational entropy and resulting ‘density’ map (the gene density map displays the number of genes assigned to every particular CGG. In this case it is day-of-development density map and displays the number of days assigned to every particular CDG) – helped elucidating stages of brain development and were used along with obtained on Step D signaling pathways and biological processes in hierarchical network analysis represented by Step E. In Step D we elucidated the signaling pathways, grouping genes by SOM method in 30 CGGs and creating corresponding gene networks using Ingenuity Pathway Analysis (IPA®) program (Ingenuity Systems Inc., Redwood City, CA, USA), which assigned the conventional signaling pathways (CSPs) for each of 30 CGGs. Analysis of obtained pathways with IPA led to listing biological processes inherent in each of them. Step E was involved in multilevel hierarchical network clustering using BiologicalNetworks (Baitaluk et al. 2006b; Kozhenkov et al. 2010) – and data obtained by Steps B–D. Analysis of hierarchical networks showed gene–TF interaction on different levels. Results of Step E allowed analyzing specific gene–TF interactions, for example, for neurological disorders, expanded the former along the length of time series. This was implemented in Step F. Finally, Step G showed further directions for drug design based on analysis of hierarchical networks.

Sample preparation, microarray and data analysis

Sample preparation and preliminary filtering and normalization were provided by J. D. H. Stead et al. (Carlton University, Ottawa, Canada) as described in the article (Stead et al. 2006). The samples were taken from three different regions of the brains of Sprague Dawley® rats from the Charles River Laboratories (Wilmington, MA, USA) – CTX, HPC and HYP – on different days of development. Each region included several time points: CTX – 11 (from prenatal day E16 to postnatal day P90), HPC – 7 (postnatal time points from day P01 to P90) and HYP – 9 (from prenatal day E18 to postnatal day P90). Six samples were taken for each time point. Affymetrix GeneChip® Array RAE230A (Santa Clara, CA, USA) was used. Each probe set originally included 10 063 genes that were subjected to statistical analysis (Dai et al. 2005; Stead et al. 2006). The microarray was validated by Stead et al. (2006) , using reverse transcription polymerase chain reaction of 33 genes. We calculated the average for each probe from the six sample replicas of each time point.

The rat microarray data was obtained from J. D. H. Stead (J. D. H. Stead, personal communication). In the original article (Stead et al. 2006) authors filtered the initial set of around 10 000 genes using levels of detection for all six biological replicates. This filtration brought 6653 genes that undergone two-way analysis of variance (anova). Total 87.8% of these genes (5844) showed significant differences in expression between brain regions, and 97.2% (6470) genes showed significant differences in expression over time. We deleted expressed sequence tag genes with 4819 genes remaining, and then the averaged expression values for each time point were normalized along the time series separately for each probe for each region of the brain. This operation allowed for comparing gene expression within different expression ranges; so that genes with low expressions would not be discriminated.

The human microarray data (Somel et al. 2010) was obtained from the NCBI GEO Datasets, ID GSE 18069 and subjected to the processing described above.

Self-organized map generation

For clustering microarray data we selected the method of SOMs (Kohonen 2000) because it is a two-way clustering method that allows clustering of both genes and samples (i.e. time points), taking into account the relationship between both gene and sample clusters. SOMs allow visualization, multidimensional scaling and data reduction from thousands of genes to hundreds of gene groups [we used implementation of GEDI (http://www.childrenshospital.org/research/ingber/GEDI/gedihome.htm) (Eichler et al. 2003; Guo et al. 2006)], while clustering performance is comparable with other clustering techniques (Tsigelny et al. 2008). To generate SOMs one needs to set several parameters. First of all it is a number of clusters that is represented as a two-dimensional grid map and number of training iterations. Also other parameters such as neighborhood, learning factor and distance metrics, should be assigned. GEDI program projects genes with the same expression profile along time points in the same location of the map, creating ‘coherent-gene group’ (CGG) and placing genes with minimal distance metrics in neighborhood vicinity (Eichler et al. 2003) (Fig. S2, left). Moreover, GEDI calculates expression centroid for each CGG and assigns a color according to a color spectrum of gene expression. Derived heatmap-like image was used to calculate informational entropy for each time point. Using a 26 × 25 map and Euclidean distance metric, we generated 650 CGGs for each day of development.

Elucidation of zones with highly-expressed genes

Differentially expressed genes within each brain region were derived by calculating expression fold changes (the ratios between the minimal and maximal expressions along the time series) for each probe along all time points. Because not every brain region contained the same time points, we examined only seven time points for postnatal period; 1086 genes with a fold change ≥2.5 were considered differentially expressed and used for further analysis. Such a cutoff value is used often in microarray study (Avasarala et al. 2008; Li et al. 2007; Sarkar et al. 2008).

These 1086 genes were then mapped into the SOM, where they fit to 138 (out of 650 totally) CGGs that were grouped in several zones (outlined in white in Figs 1 and S3). We considered CGG ‘active’ if it contained 50% and more of genes having 2.5-fold change of expression. This percent had been chosen after analysis of the dependency of the average number of genes from percentage of active genes in a CGG. The inflexion point of the curve was found if 50% of CGG's genes were active (see upper right chart of Fig. S3).We conducted additional two-way anova analysis of the initial gene set using the program MeV (Saeed et al. 2006). In the zones selected (including both the genes with 2.5-fold change and other genes) 97.6% of genes for zone 1 and 98.3% for zone 2 (that were used in further analysis) were considered significant by anova analysis. The mapped SOMs for CTX on day P01 are shown in Fig. S3. Distribution of differentially expressed CGGs and genes along the time series in the rat cortex is presented in Tables S1 and S2.

image

Figure 1. Self-organized maps ( SOMs) of gene expression: (a) during postnatal development of rat brain for CTX, HPC and HYP. The method of SOMs identifies from microarray data the genes exhibiting correlated expression profiles referred to as ‘CGG’. Each pixel of the SOM represents a group of genes having correlated profile of expression in time. A color code (dark-blue to dark-red) denotes the centroid (average) of expression values for the entire CGG at this pixel. The program further distributes them on two dimensional maps (26 × 25 grid), positioning closer to each other the groups with the gene-expression profiles with the highest similarity. Outlined by white lines are the CGGs in which the activity of more than 50% of the genes change their expression more than 2.5 folds during development. It is clearly seen that these CGGs create several connected areas that we called ‘zones’ enumerated with figures near them. Such display visualizes the CGGs with the most profound change in expression during brain development. (b) Higher-level ‘second-order’ analysis of SOMs where the relationships of rat CTX, HPC and HYP transcriptome during development are represented as a single square map. A single building block (pixel) gives these relationships for each day of development. Close-up of the pixel [1,3] at the left explains the contents of each of the pixels, including the entire expression profile for the specified region of the brain during development. At late prenatal stages HYP data are close to CTX data but segregate from CTX and HPC at early postnatal stages. Between days P7 and P14 the transcriptome of CTX and HPC evolve in parallel, in distinction to HYP. (c) The same analysis for the development of human CTX. The data at postnatal days P004 and P034 are close to each other as well as those for postnatal days P094, P204, P443 and P787, while data of the day P002 (which is close to birth) are distinct.

Download figure to PowerPoint

Informational entropy calculations

To estimate the complexity of brain growth and organization during development, informational entropy was used, which value is reciprocal to each system organization level (Kouznetsova & Rakov 1987). From the obtained set of SOMs, image histograms were generated for each time point (Davies 2005; Kapur et al. 1985). On the image histograms the horizontal axis represents the intensity variations of the image and the vertical axis represents the number of pixels captured in each intensity zone. The transcription intensities of gene groups were calculated as centroids that are the multivariate equivalent of the mean intensities. There were 11 histograms for the CTX, 7 for the HPC and 9 for the HYP. The histogram for the SOM of day P07 is shown in the top left corner of Fig. S2. The two peaks of intensity in two groups appeared in this histogram: 47 units collecting 81 pixels and 170 units collecting 35 pixels of the total 650.

The maximum gene-expression centroid value for CGGs in the SOM dataset was 10.85 and the minimum 4.91, a span of 5.94. The span was divided by 0.5 intervals, providing 13 samplings, and plotted across the number of intensities for each sampling. The 0.5 intervals for the sampling were chosen to simplify the calculations; further study indicated no significant difference in entropy profiles between 13 and the higher-value samplings, and explained the image patterns. From these histograms, probability was calculated for each sampling using the formula pi = n/N, where n is the number of intensities that fell in the i-th sampling and N is the total number of intensities. These probabilities were used in the Tsallis entropy calculation (eqn 1 below) (Tsallis 1988).

  • display math(1)

In the equation, p denotes the probability distribution of the SOM and q is a parameter that measures non-extensivity. This definition is useful in distinguishing developmental changes related to the general growth of the system from those related to its self-organization. When a parameter is ‘extensive’, it is dependent on the size of the system (e.g. the mass of developing brain); when a parameter is ‘non-extensive’, it is not dependent on the size of the system (e.g. brain-tissue differentiation and organization). The developing brain possesses both properties. It is constantly growing in size (extensivity) and simultaneously developing functionally important structures, including anatomical changes in shape, cell-type diversification and synapse formation (non-extensivity).

Tsallis entropy was chosen because of its suitability for calculation of non-extensive or mixed entropy. Parameter q is thought to quantify the degree of departure from extensivity: If q = 0, the system is non-extensive, whereas if q = 1, the system is extensive, and its entropy is equivalent to extensive entropies such as Boltzmann–Gibbs or Shannon entropy. An entropic index of q = 0.25 was selected because lower the q lesser the extensivity it has and less the system entropy depends on its size; the change in entropy therefore corresponds to the change in morphology and other function-related characteristics of the brain but not to the increased physical dimensions. Tsallis entropy was calculated separately for each set of gene-expression data for the rat CTX, HPC and HYP as described above to create the profile chart.

Signaling pathways analysis

The sets of genes over-expressed during each day of development (extracted from 1 of 11 6 × 5 SOMs generated by GEDI for the rat CTX – Fig. S5) were submitted to the IPA® program (Ingenuity Systems Inc.), which assigned the conventional signaling pathways (CSP) for each CGG. The criterion for selection of a specific pathway as corresponding to the genes on a given day of development was the P value threshold, so that the selected set of over-expressed genes from this day was not randomly included in the existing scheme of the CSP.

Gene–TFs networks elucidation

Genes exhibiting correlated expression patterns may be co-regulated by common TFs or may be functionally related, forming a functional module or a molecular complex (Fessele et al. 2002). In an attempt to identify those genes that are considered significant during the early stages of brain development, a systematic, large-scale search for TFs, their respective binding sites [transcription factor binding sites (TFBSs)] and for a phylogenetically conserved promoter structures was conducted.

In orthologous promoters, elements important for regulation of a given gene are expected to be conserved throughout evolution. Starting with a set of gene pairs found to be differentially expressed in the early stages of brain development and significantly co-regulated or functionally related (i.e. members of CGGs), orthologous promoter regions were identified from three species (H. sapiens, Mus musculus and Rattus norvegicus), using BiologicalNetworks – a comparative genomics program tool (Baitaluk et al. 2006b; Kozhenkov et al. 2010). A systematic approach was taken to identify promoter frameworks shared among over-expressed genes in the early stages of brain development. Transcription factor binding sites common to promoter regions across species were first identified for individual genes. Using a modified phylogenetic footprinting method, a search was conducted for TFBSs that are enriched in the region of 6000 bases upstream of each transcription start site to 500 bases downstream of each transcription start of every gene in the gene pairs found and are conserved in the mouse, rat and human genomes. Binding sites and correspondent TFs were filtered for p < 10−3 and examined for consistency (Fig. S13).

The parameter settings of the TFBS selection, strand orientation and order were strictly determined by sequence analysis of the promoter sets; only two parameters were subsequently optimized manually. The stringency of evolutionary conservation by selecting a subset of the six available species played a central role in the analysis because only TFBSs that were conserved in the selected species were scored. Minimal distance ranges between matrices were reduced (from the default, 10 nt to 1 nt) to include conserved TFBSs identified by visual inspection. Matrix similarities were set at default values and fractionally adjusted (default −0.1, −0.2 or −0.5) when the reduction was allowed for detection of evolutionarily conserved TFBS sets that were missed with the default settings. The evolutionarily conserved frameworks were determined using the modified phylogenetic footprinting method on the predefined TFBS subsets. Models reflecting the frameworks were then built using IntegromeDB data integration system (Baitaluk & Ponomarenko 2010; Baitaluk et al. 2006a) and BiologicalNetworks analysis environment (Baitaluk et al. 2006b; Kozhenkov et al. 2010) and optimized for all three aforementioned species (Fig. S13).

This approach eventually yielded X potential TFs within the set of Y genes, and among the promoter frameworks of conserved TFBSs that originated in early brain development. These specific genes were also located in the human, mouse and rat promoters (Fig. S13) by searching the specific proximal promoter sequences (a total of Z sequences).

To find novel potential interactors of significant genes and TFs in early brain development, the BiologicalNetworks integrated database (Baitaluk & Ponomarenko 2010; Baitaluk et al. 2006a), dedicated to molecular interactions and including both physical and functional interactions among genes and proteins, was used. This database weighs and integrates information from numerous sources, including experimental repositories, computational prediction methods and public text collections, thus acting as a upper-level-database that maps all interaction evidence onto a common set of genomes and proteins.

Two more searches were conducted: the first was for physical interactions and phosphorylation events among TFs, kinases and general proteins; the second, for newly found TFs and proteins that are co-expressed with any of the other TFs or are co-regulated with target genes (this latter search was done in the GEO compendium). The integrated picture of co-expressed genes and proteins together with their interactors, as well as the bindings of TFs with their target genes, is represented in Fig. S13.

Results

  1. Top of page
  2. Abstract
  3. Materials and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. References
  8. Supporting Information

Coherent gene groups expression in brain development

Computational analysis of CGGs

Classical genetic analysis follows the one disease–one gene paradigm tying a given gene to a given function. Yet recent genome sequencing of thousands of brain pathologies, together with contemporary analysis of microarray data, point to a broad diversity of often-rare gene-determination and gene-expression patterns (Sebat et al. 2009). Our hypothesis is that, at any step of development, CGGs behave as coding units, as introduced earlier (Tsigelny et al. 2005a).

The transcriptome data we used originated from microarray data of prenatal (E) and postnatal (P) rat brain (Stead et al. 2006) and postnatal (P) human brain (Somel et al. 2010). The rat original dataset was filtered, resulting in 4819 genes, and normalized. Using GEDI, we obtained SOMs of 650 groups representing these genes for each pre- and postnatal day. The SOM reflects the general relationships between the brain genes studied during development. A set of SOMs was generated for each available day of development for the CTX, HPC and HYP. Fig. 1a shows the SOMs for the postnatal period. White lines outline the five zones of CGGs that exhibit significant differences (≥2.5-fold change) in expression during development (Avasarala et al. 2008; Li et al. 2007; Sarkar et al. 2008). The further analysis of these zones showed that genes in them switched on and off at different times of development thus turning on and off the pathways responsible for different processes and functions. Successive SOM images showed steady changes in color patterns during development. Same CGGs do not vary their position on SOMs, indicating that the same pixel on all maps represents the same CGG through the entire time series. Figure S2 explains the concept of SOMs. The outlined square with four subsquares from the SOM corresponds to the four groups of genes having similar temporal profiles. These groups are neighbors on the map. Interestingly, SOM patterns are similar for CTX and HPC but notably different for HYP (Fig. 1a). This observation is consistent with CTX and HPC, but not HYP, containing cortical neurons generated within the ventricular zone.

Second-level gene group analysis in brain development

We also applied the SOM method in a higher-level analysis. We attempted to present as a single square map (Fig. 1b) the relationships between transcriptomes of the studied brain regions represented by a CGG for each day of development. We used the previously created 650 CGGs with their expression values as a matrix of input data for each day of development for CTX, HYP and HPC. We presented them as pixels on a 5 × 5 SOM grid and we called them coherent day groups (CDG). The close-up of CDG [1,3] on the left of Fig. 1b explains the contents of each CDG, including the entire expression profile for the specified region of the brain during development.

The resulting second-level SOMs (Fig. 1b) showed the relationships between the studied-day transcriptomes. Late prenatal CDG (E18 and E20) of HYP were close to CTX late prenatal CDG (the data for prenatal HPC CDG were missing, but we considered them close to CTX CDG at these days), but they become separated from CTX and HPC soon after birth. At postnatal day P14, CTX and HPC CDG separated from each other, showing qualitative transformation of their transcriptomes between days P07 and P14. The HYP CDG also separated from CTX and HPC between days P07 and P14. They were very distant from CTX and HPC locations on the second-level SOM; the resultant blanks on the map were interpreted as indicating discrete changes in the evolution of gene-expression patterns in the brain regions investigated. We also conducted CDG analysis for human cortex-expression data (Fig. 1c). We used the gene-expression datasets for different times of human cortex development (Somel et al. 2010). This set contained the data for the postnatal days P002–P787 (approximately 2 years) of human development. Note that human day-transcriptome data were grouped by SOMs into three definite CDG. The first CDG contained a single day, P002; the second contained P004 and P034; and the third contained the remaining days (in a 2-year period), starting from day P094. The map blanks again indicated that discrete changes in the evolution of gene-expression patterns have occurred during cortical development. The methods we used were justified by Rice and Barone (2000) who examined various data about cortex development from different species. Last, we found a correspondence of high-order SOMs at postnatal day P07 in rat with embryonic (prenatal) day E145 in human and at day P14 in rat with day E195 in human (Fig. S4).

Both rat and human results support the stepwise development of mammalian cortex, in which the stages of gradual (extensive) development, characterized by growth in size and mass, are interspaced by stages of significant qualitative changes in transcriptomes (non-extensive or informational development). Consistent with this analysis, qualitative changes in the neural organization of the human cortex take place as changes in fetal activity occur between prenatal days E145 (E4.8 months) and E195 (E6.5 months) (Lagercrantz et al. 2010). Such information may be used when selecting the strategies of pharmacological treatments.

Informational entropy and brain morphological development

Informational entropy is an important parameter extracted from the first-level SOMs. This parameter, introduced by Shannon (1948) is used in calculating the informational content of an image. Each pixel of a SOM (Fig. 1a) carries information on CGG expression, and pixel positions define relationships between CGGs in a system. We used SOM informational content as a parameter for the complexity of the CGG interrelationships. In general, informational entropy is a measure of the complexity and self-organization of an organ (Kouznetsova & Rakov 1987; Tsigelny et al. 2008). In other words, the higher the organization, the lower the entropy. For SOM analysis we used Tsallis entropy (Tsallis 1988) rather than the classical Shannon entropy. In biology Tsallis entropy is used to define integral properties of developing organs and systems and evolutionary processes (Sieniutycz & Jeżowski 2009). Tsallis entropy is used to study a system's non-extensive (depending on connectivity) rather than extensive (depending on size) entropy. The developing brain possesses both characteristics, but in terms of brain function, the important one is connectivity and its related non-extensive entropy. Tsallis entropy for an each-day SOM is actually the informational entropy reflecting the internal relation of CGGs during development. High levels of entropy correspond to more versatility of CGG expression, and lower levels correspond to higher organization of the entire system. Such a simplistic description obviously cannot account for organ development in entropic terms, but it offers a gestalt pattern of the entire gene set studied during brain development.

We calculated correlation coefficients for CTX reversed-entropy curve and two published plots relevant to neural development that appeared sufficiently high (0.60 and 0.57), and superimposed all three curves scaled to plot (Fig. 2a). The first curve, from Bandeira et al. (2009), represents the total number of neurons evaluated after in vitro homogenization of the cortex by counting the number of nuclei stained by a neuronal antigen (to distinguish them from glial cell nuclei). The second curve (Furtak et al. 2007) represents the quantitative developmental evolution of differentiating cortical neurons measured by Golgi-staining method (red line). The first curve reflects the general evolution of the global number of neurons in all studied regions of the brain and correlates with the CTX reversed-entropy curve (correlation coefficient is 0.60), and the second also correlates with the CTX reversed-entropy curve (correlation coefficient 0.57). Interestingly, the Golgi-stained cortical neurons of the CTX second curve are pyramidal cells that constitute around half the Golgi-impregnated cells in the studied perirhinal cortex and those possessing the largest and most complex dendritic arborization (Furtak et al. 2007). This correlation supports the interpretation that the peak of the reversed informational entropy curve of CGG expression corresponds to the differentiation of cortical neurons (at peak), concomitant with the progressive developmental organization of their connectivity (late decline). We compared the entropy plot to the expressions of three standard markers of central nervous system (CNS) development: Myc for neural proliferation, Jun for neuronal migration and Fos for neural activity. Figure 2b illustrates the superimposition of the temporal profiles of these genes' expression with the reversed Tsallis's entropy curve. At the first stages of development (E16–E17), the maximum of Jun activity occurred, and the reversed entropy plot reflected the envelope of migration of neuroblasts that were still not significantly diversified. The entropy of such a relatively homogeneous system was not expected to change much. Increase of reversed entropy during early postnatal days (P01–P04; Fig. 2b) correlated with the upregulation of Myc expression and reflected the envelope of neuron and glial cell proliferation with axon growth and differentiation of the neuronal cell types and their connections. The last step in neuronal circuit development is thought to be the selective stabilization of synapses with elimination of exuberant synapses and establishment of stable interneuron connections with increased synaptic activity of the neural networks. Such synapse selection is accompanied by an enhanced organization of the neural networks (Changeux et al. 1973; Luo & O'Leary 2005). We proposed that the observed increase of reversed Tsallis entropy reflects increased cortical circuit organization corresponding to the laying down of what has been called the adult CTX synaptome (de Felipe 2010) and the expression of the relevant population of synaptic genes (Bayès et al. 2011; Liu et al. 2012). Interestingly, this late cortex evolution coincided with the maximal increase of Fos upregulation, a gene expression considered a faithful indicator of activity of the developing neural network. This correspondence supported the view that this postnatal increase of neural network complexity – which becomes especially extended in humans (Changeux 1983; Liu et al. 2012; Rakic 2006) – is associated with increased intrinsic (spontaneous and evoked) neuronal activity that manifests an extensive interaction of the developing brain with external physical, biological, social and for humans, cultural environments (Castro-Caldas et al. 1998; Changeux 1983; Dehaene et al. 2010; Petanjek et al. 2011).

image

Figure 2. Tsallis entropy, neurons development and selected gene expression. (a) Evolution of reversed Tsallis entropy (green triangles) in the course of rat brain development (blue and red). Tsallis entropy is the ‘informational entropy’ that reflects the internal relation of gene groups at any given day of development and represents the measure of disorder. It is non-extensive entropy independent of system (SOM's) size and calculated using the formula inline image in which p denotes the probability distribution of interest and q is a real parameter that measures non-extensivity of the system. Reversed Tsallis entropy is opposite to disorder and represents level of structural and functional organization. High levels of entropy correspond to more versatility of the CGGs expression and lower levels correspond to higher ‘organization’ of the entire system (The reversed Tsallis entropy calculated from CTX developmental SOMs with q = 0.25), the number of neurons (blue) (from Bandeira et al. 2009) and the number of Golgi-stained cortical neurons with full dendritic arborization (red) (from Furtak et al. 2007) follow correlated developmental curves indicating a relationship between calculated entropy and gene-controlled set of neurons and synapses development events. (b) Expression of the markers: Fos as an indicator of neural activity (excitation), Myc as an index of cell proliferation (and apoptosis) and Jun as a marker for cell migration (Jaźwińska et al. 2007) and neural activity (excitation). Expression of Jun shows three peaks–two of them are related most probably to neural migration on days E16 and P14, and the last one, on days P21–P30, which is also the maximum of Fos, is most likely related to enhanced spontaneous and/or evoked neural activity.

Download figure to PowerPoint

Conventional signaling pathways (CSP) and control of CTX development

Our analysis has showed that defined patterns of multiple CGGs were differentially activated at specific time points of development. We attempted a deeper analysis of the regulatory processes controlling these patterns of gene activations, trying to group the gene-expression data within signaling pathways, automatically selecting them according to a threshold rule.

To find which canonical pathways were active on specific days of development we again used first-level SOM analysis. We examined a set of different levels of SOMs (16–625 CGGs). The level best corresponding to the current CSP description was 30 CGGs – 6 × 5 grid (Fig. S5). It gave us a reasonable description of signaling pathways (Fig. 3a) and processes (Fig. 3b) during brain development. For analysis, we used a database of the networks of genes organized in CSPs as presented by the program IPA® (Ingenuity Systems Inc.). From the SOM analysis (Fig. S5), we learned that a significant number of CSPs can be depicted as having P values of less than 0.001 of non-random possession of genes over-expressed during development (Fig. S6). Note that only a fraction of these pathways related to neural development were selected (about 10 pathways selected from around 100). The general schemes of CSPs (Fig. 3a) and processes (Fig. 3b) participating in CTX development were obtained by in silico analysis. These were automatically selected using P values for non-random participation of the upregulated CTX genes. The selected CSPs and processes included the genes upregulated in the microarray analysis with values that are more than 95% of the maximum through the time series. In other words, the pathways shown in Fig. 3a had maximal upregulation at specific stages of development.

image

Figure 3. General schemes of intracellular signaling pathways and neuronal processes obtained from SOM's analysis of rat CTX development. These pathways (a) and processes (b) were automatically selected using P values for non-random participation of the upregulated genes of CTX. Fig.  3a reveals a stepwise development of groups of signaling pathways, which confirms and expands the above mentioned second-order SOM analysis (Fig. 1b) and Tsallis entropy data (Fig. 2a, green curve). It shows a major qualitative change starting on days P04–P07 immediately following birth.

Download figure to PowerPoint

Figure 3a shows a stepwise development of groups of signaling pathways that confirmed and expanded the above-mentioned second-order SOM analysis (Fig. 1b) and Tsallis-entropy data (Fig. 2a,b), with a major qualitative change starting at days P04–P07 of rat brain development. Our analysis showed that these pathways can be assembled into five principal groups, represented by different colors (Fig. 3a): (1) the first group comprised the pathways upregulated during prenatal development (E16–E20) involving early neurogenesis and neuronal differentiation and migration; (2) the next group, the pathways that started to be upregulated prenatally and continued to be upregulated during the first postnatal days of cortex development and were still involved in neurogenesis, neuronal differentiation and migration, and early neuronal pattern formation; (3) the third group, the pathways upregulated soon after birth, at the early stages of postnatal development (P04–P07), that are involved in neurite outgrowth and synapse formation, including the effect of the early social environment on neural development; (4) the fourth group, the pathways upregulated at the latest stages of postnatal development (P07–P21) and involved in late synaptic epigenesis of cortical connections, including its long-range connectivity, assumed to be involved in conscious access (Dehaene & Changeux 2011; Dehaene et al. 2010) and (5) the fifth group, the pathways upregulated for the most part during the studied prenatal and postnatal days (E16–P21) of cortex development. The details of the pathways are given in the Materials and methods section.

Figure 3a,b summarize the activation of genes related to cortical development. At the late prenatal stages (E16–E18) we see an expected selective increase of upregulated genes engaged in neuron quantity and differentiation, proliferation of neuroglia and outgrowth and selection of neuronal processes. The neuroglia-proliferation-related genes remained upregulated up to postnatal day P01. Note the remarkable concomitant upregulation of genes related to development of neural structures and synaptic transmission after birth (P04–P30). This sudden increase of gene expression is probably related to the first breath, or ‘awakening’, of the newborn (Lagercrantz et al. 1986), caused by the dramatic change in environmental conditions from the maternal womb to the external world expressed, in particular, by activation of the noradrenergic system and related genes engaged in brain arousal (Lagercrantz et al. 1992).

An important feature of this diagram is, as anticipated from Tsallis-entropy analysis, the postnatal activation of synaptome genes (Bayès et al. 2011; Liu et al. 2012) correlated with the postnatal maximum of reverse entropy (Fig. 2a). This corresponds to the evolution of the CTX synaptic network from neuronal migration and differentiation, neurite growth and branching, and synapse formation with a maximum of fibroblast growth factor-related neural plasticity (P04–P14) and mostly lasts until activation of synaptic transmission (neurotransmission) genes (P07–P30), in particular glutamate and gamma-aminobutyric acid (GABA)-related plasticity (neurogenesis and memory), and the epigenesis of neuronal networks by selective stabilization of synapses.

Hierarchical networks of TFs regulating CTX development

The ultimate step of our approach was to investigate the connection between TFs and promoter elements engaged in the formerly identified patterns of brain gene expression and the complementary hypothesis that genes exhibiting CGGs are co-regulated by common TFs, forming ‘transcription modules’ (Fessele et al. 2002; Segal et al. 2003). To address this hypothesis, we made a systematic large-scale search for TFs, TFBSs and phylogenetically conserved promoter structures to identify transcription modules during CTX development. Orthologous promoter regions were identified by searching for TFBSs that are enriched in the region encompassing 6000 bases upstream of each transcription start site to 500 bases downstream of each translation start of every gene and that are conserved in mouse, rat and human genomes. To reduce false-positive TFBS predictions, a modified phylogenetic footprinting method (Chang et al. 2007; Halfon et al. 2002; Qiu et al. 2003) was used that searches only for genomic DNA sequences, which are conserved among species. Binding sites and corresponding TFs were filtered and visually examined for consistency (Fig. 4). Models reflecting the frameworks were constructed and optimized, yielding potential TFs and promoter frameworks of conserved TFBSs. In this study we used the IntegromeDB database, integrating data on transcriptional regulatory regions (Baitaluk et al. 2012), and BiologicalNetworks environment for inference, visualization and analysis of gene regulatory modules and networks (Baitaluk et al. 2006a,b; Kozhenkov et al. 2010). Lastly, molecular interactions among genes and proteins, in particular phosphorylation events, and existing protein–protein (TF–TF, TF-kinase and protein–protein) interactions extracted from BiologicalNetworks integrated database gave an integrated picture of co-regulated gene modules and proteins together with their TFs and TFBSs (Fig. S9a,b). Molecular functions related to the selected modules (CGGs) are presented in Table S7. One can see that genes in each of CGGs often have quite specific functions.

image

Figure 4. Strategy used to investigate gene regulation at the promoter level during the early stages of brain development. (a) Systematic search for co-expressed genes and promoter frameworks assumed to represent the co-regulation of genes at early stages of brain development. (i) Self-organized map representing groups of co-expressed genes – CGGs – on one of the days of development. ‘(Gene1 and Gene2)’ is a pair of co-expressed genes from a CGG; ‘(Gene1' and Gene2')’ – a pair of genes from another CGG; ‘(Gene1n and Gene2n)’, ‘(Gene1n',Gene2n')’ – pairs of human orthologues genes and ‘(Gene1m,Gene2m)’, ‘(Gene1m',Gene2m')’ – pairs of mouse orthologues genes (not shown). In analysis all combinatorial pairs of genes were used. (ii) Transcriptional regulation of co-expressed genes with phylogenetically conserved promoters. A hypothesis is used that co-expressed genes are co-regulated by common transcription factors binding to the promoter sequences (Segal et al. 2003; Fessele et al. 2002). Promoter analysis of co-expressed genes was performed to find common TFBSs. (iii) Co-expression (linked regulation) of genes in early stages of brain development integrated with protein–protein interaction data. (b) Integrated molecular interaction (protein–protein, protein–DNA and co-expression) network of human/mouse/rat brain development. (c) Scheme of hierarchical network including genes and TFs (day P01, zone 1, see Figs 1a and S8).

Download figure to PowerPoint

We explored the hierarchy of gene-transcription modules controlling early brain development using a TF-maximal-connectivity algorithm – thus filtering out subnetworks responsible for different levels of regulation – from the upper general level, where the minimal number of TFs control the maximal number of CGGs, to the lower level, where the control became more specific with limited numbers of genes regulated by TFs. All vertices (genes/TFs) are connected with the highest co-expression, co-regulation and interaction strength (integrated score) during development. As a result of our exploration of zones defined in Figs 1a and S2, we uncovered a set of CGGs hierarchically nested from early stages of brain development. The most-populated CGG zones 1 and 2 (Figs 1a and S8) were selected for a detailed analysis. Figure 5a–f shows network hierarchy based on the connectivity profile of TFs controlling genes in CGGs in zone 1 from the level 1 (top level, Fig. 5a) to the level 10 (bottom level, Fig. 5f). The genes' maximal expression in rat is at day E16 (the lowest level is shown in Fig. S9a,c shows a close-up), at the beginning of the prenatal period studied, but almost none of them are still active at day P30 (Fig. S9b). The genes in CGGs often exhibit tight interrelationships, for example, CGG M27 (Figs S9a,b and S10). The gene pairs within CGG M27 either are co-regulated by the same protein, are directly affected by each other, or participate in the common protein complex, etc. (Fig. S11, Table S2). Zone 2 CGGs behave in a manner opposite to zone 1 CGGs (Figs 6 and S12a,b), with the minimal gene expression at day E16 (Fig. S12a), at the beginning of the prenatal period, with almost all the genes active at day P30 (Fig. S12b).

image

Figure 5. Hierarchy of genes networks regulation based on connectivity profile of TFs controlling genes in CGG in the zone 1 (Figs 1a and S8) and involvement of schizophrenia-related CGGs. Color scheme for CGGs: blue – those containing two or more schizophrenia-related genes, light blue – those containing one schizophrenia-related gene, gray – those without schizophrenia-related genes. (a) 1st (top) level, (b) 3rd level, (c) 5th level, (d) 7th level, (e) 9th level and (f) 10th (lower) level; yellow highlighting – schizophrenia-related genes and TF, red – genes with the increased expression, green – showing no significantly increased expression on day E16. (We have to note that ‘related to schizophrenia’ genes are the regular genes involved in growth, proliferation and other cell functions. The mutations of them can skew the brain development toward the disorder).

Download figure to PowerPoint

image

Figure 6. Hierarchy of genes networks regulations based on connectivity profile of TFs controlling genes in CGG in the zone 2 (Figs 1a and S8) and involvement of schizophrenia-related CGGs. Blue – CGGs including genes related to schizophrenia. (a) 1st (top) level, (b) 3rd level, (c) 7th level and (d) 10th (bottom) level. Yellow highlighting – schizophrenia-related genes and TFs; red – genes with the increased expression, green – showing no significantly increased expression on day P30. (We have to note that ‘related to schizophrenia’ genes are the regular genes involved in growth, proliferation and other cell functions. The mutations of them can skew the brain development toward the disorder).

Download figure to PowerPoint

In the TF hierarchy (Figs 5 and 6) a top-level TF network has the most connections to CGGs. The lower networks have decreasing levels of connectivity. Each lower-level network includes a set of genes or TFs, or both, of higher-level networks. These sets of descending networks involve 8–10 levels of hierarchy depending on program separation. Our analysis uncovers nested hierarchical interactions between TFs during brain development that plausibly account for the distinct temporal evolution of CGGs, thus the ‘hierarchical coherent-gene-group model’ in this article's title. To check independently the validity of the TF/gene modules selection, we have performed a functional analysis of all TF/gene modules obtained by our method and added functional evaluation of all transcription modules (for the network related to zone 1). This analysis shows that the genes and TFs included previously in the CGGs (modules) using possible protein–protein interactions also possess the functional connections required to perform a specific function in the brain (Table S3). For example, CGG M27 mentioned above functionally is involved in forebrain development (P value 10−6).

Discussion

  1. Top of page
  2. Abstract
  3. Materials and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. References
  8. Supporting Information

This analysis identified the occurrence of CGGs that differ in their developmental evolution in rat CTX, HPC and HYP, elucidated the signaling pathways and TFs controlling CGGs assumed to behave as functionally coding units with distinct patterns, in particular those controlled by the current activity of the neural network and last uncovered nested hierarchical interactions between TFs during brain development. In the following discussion we examined first the plausible contribution of the TFs network to ASD and schizophrenia, then we proposed a novel, although still hypothetical, strategy of drug design based on a hierarchical network of TFs, paving the way to a new category of pharmacological tools administered at sensitive periods of development.

Transcription factors contributing to brain development and their possible relation to ASDs and schizophrenia

We have to emphasize that all the elucidated TFs related to disorders participate in a broad set of regulations within the organism, especially related to its development. Direct implications of some of them to a possible strategy of drug design for brain diseases have to take in consideration this fact. The analysis below, to our opinion, may set a precedent of possible disease-related selection of TFs during brain development with the additional possibility to elucidate the specific time points of possible pharmacological intervention that can be validated by experiments on animal models. We examined how genetic predisposition to two well-known brain disorders – ASDs and schizophrenia – can be framed within the gene–TF hierarchical networks aiming to concrete suggestions about their genetic origin and for the design of pharmacological intervention at definite levels in these networks to prevent the emergence of the pathological phenotypes. The following TFs were identified at different levels of the hierarchy (Table 1).

Table 1. Transcription factors involved in the control of CGCs from cluster 1 at different levels of TFs hierarchy
Level 1Level 3Level 5Level 7
  1. Tables for the next levels are shown in the supplement material.

  2. a

    Transcription factors related to autism.

  3. b

    Transcription factors related to schizophrenia.

GATA-1AP-1aACAATACAAT
 c-Myc:MaxAP-1aAHR
 GATA-1AP-4AHRHIF
 Pax-4CASP1AP-1a
 ZF5c-Myc:MaxCASP1
  CRISP2c-Ets-1
  E2F1bc-Myc:Max
  GATA1CREB1
  HNF-1bCRISP2
  NkxaCSX
  Pax-4E2F1*
  PolyGATA1
  PSG2Hand1:E47a
  RFX1HNF-1
  SLC22A1HNF4A
  SREBP-1MAF
  ZEB1MUSK
  ZF5MYEF2
   Nkxa
   Pax-4
   Poly
   PSG2
   REL
   RFX1
   RNASEH2A
   RORalfa-1
   SLC22A1
   SREBF1
   TBX5
   TFAP4
   YY1
   ZEB1
   ZFP161

The top level of the hierarchical network comprised 10 sublevels (Fig. 5a). In further description we omitted some sublevels for brevity.

  1. Sublevel one included the TF GATA-1 (GATA binding factor 1) and nine CGGs, four of which included schizophrenia-related genes. The GATA-1 interacts with histone deacetylases and other proteins involved in chromosomic epigenetic-mechanisms and thus may behave as a master TF, switching on cell-growth mechanisms. Mutations in GATA-1 exon 2 have been found in Down syndrome – associated transient myeloproliferative disorder (Greene et al. 2003).
  2. The third sublevel (Fig. 5b) included 5 TFs and 17 CGGs, 7 of which contain schizophrenia-related genes. Note that TF AP-1 had been selected in our analysis as a TF related to schizophrenia; AP-1 is a Jun and Fos heterodimer. The AP-1 controls several neuron- and synapse-development-related genes. We showed earlier (Fig. 2b) that Fos gene is upregulated in days P14–P30 of postnatal cortex development. Our hierarchical analysis of TF activity is consistent with this fact and identifies the TFs that cause this upregulation. The AP-1 is known to be involved in CNS disorders and, particularly, schizophrenia (Pennypacker 1995). Remaining TFs (GATA-1, Myc, Pax-4 and ZF5) are known to be involved in neurogenesis regulation (Buckingham & Relaix 2007; Shibata 1998; Wei et al. 2007).
  3. The fifth sublevel (Fig. 5c) included 19 TFs and 24 CGGs, 12 of which contain genes related to schizophrenia. Additional to AP-1, we see in this group AP-4, which has functions similar to AP-1, and Oct-1 (POU2F1), related to schizophrenia (Narayan et al. 2008). The remaining TFs are presented in Table 1. Figure 6 presents TF–CGG networks for zone 2 at different levels of hierarchy.

Figure 7 shows the ASD- and schizophrenia-related CGGs that contain over-expressed genes at different stages of CTX development. For zone 1, both schizophrenia- and ASD-related CGGs are mostly active in the prenatal period of development, but in zone 2 the CGGs are mostly expressed in the postnatal period of development.

image

Figure 7. Evolution of the number of schizophrenia- and autism-related CGGs in the network during CTX development shown for zones 1 and 2 (Figs 5 and 6). Each point in the plot represents a CGG with at least one gene over-expressed. Note the expression maximums for these CGGs in prenatal period for the zone 1 and in the postnatal period for the zone 2. In the rectangles below the plot are listed the pathways that are found with a P value smaller than 10−5 to be corresponding to the zones 1 (blue) and 2 (red) on the basis of the genes included to them (described in Fig. 3a).

Download figure to PowerPoint

Tables S4–S6, show the TFs related to schizophrenia and ASDs that regulate CGGs in zones 1–3 (in the original SOMs), from which AP-1 is related to both disorders, as mentioned above. Schizophrenia-related TCF4 is specific for zone 1 (active at the prenatal stage of development) (Kwon et al. 2011; Ripke et al. 2011). In transgenic mouse brain it causes cognitive and sensorimotor-gating impairments (Brzózka et al. 2006). Transcription factors either disorder-specific or related to both disorders in the examined CGG zones 1–3 are presented in Tables S4–S6 and Fig. S14.

An interesting feature emerged from this analysis associated with the heterodimeric structure of many TFs. For instance, TFs E2 (E2A) and MyoD associate into a DNA-binding heterodimer. The E2A belongs to zone 3, which has the same expression timing as zone 2, with the maximum in the postnatal period. To form the heterodimer both TFs have to be concomitantly over-expressed, and this occurred only during postnatal development of the cortex. Interestingly, another type of TF heterodimer appeared in our networks. Peroxisome proliferator-activated receptor (PPAR) is listed as a TF related to schizophrenia in zone 1. Dimer PPAR–Retinoid × receptor (RXR) is listed in both zone 1 and zone 2. A relation between schizophrenia and retinoid signaling in brain development had been noted in some patients (van Neerven et al. 2008).

Thus critical periods in cortex development resulted from temporal unfolding of the hierarchical network of TFs and determined the subsequent neural phenotype of the diseased brain. One period was in the late prenatal stage of development and another in the early stages of postnatal development, supporting the idea that ASDs and schizophrenia are disorders that start in the prenatal period, develop further in the early postnatal period and manifest themselves in the neural and behavioral phenotypes later in the postnatal period (Insel 2010). Different sets of genes control brain development during these separate periods. This conclusion has several implications. First, important temporal delays possibly occur between the expression of the mutated TF and the manifestation of the behavioral defect. These delays might vary with the type of TF in the hierarchy that is mutated. Thus drug treatments targeted to the diseased phenotype at the moment it becomes expressed are expected to be inefficient. Administration of drugs targeted to a given TF's hetero-oligomer must coincide with the time of development when the TF of interest is expressed and when it forms functional heterodimers binding to specific DNA sequences. Last, some of these TFs might be regulated by activity of the nerve cell in which they are expressed – for example, MEF2 is activated by neuronal activity (Flavell et al. 2006) – thus accounting for the known effects of the physical, social and cultural environment in the development of the behavioral phenotype of the genetically predisposed patient.

At last it is often noted that many of the mutations identified as predispositions of both autism and schizophrenia concern genes dedicated to synaptic development. Yet the pathological behavior does not resemble the mental retardations or intellectual disabilities expected for global synaptic deficits. Interestingly, brain imaging studies (in particular, by Diffusion Tensor Imaging Tractography) suggest an overproduction (instead of a loss) of connections, in particular, of long range axonal fibers, in some privileged areas of the cortex (Pugliese et al. 2009; Karlsgodt et al. 2008, 2012). This overproduction of connections has been attributed to alterations of the synapse elimination process elicited by some of the predisposition genes (Karlsgodt et al. 2008). In this context one may mention that the long-range horizontal connectivity of the CTX hypothesized to contribute to access to consciousness (Dehaene & Changeux 2011; Dehaene et al. 2010) is found selectively altered in schizophrenic and ASD patients (rev Dehaene & Changeux 2011). One among other interpretations of the pathological phenotype is that the long-range connections would be more vulnerable than the short-range ones to the pathological genes (because of difficulties in axonal transport?) thus introducing selective alterations of the behavior that identify the disease.

Further directions for drug design based on hierarchical gene–TF-network analysis

The novel strategy we proposed for designing drugs for ASDs and schizophrenia relies on synthesis of compounds able to block (or enhance) in a steric (or allosteric) manner (Changeux 2012) the activity of a given TF within the hierarchical network described above (Fig. 8a). The following example illustrates how the information obtained from the hierarchical TF-network analysis can be used for concrete drug design. Table 1 shows the TF E2A, which belongs to the CGG zone 3. The genes in this zone are expressed mostly at earlier stages of the prenatal period. E2A is known to form heterodimers. Protein-binding databases (Prieto & Las Rivas 2006) offer a number of possible TF partners to E2A. Figure 8b shows a fragment of the connectome of this TF, one that includes achaete-scute homolog 1 (ASCL1). Its heterodimer with E2A is known to activate expression of Stathmin1 gene (STMN1), which causes abnormal axonal arborizations (Yamada et al. 2010), a defect related to schizophrenia. In this case the design of a pharmacological agent that inhibits the function of such a dimer offers a plausible strategy to treat schizophrenia. One possibility could be to synthesize a compound that mimics one of the members of the contact pairs, preventing or facilitating the other partner from the proper association (Kouznetsova et al. 2011; Tsigelny et al. 2011). Another approach could be to select possible attachment points for binding a compound acting as allosteric (negative–positive) modulator of the heterodimer structure, thus blocking or enhancing its proper function. These strategies can be experimentally tested, in a first instance, on mouse models, which carry mutations homologous, for instance, to those identified in ASD patients and display distinct neural and behavioral phenotypes (Ey et al. 2011). In animal studies, the drugs targeted to a particular TF shall be tested at defined periods of pre- and postnatal development and their efficacy and timing of application shall be evaluated by following the behavioral (such as hyperactivity, reduced social interaction and ultrasonic communication (Schmeisser et al. 2012; Jamain et al. 2008) and neurobiological phenotypes. If successful, the strategy shall eventually be applied to human patients.

image

Figure 8. Hierarchy of CGGs – TFs and a fragment of E2A connectome. (a) Diagram illustrating the hierarchy of TFs and CGGs networks and the novel strategy of drug design based on hierarchical gene–TF network analysis: blue − schizophrenia-related and red − autism-related CGGs and TFs. Some of CGGs and TFs are common for both disorders. Some of them are unique for each disorder. Drug can be administered at different level of hierarchy and delivered either to a set of possible targets or to the selected CGG. (b) Fragment of the connectome for the transcription factor E2A using protein-binding databases.

Download figure to PowerPoint

Also new diagnostic techniques for schizophrenia and ASD could be imagined such as the transient expression of given TFs possibly concerned by the disease using in vivo (before the expression of the behavioral phenotype) brain-imaging techniques like PET with specific radioligands (Gulyas & Halldin 2012). Moreover a close analysis of the transcriptional hierarchy may eventually show unexpected branches of the TF expressions tree in peripheral tissues thus offering biological assays that would complement the currently used genetic and/or behavioral methods.

Conclusions

  1. Top of page
  2. Abstract
  3. Materials and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. References
  8. Supporting Information

We have adopted a novel strategy, different from the one gene–one phenotype approach, to describe and understand the abundant data available on brain gene expression in simple terms. It was based on the concept of CGGs (Tsigelny et al. 2005b, 2008), which elaborated a hierarchical model of gene-expression patterns during brain development. A convenient method to separate the elementary bricks from the data of the entire genome expression is the SOM approach, already successfully used to describe kidney development (Tsigelny et al. 2008) and transient embryonic zones of the CTX (Ayoub et al. 2011). Analysis of the informational entropy of SOMs during development showed a significant correlation with experimental results in cortical neurogenesis. SOMs offer a convenient paradigm to account for gene parsimony (Changeux 1983; Konopka et al. 2012) on the basis of the combinatorics of a relatively small number of genes. Elucidation of the signaling pathways and processes related to the different stages of development was achieved by selecting an optimal SOM size. The strategy underlines the different timing of gene-expression patterns in relation to their contribution to neurogenesis and gliogenesis, neuronal differentiation, synaptogenesis and maturation and shows striking differences between CTX, HPC and HYP. It further shows the constant intricacies between genes and environment during postnatal development, extension of which from primitive mammals through chimpanzees and up to humans suggests that a delayed and enriched expression of synaptome genes in humans compared with rat (Bayès et al. 2011; Liu et al. 2012; Konopka et al. 2012) is plausibly associated with cognitive development and an interaction with the social and cultural environment (Changeux 1983; Dehaene & Changeux 2011). Exploring the networks of CGGs and TFs related to psychotic disorders, we found that CGGs form specified zones that clearly match the signaling pathways and their timing.

One of the most original features of our SOM and CGG–TF analyses is the attempt to identify the underlying hierarchical tree of gene–TF networks that determines the patterns of genes expressed during brain development. Some master TFs at the hierarchy's top level are connected to a maximal number of CGGs, and a steadily decreasing number of connections down to bottom levels are responsible for the precise tuning of development-related gene expression and signaling. This scheme explains the frequent pleiotropic effect of these genes (Sebat et al. 2009) and further imposes constraints upon development. For instance, some TFs active as heterodimers often have one dimer partner member participating at almost all levels of the hierarchy while the other member is active only at the specified levels. In other words, ‘partners make patterns’ (Kerszberg & Changeux 1994b). Most importantly, enhanced or decreased activity of a few master TFs in this hierarchical scheme could account for the paradox of non-linear evolution of gene vs. neural complexity, particularly flagrant in the case of human evolution (Changeux 1983; Liu et al. 2012).

Our analysis further showed common or related CGGs in ASDs and schizophrenia with gene–TF networks exhibiting two maxima of expression, one in the prenatal period and the other at early stages of postnatal development. These results support the view that these disorders are developmental and start in the prenatal period, further develop in the postnatal period and subsequently reach the neural and behavioral phenotypes of the disorder (Insel 2010, and references therein), with different sets of genes controlling each period. In zones 1–3 (from 35 TFs selected as controlling CGGs), 21 TFs were related exclusively to schizophrenia, 5 exclusively to ASDs and 9 were common to both (these numbers may change with the discovery of new relations for novel specific genes).

To uncover new therapeutic targets for drug design for these diseases, we propose a strategy based on the pattern of expression of the concerned TFs and their development timing. Orthosteric or allosteric ligands targeted to specific TF oligomers may interfere with, and even restore, pathological evolution of mutated TFs in genetically predisposed patients when administered at relevant sensitive stages of brain development. This strategy takes into account TFs' genetic diversity, which can be obviated by their inclusion in common branches of the TF developmental tree. The practical significance of our strategy is that it will show new pharmacological agents strikingly different from those commonly used for psychotic patients.

References

  1. Top of page
  2. Abstract
  3. Materials and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. References
  8. Supporting Information
  • Avasarala, J.R., Chittur, S.V., George, A.D. & Tine, J.A. (2008) Microarray analysis in B cells among siblings with/without MS–role for transcription factor TCF2. BMC Med Genomics 1, 2.
  • Ayoub, A.E., Oh, S., Xie, Y., Leng, J., Cotney, J., Dominguez, M.D., Noonan, J.P. & Rakic, P. (2011) Transcriptional programs in transient embryonic zones of the cerebral cortex defined by high-resolution mRNA-sequencing. Proc Natl Acad Sci U S A 108, 1495014955.
  • Baitaluk, M. & Ponomarenko, J. (2010) Semantic integration of data on transcriptional regulation. Bioinformatics 26, 16511661.
  • Baitaluk, M., Qian, X., Godbole, S., Raval, A., Ray, A. & Gupta, A. (2006a) PathSys: integrating molecular interaction graphs for systems biology. BMC Bioinformatics 7, 55.
  • Baitaluk, M., Sedova, M., Ray, A. & Gupta, A. (2006b) BiologicalNetworks: visualization and analysis tool for systems biology. Nucleic Acids Res 34, W466W471.
  • Baitaluk, M., Kozhenkov, S., Dubinina, Y. & Ponomarenko, J. (2012) IntegromeDB: an integral system and biological search engine. BMC Genomics 13, 35.
  • Bandeira, F., Lent, R. & Herculano-Houzel, S. (2009) Changing numbers of neuronal and non-neuronal cells underlie postnatal brain growth in the rat. Proc Natl Acad Sci U S A 106, 1410814113.
  • Bayès, A., van de Lagemaat, L.N., Collins, M.O., Croning, M.D., Whittle, I.R., Choudhary, J.S. & Grant, S.G. (2011) Characterization of the proteome, diseases and evolution of the human postsynaptic density. Nat Neurosci 14, 1921.
  • Ben-David, E. & Shifman, S. (2012) Networks of neuronal genes affected by common and rare variants in Autism Spectrum Disorders. PLoS Genet 8: e1002556.
  • Benoit, P. & Changeux, J.-P. (1975) Consequences of tenotomy on the evolution of multiinnervation in developing rat soleus muscle. Brain Res 99, 354358.
  • Benoit, P. & Changeux, J.-P. (1978) Consequences of blocking the nerve with a local anaesthetic on the evolution of multiinnervation at the regenerating neuromuscular junction of the rat. Brain Res 149, 8996.
  • Britten, R.J. & Davidson, E.H. (1969) Gene regulation for higher cells: a theory. Science 165, 349357.
  • Brunet, A., Datta, S.R. & Greenberg, M.E. (2001) Transcription-dependent and -independent control of neuronal survival by the PI3K–Akt signaling pathway. Curr Opin Neurobiol 11, 297305.
  • Brzózka, M.M., Radyushkin, K., Wichert, S.P., Ehrenreich, H. & Rossner, M.J. (2006) Cognitive and sensorimotor gating impairments in transgenic mice overexpressing the schizophrenia susceptibility geneTcf4 in the brain. J Neurosci 26, 12003120013.
  • Buckingham, M. & Relaix, F. (2007) The role of Pax genes in the development of tissues and organs: Pax3 and Pax7 regulate muscle progenitor cell functions. Annu Rev Cell Dev Biol 23, 645673.
  • Castro-Caldas, A., Petersson, K.M., Reis, A., Stone-Elander, S. & Ingvar, M. (1998) The illiterate brain. Learning to read and write during childhood influences the functional organization of the adult brain. Brain (Pt 6), 10531063.
  • Chang, L.W., Fontaine, B.R., Stormo, G.D. & Nagarajan, R. (2007) PAP: A comprehensive workbench for mammalian transcriptional regulatory sequence analysis. Nucleic Acids Res 35, W238W244.
  • Changeux, J.-P. (1973) Cholinergic receptors in electrophorus. Neurosci Res Program Bull 11, 246252.
  • Changeux, J-P. (1983) L'homme neuronal. Fayard Paris. (English edition: Changeux, J-P. Neuronal Man: The Biology of Mind. Pantheon Books, New York.
  • Changeux, J.-P. (2004) The Physiology of Truth: Neuroscience and Human Knowledge (Mind/Brain/Behavior Initiative). Belknap Press of Harvard University Press, Boston.
  • Changeux, J.-P. (2012) The Good, the True, and the Beautiful: A Neuronal Approach. Yale University Press, New Haven.
  • Changeux, J.-P. & Danchin, A. (1976) Selective stabilisation of developing synapses as a mechanism for the specification of neuronal networks. Nature 264, 705712.
  • Changeux, J.-P., Courrège, P. & Danchin, A. (1973) A theory of the epigenesis of neuronal networks by selective stabilization of synapses. Proc Natl Acad Sci U S A 70, 29742978.
  • Dai, M., Wam, P., Boyd, A.D., Kostov, G., Athey, B., Jones, E.G., Bunney, W.E., Myers, R.M., Speed, T.P., Akil, H., Watson, S.J. & Meng, F. (2005) Evolving gene/transcript definitions significantly alter the interpretation of GeneChip data. Nucleic Acids Res 32, e175.
  • Davidson, E.H. (2010) Emerging properties of animal gene regulatory networks. Nature 468, 911920.
  • Davies, E. (2005) Machine Vision: Theory, Algorithms and Practicalities. Elsevier Inc., Morgan Kaufmann Publishers, San Francisco.
  • Dehaene, S. & Changeux, J.-P. (2011) Experimental and theoretical approaches to conscious processing. Neuron 70, 200227.
  • Dehaene, S., Pegado, F., Braga, L.W., Ventura, P., Nunes Filho, G., Jobert, A., Dehaene-Lambertz, G., Kolinsky, R., Morais, J. & Cohen, L. (2010) How learning to read changes the cortical networks for vision and language. Science 330, 13591364.
  • Driever, W. & Nüsslein-Volhard, C. (1998) A gradient of bicoid protein in Drosophila embryos. Cell 54, 8393.
  • Ebersberger, I., Metzler, D., Schwarz, C. & Pääbo, S. (2002) Genomewide comparison of DNA sequences between humans and chimpanzees. Am J Hum Genet 70, 14901497.
  • Eichler, G.S., Huang, S. & Ingber, D.E. (2003) Gene Expression Dynamics Inspector (GEDI): for integrative analysis of expression profiles. Bioinformatics 19, 23212322.
  • Ey, E., Leblond, C.S. & Bourgeron, T. (2011) Behavioral profiles of mouse models for autism spectrum disorders. Autism Res 4, 516.
  • de Felipe, J. (2010) From the connectome to the synaptome: an epic love story. Science 330, 198201.
  • Fessele, S., Maier, H., Zischek, C., Nelson, P.J. & Werner, T. (2002) Regulatory context is a crucial part of gene function. Trends Genet 18, 6063.
  • Flavell, S.W., Cowan, C.W., Kim, T.K., Gree, P.L., Lin, Y., Paradis, S., Griffith, E.C., Hu, L.S., Chen, C. & Greenberg, M.E. (2006) Activity-dependent regulation of MEF2 transcription factors suppresses excitatory synapse number. Science 311, 10081012.
  • Fu, X., Giavalisco, P., Liu, X., Catchpole, G., Fu, N., Ning, Z.B., Guo, S., Yan, Z., Somel, M., Pääbo, S., Zeng, R., Willmitzer, L. & Khaitovich, P. (2011) Rapid metabolic evolution in human prefrontal cortex. Proc Natl Acad Sci U S A 108, 61816186.
  • Furtak, S.C., Moyer, J.R. & Brown, T.H. (2007) Morphology and ontogeny of rat perirhinal cortical neurons. J Comp Neurol 505, 493510.
  • Girirajan, S., Campbell, C.D. & Eichler, E.E. (2011) Human copy number variation and complex genetic disease. Annu Rev Genet 45, 203226.
  • Greene, M.E., Mundschau, G., Wechsle, J., McDevitt, M., Gamis, A., Karp, J., Gurbuxani, S., Arceci, R. & Crispino, J.D. (2003) Mutations in GATA1 in both transient myeloproliferative disorder and acute megakaryoblastic leukemia of Down syndrome. Blood Cells Mol Dis 31, 351356.
  • Gulyás, B. & Halldin, C. (2012) New PET radiopharmaceuticals beyond FDG for brain tumor imaging. Q J Nucl Med Mol Imaging 56, 173190.
  • Guo, Y., Eichler, G.S., Feng, Y., Ingber, D.E. & Huang, S. (2006) Towards a holistic, yet gene-centered analysis of gene expression profiles: a case study of human lung cancers. J Biomed Biotechnol 2006, 111.
  • Halfon, M.S., Grad, Y., Church, G.M. & Michelson, A.M. (2002) Computation-based discovery of related transcriptional regulatory modules and motifs using an experimentally validated combinatorial model. Genome Res 12, 10191028.
  • Insel, T.R. (2010) Rethinking schizophrenia. Nature 468, 187193.
  • Jacob, F. & Monod, J. (1961) Genetic regulatory mechanisms in the synthesis of proteins. J Mol Biol 3, 318356.
  • Jamain, S., Radyushkin, K., Hammerschmidt, K., Granon, S., Boretius, S., Varoqueaux, F., Ramanantsoa, N., Gallego, J., Ronnenberg, A., Winter, D., Frahm, J., Fischer, J., Bourgeron, T., Ehrenreich, H. & Brose, N. (2008) Reduced social interaction and ultrasonic communication in a mouse model of monogenic heritable autism. Proc Natl Acad Sci U S A 105, 17101715.
  • Jaźwińska, A., Badakov, R. & Keating, M.T. (2007) Activin-betaA signaling is required for zebrafish fin regeneration. Curr Biol 17: 13901395.
  • Jiang, J. & Levine, M. (1993) Binding affinities and cooperative interactions with bHLH activators delimit threshold responses to the dorsal gradient morphogen. Cell 72, 741752.
  • Jiang, Z., Tang, H., Ventura, M., Cardone, M.F., Marques-Bonet, T., She, X., Pevzne, P.A. & Eichler, E.E. (2007) Ancestral reconstruction of segmental duplications reveals punctuated cores of human genome evolution. Nat Genet 39, 13611368.
  • Johnson, M.H. (2011) Developmental Cognitive Neuroscience, 3rd edn. Wiley-Blackwell, Hoboken, New Jersey, USA, pp. 308.
  • Kang, H.Y., Kawasawa, Y.I., Cheng, F. et al. (2011) Spacio-temporal transcriptome of the human brain. Nature 478, 483489.
  • Kapur, J.N., Sahoo, P.K. & Wong, A.C. (1985) A new method for gray-level picture thresholding using the entropy of the histogram. Comput Vision Graph 29, 273285.
  • Karlsgodt, K.H., Sun, D., Jimenez, A.M., Lutkenhoff, E.S., Willhite, R., van Erp, T.G. & Cannon, T.D. (2008) Developmental disruptions in neural connectivity in the pathophysiology of schizophrenia. Dev Psychopathol 20, 12971327.
  • Karlsgodt, K.H., Jacobson, S.C., Seal, M. & Fusar-Poli, P. (2012) The relationship of developmental changes in white matter to the onset of psychosis. Curr Pharm Des 18, 422433.
  • Kerszberg, M. & Changeux, J.-P. (1994a) A model for reading morphogenetic gradients: autocatalysis and competition at the gene level. Proc Natl Acad Sci U S A 91, 58235827.
  • Kerszberg, M. & Changeux, J.-P. (1994b) Partners make patterns in morphogenesis. Curr Biol 4, 10461047.
  • Kerszberg, M. & Changeux, J.-P. (1998) A simple molecular model of neurulation. Bioessays 20, 758770.
  • Kirov, G., Pocklington, A.J., Holmans, P. et al. (2012) De novo CNV analysis implicates specific abnormalities of postsynaptic signalling complexes in the pathogenesis of schizophrenia. Mol Psychiatry 17, 142153.
  • Kohonen, T. (2000) Self-Organized Maps, 3 edn. Springler, Berlin.
  • Konopka, G., Friedrich, T., Davis-Turak, J., Winden, K., Oldham, M.C., Gao, F., Chen, L., Wang, G.-Z., Luo, R., Preuss, T.M. & Geschwind, D.H. (2012) Human-specific transcriptional networks in the brain. Neuron 75, 601617.
  • Kouznetsova, V.L. & Rakov, M.A. (1987) Self-Organization in Technical Systems. Naukova Dumka, Kiev (in Russian).
  • Kouznetsova, V.L., Tsigelny, I.F., Nagle, M.A. & Nigam, S.K. (2011) Elucidation of common pharmacophores from analysis of targeted metabolites transported by the multispecific drug transporter-Organic anion transporter1 (Oat1). Bioorg Med Chem 19, 33203340.
  • Kozhenkov, S., Dubinina, Y., Sedova, M., Gupta, A., Ponomarenko, J. & Baitaluk, M. (2010) BiologicalNetworks 2.0–an integrative view of genome biology data. BMC Bioinformatics 11, 610.
  • Kwon, E., Wang, W. & Tsai, L.-H. (2011) Validation of schizophrenia-associated genes CSMD1, C10orf26, CACNA1C and TCF4 as miR-137 targets. Mol Psychiatry 20, 12.
  • Lagercrantz, H. (2009) The birth of consciousness. Early Hum Dev 85 (Suppl. 10), S57S58.
  • Lagercrantz, H. (2012) Are cardiorespiratory complications a question of epigenetics? Proc Natl Acad Sci U S A 109, 21922193.
  • Lagercrantz, H., Nilsson, E., Redham, I. & Hjemdahl, P. (1986) Plasma catecholamines following nursing procedures in a neonatal ward. Early Hum Dev 14, 6165.
  • Lagercrantz, H., Pequignot, J., Pequignot, J.M. & Peyrin, L. (1992) The first breaths of air stimulate noradrenaline turnover in the brain of the newborn rat. Acta Physiol Scand 144, 433438.
  • Lagercrantz, H., Hanson, M.A., Ment, L.R. & Peebles, D.M. (2010) The newborn brain, 2 edn. Cambridge University Press, Cambridge.
  • Larson, E.B., Akkentli, F., Edwards, S., Graham, D.L., Simmons, D.L., Alibhai, I.N., Nestler, E.J. & Self, D.W. (2010) Striatal regulation of ΔFosB, FosB, and cFos during cocaine self-administration and withdrawal. J Neurochem 115, 112122.
  • Le Bras, B., Barallobre, M.J., Homman-Ludiye, J., Ny, A., Wyns, S., Tammela, T., Haiko, P., Karkkainen, M.J., Yuan, L., Muriel, M.P., Chatzopoulou, E., Bréant, C., Zalc, B., Carmeliet, P., Alitalo, K., Eichmann, A. & Thomas, J.L. (2006) VEGF-C is a trophic factor for neural progenitors in the vertebrate embryonic brain. Nat Neurosci 9, 340348.
  • Li, E. & Davidson, E.H. (2009) Building developmental gene regulatory networks. Birth Defects Res C Embryo Today 87, 123130.
  • Li, L., Li, Q., Rohlin, L., Kim, U.M., Salmon, K., Rejtar, T., Gunsalus, R.P., Karger, B.L. & Ferry, J.G. (2007) Quantitative proteomic and microarray analysis of the archaeon Methanosarcina acetivorans grown with acetate versus methanol. J Proteome Res 6, 759771.
  • Liu, X., Bushnell, D.A., Wang, D., Calero, G. & Kornberg, R.D. (2010) Structure of an RNA polymerase II-TFIIB complex and the transcription initiation mechanism. Science 327, 206209.
  • Liu, X., Somel, M., Tang, L., Yan, Z., Jiang, X., Guo, S., Yuan, Y., He, L., Oleksiak, A., Zhang, Y., Li, N., Hu, Y., Chen, W., Qiu, Z., Pääbo, S. & Khaitovich, P. (2012) Extension of cortical synaptic development distinguishes humans from chimpanzees and macaques. Genome Res 22, 611622.
  • Luo, L. & O'Leary, D.D. (2005) Axon retraction and degeneration in development and disease. Annu Rev Neurosci 28, 127156.
  • Mannervik, M., Nibu, Y., Zhang, H. & Levine, M. (1999) Transcriptional coregulators in development. Science 284, 606609.
  • Monod, J. & Jacob, F. (1961) Teleonomic mechanisms in cellular metabolism, growth, and differentiation. Cold Spring Harb Symp Quant Biol 26, 389401.
  • Narayan, S., Tang, B., Head, S.R., Gilmartin, T.J., Sutcliffe, J.G., Dean, B. & Thomas, E.A. (2008) Molecular profiles of schizophrenia in the CNS at different stages of illness. Brain Res 1239, 235248.
  • van Neerven, S., Kampmann, E. & Mey, J. (2008) RAR/RXR and PPAR/RXR signaling in neurological and psychiatric diseases. Prog Neurobiol 85, 433451.
  • Noll, M. (1993) Evolution and role of Pax genes. Curr Opin Genet Dev 3, 595605.
  • Pennypacker, K.R. (1995) AP-1 transcription factor complexes in CNS disorders and development. J Fla Med Assoc 82, 551554.
  • Petanjek, Z., Judaš, M., Šimic, G., Rasin, M.R., Uylings, H.B., Rakic, P. & Kostovic, I. (2011) Extraordinary neoteny of synaptic spines in the human prefrontal cortex. Proc Natl Acad Sci U S A 108, 1328113286.
  • Prieto, C. & De Las Rivas, J. (2006) APID: Agile Protein Interaction DataAnalyzer. Nucleic Acids Res 34, W298W302.
  • Pugliese, L., Catani, M., Ameis, S., Dell'Acqua, F., Thiebaut de Schotten, M., Murphy, C., Robertson, D., Deeley, Q., Daly, E. & Murphy, D.G. (2009) The anatomy of extended limbic pathways in Asperger syndrome: a preliminary diffusion tensor imaging tractography study. Neuroimage 47, 427434.
  • Purves, D. & Lichtman, J.W. (1980) Elimination of synapses in the developing nervous system. Science 210, 153157.
  • Qiu, P., Qin, L., Sorrentino, R.P., Greene, J.R., Wang, L. & Partridge, N.C. (2003) Comparative promoter analysis and its application in analysis of PTH-regulated gene expression. J Mol Biol 326, 13273136.
  • Rakic, P. (2006) A century of progress in corticoneurogenesis: from silver impregnation to genetic engineering. Cereb Cortex 16 (Suppl. 1), i3i17.
  • Rice, D. & Barone, S. (2000) Critical periods of vulnerability for the developing nervous system: evidence from humans and animal models. Environ Health Perspect 108, 511533.
  • Ripke, S., Sanders, A.R., Kendler, K.S., Levinson, D.F., Sklar, P., Holmans, P.A. et al. (2011) Genome-wide association study identifies five new schizophrenia loci. Nat Genet 43, 969976.
  • Saeed, A.I., Bhagabati, N.K., Braisted, J.C., Liang, W., Sharov, V., Howe, E.A., Li, J., Thiagarajan, M., White, J.A. & Quackenbush, J. (2006) TM4 microarray software suite. Methods Enzymol 411, 134193.
  • Sarkar, S., Kalia, V., Haining, W.N., Konieczny, B.T., Subramaniam, S. & Ahmed, R. (2008) Functional and genomic profiling of effector CD8 T cell subsets with distinct memory fates. J Exp Med 205, 625640.
  • Schmeisser, M.J., Ey, E., Wegener, S. et al. (2012) Autistic-like behaviours and hyperactivity in mice lacking ProSAP1/Shank2. Nature 486, 256260.
  • Sebat, J., Levy, D.L. & McCarthy, S.E. (2009) Rare structural variants in schizophrenia: one disorder, multiple mutations; one mutation, multiple disorders. Trends Genet 25, 528535.
  • Segal, E., Shapira, M., Regev, A., Pe'er, D., Botstein, D., Koller, D. & Friedman, N. (2003) Module networks: identifying regulatory modules and their condition specific regulators from gene expression data. Nat Genet 34, 166176.
  • Shannon, C.E. (1948) A mathematical theory of communication. AT&T Tech J (Bell Syst Tech J) 27, 379423 623–656.
  • Shibata, K., Ishimura, A. & Maéno, M. (1998) GATA-1 inhibits the formation of notochord and neural tissue in Xenopus embryo. Biochem Biophys Res Commun 252, 241248.
  • Sieniutycz, S. & Jeżowski, J. (2009) Energy Optimization in Process Systems E-book. Elsevier, Boston.
  • Somel, M., Franz, H., Yan, Z., Lorenc, A., Guo, S., Giger, T., Kelso, J., Nickel, B., Dannemann, M., Bahn, S., Webster, M.J., Weickert, C.S., Lachmann, M., Pääbo, S. & Khaitovich, P. (2009) Transcriptional neoteny in the human brain. Proc Natl Acad Sci U S A 106, 57435748.
  • Somel, M., Guo, S., Fu, N., Yan, Z., Hu, H.Y., Xu, Y., Yuan, Y., Ning, Z., Hu, Y., Menzel, C., Hu, H., Lachmann, M., Zeng, R., Chen, W. & Khaitovich, P. (2010) MicroRNA, mRNA, and protein expression link development and aging in human and macaque brain. Genome Res 20, 12071218.
  • Stead, J.D., Neal, C., Meng, F., Wang, Y., Evans, S., Vazquez, D.M., Watson, S.J. & Akil, H. (2006) Transcriptional profiling of the developing rat brain reveals that the most dramatic regional differentiation in gene expression occurs postpartum. J Neurosci 26, 345353.
  • Stretavan, D.W., Shatz, C.J. & Stryker, M.P. (1988) Modification of retinal ganglion cell axon morphology by prenatal infusion of tetrodotoxin. Nature 336, 468471.
  • Thayer, M.J., Tapscott, S.J., Davis, R.L., Wright, W.E., Lassar, A.B. & Weintraub, H. (1989) Positive autoregulation of the myogenic determination gene MyoD1. Cell 58, 241248.
  • Torkamani, A., Dean, B., Schork, N.J. & Thomas, E.A. (2010) Coexpression network analysis of neural tissue reveals perturbations in developmental processes in schizophrenia. Genome Res 20, 403412.
  • Tsallis, C. (1988) Possible generalization of Boltzmann–Gibbs statistics. J Stat Phys 52, 479487.
  • Tsigelny, I., Burton, D.W., Sharikov, Y., Hastings, R.H. & Deftos, L.J. (2005a) Coherent expression chromosome cluster analysis reveals differential regulatory functions of amino-terminal and distal parathyroid hormone-related protein domains in prostate carcinoma. J Biomed Biotechnol 4, 353363.
  • Tsigelny, I., Hotchko, M., Yuan, J.X. & Keller, S.H. (2005b) Identification of molecular determinants that modulate trafficking of DeltaF508 CFTR, the mutant ABC transporter associated with cystic fibrosis. Cell Biochem Biophys 42, 4153.
  • Tsigelny, I.F., Kouznetsova, V.L., Sweeney, D.E., Wu, W., Bush, K.T. & Nigam, S.K. (2008) Analysis of metagene portraits reveals distinct transitions during kidney organogenesis. Sci Signal 1, ra16.
  • Tsigelny, I.F., Kovalskyy, D., Kouznetsova, V.L., Balinskyi, O., Sharikov, Y., Bhatnagar, V. & Nigam, S.K. (2011) Conformational changes of the multispecific transporter organic anion transporter 1 (OAT1/SLC22A6) suggests a molecular mechanism for initial stages of drug and metabolite transport. Cell Biochem Biophys 61, 251259.
  • Ueno, M., Tomita, S., Ueki, M., Iwanaga, Y., Huang, C.-I., Onodera, M., Maekava, N., Gonzalez, F.J. & Sakamoto, H. (2006) Two pathways of apoptosis are simultaneously induced in the embryonal brains of neural cell-specific HIF-1α-deficient mice. Histochem Cell Biol 125, 535544.
  • Walther, C. & Gruss P. (1991) Pax-6, a murine paired box gene, is expressed in the developing CNS. Development 113, 14351449.
  • Wei, K., Chen, J., Akrami, K., Galbraith, G.C., Lopez, I.A. & Chen, F. (2007) Neural crest cell deficiency of c-myc causes skull and hearing defects. Genesis 45, 382390.
  • Wolpert, L. (1969) Positional information and the spatial pattern of cellular differentiation. J Theor Biol 25, 147.
  • Yamada, K., Matsuzaki, S., Hattori, T., Kuwahara, R., Taniguchi, M., Hashimoto, H., Shintani, N., Baba, A., Kumamoto, N., Yamada, K., Yoshikawa, T., Katayama, T. & Tohyama, M. (2010) Increased stathmin1 expression in the dentate gyrus of mice causes abnormal axonal arborizations. PLoS One 5, e8596.
  • Yu, Y., Platoshyn, O., Safrina, O., Tsigelny, I., Yuan, J.X. & Keller, S.H. (2007) Cystic fibrosis transmembrane conductance regulator (CFTR) functionality is dependent on coatomer protein I (COPI). Biol Cell 99, 433444.

Supporting Information

  1. Top of page
  2. Abstract
  3. Materials and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. References
  8. Supporting Information
FilenameFormatSizeDescription
gbb12005-sup-0001-Figure S1.pptPowerPoint presentation912KFigure S1: Strategy used for a hierarchical CGG modeling of brain development to aid design of drugs to treat autism and schizophrenia. (a) Input microarray data analysis and SOM clustering. (b) Extraction of differentially expressed genes from SOM CGGs. (c) Information Entropy during brain growth and development was calculated as a value reciprocal to system's organization level. (d) Enrichment of CGGs in conventional signaling pathways and processed controling cerebral cortex development. (e) Modular gene-transcription factor regulatory network. (f) Functional/disease assessment of modules and regulating TFs involved in neurodisorders. (g) Future pharmacophore development and drug design based on hierarchical modular TF–gene network analysis.
gbb12005-sup-0002-Figure S2.pptPowerPoint presentation251KFigure S2: Explanation of self-organized map (SOM) construction: the first-level SOM. Genes with the similar profiles are mapped in the same pixel of the SOM, creating CGGs. Genes with profiles that are close enough but differ by some margin are mapped in adjacent pixels. The four CGGs circled on the SOM (with SOM coordinates [10,16], [11,16], [10,17] and [11,17]), a close-up of which is shown in the center of the figure, have profiles similar in appearance to those seen on the charts for each CGG. The CGG's colors represent the average expression value of each CGG for a different time point (day P07 is shown). Each CGG contains different numbers of genes, and its intensity is calculated as the centroid C of a finite set of k points x1, x2, …, xk to find the average of the points of the set (see eqn 1). Then color the span of centroids was calculated and a color code was assigned to each CGG for each time point according to its centroid.
gbb12005-sup-0003-Figure S3.pptPowerPoint presentation234KFigure S3: Explanation of zone selection. Each CGG contained a different number of genes with folds exceeding the 2.5 threshold, ranging from every gene in the CGG to only one. The following percentages of genes within each CGG with an expression over the threshold were considered at length: 100%, 75%, 50%, 33%, 25% and 9%. There are: (1) 46 CGGs containing 447 genes, each of which consists 100% of genes with an expression fold above the threshold, (2) 64 CGGs containing 634 genes, each CGG with 75% of genes above the threshold, (3) 91 CGGs containing 789 genes, each CGG with 50% of genes (394 genes) above the threshold, (4) 112 CGGs containing 835 genes, each CGG with 33% of genes above the threshold, (5) 120 CGGs containing 930 genes, each CGG with 25% of genes above the threshold and 135 CGGs containing 1073 genes, each CGG with only 9% of genes above the threshold; (this is illustrated in the curve of the chart at top left corner of the figure). Because the 50% grouping was located at the inflexion point of the curve, further analysis was completed using this option. Mapping of these 50% CGGs into the SOM created five zones located in its different areas. The resulting figures are summarized in Table S1 and illustrated here. See details in Materials and Methods section.
gbb12005-sup-0004-Figure S4.pptPowerPoint presentation107KFigure S4: Correlations between human and rat cortex development. Using the method described in (Rice & Barone 2000), we created a plot of possible correspondence of the courses of rat and human cortex development. The horizontal axis – days of rat brain development; the vertical – days of human brain development. PC, postcoitum. The red star indicates the day of birth of the human child (approximately 270 days PC) that corresponds to day 47 PC of rat. In different terms we can say that first postnatal day P01 of human corresponds with some level of reliability to postnatal day P28 of rat. Approximately P90 day of human corresponds to the P34 of rat. The program (translatingtime.net gives the limit for rat translating 35 PC days). The rest is our extrapolation.
gbb12005-sup-0005-Figure S5.pptPowerPoint presentation112KFigure S5: Thirty coherent-gene groups SOM used for elucidation of signaling pathways and processes in the studied genes set.
gbb12005-sup-0006-Figure S6.pptPowerPoint presentation234KFigure S6: Canonical signaling pathways activity during development. The plots show the P values supporting the statement that the pathway is activated (−log(P value) = 3 corresponding to the P value = 0.001).The following signaling pathways are selected: (a) Notch, (b) fibroblast growth factor (FGF), (c) apoptosis, (d) neuregulin, (e) Myc-mediated apoptosis, (f) integrin, (g) axonal guidance, (h) NF-kB, (i) IGF1, (j) endothelin-1, (k) glutamate receptor in neurons, (l) GABA-receptor in neurins, (m) CREB in neurons and (n) G-beta-gamma in neurons
gbb12005-sup-0007-Figure S7.pptPowerPoint presentation429KFigure S7: Glutamate metabotropic receptors during cortex development. (a) Expression of different glutamate metabotropic receptors for rat cortex development. Maximum expression occurs on P07 for GRM2, P14 GRM7 and on P21–P90 for GRM2. (b) The gene network involved in GRM3. Courtesy of Ingenuity Systems, Inc., Redwood City, CA, USA.
gbb12005-sup-0008-Figure S8.pptPowerPoint presentation1098KFigure S8: Cortex development. Illustration of TF-gene networks from different zones for different days of development. The example shows that genes and TFs are upregulated in the zone 1 and downregulated in the zone 2 on prenatal day E17, while vicevesrsa on postnatal dayP30.
gbb12005-sup-0009-Figure S9.pptPowerPoint presentation3771KFigure S9: Transcription factor–gene networks for the zone 1. Red – genes (ovals) and TFs (triangles) with increased expression; green – with no significantly increase expression (TFs are colored only when they are present in the initial gene set). (a) The network for the day E16 of development. (b) The network for the day P30. (c) Close-up of the central part of the lower (10th) level TF-gene network for the day E16; highlighted blue are CGGs having two and more schizophrenia-related genes; light-blue – having at least one schizophrenia-related gene; highlighted yellow – schizophrenia-related TFs.
gbb12005-sup-0010-Figure S10.pptPowerPoint presentation646KFigure S10: Close-up of the CGG M27 in zone 1 for day E16 (Fig. S9a) showing that genes have a number of interrelations. Red ovals – upregulated genes, green – not upregulated.
gbb12005-sup-0011-Figure S11.pptPowerPoint presentation528KFigure S11: Gene network connecting the genes within the CGG M27, created using IPA® (Ingenuity Systems, Inc., Santa Clara, CA, USA).
gbb12005-sup-0012-Figure S12.pptPowerPoint presentation1598KFigure S12: Transcription factor–gene networks for zone 2. Red – genes and TFs with increased expression; green – with no significantly increase expression. (TFs are colored only when they are present in the initial gene set; ovals in CGGs – genes: red – upregulatetd, green – not upregulated). (a) The network for the day E16 of development. (b) The network for the day P30 of development.
gbb12005-sup-0013-Figure S13.pptPowerPoint presentation329KFigure S13: Predicted transcription factor binding sites for genes essential in early stages of brain development and evolutionarily conserved in human, mouse and rat; (a) 15-kb proximal promoter region is displayed, exons, introns, the transcription start site, repetitive elements, conserved sequence regions and transcription factor binding sites, are color-coded. (b) Close-up of transcription factor-binding sites on the genes is significant for early stages of brain development.
gbb12005-sup-0014-Figure S14.pptPowerPoint presentation185KFigure S14: Venn diagrams that show relationships between different sets of TFs included in hierarchical gene–TF networks (see examples in Figs 5 and 6) for zones 1–3 (see Fig. 1a and S8). (a) Venn diagram of sets of TFs for schizophrenia and autism; one can see that there are many common TFs involved. (b and c) Venn diagram for TFs derived from three zones on bottom (most comprehensive) level of the hierarchical networks: (b) schizophrenia-related TFs. (c) autism-related. See related tables of genes in Supporting Information.
gbb12005-sup-0015-TableS1.docWord document28KTable S1: Coherent gene group with at least 50% of differentially expressed genes with the folds greater than 2.5 threshold in the selected clusters.
gbb12005-sup-0016-TableS2.docWord document29KTable S2: Number of CGGs with differentially expressed genes (from Table S1) active in specific days of development.
gbb12005-sup-0017-TableS3.docWord document47KTable S3: Interactions between the genes from the CGG M27 in zone 1, SOM coordinates [10,1].
gbb12005-sup-0018-TableS4.docWord document41KTable S4: Transcription factors related to schizophrenia and ASD in zones 1–3 (Fig. 8a)
gbb12005-sup-0019-TableS5.docWord document36KTable S5: TFs related to schizophrenia in zones 1–3 (Fig. 8b)
gbb12005-sup-0020-TableS6.docWord document29KTable S6: TFs related to ASD in zones 1–3 (Fig. 8c)
gbb12005-sup-0021-TableS7.xlsExcel spreadsheet45KTable S7: Functions of the genes in the selected CGGs of cluster 1

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.