The present findings revealed that in the frontopolar cortex of depressed individuals that died by suicide, networks of gene products exist that appear to be dysregulated relative to the non-depressed cohort that died quickly of causes other than suicide. We used two types of analysis to provide insight into how the biology of the depressed suicide brain might be different from normal controls. The first, pathway analysis, provided the gene networks shown in which was based on the “reading” of the scientific literature by the software. The second set of networks (–) was created by an analysis of gene expression correlations in controls and in depressed individuals that died by suicide. These analyses indicated agreement in the biological processes that were implicated as being different in the depressed suicide brain. The potential biological processes that have been implicated for these networks are listed in (and Table S3
). Although the functional relevance of a few processes are obviously difficult to reconcile with known brain functions (myeloid leukocyte differentiation for example), the overwhelming majority are involved in regulatory or developmental process, synaptic communication and cell to cell interactions.
These conclusions are based on a computational analysis which relies on identifying functional relationships that have been established to various degrees of certainty in the biomedical literature. The analysis enables the effective access of 21 million PubMed abstracts and 61 full text journals that cover mammalian biology, a task that is obviously not possible by conventional reading of the scientific literature. Although this analysis, at face value appears useful and exciting, there is, simply by sheer number of possible interactions analyzed, the question of whether or not this analysis is valid. First, some proteins have many functions and will therefore link with many others. Enzymes may have many targets or substrates and so relationships that are reported by this analysis may not be valid within a certain cell type that does not express a target molecule. The analysis is only as good as the present day knowledge of the protein interactions that are reported. There is also no “quality control” for the validity of the data contained in the publications. These limitations make the GO analysis particularly important as it identifies the overall implications of the gene interactions that are being reported.
These caveats notwithstanding, our data/analysis is remarkably consistent in that each GO analysis generated similar functional groups, those involved in cell structure and communication. As well, we showed that the pathway analysis of a random selection of the same number of genes did not provide apparently biologically relevant networks. This conclusion is based on the pathway analysis of the 100 random probe set lists of the same size where we reconstructed relationship networks and then assessed number of proteins and relationships in these networks. A statistical analysis showed that random networks identified by the pathway analysis were significantly smaller and less connected than the networks generated from the experimentally derived DEG list. This indicates that the interrelations found through the pathway analysis likely represent true biological differences between controls and depressed suicide samples. The function of this network is implicated in cell to cell communication as many of the processes suggested are involved in cell adhesion, cell morphology and synapse formation. A similar approach, commonly referred to as “bootstrapping” is a standard statistical procedure in other studies 
but to our knowledge has not been done before in an analysis of this kind.
The pathway analysis also identified five genes that appeared as “network hubs” with connectivity higher than 20: RAC1, CTNNB1, STAT3, EP300 and PTK2. These genes are of interest as they represent the central points in cellular machinery that have relationships with a variety of other proteins. For example, there are more than 1500 relationships currently known for each of the five aforementioned proteins. Importantly, each of these has an established role in nervous system processes, having been implicated in nervous system development, neuronal migration and differentiation, neuroprotection and neuronal plasticity ().
It is also interesting that 11 transcription factors were identified that regulate at least two genes from the initial network. Of these, 8 have well established roles in nervous system functioning. Specifically, SP1, SP3 and RELA are involved in neurite growth, myelination and neuron survival 
, and TCF4 is important for nervous system development, axon morphogenesis and oligodendrocyte differentiation 
. FOXO3 
regulates neural progenitor stem cell proliferation as well as the induction of genomic death responses upon its’ translocation from the cytosol to the nucleus in response to excitotoxic stimuli 
. MeCP2 is one of the central factors involved in gene regulation through differential CpG methylation in various tissues and organs, including the nervous system 
. Finally, CREB1 and EGR1 have well established roles in synaptic plasticity, learning and memory 
We also found 7 genes that were under control of at least 2 transcription factors implicated in the direct relationship network in . Of these, 5 have a well-established role in nervous system functioning. FAS is involved in neuronal development and degeneration 
, CCND1 is important for neuronal cells proliferation 
, and GFAP and ERBB2 are involved in glia and Schwann cell function, as well as synapse formation and maturation 
. Furthermore, VEGFA is involved in angiogenesis and is also implicated in development of amyotrophic lateral sclerosis (ALS) 
. How any of these factors contribute to the underlying depressive behavior is unclear, but the wide ranging effects on the expression of these transcription factors suggest broad disturbances in gene function.
Networks Identified through Correlation Analyses
As we have previously shown in a much more limited analysis, gene expression in MDD/suicide brain seems to be much less coordinated than in control samples 
. In the correlation analyses done here we found 43 correlated genes in the control (non-depressed) class, forming two distinct sub-networks with 134 correlated relationships. In the depressed/suicide condition, by contrast, only 21 proteins were significantly correlated to one another. The loss of coordinated expression seemed to be more profound among genes that were up-regulated in MDD/suicide group. Specifically, in the depressed/suicide PC networks, 12 DEG’s were up-regulated in the depressed/suicide class and 33 were down-regulated (approx. 1
3). In MDD/suicide PC network, in contrast, only 2 DEG’s were up-regulated and 19 were down-regulated (approx. 1
10). This suggests that the down-regulation is “concerted”, whereas genes that were up-regulated in expression appear to do so in a more “random” manner. Overall this analysis suggests that in depressed/suicide individuals there is a wide ranging loss of organized expression among genes that are important for determining the wiring neural networks.
A number of genes were down-regulated in depressed/suicide tissue, but were not present in the network () that nonetheless would be predicted to have wide ranging effects on synaptic function. One of these, GNAS (stimulatory alpha subunit of G protein), is involved in many neurotransmitter signaling cascades, including 5-HT and dopamine receptors. Another is RTN4 or neurite outgrowth inhibitor, a regulator of apoptosis and was implicated in neurodevelopmental processes 
. As well, the expression of synaptosomal-associated protein 25 (SNAP25), which is involved in synaptic vesicle membrane docking and fusion and regulation of neurotransmitter release was reduced. The down-regulation of this protein has also been implicated in several psychiatric disorders 
. In addition, SPARCL1 or hevin is a putative extracellular matrix glycoprotein that binds calcium and plays an important role in the developing nervous system 
. Finally, GLUL or glutamate-ammonia ligase clears L-glutamate, the major neurotransmitter in the central nervous system, from neuronal synapses 
. Overall these changes implicate a profound disturbance in excitatory amino transmission irrespective of any other changes in gene expression within the networks identified. The fact that they appear to be altered in expression but are not represented in the largest network identified () suggests a wider disturbance in the gene expression than was identified by pathway analysis.
One of the limitations of this study is the small n of both the control and sample groups. This was deliberate as we choose to analyze a small number of well-matched samples (similar age RIN sex etc) rather than a larger more variable cohort. For the correlation analysis, where hundreds of correlations were compared, the α error may be potentially high, however, this ought to be comparable in both groups. Thus our results showing many fewer correlations in depressed/suicide cannot be attributed to the number of genes compared. We should also note that some genes that we and others have found to be down-regulated (BDNF, GABRA1 
) were not found in these analyses (although they were near the cut-off used here). This is likely attributed to the fact that although microarray analysis is very reproducible is not very sensitive in detecting small changes in gene expression. This fact can be observed in where a 1.3 fold change in expression detected on the microarray correlated to more than 2 cycles of change in the QPCR data. Thus the differences reported here are likely the largest differences in gene expression between these two groups. Finally, our cohort is also confounded by the fact that the samples come from those who also committed suicide. Thus our findings may be relevant to suicidality and/or depression. Recent studies by Turecki and others have shown that there may be gene expression patterns that are associated with suicidality 
albeit with some overlap in those associated MDD. This overlap is a significant confound in “teasing out” the gene expression that is unique to these two psychiatric states. Studies of gene expression in those with MDD who did not die from suicide found two highly dysregulated genes including stresscopin, a neuropeptide involved in stress responses and Forkhead box D3 (FOXD3), a transcription factor as well as factors related to synapse formation 
. The finding of another Forkhead box transcription factor is similar to the data reported here where we found FOXO3 and many factors related to synaptogenesis/maintenance. This suggests that our data speak more to depressive syndrome rather suicidality.
In summary, the present findings indicate that among depressed individuals that died by suicide, profound alterations exist in the expression of genes that control synaptic function, cell adhesion and cytoarchitecture. They also extend and support our observation 
that coordinated gene expression is apparently disturbed in the MDD/suicide samples in comparison to normal controls. Interestingly, we also found that in mice acute and chronic stressors can also alter coordinated gene expression of the GABAA
receptor gene cassette 
. As stressful events may be a precipitating factor in the development of MDD, it might be important to identify the biochemical and/or epigenetic processes that disturb normal gene expression. These data also provide a number of new targets for interventions that could help treat MDD.