This is the first joint study of whole-genome mRNA and miRNA profiling using individual human brain RNA from autopsies of HAD and HIV non-dementia patients. In this study, we initially compared mRNA and miRNA data at the clustering, gene ontology (including cellular components and biological processes), and pathway levels. Following that, SA-BNs correlating miRNAs and mRNAs by their expression levels were performed to validate the accurate prediction of genes potentially targeted by dysregulated miRNAs.
The clustering and gene ontology results showed excellent functional concordance between mRNA and miRNA, demonstrating the significant involvement of neuronal cellular components and biological processes such as: signal transduction, transcriptional regulation, metabolism, response to stimuli, cell cycle/apoptosis, protein modification, neuronal processes and ion transport, respectively. This intrinsic functional consistency between miRNA and mRNA has given an extra power to our findings in relation to understanding the genomic basis of HIV neuropathogenesis and HIV-mediated neurodegeneration. Moreover, the DE miRNAs were more robust than their mRNA counterparts in providing a comprehensive snapshot of cellular components and biological processes involved in neuropathology and neurodegeneration. Compared to DE miRNAs, DE mRNAs could only predict elemental functional pathways and processes related to neuropathology. DE miRNAs revealed the participation of additional cellular components (eg. ion channel, actin cytoskeleton, tight junction) and biological processes (cell adhesion, tissue/embryonic development, learning), which also concurs with biological processes of mRNA. Interestingly, these findings are consistent with study, which has been done using CSF of HIVE patients . The most plausible explanation for the comprehensiveness of miRNA coverage as compared to their mRNA counterpart is that a single miRNA or the miRNAs belonging to the same family in the cluster can target several hundred genes within a biological process or pathway. Therefore, it is not surprising that miRNA gives broader information compared to mRNA. Secondly, there are many other gene regulational mechanisms apart from miRNA and together they can compensate the effect of miRNA dysregulation to some degree, consequently the miRNA regulation effect can only be seen after certain stage of disease progression when other mechanisms are not sufficient any more .
Pathway analysis of the joint mRNA and miRNA results provided the first in vivo evidence of significant involvement of axon guidance pathway and its downstream signalling pathways (such as MAPK signalling pathway and Ca2+ signalling pathway, etc.) on both transcriptional level and regulation level. Axon guidance pledges precise path finding and defines their termination zones and synaptic partners [42, 43], which is fundamental to neuronal development and networks. In addition, misrouted fibers have been shown in AD and PD’s brains . Furthermore, it is well known that HIV envelope glycoprotein (GP120) can cause axonal degeneration  and recently axon damage has been claimed as a key predictor of outcome in multiple neurological disorders, including HAD . Axon guidance pathway contains four prominent families of ligands (ephrin, netrin, semaphorin and slit proteins), their receptors and downstream signalling proteins. The role of axon guidance pathway molecules in the maintenance and plasticity of neural circuits has been reported . Moreover, the variations (single-nucleotide polymorphisms) in axon guidance pathway genes have been reported to predict PD outcomes . Significantly, 9 of our DE miRNAs have been found targeting this pathway according to TargetScan results; and more importantly these results support previous observations with 3 out of 4 ligands/receptors being dysregulated in our mRNA studies, including ephrin(EPH) receptor A4 (EPHA4), netrin G2 (NTNG2), and semaphoring 3A (SEMA3A), strongly suggesting the impairment of axon guidance pathway in HAD brains in vivo.
Moreover, our results highlight axon guidance downstream signalling pathways, which allow precise patterns of connectivity within the CNS. For instance, the MAPK pathway comes out significant in both our mRNA and miRNA profiling. Studies have shown that the activation of MAPK is necessary for axon guidance , and it contributes to netrin signalling in axon guidance . Besides, netrin-dependent axon outgrowth and orientation can be antagonized by inhibition of ERK-1/2 . The role of MAPK pathway in HIV infection has also been well documented. For instance, it has been reported that the MAPK pathway plays a crucial role in HIV-1 replication and virulence  and is one of the transcriptional signatures in HIV+ long-term non-progressors . In addition, the binding of HIV-1 GP120 to CD4 receptors on T cells can activate the MAPK pathway and induce transcription of cytokine and chemokine genes . Interestingly, MAPK pathway was targeted by 3 DE miRNAs and it includes 11 of our DE genes, such as RPS6KA1, FLNA, RRAS2, and MAP2K4 etc., each of which play an important role in MAPK signalling. MAPK signalling cross-talks with the Jak-STAT signalling pathway at multiple levels. In mammals, the Jak-STAT signalling pathway is the principal signalling mechanism for cytokines and growth factors and therefore plays a key role in cell proliferation, differentiation, cell migration and apoptosis. Supporting this relationship, Jak-STAT was highlighted in our analysis as well. We found 7 genes significantly dysregulated in that category (including STAT) and it was targeted by DE miR-19a. Dysregulation of Jak signalling might result in inflammation , which is commonly accepted as an important mediator in the pathogenesis of neurodegeneration. VEGF signalling pathway is another significant pathway revealed by our results, and it closely links to MAPK signalling pathway as well. Via activating MAPK signalling pathway, VEGF can exert direct effect on multiple types of neuronal cells, including neurons, astrocytes, and microglias . VEGF also has been reported to be involved in vascular permeability  and several studies have shown the potential utility of inhibiting VEGF signalling pathway in reducing BBB disruption [57, 58]. Besides, Ca2+ can mediate guidance receptor signalling in vitro and change in Ca2+ concentration can signal growth cone turning . Equivalently, guidance cues can also trigger Ca2+ influx and alteration in Ca2+ concentration or slope its gradient, thereby influencing the outcome of growth cone behavior (attraction or repulsion) [61, 62]. Our studies have demonstrated several genes related to Ca2+ transport/signalling dysregulated, including ATP2B4, which play a critical role in intracellular calcium homeostasis.
In addition, endocytosis is another critical aspect of guidance receptor activation and signalling . Nine of our DE miRNAs were found targeting this pathway and several key genes were found dysregulated. Efficient cell detachment needs the endocytosis of the ephrin-Eph complex, or even bidirectional endocytosis for ephrinB-EphB-induced repulsive guidance. In addition, endocytosis also plays a role in regulating the senstivity of the growth cone correspondent to a repulsive cue. Again, 9 of 68 of our DE miRNAs targeted endocytosis pathway. Our mRNA study also revealed dysregulation of Ras related protein (RAB4B) and EHD protein (EHD4), which are important components of endocytosis pathway. We also found ADAM22 dysregulated, whose family member ADAM10 has been reported to play a role in converting initial adhesive interaction into repulsion and therefore providing an effective strategy for axon detachment and attenuation of signalling [63, 64].
Further, our miRNA and mRNA Bayesian correlation analysis has provided an unambiguous snapshot of miRNA and mRNA functional interactions and their biological significance. Sophisticated Bayesian Structure learning approach defines miRNA-mRNA interactions based on their relative expression of all of these molecules in each condition. This network-based approach identified these key interactions with very high confidence. These interactions define the network topography that is provided by Bayesian statistics and is substantially more rigorous than individual correlations that can be defined conventionally. These relationships, therefore, are more likely to be meaningful at the system level compared to reporter assay. We identified 195 positive and negative statistically significant correlations (P<0.05) between our DE mRNAs and miRNAs. It strengthens the correlation and shows the credibility of Bayesian analysis we have performed for establishing this intrinsic functional relationship between mRNA and miRNA. Among them, two thirds of functional correlations are from 5 DE miRNAs: miR-137, 153, 218, 376a and 299-5p. They all changed greater than 2 fold. Since all of them have been reported to be involved in classical neurodegeneration, for instance, miR-376a has been reported to mutate in Huntington’s disease (HD) brains ; miR-299-5p has been reported to be dysregulated in multiple sclerosis (MS) brains  and more interestingly, all of them have been shown to be dysregulated in the AD brains . In particular, miR-137, 153 and 218, which can target more than 5 neurodegeneration-related pathways, implying their functional relevance to the observations noted in this study. The miR-137 has been shown to be enriched in neurons, especially within the dentate gyrus and the molecular layer of adult hippocampus  and studies have shown that it plays an important role in modulating neuronal cell proliferation and differentiation [68–71]. Moreover, it has been shown to be genetically associated with schizophrenia . Recently, it has been shown dysregulated in the CSF of HIVE patients , which is consistent with the expression of miRNA-137 in the frontal cortex in our study in HAD patients. In addition, the miR-218 is enriched in hippocampus  and altered miR-218 expression has been reported in HD and MS brains [66, 73]. The miR-153 has been shown to play important role in AD and PD pathogenesis [74, 75]. It can downregulate the expression of APP protein in vivo, suggesting its possible role in AD pathogenesis . Moreover, it can regulate α-synuclein, which is the primary structural components of Lewy bodies, indicating its role in PD pathophysiological process . The genes that correlated to are all involved in several significant pathways discussed above. In particular, SPRED1, MAP2K4 and DIRAS2, they all correlated with 3 out of 3 miRNAs (miR-137, 153 and 218) and they all appear to be involved in MAPK signalling pathway, which strongly indicate the participation of MAPK pathway in HAD pathogenesis and is consistent with our previous proteomic studies .
Although our study is the first comprehensive parallel genome-wide mRNA and miRNA profiling of HIV infected human brains, there are still limitations: 1. In future, a bigger sample size, blood and CSF samples will be needed to further validate these findings, and confirm the clinical value of this findings; 2. These findings are based on the whole human brain cortex rather than specific cell types (eg. neurons) due to lack of plausible and effective methodologies for perfect cell separation. Although Laser Capture Microdissection (LCM) is currently used in studying cell types, there are significant limitations in profiling single cell type : A. Due to the minimum laser spot size (7.5 μm), it poses a limit to the precision of single cell micro-dissection without contaminating fragments of adjacent cells. B. Due to its suboptimal optical resolution on uncovered sections, it will compromise cell borders distinction and result in cytoplasmic compartment loss, which is crucial for our mRNA analysis. Although immunohistochemical (IHC) staining will circumvent this problem to some degree, it is impossible to recover cytoplasmic compartment precisely without contamination. Moreover, IHC procedures, tissue fixation and LCM capturing of cells dramatically affect RNA yield. C. Sectioning will generate large number of attached fragments, which might alter expression profiles greatly. In addition, due to the lack of the cytoplasm or even the nucleus, the genomic information will be considerably compromised.
Overall, our study provides a strong foundation and durable framework for systematic large-scale studies on HIV-infected adult brain to define functional genomic phenotypes of neurodegenerative diseases and functional networks between miRNA and mRNA, which may lead to the development of new generation of prognostic and diagnostic markers and therapeutic intervention strategies for viral and non-viral neurodegenerative diseases.