The pathophysiological process of neurodegenerative disorders such as Alzheimer's disease (AD) begins well before the diagnosis of clinical dementia. AD is characterized pathologically by the presence of insoluble plaques and tangles composed of beta-amyloid (Aβ) formed by sequential amyloid precursor protein (APP) proteolysis and hyperphosphorlyated Tau (pTau) proteins (Hardy and Allsop, 1991; Walsh and Selkoe, 2004; Cole and Vassar, 2008). The accumulation of Aβ has been the major focus of AD research and has been shown to interfere with long term potentiation which is required for neuronal signaling, and is implicated in pro-apoptotic signaling leading to neuronal loss (Chapman et al., 1999; Roth, 2001). The majority of AD patients are asymptomatic during the pre-clinical stages of the pathological process which is believed to be a period of approximately 17 years (Villemagne et al., 2013). Therefore, early diagnosis of AD is required before or during the pre-clinical phase in order that therapeutic intervention, or the use of disease modifying drugs, can be administered.

The main biomarker targets currently employed for AD diagnosis are measurements of Aβ and pTau (Hansson et al., 2010; Prvulovic and Hampel, 2011; Watt et al., 2011; Rosenmann, 2012). MicroRNA (miRNA) are a class of non-coding RNA of approximately 22 nucleotides in length, and are known to regulate post-translational transcription. Expression profiling of miRNA levels represents a new class of potential biomarkers that are currently being investigated for the diagnosis of a number of diseases (Skog et al., 2008; Taylor and Gercel-Taylor, 2008; Baraniskin et al., 2012b; Geekiyanage et al., 2012; Jones et al., 2012). miRNAs are derived from RNA hairpins comprising of precursor miRNA and processed by endoribonucleases (Dicer and Drosha) to form mature miRNA fragment (Krol et al., 2010). The mature miRNA is incorporated into the RNA-induced silencing complex (RISC) which binds to complementary sites in the 3′ untranslated region of their mRNA targets resulting in downregulation of gene expression (He and Hannon, 2004). They can be secreted into biological fluids where they can be detected and profiled using methods including quantitative real-time PCR (qRT-PCR), microarrays, and more recently by deep sequencing technologies.

Here, we review the current literature to highlight the diagnostic potential to screen for neurodegeneration using gene expression profiling in biological fluids. In particular, we have focused on the potential of profiling miRNA expression associated with AD and evaluate the current deep sequencing platforms suitable for biomarker discovery including the implementation into clinical diagnostic laboratories.

It has been revealed that the highest expression of tissue specific miRNA is found in the brain (Babak et al., 2004; Sempere et al., 2004; Schonrock et al., 2010). The significance of miRNA and their conclusive biological functions were gradually discovered using knockout mouse models (Schaefer et al., 2007). For example, Dicer knockout mice have been shown to deregulate miRNA processing, leading to defects in neuronal development and underdevelopment of the brain, demonstrating a role for miRNA in neurogenesis (Schaefer et al., 2007; Kawase-Koga et al., 2009; Huang et al., 2010). The first reports translating early miRNA studies to the brain observed a number of brain enriched miRNA such as miR-9, miR-29a, miR-125, miR-128, miR-134, and miR-137 (Table 1). With respect to AD, a number of deregulated miRNA have been identified to correlate with disease, including miR-9, miR-20a, and miR-132 (Makeyev et al., 2007; Cogswell et al., 2008; Hebert et al., 2009, 2012). While, synthetic miRNA precursors, miR-20a, miR-17-5p, and miR-106b, when co-transfected in HeLa cell lines, inhibited APP protein translation (Hebert et al., 2009). Highly abundant and brain enriched miRNAs found to be deregulated in AD models (human and mice models) are summarized in Table 1.

Extracellular miRNA originating from specific tissues such as the brain and cancerous tissues can be released into biological fluids for example, Cerebral spinal fluid (CSF) (Cogswell et al., 2008; Baraniskin et al., 2012b), blood (Hunter et al., 2008; Baraniskin et al., 2011), saliva (Patel et al., 2011; Ogawa et al., 2013) and urine (Gilad et al., 2008; Qi et al., 2012). Despite the presence of high RNase activity, circulating miRNAs are protected from degradation by binding to RNA binding proteins such as lipoproteins (Arroyo et al., 2011; Vickers and Remaley, 2012) or contained in membrane derived microvesicles, in particular, exosomes (Mitchell et al., 2008; Arroyo et al., 2011). Exosomes are formed within multi-vesicular bodies (MVBs) in the endosomal system, which co-ordinates the transport of cargo between the plasma membrane, trans-Golgi network (TGN) and lysosomal system [reviewed in (Bellingham et al., 2012b)]. Exosomes can function as an intercellular delivery mechanism by sending miRNA cargo to initiate cell-to-cell communication as they shuttle between neighboring and distant cells (Valadi et al., 2007).

Profiles of deregulated miRNA isolated from peripheral blood (Jin et al., 2008) and serum (Skog et al., 2008; Taylor and Gercel-Taylor, 2008; Edbauer et al., 2010) have been demonstrated and suggest they have diagnostic potential for human diseases such as cancer. For example, the pathology of various cancers and miRNA originating from the site of metastasis correlate positively in both plasma and serum (Skog et al., 2008; Tsujiura et al., 2010). The use of miRNA expression levels as biomarkers can be applied to other human diseases and has not been thoroughly investigated in neurodegenerative diseases such as AD. There are some limitations to this approach, in particular the transport of brain specific miRNA through the blood brain barrier (BBB) into the circulation system of which the current mechanism is unknown. The BBB serves as a tight control point that has specialized molecular machinery to regulate the transport of nutrients and macromolecules, while ensuring viruses and bacteria do not cross the barrier (Begley and Brightman, 2003). On the contrary, under normal conditions, Aβ is cleared from the brain and transported across the BBB mediated by low-density lipoprotein receptor-related protein (Tanzi et al., 2004), suggesting the possibility of other neurodegenerative disease markers crossing the BBB.

There are two possible mechanisms by which miRNAs are able to transport through the BBB. Firstly, a number of neurological diseases such as Multiple Sclerosis and meningitis are well known conditions that weaken and eventually disrupt the BBB, consequently permitting a non-specific release of cellular factors and nucleic acids (Begley and Brightman, 2003). Moreover, thinning and perforations of the vascular basement membrane have been observed in post-mortem brains of late-stage AD patients (Blennow et al., 1990; Berzin et al., 2000; Zipser et al., 2007). Secondly, exosomes and microvesicles may play an important role as carriers of miRNA across the endothelial cellular layers of the BBB in order to communicate between the brain and distant organs via biological fluids (Haqqani et al., 2013). A proposed mechanism involves the transcytosis of extracellular vesicles, such as exosomes, across endothelial cells of the BBB by receptor-mediated endocytosis and releasing exosomal contents into circulation. As a result, the contents of the vesicles can be used as biomarkers reflective of the brain (Haqqani et al., 2013). miRNAs found highly abundant in the brain have been detected in human biological fluids such as plasma, urine, and CSF (summarized in Table 1).

Implementing deep sequencing technology represents a powerful and innovative approach to discover differentially expressed miRNAs in neurodegenerative diseases (Brase et al., 2010; Debey-Pascher et al., 2012). The advantage of using deep sequencing, unlike traditional Sanger sequencing, lies upon the capability to simultaneously process millions of independent sequencing events. This offers billions of nucleotide information within a single experiment (Shendure et al., 2004; Church, 2006). Deep sequencing experiments enable comprehensive analyses of large amounts of sequence data, resulting in dramatically accelerated research compared to traditional labor-intensive efforts and is a powerful approach to determine accurate encoded-information from nucleotide fragments (Hall, 2007; Shendure and Ji, 2008; Tucker et al., 2009).

Typically, the data generated from benchtop deep sequencing instruments (e.g., high assembly coverage by 454 GS Junior due to long read length) are generally sufficient to obtain disease specific profiles from individual samples (Loman et al., 2012b). Ultimately, the choice of platform will depend on its performance metrics (i.e., read length, accuracy and data output) to complement the type of study being undertaken. For clinical practice, platforms such as the Ion PGM and MiSeq offer the best value for money, more flexibility, accuracy, adequate throughput and coverage depths for study of miRNA.

To incorporate miRNA deep sequencing for clinical practice, a simple and standardized workflow for the routine biomarker and diagnostic screening needs to be defined (Barzon et al., 2011; Natrajan and Reis, 2011; Radford et al., 2012; Rizzo and Buck, 2012). A typical workflow involved in performing miRNA-sequencing from sample preparation to data analysis is outlined in Figure 1. The sequencing workflow consists of the following steps: (i) miRNA isolation; (ii) Library preparation and enrichment of templates containing size selected libraries of the appropriate fragment length; (iii) and Sequencing reaction. Isolation of low abundance miRNA present in body fluids (e.g., blood and urine) can be obtained by the use of RNA isolation columns that specifically enrich for miRNA (Mitchell et al., 2008; Arroyo et al., 2011). Preparation of isolated miRNA for sequencing includes the assessment of quality (i.e., RNA integrity number; RIN), profiling and quantification of small RNA species, in particular miRNA (Ozsolak and Milos, 2010). Preparation of sample library often requires barcoding, which involves ligation of different adaptors of identifiable sequences to either end of each sample. This method is to increase capacity as it allows multiplexing of samples in the same sequencing reaction (Chen et al., 2012). Template preparation for sequencing and the sequencing reaction is platform specific as seen in Table 2.

Another approach to increase throughput is by capturing and sequencing disease specific targets known as sequence enrichment. In clinical practice, the factors to consider while sequencing for an enriched population of targets may involve the size of the region of interest, capture efficiency of interest region, average read depth coverage of the targeted region, distribution of coverage and its sensitivity and specificity (Mamanova et al., 2010). The quality control involved would require a set of positive and negative control references to establish a standard depth and quality percentage of targeted regions for determining the accuracy of enrichment (Harismendy et al., 2009; Teer et al., 2010). Furthermore, reference controls eliminate any experimental problems which may arise from sequence variations that affect hybridization to biotinylated probes during enrichment, causing low capture efficiency (Schwartz et al., 2011). Hence, quality control measures should be routinely assessed before proceeding to analyse samples for a diagnosis to eliminate false positives which may be caused by poor miRNA quality or technical artefacts (Fehniger et al., 2010; Lee et al., 2010). Furthermore, instruments that provide a fast turn-around time will be suitable in diagnostic laboratories in order to produce diagnostic reports in a timely manner (Tucker et al., 2009). In clinical practice, a high throughput strategy may involve using deep sequencing platforms to screen for AD biomarkers by indexing a large number of patient samples and pooling samples into one sequencing run. Indexing patient samples and batch testing in a diagnostic laboratory would be a cost-effective and practical approach to handle large population screening.

All deep sequencing instruments produce a massive quantity of raw data that requires extensive computational tools to process the information (Zhang et al., 2011). Open-source tools and in-house (Perl) scripts are available for handling large quantities of sequencing data by providing an integrated and streamlined analysis workflow (Nix et al., 2010). A fundamental workflow for post-analytic miRNA data analysis (Figure 1) consists of four fundamental steps: (i) Sequence assessment; (ii) Sequence alignment; (iii) Profiling of miRNA associated with neurodegenerative disease; (iv) and Final selection. This workflow requires databases and tools such as FastQC (quality control check), miRbase/miRandola (miRNA database), bowtie (short read aligner) and EdgeR/DESeq (normalization) (Horner et al., 2010; Kozomara and Griffiths-Jones, 2011; Russo et al., 2012). In order to make data analysis more accessible to the diagnostic end-user, there are a number of commercially available software packages that aim to make analysis uncomplicated by providing user-friendly graphical interfaces (Richter and Sexton, 2009). This includes Partek Genomics Suite, CLC Genomics Workbench, Ingenuity Systems and platform-specific software (e.g., Ion Torrent Suite and Illumina GenomeStudio). These commercial software tools have less flexibility and scalability in terms of their parameters' settings, but are suitable for small-scale studies to establish miRNA profiles and perform differential expression analysis in order to detect disease specific miRNA (Meldrum et al., 2011).

Early diagnosis of AD is critical as it is hypothesized that the pathology of the disease occurs up to 20 years before cognitive decline. Many groups have searched for protein biomarkers in blood, plasma, serum and CSF have not yielded a reliable, sensitive, and specific candidate biomarker marker for AD diagnosis. Due to the sensitivity of deep sequencing, it is possible to detect genetic modifications in disease, in particular deregulated miRNA. The possibilities of profiling miRNA in CSF, urine, blood, plasma and serum have been explored with some successes in cancer and may be applicable to AD and other neurodegenerative disorders. In addition, the genetic information contained in circulating exosomes may provide a highly specific readout of disease and should be further investigated. Deep sequencing of miRNA together with high-throughput validation methods will complement diagnostic testing and represents a vital step toward developing a cost-effective, non-invasive and low risk diagnostic test to detect the onset and monitor various stages of AD. Furthermore, the development of a diagnostic test comprising of a profile of RNA biomarkers associated with AD, and has potential for other neurodegenerative diseases such as Parkinson's, Prion and Huntington's diseases.