Background
Neurodegenerative diseases (NDs) have become a serious medical and social problem in developed countries, owing to significantly increased lifespans. For example, Alzheimer’s disease (AD) is now the fifth leading cause of death for those aged 65 years and older in the United States [
1]. Despite intense research and numerous clinical trials, no effective treatments are currently available for NDs. One key element in the development of new therapies is the availability of diagnostic tools to help identify disease early, stratify patients for clinical trials, and monitor disease progression during treatment. In this respect, the most advanced diagnostic approaches for NDs, such as analysis of proteins in cerebrospinal fluid (CSF) and various imaging techniques [
2,
3], are highly promising but not suitable for primary screening and monitoring purposes, owing to their invasiveness and high cost. Minimally invasive, cost-effective biomarkers of NDs would be very helpful in the advancement of ND diagnosis and treatment [
4‐
8]. Several factors complicate the search for such biomarkers [
7,
9]. First, mechanisms of ND initiation and development are not well understood; second, NDs may develop without any signs or symptoms for 10–20 years prior to clinical manifestation; and third, commonly occurring comorbidities and symptomatic overlap, such as cognitive impairment, complicate the differential diagnosis of NDs (
see [
10,
11] for detailed discussion of the need and the roadmap for the development of blood-based biomarkers for AD and other dementias).
Recently, several groups have proposed the use of microRNAs (miRNAs) circulating in plasma or serum for ND detection [
12‐
23]. miRNAs are small molecules (~22 nucleotides) that play important roles in gene regulation by binding to complementary regions of messenger transcripts and repressing their translation or regulating their degradation [
24,
25]. On the basis of sequence complementarity analysis, an individual miRNA can bind to and regulate > 100 messenger RNAs (mRNAs), and an mRNA can be regulated by multiple miRNAs; thus, as potential biomarkers, miRNAs are reflective of multiple cellular processes. Over 2000 miRNAs have been discovered in human cells to date, and many of these miRNAs are specific to or overexpressed in certain organs, tissues, and cells [
25‐
30]. Some miRNAs, including those that are cell-specific, can be enriched in particular cellular compartments, such as neurites and synapses [
31‐
37]. miRNAs can be secreted or excreted into the extracellular space [
38‐
41] and are detectable in plasma and serum [
14,
42‐
44].
Intracellular concentrations and rates of secretion of miRNAs can be dramatically affected by physiological and pathological cellular processes [
39,
45‐
47]. In the case of mild cognitive impairment (MCI), a heterogeneous syndrome characteristic of early stages of various NDs, we have previously shown that the condition can be detected by analysis of miRNAs enriched in synapses of brain regions affected in early AD, such as the hippocampus [
14,
48]. The concentration of a brain-enriched miRNA in bodily fluids depends on intrinsic factors such as its expression; its intracellular localization; and disease-associated changes in expression, metabolism, and secretion. It also depends on extrinsic factors such as blood supply to a particular brain area, changes in blood-brain barrier permeability, and miRNA stability in circulation.
To compensate for the impact of factors unrelated to a specific ND and to account for smaller changes in miRNA concentrations accompanying slowly developing pathologies (as compared with those in acute diseases such as stroke), other brain-enriched miRNAs can be used for normalization of the miRNA biomarkers. This miRNA “pair” approach (
see below) has led to the discovery of miR-132 and miR-134 families of miRNA biomarker pairs capable of differentiating MCI from age-matched controls with approximately 0.90 accuracy [
48,
49]. Interestingly, these miRNA pairs proved less effective as biomarkers for detecting AD dementia stages [
49], possibly owing to the loss of synapses and death of neurons in the hippocampus causing the amount of hippocampal synaptic miRNAs to decrease in plasma as the disease progresses.
The present study was designed to evaluate circulating brain-enriched miRNAs as potential biomarkers for detection and differentiation of AD, Parkinson’s disease (PD), frontotemporal lobar degeneration (frontotemporal dementia [FTD]), and amyotrophic lateral sclerosis (ALS).
Discussion
Previously, we proposed a novel approach to the development of biomarkers in the area of neurodegenerative, neurodevelopmental, and neurological diseases based on analysis of brain-enriched miRNAs circulating in plasma, and we demonstrated the viability of the concept for early AD detection with the discovery of two miRNA families capable of detecting MCI and pre-MCI with high accuracy [
48,
49]. In the present study, we used this approach for the detection and differentiation of four NDs: AD, PD, FTD, and ALS.
The most common approaches of searching for circulating miRNA biomarkers to detect a specific pathology are based on the analysis of as many plasma/serum miRNAs as technically feasible using miRNA arrays or next-generation sequencing, followed by qRT-PCR of the identified candidates, and normalization of miRNA(s) whose concentration significantly changes in a pathology sample per minimally variable miRNA or average of all miRNAs tested. There are several disadvantages to these approaches: (1) their sensitivity is significantly lower than that of qRT-PCR, and thus many brain-enriched miRNAs are not reliably detected in plasma, which precludes their further analysis; (2) on one hand, the variability of these methods is high, and many potential candidate miRNA biomarkers are not confirmed by qRT-PCR, and on the other hand, some promising biomarkers are not selected as candidates; and (3) many potential normalizer miRNAs are not uniformly expressed in various pathologies. The latter consideration is especially important in NDs in elderly people because there is a high chance of comorbidities as well as use of various medications. For example, plasma concentration of miR-16, which is widely used as a normalizer in other indications, is changed in patients with AD ([
89,
90] and our unpublished data). Other considerations are potential changes of brain blood supply or blood-brain barrier permeability, which would affect concentrations of brain-enriched miRNAs in plasma. Thus, in our earlier studies, we developed a different approach based on miRNA pairs, consisting of one miRNA enriched in synapses of a brain region affected by the disease and another miRNA enriched in a different brain region or cell type, such as glial cells. In effective miRNA pairs, miRNAs are frequently highly correlated [
49], decreasing intersubject variability. By combining two or three effective miRNA pairs into a single miRNA classifier, we achieve greater accuracy. Other groups have also identified miRNA pairs as effective biomarkers in the context of cancer diagnosis and prognosis [
91‐
93].
The following factors, which may complicate the search for miRNA biomarkers, should be considered: (1) data on miRNA enrichment in the brain and the different brain regions is still limited, and some brain-enriched miRNAs can be expressed, although at different levels, in several brain regions; (2) ND progression can substantially change both underlying processes, such as synapse dysfunction and destruction in early disease stages and neuronal death in late disease stages, and brain regions involved in a disease due to the expansion of pathologic processes to new brain regions; (3) NDs exhibit heterogeneous clinical symptoms and different brain pathologies (in this respect, analysis of larger cohorts of pathologically homogeneous ND groups would be valuable); and (4) development of neurodegeneration can result in changes in other tissues and organ systems, such as the muscle fibers in ALS.
The data obtained in the present study support our minimally invasive approach to the detection of NDs based on the analysis of circulating brain-enriched miRNA in plasma. Individual miRNA pairs tested in this study classified diseases with accuracies > 0.80, and combinations of several miRNA pairs demonstrated accuracies of up to 0.90. Some miRNA pairs proved to be effective in detecting more than one ND, such as AD and FTD, which partially affect overlapping brain regions. Effective classifiers for ALS detection in addition to brain-enriched miRNAs included muscle-enriched miR-206, which is also highly expressed in cerebellum [
72,
94].
As expected, miRNA classifiers for AD dementia were different from the optimal miRNA classifiers for MCI and pre-MCI [
48,
49]. MCI is a heterogeneous syndrome characteristic of many NDs. During the progression from MCI to the dementia stage of AD, the ratio of miRNA levels in bodily fluids can change as a result of several factors. Hippocampal synapse and neurite degeneration during early stages of the disease leads to reduced numbers of synapses and neurites in later stages of AD, resulting in diminished amounts of miRNAs being excreted from synapses and neurites of the hippocampus; in addition, because of the increased neuronal cell death that occurs during the later stages of AD, the concentration of neuronal cell body miRNAs in bodily fluids increases. As the disease progresses, new brain regions become involved in the pathology, compromising the reliability of the normalizer miRNAs. This results in numerator miRNAs of effective biomarker pairs for early AD (e.g., miR-128a and miR-874) becoming denominators in effective pairs for later disease stages. Concurrently, because the pathology expands to new brain regions (e.g., cortex and midbrain), synaptic miRNAs from those regions, such as miR-107 and miR-9*, become good biomarker numerators (Additional file
4A). These events can be of informative value for disease monitoring. For example, changes in relative concentrations of miRNAs enriched in different brain regions or different cell types (e.g., neurons and glial cells) may be used as an indicator of disease progression.
Our analysis further revealed miRNA classifiers differentiating AD from FTD with accuracy > 0.75 and all other NDs from each other with accuracy in the range of 0.80–0.90, a highly promising outcome for such heterogeneous diseases, which are in many cases accompanied by other related comorbidities.
Recent data indicate that brain aging and AD development are sex-dependent phenomena [
95‐
98]. Analysis of the data presented here reveals sex dependence of certain brain-enriched miRNAs, which distinguished NDs from control in sex-specific subsets with significantly better accuracy than in the total (female plus male) sets (Table
2, section B). The effect is particularly notable for female participants, such as that accuracy is 0.98 for female and 0.83 for all patients with ALS. More detailed, larger studies are needed to further substantiate this observation and to investigate potential physiological mechanisms underlying this phenomenon. In particular, it has to be determined whether these sex-dependent differences are due to differential expression, secretion, or excretion of some brain-enriched miRNAs, intrinsic sex-dependent differences associated with NDs, or both.
It is worth noting that the approach of the present study was based on analysis of brain-enriched miRNAs in the total pool of miRNAs extracted from plasma. Recently, exosomes, and in particular neuronally derived exosomes, have been studied as potential biomarkers of NDs [
99‐
101]. The promise of exosomes as biomarkers of pathology is due largely to their protein content and membrane composition because there is a high degree of variability in miRNA distribution across exosome populations, and contradictory data exist on the fraction of miRNAs circulating in exosomes as compared with other forms of cell-free miRNAs in the bloodstream [
47,
102,
103].
Limitations
A limitation of this study is that all participants were recruited at a single clinical site. Larger, multicenter studies are needed to further evaluate the utility of the approach described herein. A further limitation of the study is that the blood samples were collected from symptomatic patients only. Longitudinal studies are needed to assess the prognostic value of the biomarker candidates in presymptomatic patients.
Acknowledgements
We thank members of the Center for Neurodegenerative Disease Research at the University of Pennsylvania who contributed to this work, as well as the many patients studied and their families, for making the reported research possible. We also thank Drs. Michael Kiefer and Gaspar Taroncher-Oldenburg for their help with editing the manuscript.