Brain-enriched microRNAs circulating in plasma as novel biomarkers for Rett syndrome

Rett syndrome (RTT) is a severe neurodevelopmental disorder caused by mutations in the X-linked gene MECP2 (methyl-CpG-binding protein 2). Minimally invasive and accurate biomarkers of disease progression and treatment response could facilitate screening of therapeutic compounds in animal models, enrollment of better-defined participants into clinical trials, and treatment monitoring. In this study, we used a targeted approach based on analysis of brain-enriched microRNAs (miRNAs) circulating in plasma to identify miRNA biomarkers of RTT using Mecp2-mutant mice as a model system and human plasma samples. An “miRNA pair” approach, i.e. the ratio between two miRNAs, was used for data normalization. Specific miRNA pairs and their combinations (classifiers) analyzed in plasma differentiated wild-type from Mecp2 male and female mice with >90% accuracy. Individual miRNA pairs were more effective in distinguishing male (homozygous) animals than female (heterozygous) animals, suggesting that disease severity correlated with the levels of the miRNA biomarkers. In the human study, 30 RTT patients were compared with age-matched controls. The results of this study showed that miRNA classifiers were able to differentiate RTT patients from controls with 85–100% sensitivity. In addition, a comparison of various age groups demonstrated that the dynamics in levels of miRNAs appear to be associated with disease development (involvement of liver, muscle and lipid metabolism in the pathology). Importantly, certain miRNA biomarker pairs were common to both the animal models and human subjects, indicating the similarity between the underlying pathological processes. The data generated in this feasibility study suggest that circulating miRNAs have the potential to be developed as markers of RTT progression and treatment response. Larger clinical studies are needed to further evaluate the findings presented here.


Results
RTT mouse models study miRNA selection. Table 1 lists the miRNAs used in the present study. Because loss of MECP2 affects neurons throughout the neuraxis, miRNAs that have been shown to be enriched in synapses of both forebrain and brainstem regions were used, including miR-107, miR-132, miR-411 and miR-491-5p [35][36][37][38][39][40]. Additional miRNAs enriched in several brain regions (miR-323-3p) or predominantly in the pituitary gland (miR-335-5p, miR-370), as well as miR-16, a ubiquitously expressed miRNA involved in apoptosis regulation, were also analyzed. Predictive analysis indicates that some of the selected miRNAs may affect expression of RTT-related proteins ( Table 1).
Male RTT model. In the first experiment, the levels of eight miRNAs were measured in plasma samples from 11 Mecp2 tm1.1Jae Null and 9 wild-type (Wt) male mice aged 5-6 weeks. One control sample did not meet the quality control parameters and was thus excluded from the analysis. The dot plots, receiver operator characteristic (ROC) curves, sensitivity, specificity and accuracy for the resulting miRNA pairs are presented in Fig 1 and S1 Fig. Several miRNA pairs and their combinations distinguished Wt from Mecp2 tm1.1Jae mice with 89-95% accuracy. These results were encouraging, particularly because two effective numerators in the miRNA pairs, namely, miR-107 and miR-491-5p, are expressed in synapses of the forebrain and brainstem. pairs first identified in the initial experiment with Mecp2 tm1.1Jae mice. The results obtained from both mouse strains were very similar. However, in the Mecp2 tm1.1Bird animals, additional miRNA pairs were identified that could effectively differentiate between Wt and Mecp2 Null mice (S1 Fig).
Although the sample sizes in both the pilot study and the confirmatory study were relatively small (n = 19 and n = 18 samples, respectively), the reproducibility of the findings from these two different RTT mouse models obtained by different labs strengthened the conclusion that plasma levels of the selected brain-enriched miRNAs can reliably distinguish between Wt and Null animals.
Female RTT model. We then compared the levels of the eight miRNAs in two cohorts of female heterozygous (Het) and Wt mice: Mecp2 tm1.1Jae Het n = 6, Wt n = 7 (6 months old); and Mecp2 tm1.1Bird Het n = 10, Wt n = 10 (18 weeks old). The two pairs of mutant and Wt cohorts were tested at different ages and, therefore, were analyzed independently. Fig 1 and S2 Fig present the data from these two experiments as dot plots, ROC curves, sensitivity, specificity and accuracy. As expected, in the Het mice, the area under the ROC curve (AUC) and associated accuracy, which represented the ability of miRNA pairs to distinguish mutant from Wt mice, were lower than those in the Null mice since female heterozygotes retain some normal Mecp2 expression. However, classifiers consisting of combined miRNA pairs provided >90% of the overall accuracy in the female Het Mecp2 tm1.1Jae and Mecp2 tm1.1Bird mouse models. The AUC values for all four experiments are summarized in Table 2. Notably, certain miRNA pairs effectively distinguished between mutant and Wt mice consistently across all four animal models (Mecp2 tm1.1Jae Null and Het and Mecp2 tm1.1Bird Null and Het).

Human RTT study
Thirty RTT patients and 30 female age-matched control (AMC) participants were enrolled in the study (Table 3 and S1 Table). Since larger volumes of human plasma were available, 19 miRNAs (Table 1) were tested, 7 of which were identical to those tested in the mouse model studies, and 4 were additional miRNAs that were shown to inhibit MECP2 synthesis (http:// mirtarbase.mbc.nctu.edu.tw/php/search.php). Further, inflammatory miR-146a and miR-155, liver-enriched miR-122, and muscle-enriched miR-206, as well as additional brain-enriched miRNAs were tested (Table 1). RTT patient groups stratified according to age, namely, 2-5 years old, 6-15 years old, and >15 years old, were effectively differentiated from the AMCs by miRNA pairs and classifiers with an accuracy up to 100%, 94-97%, and 100%, respectively (Fig  2A-2C). Moreover, disease progression was reflected by changes in levels of miRNAs enriched in organs involved in a specific pathology. For example, the appearance of miR-122 as the numerator reflected the development of liver pathology. Table 2 demonstrates that certain miRNA pairs were common among different animal RTT models or human RTT in all, or at least some, age cohorts. The pathological progression was additionally demonstrated by the differentiation of the RTT groups from each other (S3 Fig). miR-122, as a biomarker of liver pathology, was the numerator that performed best in separating RTT groups of various ages. Fig 3 demonstrates that the dynamics of age-dependent changes in the plasma levels of the tested miRNAs differed substantially between the RTT and AMC groups. S2 Table provides a summary of miRNA pairs that effectively differentiated RTT age groups but did not differentiate the respective AMC groups and vice versa. In addition, we subsequently investigated whether some miRNA pairs and/or classifiers were able to detect the various pathological processes associated with RTT. Table 4 demonstrates that patients with seizures could be differentiated from patients with no seizures with up to 78% accuracy. Ten RTT patients with walking problems were effectively differentiated from 20 ambulatory RTT patients (Table 5). Though, at present, the cause and effect relationship between walking problems associated with RTT and muscle-specific pathology remains to be determined, the best numerator in this case was miR-206, which is highly enriched in muscle cells. Furthermore, although liver pathology is often present in RTT patients, the alanine transaminase (ALT) levels were higher than those of the AMCs in only six RTT patients. These patients were effectively separated from the other RTT patients by multiple miRNA pairs, with liver-specific miR-122 being found to be the best numerator (Table 6). Cholesterol, the metabolic perturbations of which are characteristic of some patients with MeCP2 mutations, was also measured in all 30 RTT patients. Six and three patients had higher and lower than normal cholesterol levels, respectively (Table 3). Table 7 demonstrates that six patients with higher than normal cholesterol levels were differentiated from the other RTT patients with up to 82% accuracy.

Discussion
The goal of this study was to determine whether circulating cell-free miRNAs in plasma can be used to reliably detect RTT patients, monitor disease development, and evaluate the different aspects of this disease, such as liver and muscle pathology. This study initially implemented animal RTT models to distinguish between Wt and Mecp2 mutant mice as a step towards examining the utility of this approach in monitoring disease progression and treatment response in preclinical RTT models. Loss of MECP2 can alter the transcription of many genes, including those encoding specific brain miRNAs [24], such as miR-132, an inhibitor of MECP2 expression [24,[45][46][47]. Altered transcription and secretion of brain miRNAs may, in turn, lead to changes in the concentrations of these miRNAs in plasma. In fact, it has previously been shown that analysis of plasma levels of miRNAs enriched in neurites and synapses of brain regions affected by neurodegeneration yields effective miRNA biomarkers for detection and monitoring of neurodegenerative diseases compared to healthy controls [20,21,23]. Highly conserved miRNA biomarkers are advantageous because the same assay can be applied to both animal models and humans. Since plasma concentrations of brain-enriched miRNAs are relatively low, miRNA array and next-generation sequencing (NGS) approaches are not sufficiently sensitive for reliable detection of these miRNAs [23,48], whereas the approach used in this study is based on targeted analysis of miRNAs by individual RT-qPCR. An additional challenge is that only a small volume of plasma (approximately 0.2 ml), which is sufficient to analyze only eight miRNAs by RT-qPCR, can be obtained from each mouse. Furthermore, changes in miRNA expression, not only in the brain but also in other organs, can affect plasma levels of ubiquitous miRNAs. Based on our previous data and analysis of the literature [23], in this study we tested seven brain-enriched miRNAs, as well as miR-16, a ubiquitous miRNA involved in apoptosis regulation, in four RTT murine models. In order to account for the potential effects of extraneous factors, such as plasma collection and specimen handling, samples from three different sites were tested. The results supported the hypothesis that miRNA pairs and their combinations, including several miRNA pairs that were uniformly  [47] demonstrated that MECP2 regulates biogenesis of some miRNAs and found changes in their expression, such as increases in miR-107 and miR-122 and decreases in miR-335-5p, miR-132 and miR-411 expression in the brain, due to MECP2 mutations. These changes in miRNA expression may explain the effectiveness of certain miRNA pairs in our studies, for example miR-107/miR-335-5p, and miR-107/miR-132. On the other hand, although according to Cheng et al. expression of miR-323-3p increases, this miRNA proved to be a good denominator both in our animal and human studies. This discrepancy can be explained by various factors, including differences in genotypes and the ages of the animals tested in the two studies. Additional studies should be performed to address these questions.
Consistent with our hypothesis, individual miRNA pairs are more effective in Null (male) RTT models than in Het (female) models, indicating that disease severity correlates with miRNA ratios, with the only exception being miRNA pairs with miR-132 as the numerator (miR-132/miR-335-5p and miR-132/miR-323-3p; Table 2), which were equally effective in the Null and Het models. Circulating brain-enriched miRNAs as biomarkers for Rett syndrome We then extended our investigation to human patients, and the data obtained both supported and expanded the results of the mouse models studies due to the testing of more miR-NAs, the analysis of RTT patients at several stages of pathological development, and the comparison of patients with different RTT symptoms. The most important result was that the same brain-enriched miRNAs formed biomarker pairs that could be used to detect RTT in both animal models and humans (Table 2). In addition, the human data demonstrated the applicability of our approach to the detection of liver and muscle pathologies and of changes in cholesterol metabolism. The analysis of various age groups indicated the importance of using AMCs for data analysis. The results generated in this study are preliminary due to the relatively small number of study participants, and larger studies are needed to select the best miRNA biomarkers; however, the similarity between the data obtained in the 4 animal models and those obtained in the human patients is highly promising.
In summary, the data presented here demonstrated that RTT biomarker development based on analysis of plasma miRNAs is feasible. Moreover, our finding that miRNA biomarker candidates more effectively differentiated male rather than female Mecp2 mutants from controls in the mouse models studies, as well as in the human studies, suggested that these markers correlate with disease severity. This result, in turn, suggested that miRNA pairs have the potential to be developed as prognostic biomarkers for RTT and as markers of disease progression and treatment response. Finally, our data suggested that this approach could also prove fruitful for other neurodevelopmental disorders. Circulating brain-enriched miRNAs as biomarkers for Rett syndrome

Animal models and plasma collection
Two mouse models of RTT, Mecp2 tm1.1Jae (deletion of exon 3) 14 and Mecp2 tm1.1Bird (deletion of exons 3 and 4) 15 , were tested. Similar to RTT patients, these animals appear normal at birth and then develop RTT-like symptoms, including reduced brain size and weight, hypoactivity, abnormalities in locomotion (altered gait, hindlimb clasping), anxiety, cognitive deficits, and occasional hard respiration. MECP2 is located on the X chromosome, and the severity of RTT depends on the type of mutation (MECP2 can be completely or partially inactivated) and variability in X chromosome inactivation. The disease is much more severe in males. In the Mecp2 tm1.1Bird Null male model, mice appear normal until postnatal week 3, with symptoms progressing between postnatal weeks 3 and 5, leading to death around postnatal week 8. Mecp2 tm1.1Bird Het females are asymptomatic up to postnatal week 6 when initial symptoms appear, and animals die by 11 Mice were anesthetized with isoflurane gas. Whole blood was collected into K2EDTA tubes via cardiac puncture. Samples were centrifuged at 2000 g for 15 minutes at 4˚C, and approximately 0.2 ml of plasma was transferred to Eppendorf tubes and frozen on dry ice. All animal studies were conducted in accordance with the relevant guidelines and regulations and were approved by the corresponding IACUCs of Case Western Reserve University School of Medicine, Jackson Labs, and Psychogenics.

Study participants and plasma collection
All patients with RTT were recruited from the Tri State Rett Syndrome Center (NY) database. The parents were informed about the study prior to their scheduled annual health care maintenance visit, and the patients were recruited during the visit. Written, informed consent was provided by all patients and/or their legal guardians. The study protocol and procedures were approved by the Institutional Review Board of Albert Einstein College of Medicine, Montefiore Medical Center and were conducted in accordance with the Declaration of Helsinki.
All study participants with RTT were experiencing their usual, baseline health at the time of their recruitment and of the blood draw. Control study participants were recruited from the pediatric waiting area at the Children's Hospital at Montefiore from among healthy siblings (with no neurological developmental problems and who were not on medications), who were accompanying their brothers/sisters to their medical appointments. Control study participants were age-and gender-matched to the recruited patients with RTT. The clinical information regarding the study participants is summarized in Table 3. Five high functioning (HF) patients were excluded from the analysis due to the small sample size and wide age distribution.
Whole blood from 30 RTT patients and 30 AMCs was collected via venipuncture into K2EDTA vacutainers, and the plasma was separated as described above and frozen at -70˚C.

RNA isolation and RT-qPCR
In each experiment, RNA was extracted from 200 μl of plasma (two preps obtained from the same human study participant were then combined) using a TRIzol treatment and silica (Ambiom Glass Fiber Microcolumn) binding protocol (http://asuragen.com/wp-content/ uploads/2016/05/biomarkers.pdf). Single-target quantitative RT-PCR (RT-qPCR) was performed using the TaqMan Reverse Transcription Kit and miRNA-specific stem-loop primers (ThermoFisher). The RT step was performed in triplicate, and 2 μl of the original plasma equivalents was present in the final PCR. Placental RNA was used as a "positive control", and no-template controls were used as a "negative control" for each run. Calibration curves were obtained for each miRNA tested. Quality control of the miRNA preps was performed by testing two ubiquitous miRNAs, namely, miR-16 and miR-27a, known to have low variability in each plasma preparation.

Statistical methods
All Ct values were recalculated for 1 μl of plasma based on the calibration curves. DiamiR's custom software (20) was used for all calculations, including the identification of miRNA pairs capable of differentiating control and mutant Mecp2 mice based on their concentration ratios (ΔCt) in plasma. Mann-Whitney U tests were used to evaluate the significance of the differences between any two experimental groups based on the various biomarker miRNA pairs. Receiver operating characteristic (ROC) curves were constructed, and the AUC values were calculated to evaluate the sensitivity and specificity of the miRNA pairs. The sensitivity and specificity are reported for the cutoff points on the ROC curves that provided the best overall accuracy.
Supporting information S1 Table. RTT and AMC groups: Number of participants and age ranges (average ± standard deviation). (TIF) S2 Table. A) miRNA pairs that differentiate AMC groups do not distinguish the corresponding RTT age groups. B) miRNA pairs that differentiate RTT age groups do not distinguish the corresponding AMC groups. (PDF)