Validation of plasma microRNAs as biomarkers for myotonic dystrophy type 1

Non-invasive and simple to measure biomarkers are still an unmet need for myotonic dystrophy type 1 (DM1). Indeed, muscle biopsies can be extremely informative, but their invasive nature limits their application. Extracellular microRNAs are emerging humoral biomarkers and preliminary studies identified a group of miRNAs that are deregulated in the plasma or serum of small groups of DM1 patients. Here we adopted very stringent selection and normalization criteria to validate or disprove these miRNAs in 103 DM1 patients and 111 matched controls. We confirmed that 8 miRNAs out of 12 were significantly deregulated in DM1 patients: miR-1, miR-27b, miR-133a, miR-133b, miR-206, miR-140-3p, miR-454 and miR-574. The levels of these miRNAs, alone or in combination, discriminated DM1 from controls significantly, and correlated with both skeletal muscle strength and creatine kinase values. Interestingly, miR-133b levels were significantly higher in DM1 female patients. Finally, the identified miRNAs were also deregulated in the plasma of a small group (n = 30) of DM2 patients. In conclusion, this study proposes that miRNAs might be useful as DM1 humoral biomarkers.

are deregulated in the plasma or serum of DM1 patients. These studies indicate that the levels of specific miRNAs correlate with loss of muscle strength and disease stage, suggesting a diagnostic and/or prognostic potential.
Although very promising, the significance of these studies is limited by the low numerosity of patients analyzed: 36 for the study of Perfetti et al. 23 and 23 for the study of Koutsoulidou et al. 24 . Given the need of humoral biomarkers of prognostic value and also potentially useful as outcome measure for future therapeutic trials in DM1 patients, in this work we have analyzed the miRNAs deregulated in DM1 in a larger and independent patients group, to corroborate or disprove the existing data.

Results
Validation of deregulated plasma miRNAs in DM1. Platelet-free plasma samples were harvested from 103 DM1, and 111 age-and sex-matched control (CTR) patients, displaying no neuromuscular disorders. DM1 patients presented the expected hallmarks of myotonic dystrophy, summarized in Table 1: myotonia, cataract and loss of muscle strength 4 . Most of the patients were in stage 3 of the disease (MIRS score) 25 and the pathological CTG expansions ranged from 65 to 1400. Of note, there was no overlap between the patients and controls analyzed in this study and the ones analyzed in our previous investigation 23 . Total RNA was extracted from plasma samples and miRNAs were measured by qPCR. For this validation study, we measured miR-133a, miR-193b, miR-191, miR-140-3p, miR-27b, miR-454, miR-574, miR-885-5p and miR-886-3p, previously described in Perfetti et al. 23 , and miR-1, miR-206 and miR-133b identified by Koutsoulidou and collaborators 24 . Normalization is a critical issue in all circulating miRNA studies 26 . Three normalizers were used, one exogenous (cel-miR-39) and two endogenous (miR-106a and miR-17-5p). Exogenous spike-ins, such as cel-miR-39, have been widely used for normalization purposes 11,26-28 and we previously described the use of miR-106a and miR-17-5p as stable endogenous normalizers in DM1 patients 23 . Additionally, our previously published profiling study displayed no evidence of global changes of plasma miRNAs associated to DM1 23 .
The high number of DM1 patients and of miRNAs tested required several qPCR sessions. To make sure to have comparable results, we used a reference sample obtained by pooling a fraction of all control preparations. By measuring this reference sample in each qPCR session, we minimized technical variability throughout the study.
Given the importance of the normalization step and in order to maximize specificity, we considered as validated only those miRNAs that displayed significant differences after normalization with all of the following normalizers: (1) the average of cel-miR-39, miR-106a and miR-17-5p (Fig. 1); (2) exogenous cel-miR-39 only (Supplementary Table S1); (3) the average of the two endogenous miRNAs, miR-106a and miR-17-5p (Supplementary Table S1). Using these stringent criteria, we confirmed that 8 miRNAs out of 12 were significantly deregulated in DM1 patients. In particular, miR-1, -133a, -133b, -206, -140-3p, -454 and miR-574 were up-regulated, while miR-27b was the only down-modulated one. Notably, although lower standard errors and p values were generally observed upon normalization with three normalizers averaged, values obtained with all normalization methods were very similar, confirming the robustness of the data.
In particular, considering a cut-off value of 2.55, sensitivities of 81.6%, 82.3%, and 82.5% as well as specificities of 70.3%, 71.2% and 71.2% were observed for the DM1-miRNAs score, the myomiR score and the miR-133a/b score, respectively.
We also run a statistical analysis to identify outliers. Supplementary Figure S2 shows that, even after exclusion of the outliers, all differences remained statistically significant. Finally, we also found that, using different normalization procedures, similar AUC values were obtained for all scores, confirming the robustness of the data (Supplementary Table S2).
In conclusion, all three scores were useful in differentiating DM1 patients from controls.    Correlation with clinically relevant parameters. We assessed the correlation of the identified scores with clinically relevant parameters. We found that, analyzing all subjects, miR-133b levels and all three scores displayed a weak but significant inverse correlation with muscle strength (MRC, Figures 2, 3, 4 and 5, panels c), and a weak direct correlation with CK values (Figures 2, 3, 4 and 5, panels d). One limitation of the latter correlation is constituted by the fact that exercise in the days before testing was not recorded. Moreover, we also found that miR-133b levels were significantly higher in female DM1 patients compared to DM1 males, while no difference was observed in control subjects (Fig. 6). This difference was present with all three normalization procedures.
Next, to better investigate the relationship between clinically relevant variables in DM1 patients that were most affected by disease, i.e. with MIRS > 2, a multiple linear regression was performed analyzing CK values, age of onset, number of triplet expansions and muscle strength. Statistically significant associations are reported in Table 2. We identified the association of muscle strength with miR-133b levels and all three scores. Moreover, CK values were associated to myomiR-score and miR-133a/b score.
We can conclude that all the identified scores display good combinations of sensitivity, specificity and correlation with clinically relevant features. Considering the minimization of the number of assays as an important parameter for the transferability to the clinical practice, also a miR-133a/b score obtained using only the two internal normalizers (Supplementary Table S4) displayed features very similar to these obtained using the three normalizers (Fig. 5).
DM1-deregulated miRNAs are similarly modulated in DM2 patients. In order to assess whether the miRNA deregulations found in DM1 patients were also observed in DM2 patients, the plasma of 30 DM2 patients and 30 age and sex matched control subjects was assayed. Figure 7a shows that, with the exception of miR-27b, all the other miRNAs were significantly increased, further confirming the similarity of the two DM diseases.
Next we tested the differentiating value of the identified miRNAs in DM2 patients, calculating the same scores as for DM1, keeping out miR-27b. It was found that all three scores efficiently differentiated DM2 patients from controls (Fig. 7b,c and d).

Discussion
DM1 is the most prevalent muscular dystrophy in adults; nevertheless there are very few humoral biomarkers that can facilitate the assessment of the patient clinical status 4 .
Plasma/serum miRNA hold a great potential as biomarkers in many diseases, including muscle dystrophies 13 . In spite of this, several hurdles exist for the successful application of circulating miRNAs as biomarkers. Indeed, one of the main issues in this kind of studies is that usually small patient groups are analyzed 11,21 , particularly in case of rare diseases, such as DM1. Considering the great variability existing from one patient to another and the absence of standardized procedures, this could lead to unreproducible tests not suitable for diagnostic routine. To overcome this hurdle, in this study we measured the same miRNAs previously found deregulated in plasma or serum of DM1 patients 23,24 in a larger and independent group, in order to corroborate or disprove the existing data. Here we analyzed more than 200 subjects between DM1 patients and controls. Albeit still limited in size, this already represents a big step forward compared to similar studies performed in rare diseases 11,21 . Additionally, an accurate normalization method 23 and stringent inclusion criteria were adopted. This approach allowed us to validate eight previously identified miRNAs, whereas four miRNAs were not confirmed, proving the importance of an accurate validation procedure. It is also worth noting that one of the RNA that were not confirmed, miR-886, is not even properly a miRNA, since it is identical to a fragment of Vault RNA 2 (VTRNA2) 29 , further supporting the superior performances of miRNAs as biomarkers. Finally, it is also worth noting that small studies also display low sensitivity. Indeed, several myomiR identified by Koutsoulidou, et al. 24 escaped identification in our previous study 23 . In this study, we validated the myomiR miR-1, 133a,-133b and -206 as increased in the plasma of DM1 patients.
Both miR-133a/b and myomiR scores correlated, although weakly, with clinical parameters of muscle involvement, supporting the hypothesis that myomiR dysregulation in the plasma of DM1 patients is specifically related to mechanisms of muscle damage. Whether this was due to passive release from damaged myofibers, it is not known. However, unlike Duchenne muscle dystrophy, necrosis is not a DM1 disease hallmark and this increase might be due, at least in part, to active mechanisms. Interestingly, while myomiR levels are not modulated in DM1 skeletal muscle biopsies, their intracellular distribution is aberrant 15 , possibly leading also to increased extracellular release.
Also interesting is the fact that miR-133b elevation was particularly prominent in DM1 female patients. While the reason for this gender difference is unknown, it is not surprising since gender is emerging as an important factor influencing DM1 clinical profile 30,31 . Of note, it has been found that miR-133b stimulates ovarian estradiol synthesis 32 . Whether this is related to the endocrinological abnormalities observed in DM1 patients remains to be elucidated 1,2 .
Other non muscle-specific miRNAs, namely miR-140, -27b, -454 and -574 were also validated as deregulated in DM1 plasma. Indeed, since DM1 is a multisystemic disorder 3,33 , it is possible that the tissue of origin of   Table 2. Multiple regression analysis, DM1 patients with stage of disease higher than 2. a Δ MRC is the difference between MRC megascore reference value (150), and MRC of each patient.
these miRNAs might not be the skeletal muscle. According to the human miRNA tissue atlas (https://ccb-web. cs.uni-saarland.de/tissueatlas), they are expressed to different extent in multiple tissues, including brain, nerve, spinal cord, thyroid and epididymis. Thus, the plasma levels of these miRNAs may reflect the global clinical state of the patient, rather than that of a specific tissue. Notably, the new plasma signature has also been confirmed in a small group of DM2 patients. While confirmatory studies in larger patient groups are necessary, this finding suggests that similar mechanisms underpinning miRNA release in the blood might be shared by DM1 and 2. Standardized procedures and more refined quantitative measurements might also allow investigating possible differences between DM1 and DM2 in the extent of dysregulation of the identified miRNAs. As for DM1, miRNAs deregulated in the plasma of DM2 patients have not been reported to be deregulated in skeletal muscle biopsies 16 . Again, this is not surprising, since plasma miRNA levels are regulated by the combination of active and passive release of miRNAs in the bloodstream by all the tissues of the organism. Three potentially useful scores were identified: the most simple is the miR133a/b that consists of only 2 miR-NAs and that holds true even if only 2 normalizers are used, facilitating the potential transfer to the clinical practice. On the other side, the DM1-miRNA score, constituted by 8 miRNAs, is more complex, but displays better correlation with clinically relevant parameters. Moreover, it is not constituted only by myomiRs, that have been found to be deregulated in other dystrophies as well 11,[20][21][22] . Thus, it holds the potential to be more DM-specific. However, this remains to be determined experimentally. Also important will be a longitudinal study, to ascertain how the identified scores change over time during DM1 disease evolution.
From a technical point of view, the identified miRNAs are little or not expressed in erythrocytes, minimizing the potentially distortive effect of hemolysis 34 . Very important for the potential transferability to the clinics will be the development of absolute quantification assays and the identification of reference values.

Methods
to confirm DM diagnosis, as previously described 36 . Stage of disease was determined using Muscular Impairment Rating Scale (MIRS) 25 . Five-point MRC scale (Medical Research Council) in the upper and lower limbs for a total maximum score of 150 (MRC megascore) was used to evaluate muscle strength 37 .
Blood samples were obtained from 103 DM1 patients, and 111 sex and age matched subjects displaying no overt sign of neuromuscular disorders (Table 1). Moreover, 30 DM2 patients (Supplementary Table S3) and 30 age and sex matched controls were also recruited (54.3 ± 2.3 years old, 18 males and 12 females). The control groups of DM1 and DM2 were only partially overlapping. EDTA-tubes were used for plasma preparation. Cell-and platelet-free plasma was prepared as previously described 23 . Potential hemolysis was tested measuring plasma absorbance at 570 and 600 nm wavelengths, where oxy-, deoxy-and carboxy-hemoglobin display similar absorbance 38 . All samples had hemoglobin < 10 mg/ml with the exception of one (DM1-119) that was discarded.
RNA isolation, miRNA measurement and miRNA scores. Total RNA was extracted as previously described using NucleoSpin miRNA Plasma columns (Macherey-Nagel) 23 . Preparations were spiked in with exogenous cel-miR-39 to assess the efficiency of RNA extraction. miRNAs were measured by qPCR using TaqMan microRNA assays, performed in duplicate according to the manufacturer instruction (Life Technologies). Raw Ct values are shown in Supplementary Table S4.
Relative expression was calculated using the comparative Ct method 39 . To calculate ∆ Cts values, the average of 3 normalizers was used: spike in cel-miR-39 and 2 endogenous stable miRNAs, miR-106a and miR-17-5p, as previously described 23 . Next, to calculate ∆ ∆ Ct values, DM1 and individual controls were all compared to a reference pool of CTR RNAs (control mix), measured in quadruplicate, ensuring comparable results throughout the study. Intra-assay variation for each miRNA is indicated in Supplementary Table S5. For the score calculations, ∆ ∆ Ct values were converted to linear values using the formula 2 (−∆∆Ct) for the up-regulations and the formula − 2 (−∆∆Ct) for down-modulations 39 . Next "DM1-miRNAs score" was calculated averaging the fold changes obtained for all validated miRNAs. Since miR-27b was down-modulated in DM1, in this case the sign was inverted. For the "myomiR score" calculation, the fold changes obtained for miR-1, -133a, -133b and -206 were averaged. For the "miR-133a/b score" calculation, the fold changes obtained for miR-133a and b were averaged. Statistical analysis. Continuous variables are expressed as mean ± standard error (SE) unless indicated differently. The box plots represent data divided in quartiles. For group-wise comparisons, Mann-Whitney U test was used. The ability to discriminate between the DM1 and control groups was determined by the receiver operating characteristic curve, and the area under the curve was calculated. Spearman rank correlation was used to compare miRNA levels with clinical characteristics. For multiple linear regression, absolute ∆ ∆ Ct were used to study the association between all the scores and clinical characteristics. The normality of the scores was assessed through histogram visualization. Outliers were identified by Tukey's test. Bonferroni's method was used to correct for multiple testing. All tests were performed 2-sided and a p ≤ 0.05 was considered as statistically significant. For statistical analysis GraphPad Prism v. 7.0 (GraphPad Software Inc.) and R 40 software were used.