miRNA Biomarkers Diagnose Amyotrophic Lateral Sclerosis in Circulating Blood Authors: Rachael Anne Dunlop, Paul Alan Cox, Paul Mehta, Elijah W. Stommel, Sandra Anne Banack Published: 2025 ABSTRACT A rapid, accurate diagnostic test for amyotrophic lateral sclerosis (ALS) would reduce diagnostic delays and improve patient outcomes. We extracted eight circulating miRNAs from 788 blood plasma samples and analyzed them using qPCR for ALS diagnostic accuracy. The biomarker parameters were established previously using 449 individual blood samples and applied prospectively to an independent cohort for validation of a predictive model. The primary outcome was ALS classification accuracy as measured by diagnostic sensitivity, specificity, positive and negative predictive values (PPV, NPV), and area under the curve (AUC). The secondary outcome was comparative fold-regulation values determined prior to data collection. The diagnostic test had an AUC of 0.98 (95% CI 0.97–0.99), with 97% sensitivity (95% CI 96–98), 93% specificity (95% CI 90–96), 93% PPV (95% CI 91–96), and 97% NPV (95% CI 96–98). The fold-regulation values exceeded or were equal to prior calculated values. Streamlined methods resulted in higher diagnostic accuracy, cut both assay time and cost, reduced technical barriers, and enhances the feasibility for widespread clinical adoption. The high accuracy of this diagnostic biomarker suggests that continued evaluation is warranted.Supplementary InformationThe online version contains supplementary material available at 10.1007/s12035-025-05579-w. Introduction Amyotrophic lateral sclerosis (ALS) is a rare, progressive, and fatal neuromuscular disease with no known cure. Data from the National ALS Registry (2018) suggests an incidence of 1.6 per 100,000 USA population, with prevalence of 9.1 per 100,000 population, and 30,000 active cases per year in the United States [1]. However, because ALS is a notifiable disease in only a handful of US states, data for more recent years must be estimated. A recent study estimated a total of 33,000 ALS cases in 2022, with a 10% predicted increase in cases by 2030 to 36,308. The largest estimated increase is for population ages 66 years and older, with a 25% increase, from 16,349 cases in 2022 to 20,438 cases in 2030 [1]. ALS typically manifests in mid-to-late adulthood, progresses rapidly, and carries a median survival time of two to five years [2]. Some patients experience significantly longer survival, but the disease typically progresses rapidly, shortening their lifespan to as little as 6 months [3, 4]. The clinical presentation of ALS is heterogeneous, with distal limb muscle weakness being more common than bulbar onset [5]. Approximately 10% of ALS cases are familial, while the remaining 90% are considered sporadic [6]. At present, care for ALS patients remains primarily supportive and palliative, as neither a cure nor consistently effective therapeutic interventions are available [5]. ALS is diagnosed based on careful clinical assessment by a skilled neurologist using the revised El Escorial Criteria and a battery of tests to exclude other possible ALS-mimic diseases [7]. In the early stages of ALS, symptoms overlap with other neurological diseases and establishing disease progression is important to the differential diagnosis. Patients are frequently shuttled between various medical specialists before seeing a neurologist with ALS experience [8]. Misdiagnosis rates are high (13–68%) rendering patients vulnerable to unnecessary medical procedures along the path to diagnosis [8]. Goyal et al. [9] found that although most patients seek professional help within 3.8 months (± 5.2 SD) of symptom onset, the mean time of a definitive ALS diagnosis was 8.2 months (± 12.5 SD) but that time increased to 10.4 months (± 17.9 SD) for patients who were originally misdiagnosed. Diagnostic delays also differ by race [10]. Noteworthy is the high standard deviations associated with these means indicating substantial variation for individual patients. Just as important as receiving a rapid and accurate diagnosis of ALS is ruling out the disease for patients who do not have ALS. The stress and anxiety expressed by patients waiting for a diagnosis is great. Furthermore, tools that can complement existing diagnostic methods will allow earlier intervention and improve patient outcomes because treatments are likely to be more effective [11, 12]. For biomarkers to achieve widespread clinical adoption as diagnostic tools, they should ideally meet three key criteria: (1) minimal invasiveness for sample collection, (2) rapid processing, and (3) cost-effective analysis. We have previously identified an 8-miRNA ALS diagnostic [13] using neural enriched extracellular vesicles (NEE) that satisfies the first criterion—plasma obtained via a routine venous blood draw. However, the initial step of isolating and enriching NEE remains technically demanding, time-intensive, and costly at this time. We hypothesized that the identified dysregulated miRNAs would not be found in circulating blood plasma with similar fold regulation and diagnostic power as those derived from the cargo within NEE. We here present data that disproves our hypothesis and instead demonstrates that the 8-miRNA fingerprint is present in plasma as circulating miRNA, and the signal is even more robust than what we previously reported for NEE. This finding facilitates a simplified method for biomarker commercialization, hastening access for primary care doctors and patients. Methods Plasma Samples Plasma samples were tested from both ALS patients and controls (without a neurodegenerative disease diagnosis; Table 1, eMethods-Supplement 1). The majority of ALS samples (n = 390) were obtained from the U.S. National ALS Biorepository, a component of the National ALS Registry, maintained by the Centers for Disease Control and Prevention, and the Agency for Toxic Substances and Disease Registry (CDC, Advarra IRB_Pro00053269, collected between 2017 and 2018) with three ALS blood samples obtained through a Phase IIa clinical trial (NCT03580616, 2018-06-26, Dartmouth Institutional Review Board D18095, collected between 2019 and 2022). Control samples (n = 395) were obtained from Innovative Research Inc (Novi, MI, FDA Approval, #3003372368, collected across the USA prior to 2022) and Precision for Medicine (Norton, MA, CR00425931, collected across the USA from 2020 to 2022) and were all reported to be from normal controls without evidence of neurological disease. This study used only de-identified participant data and Advarra IRB (Pro00053269) determined that it "does not meet the DHHS definition of human subjects research under 45 CFR 46 and, therefore, does not require IRB oversight”. Sample collections were overseen by each respective institution (US National ALS Biorepository, CDC, Advarra Institutional Review Board, Pro00053269; Phase IIa clinical trial, NCT03580616 (2018-06-26), Dartmouth IRB D18095; Innovative Research Inc., Novi, MI, Food and Drug Administration (FDA) approval, #3003372368; Precision for Medicine, Norton, MA, CR00425931). Table 1Demographic data from ALS patient and control blood plasma samplesALSControlSex (F/M)159/234120/275Age range18–2931630–39121840–49367650–5910719560–691508870–7978280–8970 Table 1Demographic data from ALS patient and control blood plasma samplesALSControlSex (F/M)159/234120/275Age range18–2931630–39121840–49367650–5910719560–691508870–7978280–8970 Demographic data from ALS patient and control blood plasma samples All blood samples were collected in K2EDTA tubes and samples were blinded before being processed and analyzed. Samples from five different cohorts were analyzed, each cohort with unique patients and no patient overlap between cohorts. RNA Extraction/cDNA Synthesis We extracted total RNA from circulating blood plasma without first isolating NEE, as previously published [13–15]. We used a slightly different protocol for total RNA extraction which varies from the original protocol by the volume of lysis reagent and number and volume of ethanol washes (eMethods-Supplement 1). cDNA was synthesized as described previously [16]. Five-Stage Quality Control (QC) For every sample, five stages of QC were conducted prior to downstream target qPCR (eTable1-Supplement 1). qPCR of UniSp2, 4 and 5 spike-ins to measure success of RNA extraction: qPCR of UniSp2, 4 and 5 spike-ins to measure success of RNA extraction: qPCR of UniSp2, 4 and 5 spike-ins to measure success of RNA extraction: Each individual sample was assayed via qPCR for the following spike-ins; UniSp2, 4 and 5 where a “pass” was given if the SD for an individual sample varied less than three SD from the mean raw Cq for all samples in an amplicon group. 2.qPCR of UniSp6 and Caenorhabditis elegans cel-miR-39-3p spike-ins to measure efficacy of cDNA synthesis: 2.qPCR of UniSp6 and Caenorhabditis elegans cel-miR-39-3p spike-ins to measure efficacy of cDNA synthesis: qPCR of UniSp6 and Caenorhabditis elegans cel-miR-39-3p spike-ins to measure efficacy of cDNA synthesis: Each individual sample was assayed via qPCR for UniSp6 and cel-miR-39-3p. A “pass” was given if the SD for an individual sample varied less than three SDs from the mean raw Cq for all samples in an amplicon group. All samples passed spike-in QC and were cleared to proceed to the next QC stage. 3.Quantification of RNA: 3.Quantification of RNA: Quantification of RNA: We used qPCR to determine whether there was sufficient cDNA in each sample to proceed with downstream reactions. Five miRNAs that constitute a broad dynamic range (i.e., from low to high copy-number in human plasma) were quantified using qPCR; miR-142-3p, miR-451a, miR-23a-3p, miR-30c-5p, and miR-103a-3p. The presence of these targets, collectively referred to as “sample signal”, with raw Cqs less than 34 constitutes a “pass.” All samples passed this criterion and were cleared for use in the next stage of QC. 4.Hemolysis calculation: 4.Hemolysis calculation: Hemolysis calculation: Certain miRNAs are enriched in red blood cells, and if hemolysis occurs during plasma processing, the release of these miRNAs into the plasma can artefactually affect results. Hemolysis was quantified by qPCR of miR-451a (released during lysis of erythrocytes) and miR-23a-3p (stable in human plasma) [17] then a delta Cq (ΔCq) (miR-23a-3p – miR-451a) was calculated [18]. To determine the threshold where hemolysis deleteriously impacts our data, we conducted a preliminary study using plasma samples of varying degrees of pink (as determined by visible inspection), and qPCR of our eight miRNAs (Dunlop and Wood, 2024, data not shown). A ΔCq greater than or equal to 8 was sufficient to impact raw Cqs such that samples have the potential to be misclassified. Therefore, samples with ΔCq ≥ 8 were excluded from downstream experiments. 5.Amplification efficiency calculations: 5.Amplification efficiency calculations: Amplification efficiency calculations: Individual amplification efficiencies for all amplicons were calculated in LinRegPCR [19] by performing linear regression on the log-linear phase of the amplification curve derived from raw fluorescence data. An average PCR efficiency was calculated for all samples in each amplicon group (eTable2-Supplement 1). Quantitative Real-Time PCR (qPCR) qPCR was conducted as described previously [16]. Relative quantification using, gene-fold expression, and fold regulation was determined as described previously [13]. Sample Classification Algorithms Since machine learning classification algorithms are sensitive to sample size [20], we combined the complete dataset from all five cohorts. Analyses were performed using R Statistical Software v4.1.2 [21] using four library packages: data.Table1.17.0 [22]; glmnet4.1–8 [23, 24]; knitr1.49 [25]; caret7.0–1 [26]. We conducted a logistic regression on ΔCq data (not adjusted for interplate calibrators) with a Lasso feature selection using the following model parameters: seed set 42, multinomial family, MSE measure, Z-score standardization, relaxed glm fit without penalty, lambda sequence (−4, −1, length.out = 1000), 40 nfolds, 1e−7 convergence threshold, and 1e5 maximum iterations (eMethods-Supplement 1). All 8-miRNA identified as our ALS fingerprint were considered for model inclusion (miR-151a-3p, miR-151a-5p, miR-4454, miR-146a-5p, miR-10b-5p, miR-199a-3p, miR-199a-5p, miR-29b-3p). Results Consistent with ALS disease demographics, the majority of ALS patient blood samples tested were from male participants (234/393, 60%) as were the controls (275/395, 70%; Table 1). Age at disease onset was recorded by decade with 65% of ALS patients between 50 and 69 (27% 50–59; 38% 60–69) similar to controls individuals with 72% between the ages of 50 and 69 (49% 50–59; 22% 60–69). Eight miRNAs used to differentiate ALS positive plasma from donor samples who do not have ALS have previously been identified via NEE extraction from plasma, next-generation sequencing, and qPCR [13–15]. The aim of the current study was to verify that the 8-miRNA fingerprint identified in NEE could not be reproduced using circulating miRNA. Surprisingly, we report that the 8-miRNAs fingerprint did precisely replicate with fold-regulation values in circulating miRNA equal to or more pronounced than published previously (Table 2, Fig. 1). These results have now been repeated with five independent cohorts representing different patients in each group with a combined sample size of 788. Table 2Fold regulation values for eight ALS-related miRNAs comparing circulating blood plasma (new method) and neural-enriched extracellular vesicles (original method). Patients were different between cohorts and therefore there was no overlap between cohorts. The direction of fold-regulation remained the same between group comparisons (control relative to ALS), but the magnitude of the fold changes was increased in miR-4454 and miR-10b-5p in the newly developed methodOriginal method (NEE miRNA)New method (circulating miRNA) Banack et al. 2020 Cohort 1 Banack et al. 2020 Cohort 2 Banack et al. 2022 Cohort 3 Banack et al. 2024 Cohort 4 Cohort 1Cohort 2Cohort 3Cohort 4Cohort 5151a-3p1.52.21.12.13.51.83.03.03.1151a-5p1.43.21.11.82.32.22.43.44.5146a-5p1.21.41.11.62.31.52.21.51.54454(− 1.7)(− 1.8)(− 2.7)(− 2.6)(− 10.3)(− 12.3)(− 9.0)(− 4.6)(− 10.0)10b-5p(− 2.1)(− 7.0)(− 3.2)(− 7.4)(− 23.0)(− 14.0)(− 14.8)(− 14.5)(− 24.4)199a-3p1.42.712.02.71.93.42.82.6199a-5p1.94.21.12.53.82.86.75.64.629b-3p(− 1.7)(− 1.7)(− 1.5)(− 1.4)(− 1.3)(− 1.2)(− 1.5)(− 1.2)(− 1.4)TotalsTotals# ALS101050119189201325982100393# Controls101050150220201336082100395Fig. 1Box and whisker plots of 8-miRNA included as variables in a cross-validation logistic regression model for an ALS diagnostic test based on 788 ALS patients and controls. The center line represents the median of each group. The box represents the 25th and 75th percentiles. The whiskers depict 1.5 times each inter-quartile range. The outliers are shown as circles Table 2Fold regulation values for eight ALS-related miRNAs comparing circulating blood plasma (new method) and neural-enriched extracellular vesicles (original method). Patients were different between cohorts and therefore there was no overlap between cohorts. The direction of fold-regulation remained the same between group comparisons (control relative to ALS), but the magnitude of the fold changes was increased in miR-4454 and miR-10b-5p in the newly developed methodOriginal method (NEE miRNA)New method (circulating miRNA) Banack et al. 2020 Cohort 1 Banack et al. 2020 Cohort 2 Banack et al. 2022 Cohort 3 Banack et al. 2024 Cohort 4 Cohort 1Cohort 2Cohort 3Cohort 4Cohort 5151a-3p1.52.21.12.13.51.83.03.03.1151a-5p1.43.21.11.82.32.22.43.44.5146a-5p1.21.41.11.62.31.52.21.51.54454(− 1.7)(− 1.8)(− 2.7)(− 2.6)(− 10.3)(− 12.3)(− 9.0)(− 4.6)(− 10.0)10b-5p(− 2.1)(− 7.0)(− 3.2)(− 7.4)(− 23.0)(− 14.0)(− 14.8)(− 14.5)(− 24.4)199a-3p1.42.712.02.71.93.42.82.6199a-5p1.94.21.12.53.82.86.75.64.629b-3p(− 1.7)(− 1.7)(− 1.5)(− 1.4)(− 1.3)(− 1.2)(− 1.5)(− 1.2)(− 1.4)TotalsTotals# ALS101050119189201325982100393# Controls101050150220201336082100395 Fold regulation values for eight ALS-related miRNAs comparing circulating blood plasma (new method) and neural-enriched extracellular vesicles (original method). Patients were different between cohorts and therefore there was no overlap between cohorts. The direction of fold-regulation remained the same between group comparisons (control relative to ALS), but the magnitude of the fold changes was increased in miR-4454 and miR-10b-5p in the newly developed method Box and whisker plots of 8-miRNA included as variables in a cross-validation logistic regression model for an ALS diagnostic test based on 788 ALS patients and controls. The center line represents the median of each group. The box represents the 25th and 75th percentiles. The whiskers depict 1.5 times each inter-quartile range. The outliers are shown as circles We used a cross-validation logistic regression model to evaluate classification accuracy on samples from 393 ALS patients and samples from 395 healthy controls (i.e., without evidence of neurological disease). The model identified six miRNAs of the eight entered (miR-151a-3p, miR-151a-5p, miR-4454, miR-10b-5p, miR-199a-5p, miR-29b-3p) as sufficient to classify the 788 samples with 0.98 AUC (95% CI 0.97–0.99), 95% balanced accuracy (95% CI 93–97), 97% sensitivity (95% CI 96–98), and 93% specificity (95% CI 90–96). The positive and negative predictive values of the model were 0.93 (95% CI 0.91–0.96), and 0.97 (95% CI 0.96–0.98), respectively. Of interest is genetic data from 29 ALS patients in this dataset (7%) that had known ALS genetic mutations (20 C9orf, 3 SOD1, 1 ALS2, 1 VAPD, 2 SETX, 2 “Other”). All individuals with these genetic mutations correctly classified as ALS with this diagnostic test. Discussion There are no sporadic-ALS diagnostic biomarkers currently available and the need for rapid, accurate and minimally invasive biomarkers, from the perspective of ALS patients, is urgent. Rates of misdiagnosis for ALS are high and time from symptom onset to diagnosis can exceed a year, as physicians document clinical evidence of disease progression while also ensuring that no other reasonable alternative diagnosis is apparent [8]. Furthermore, ALS is a heterogeneous disease, and although a spinal presentation is most common (65%), it is also twice as likely to exceed 12 months time to diagnosis [27]. Physicians reported that rapid ALS diagnosis can be difficult since symptoms overlap with other conditions and the diagnostic delay is exacerbated by slow referral rates to specialists [9]. Delayed diagnosis can also be a barrier to successful clinical trials [28]. We report a robust ALS diagnostic test with high sensitivity (97%) and specificity (93%) and accuracy as measured by the receiver operating characteristic measure of area under the curve, 0.98 AUC (95% CI 0.97–0.99). Neurofilament light protein, which indicates general neuronal damage, is being successfully used in clinical drug trials as a secondary outcome measure for disease progression [29] but has limited value as an ALS diagnostic marker since it can also be elevated in other neurological conditions [30, 31]. The protein TDP-43 is present in the cytoplasm of neurons and as glial inclusions in both ALS and frontal temporal dementia, and has been investigated as a potential diagnostic biomarker, but the attempt to measure disease-specific forms in blood or cerebrospinal fluid so far has been unsuccessful [32]. Serum cardiac troponin-T is an interesting ALS biomarker candidate as it correlates with respiratory function and can detect progressive loss of vital capacity [33]. ALS, peer-reviewed, primary research papers since 2022 report using a variety of methods to obtain miRNA expression from a broad array of human samples including iPSCs, mesenchymal stem cells, iPSC-derived motor neurons, skeletal muscle, cerebrospinal fluid, and brain (e.g., [34–41]). Blood-based studies are also diverse using plasma, serum, whole blood, leukocytes, and lymphoblasts making comparisons to our research difficult (e.g., [36, 38, 42–45],). Evaluating only studies using blood plasma or serum and limiting the quantification methods to PCR (either qPCR or digital-PCR) still introduces a multitude of variability in results. This may be attributed to method of sample collection and storage, RNA extraction, extent of quality control, number of reference genes used for data normalization methods, and others [13]. Indeed, Wang et al. [46] observed higher miRNA concentrations in serum samples compared to the corresponding plasma samples from the same individuals. miRNA quantification can be influenced by the choice of methodology used for data analysis, and this must be considered when interpreting and comparing miRNA analyses [47]. Furthermore, a lack of detail in method reporting in many papers makes comparisons between studies difficult. Universally, samples sizes in published papers were small with ALS cohorts not exceeding 65 ALS patients making our large sample size unique (n = 393 ALS patients + 395 healthy controls) [48–53]. Researchers often plot a receiver operating characteristic curve (ROC) and report a metric known as the area under the receiver operating characteristic curve (AUC) as an estimate of the overall classification accuracy of a diagnostic test. AUC calculations, however, are known to be very sensitive to the study sample size [54]. Statisticians have conclusively demonstrated that a “small sample size leads to large inaccuracies in the estimated validation parameters associated with ROC analysis, even from well-behaved distributions” and warn, “ROC curves must be used with extreme caution unless one has a very large sample” [20]. Furthermore, the variance and error rates of the true population are not typically known making even projected 95% confidence intervals approximate estimates, highly dependent upon both the sample size and classification algorithms [20]. For these reasons, we intentionally chose not to calculate AUC metrics on previously published cohort data until we could evaluate the consistency of data across cohorts and subsequently combine the cohorts, since there was no patient overlap between cohorts, to generate a large dataset for classification algorithms ( [13], current data). We urge caution in comparing AUC values across studies as the error projection for AUC calculations from small sample sizes are generally not reported and an AUC calculated on a small sample size is unreliable and not likely to represent the population of interest [20, 54]. We have previously identified an 8-miRNA fingerprint that specifically and sensitively identified patients who have ALS using NEE extracted from plasma sourced from a standard blood draw [13]. We have successfully demonstrated that this test can identify patients at an early stage in pristine samples (early-stage disease diagnosis by a single neurologist with sample collection and processing strictly controlled) [14] and in samples sourced from a biobank at various stages of disease, diagnosed by different neurologists, and stored for many years [15]. We have also established that each of the eight miRNA identified as part of this ALS diagnostic test are involved in important biological pathways linked with ALS and neurodegeneration [13] (see also Rawat et al. [55]. However, the exact cellular origin and underlying pathogenic mechanism requires further research which could further our understanding of ALS disease etiology. In the current study, we compared results from miRNA extracted from blood plasma directly versus miRNA extracted from NEE after being enriched with immunoaffinity methods [13]. We report highly comparable data, but with larger fold-regulation changes for two down-regulated targets, thereby increasing the robustness of the assay by increasing the separation between groups. Optimization of this assay to eliminate the NEE step and improve method efficiency achieves the following: (1) cuts the processing time in half; (2) reduces the volume of plasma required by sixty-percent; (3) reduces requirement for technical expertise with the elimination of the NEE extraction and purification step; (4) reduces the likelihood of human error with the use of a liquid-handling robot; (5) increases the overall robustness of the assay, reflected in the larger fold changes for miR-4454 and miR-10b-5p; and (6) moves the assay closer to high-throughput automation. We propose that the process of identifying miRNA targets, following a NEE extraction method, allowed us to select miRNA that were disease relevant and suitable for use in disease classification (eMethods-Supplement 1). This method development process provided a more refined selection of neurally relevant miRNA leading to excellent ALS biomarker candidates, and therefore is particularly useful for biomarker discovery. Since the methods used for NEE extraction deplete the overall quantity of miRNA in the sample by as much as 90% [14, 15], the circulating miRNA method, as outlined in this paper, allow for a more robust miRNA signal to be identified once the biomarker discovery selection process has been completed. Thus, the current method, which still includes the extracellular vesicles in the blood, is better for diagnostic analyses of patient samples, but in isolation, would not have sufficiently narrowed down the selection of relevant miRNA targets. This assertion is supported by the fact that a search for relevant ALS blood-based miRNA biomarkers in circulating blood plasma has not yet produced targets with equivalent sensitivity and specificity as is reported here [52, 55]. Joilin et al. [39] further found that a correlation between differentially expressed non-coding RNA transcripts found in blood samples could not be replicated in cerebrospinal fluid and concluded that the signature initially identified in blood likely resulted from muscle atrophy. The reproducibility of our blood-based miRNA signature, now tested in 788 individuals, suggests that our process of miRNA discovery, identified through NEE, does provide a window into the brain. Supportive of this hypothesis is the finding that this diagnostic test was accurate for both sporadic and genetic ALS cases which suggests that we have identified miRNA that measure changes representative of the underlying ALS disease processes. Although the number of genetic cases in this study is small (29/393 or 7%), it is roughly consistent with the rate of genetic mutations found in the general ALS patient population [6]. Although further research is needed, this ALS diagnostic test will be of immediate benefit to patients and their primary care physicians when commercially available. Most of the blood samples used in this study were sourced from the US National ALS Biorepository which was developed by the Agency for Toxic Substances and Disease Registry (ATSDR) specifically to address the scientific gap found in other ALS biorepositories which were limited by “specimen type and availability, sample size, geographical coverage, and demographic characteristics” [56]. The pilot study for the US National ALS Biorepository began in 2013 with continued sample collections following full approval. The samples collected were both nationally representative and geographically diverse from individuals with ALS who were enrolled in the ATSDR’s congressionally-mandated National ALS Registry [56]. The US National ALS Biorepository complements the National ALS Registry which began in October of 2010 to collect population-based ALS data through self-enrolled survey modules accessed from an online portal. The samples used in this study were largely obtained from the US National ALS Biorepository and have associated self-reported data submitted through the ALS Registry. We note that the data on patient sex and age from the samples used in this study are consistent with recently published CDC database-wide studies (60% male in this study compared with 58–63% male in published data; age groups used in this study were 13% 18–49, 27% 50–59, 38% 60–69, and 22% 70 + compared with published numbers of 13% 18–49, 28% 50–59, 39% 60–69, and 20% 70 +) [57, 58]. Although the data associated with each sample in this study is limited, we expect the samples used in this study to be nationally representative across the United States as intended by the US National ALS Biorepository design [56]. Nevertheless, it is prudent to be cautious when considering the generalizability of the results. Furthermore, given the heterogeneity of ALS, the usefulness of this diagnostic biomarker would benefit from additional information related to phenotypic variables, disease onset, progression, and severity. The ability to answer these questions will require blood samples sourced from ALS clinics where extensive clinical data are also available. Comparing how the current diagnostic test accurately distinguishes ALS patients from a normal control group represents only the first step necessary for commercialization. Equally important is the ability of the diagnostic test to differentiate ALS from ALS-mimic diseases and other neurological diseases. Since miRNA can distinguish different types of cancer [59] and unique miRNA signatures have been found for Alzheimer’s disease [60] and Parkinson’s disease [61], it is a advisable to test the miRNA signature identified herein against other neurological diseases. We have previously demonstrated that the miRNAs used in the diagnostic test presented herein are statistically different from primary lateral sclerosis and Parkinson’s disease based on fold-regulation [13]. However, evaluation of the classification accuracy of ALS from other neurological diseases using machine learning tools requires a large comparative sample size before a meaningful analyses is possible [20, 54] and this work is underway in our laboratory. Nevertheless, we note that for primary care doctors and misdiagnosed patients, a quick blood test to assess the value of referring a patient to a neurologist and addressing the question, “are the symptoms neurological?” is warranted. Thus, the current research comparing 788 individual blood samples with a high degree of accuracy (95%) separating ALS patients from individuals without neurological disease could also be of diagnostic value, particularly to assess whether to transfer a patient to a neurologist early in the diagnostic process. Future studies should also evaluate how early this blood-based, diagnostic ALS test might be useful. This test was developed on blood samples from ALS patients at the time of diagnosis [14] and has been shown to accurately diagnose patients in both early and late stages of the disease (42% early-stage with ALSFRS-R scores ≥ 35, 42% late-stage with ALSFRS-R scores < 35, 16% no data [13]). Further research is required to determine if it can identify pre-symptomatic ALS patients. Given (a) that the disease process has a prodromal phase, (b) that changes in miRNA precede changes in proteins which are recognized at the time of symptom onset, and (c) that this test accurately identifies individuals carrying ALS gene mutations, we hypothesize that this test might have predictive utility. Testing ALS gene carriers prior to the development of symptoms would be a useful approach toward evaluating this hypothesis. In summary, using circulating miRNA rather than NEE-miRNA increases the biomarker sensitivity and specificity and at the same time improves the affordability of this assay owing to the smaller volume of reagents required, the reduced volume of plasma needed, and decreased requirement for technical expertise. The high degree of classification accuracy found with this large dataset along with these improved methods for diagnostic analyses make this assay a much more attractive option for commercialization. Supplementary Information Below is the link to the electronic supplementary material.ESM 1(298 KB PDF) (298 KB PDF)