Non-invasive cell-free DNA fragmentomics method and system for diagnosing facioscapulohumeral muscular dystrophy
A non-invasive cfDNA fragmentomics method using machine-learning classifiers and biomarker panels addresses the limitations of current FSHD diagnostics by analyzing DUX4-responsive genes, enabling early diagnosis and monitoring of FSHD through blood samples.
Patent Information
- Application Number
- PCT/US2026/027430
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-08
- Filing Date
- 2026-05-08
- Publication Date
- 2026-11-12
AI Technical Summary
Current diagnostic methods for Facioscapulohumeral Muscular Dystrophy (FSHD) are invasive, technically demanding, and cannot distinguish between genetically affected individuals who are clinically manifest and those who are asymptomatic, lacking information on disease activity and progression.
A non-invasive method using cell-free DNA (cfDNA) fragmentomics to analyze fragmentation patterns, nucleosome positioning, and coverage at DUX4-responsive genes to diagnose FSHD, employing machine-learning classifiers and biomarker panels to differentiate affected and unaffected subjects.
Enables early diagnosis, disease stratification, and longitudinal monitoring of FSHD through blood samples, distinguishing clinically active disease from asymptomatic carriers, and providing a pharmacodynamic biomarker for therapeutic interventions.
Smart Images

Figure US2026027430_12112026_PF_FP_ABST
Abstract
Description
Aty ref. 2848-355-PCT CU2025-209NON-INVASIVE CELL-FREE DNA FRAGMENTOMICS METHOD AND SYSTEM FOR DIAGNOSING FACIOSCAPULOHUMERAL MUSCULAR DYSTROPHYCROSS REFERENCE TO RELATED APPLICATION
[0001] This application claims the benefit of U.S. Provisional Application No.63 / 801,992, filed May 8, 2025, which is incorporated by reference herein in its entirety.FIELD OF THE INVENTION
[0002] This disclosure relates generally to the fields of molecular diagnostics, liquid biopsy, epigenomics, and neuromuscular disease. More particularly, the invention relates to methods, systems, compositions, and kits for the non-invasive diagnosis of Facioscapulohumeral Muscular Dystrophy (FSHD) through analysis of cell-free DNA (cfDNA) fragmentation patterns, also referred to herein as cfDNA fragmentomics, derived from a biological sample obtained from a subject. This disclosure provides a diagnostic platform that leverages the epigenomic information encoded within the fragment-length distributions, nucleosome positioning signatures, and coverage patterns of circulating cfDNA at genomic loci associated with the DUX4-driven transcriptional program characteristic of FSHD pathology.
[0003] The methods and systems disclosed herein are applicable to the diagnosis of both FSHD Type 1 (FSHD1) and FSHD Type 2 (FSHD2), to the monitoring of disease progression, to the assessment of therapeutic response in subjects undergoing treatment for FSHD, and to screening applications including but not limited to newborn screening. The invention occupies the intersection of next-generation sequencing technologies, computational biology, and clinical diagnostics for inherited neuromuscular disorders.BACKGROUNDA. Epidemiology and Clinical Features of FSHD
[0004] Facioscapulohumeral Muscular Dystrophy (FSHD) is the third most common form of inherited muscular dystrophy, after Duchenne muscular dystrophy and myotonic dystrophy. The estimated prevalence of FSHD ranges from approximately 1:15,000 to 1 :20,000 individuals in the general population, although recent studies suggest that the true prevalence may be higher due to underdiagnosis of mildly affected individuals and incomplete penetrance. FSHD is inherited in an autosomal dominant pattern with variable expressivity and age-dependent penetrance. The disease is characterized by progressive,Aty ref. 2848-355-PCT CU2025-209often asymmetric, weakness and wasting of the facial, scapular stabilizer, and upper arm muscles, with subsequent involvement of the trunk, pelvic girdle, and lower extremity musculature. Clinical onset typically occurs in the second decade of life, though infantileonset forms with severe phenotypes and later-onset forms with mild symptoms are well documented. Approximately 20% of genetically confirmed carriers remain clinically asymptomatic, complicating clinical diagnosis and genetic counseling.
[0005] The clinical heterogeneity of FSHD presents significant diagnostic challenges. Symptoms may initially be subtle, consisting of difficulty whistling, closing the eyes during sleep, or raising the arms above the head. Progressive scapular winging is a hallmark feature, yet it may be confused with other scapuloperoneal syndromes.Extramuscular manifestations include retinal vasculopathy in severe cases, high-frequency hearing loss, and, rarely, respiratory insufficiency. The variable age of onset, the asymmetric distribution of weakness, and the overlap with other neuromuscular conditions frequently result in diagnostic delays of years to decades. There is therefore a critical need for objective, accessible, and minimally invasive diagnostic methods that can confirm or exclude FSHD independently of the subjective clinical assessment.B. Molecular Genetics of FSHD
[0006] The molecular pathogenesis of FSHD is uniquely complex among muscular dystrophies and centers on the D4Z4 macrosatellite repeat array located in the subtelomeric region of chromosome 4q35. In unaffected individuals, the D4Z4 array consists of 11 to 100 tandemly repeated units, each approximately 3.3 kilobases (kb) in length. The D4Z4 array is present in nearly identical form on chromosome 10q26, but only the chromosome 4q35 array is associated with FSHD pathogenesis due to the requirement for a specific haplotype in the region distal to the array.
[0007] In FSHD Type 1 (FSHD1), which accounts for approximately 95% of all FSHD cases, the disease is caused by contraction of the D4Z4 repeat array on one allele of chromosome 4q35 to between 1 and 10 repeat units. This contraction must occur on a chromosomal background carrying the permissive 4qA haplotype, which provides a polyadenylation signal (PAS) immediately distal to the last D4Z4 repeat unit. The 4qA haplotype contains a pLAM region that includes a canonical polyadenylation signal sequence (ATT AAA) enabling stabilization of the mRNA transcript produced from the DUX4 retrogene encoded within the most distal D4Z4 unit. In the absence of theAty ref. 2848-355-PCT CU2025-209permissive 4qA haplotype (for instance, on the non-permissive 4qB haplotype), DUX4 mRNA is unstable and rapidly degraded, and D4Z4 contraction does not result in disease.
[0008] The inverse correlation between D4Z4 repeat array size and disease severity has been well documented. Individuals with 1 to 3 repeat units generally exhibit earlier onset and more severe disease, while those with 7 to 10 repeat units may have later onset or reduced penetrance. However, considerable phenotypic variability exists even within families carrying identical contractions, suggesting the influence of epigenetic and genetic modifier loci.
[0009] FSHD Type 2 (FSHD2) accounts for approximately 5% of FSHD cases and is genetically distinct from FSHD1. In FSHD2, the D4Z4 repeat array on chromosome 4q35 is not contracted (i.e., the array contains 11 or more repeat units), but the array nonetheless undergoes epigenetic de-repression. FSHD2 is most commonly caused by heterozygous loss-of-function mutations in the SMCHD1 gene (Structural Maintenance of Chromosomes Flexible Hinge Domain-Containing 1), located on chromosome 18p 11.32. SMCHD1 is an epigenetic regulator required for the spreading and maintenance of DNA methylation at repetitive elements, including D4Z4. Less frequently, FSHD2 is caused by mutations in DNMT3B (DNA methyltransferase 3B), which is involved in de novo DNA methylation. In either case, the result is D4Z4 hypomethylation on a permissive 4qA allele, leading to de-repression of DUX4 transcription in a manner functionally analogous to the contraction-mediated mechanism of FSHD1. Notably, SMCHD1 mutations can also act as disease modifiers in FSHD1, exacerbating disease severity when present in combination with a contracted D4Z4 array.C. DUX4 Biology and Pathogenesis
[0010] Each D4Z4 repeat unit contains a copy of the DUX4 (Double Homeobox 4) retrogene. DUX4 encodes a transcription factor that is normally expressed in the germline and in the early cleavage- stage embryo, where it plays a role in zygotic genome activation. In somatic tissues of healthy individuals, the D4Z4 array is maintained in a repressed chromatin state through multiple epigenetic mechanisms, including dense CpG methylation of the D4Z4 DNA, enrichment of the repressive histone modification trimethylation of histone H3 at lysine 9 (H3K9me3), recruitment of heterochromatin protein 1 gamma (HPly), and engagement of the cohesin complex. This multi-layered silencing ensures that DUX4 is not expressed in somatic cells, including skeletal muscle.Aty ref. 2848-355-PCT CU2025-209
[0011] In FSHD, whether caused by D4Z4 contraction (FSHD1) or SMCHD1 / DNMT3B mutation (FSHD2), the repressive chromatin structure at D4Z4 is compromised. Loss of DNA methylation, reduction of H3K9me3, and displacement of HPly and cohesin from the contracted or hypomethylated D4Z4 array result in a chromatin state permissive for transcription. DUX4 mRNA is transcribed from the most distal (telomere-proximal) D4Z4 unit and, when produced on a 4qA haplotype background, is stabilized by the polyadenylation signal in the adjacent pLAM region. The resulting DUX4 mRNA is translated into a functional transcription factor protein.
[0012] DUX4 protein contains two paired-type homeodomains that mediate its binding to specific DNA recognition sequences. The consensus DUX4 binding motif has been characterized as TAATCTAATCA and related degenerate sequences. Upon expression in differentiated skeletal muscle cells (myotubes and myofibers), DUX4 functions as a potent transcriptional activator, binding to promoters and enhancers across the genome and activating a cascade of target genes that are normally restricted to the germline, the preimplantation embryo, and certain immune-privileged tissues. The DUX4 transcriptional program includes activation of genes encoding germline-specific transcription factors (ZSCAN4, LEUTX, DUXA), chromatin remodeling factors (KDM4E), RNA processing factors, and members of the PRAMEF (Preferentially Expressed Antigen in Melanoma Family) gene family (PRAMEF1, PRAMEF2, PRAMEF12, and others). DUX4 also activates genes involved in innate immunity and inflammatory signaling, as well as members of the TRIM43 and MBD3L2 gene families.
[0013] A particularly significant aspect of DUX4 transcriptional activity is the activation of endogenous retroelements, specifically members of the HERVL (Human Endogenous Retrovirus-L) family and their associated long terminal repeats (LTRs), including the MLT2A1 subfamily. These retroelements, which are normally silenced in somatic cells, serve as alternative promoters for numerous transcripts when activated by DUX4, contributing to the aberrant transcriptional landscape of FSHD muscle cells.
[0014] The downstream pathological consequences of aberrant DUX4 expression in skeletal muscle are multifaceted. DUX4 target gene activation leads to: (a) oxidative stress through mitochondrial dysfunction and dysregulation of reactive oxygen species (ROS) homeostasis; (b) activation of apoptotic pathways, including the p53-dependent and caspase-mediated apoptotic cascades, resulting in myofiber death; (c) inhibition of myogenic differentiation through suppression of key myogenic regulators including PAX7Aty ref. 2848-355-PCT CU2025-209and MYODI, thereby impairing the regenerative capacity of muscle satellite cells; (d) activation of innate immune signaling, leading to infiltration of inflammatory cells (macrophages, T cells) into affected muscles; and (e) progressive replacement of contractile muscle tissue by fibrotic connective tissue and adipose tissue, resulting in the clinical phenotype of muscle weakness and wasting. Importantly, DUX4 expression in FSHD muscle is stochastic and occurs in only a small fraction of myonuclei at any given time (estimated at 0.1-1% of nuclei), yet the downstream effects propagate through cell-non-autonomous mechanisms to affect neighboring cells and the tissue microenvironment.D. Limitations of Current FSHD Diagnostic Methods
[0015] The current gold standard for genetic confirmation of FSHD 1 is Southern blot analysis combined with pulsed-field gel electrophoresis (PFGE). In this method, high-molecular-weight genomic DNA is extracted from peripheral blood lymphocytes, digested with restriction enzymes (typically EcoRI in combination with Bini or XapI to distinguish chromosome 4-derived from chromosome 10-derived D4Z4 fragments), separated by PFGE, and hybridized with a D4Z4-specific probe (pl3E-l 1). The resulting band pattern allows determination of the number of D4Z4 repeat units and the haplotype (4qA vs. 4qB). While this method has served as the diagnostic standard for over two decades, it suffers from significant limitations: (a) it requires large quantities (5-10 pg) of high-molecular-weight, intact genomic DNA, necessitating a fresh blood draw and careful sample handling; (b) it is technically demanding, requiring specialized equipment and expertise not available in most clinical laboratories; (c) turnaround time is typically 4 to 8 weeks; (d) interpretation can be complicated by mosaicism, complex rearrangements, and crosshybridization between chromosomes 4 and 10; and (e) it does not detect FSHD2 unless supplemented by D4Z4 methylation analysis and SMCHD1 / DNMT3B sequencing.
[0016] Alternative diagnostic approaches have been developed to address some of these limitations. Molecular combing, a technique in which individual DNA molecules are stretched on a glass surface and hybridized with fluorescent probes, allows direct visualization and measurement of D4Z4 array length at single-molecule resolution. Optical genome mapping (OGM), using platforms such as the Bionano Saphyr system, can resolve D4Z4 repeat structures through labeling of specific sequence motifs along ultra-long DNA molecules. Long read sequencing technologies, including Pacific Biosciences (PacBio) single molecule real time (SMRT) sequencing and Oxford Nanopore Technologies (ONT) nanopore sequencing, have recently demonstrated the ability to read through D4Z4 repeatAty ref. 2848-355-PCT CU2025-209arrays and determine repeat number, haplotype, and methylation status in a single assay. However, each of these alternative methods still requires extraction of high-molecular-weight genomic DNA from nucleated cells, which in practice necessitates a blood draw with specialized sample handling. Moreover, the cost of long read sequencing and OGM instruments, the computational complexity of analyzing repetitive sequences, and the limited availability of these technologies in clinical settings restrict their widespread adoption.
[0017] All of the foregoing diagnostic methods share a fundamental limitation: they analyze genomic DNA obtained directly from nucleated cells, typically peripheral blood leukocytes. This means they characterize the genetic lesion (D4Z4 contraction or SMCHD1 mutation) itself, but do not directly assess the downstream molecular consequence of that lesion-namely, the aberrant expression of DUX4 and its target gene program in skeletal muscle. Consequently, these methods cannot readily distinguish between genetically affected individuals who are clinically manifest and those who carry the genetic lesion but remain asymptomatic (non-penetrant carriers). Furthermore, they provide limited information about disease activity, disease progression, or therapeutic response, parameters that are increasingly important as DUX4-targeted therapies enter clinical trials.E. Cell-Free DNA as a Diagnostic Analyte
[0018] Cell-free DNA (cfDNA) refers to fragments of genomic DNA circulating in the bloodstream outside of cells, primarily in blood plasma. cfDNA is released into the circulation through multiple mechanisms, including apoptosis, necrosis, and active secretion from cells throughout the body. In healthy individuals, the majority of circulating cfDNA is derived from apoptosis of hematopoietic cells, particularly neutrophils, lymphocytes, and other white blood cells. However, cfDNA contributions from other tissues, including the liver, endothelium, and skeletal muscle, have been documented and can be quantified using tissue of origin deconvolution methods. The average fragment length of cfDNA is approximately 167 base pairs (bp), corresponding to the length of DNA wrapped around a single nucleosome core particle (-147 bp) plus linker DNA (-20 bp), reflecting the preferential protection of nucleosome-associated DNA from nuclease digestion during apoptosis. cfDNA concentrations in plasma of healthy individuals typically range from 1 to 10 ng per milliliter (ng / mL), though concentrationsAty ref. 2848-355-PCT CU2025-209may be significantly elevated in pathological states including cancer, trauma, autoimmune diseases, and myopathies.
[0019] In the context of muscular dystrophies and other myopathies characterized by ongoing muscle cell death, the contribution of skeletal muscle-derived cfDNA to the total circulating cfDNA pool is increased relative to healthy individuals. This increase in muscle-derived cfDNA has been demonstrated in Duchenne muscular dystrophy and other conditions associated with elevated serum creatine kinase levels, which serve as a proxy for muscle membrane damage and cellular content release. In FSHD, the ongoing apoptosis and necrosis of skeletal muscle fibers resulting from aberrant DUX4 expression is expected to release genomic DNA fragments from muscle cells into the circulation. Critically, these muscle-derived cfDNA fragments carry the epigenomic imprint, including the nucleosome positioning, chromatin accessibility, and transcription factor binding patterns-of the muscle cells from which they derive.
[0020] The field of cfDNA fragmentomics has emerged as a powerful approach for extracting biological information from cfDNA beyond its sequence content. cfDNA fragmentomics encompasses the analysis of multiple features of cfDNA fragments, including but not limited to: (a) fragment-length distributions, both globally and at specific genomic loci; (b) fragment end-motif frequencies, which are the preferred sequences at the 5' and 3' termini of cfDNA fragments, which reflect the nuclease preferences of the tissue of origin; (c) nucleosome positioning and occupancy patterns inferred from cfDNA coverage depth at single base resolution across the genome; (d) nucleosome footprinting at regulatory elements such as promoters, enhancers, and transcription factor binding sites; and (e) the Windowed Protection Score (WPS), a metric that quantifies the relative protection of DNA at a given position by nucleosomes versus linker DNA. These fragmentomic features collectively encode information about the epigenomic and transcriptomic state of the cells that contributed cfDNA to the plasma, enabling inference of gene expression, chromatin accessibility, and tissue of origin without the need for direct tissue sampling.F. Statement of the Unmet Need
[0021] Despite the significant advances in cfDNA fragmentomics for applications in oncology (cancer detection, minimal residual disease monitoring), prenatal diagnostics (noninvasive prenatal testing), and organ transplantation (allograft rejection monitoring), no cfDNA-based, noninvasive diagnostic method has been developed for FSHD. Such aAty ref. 2848-355-PCT CU2025-209method would address multiple unmet clinical needs, including: (a) providing a simple, blood-based diagnostic test that avoids the technical complexity, cost, and long turnaround time of Southern blot / PFGE; (b) enabling detection not merely of the genetic lesion but of its functional downstream consequence (the DUX4-driven transcriptional program) thereby potentially distinguishing clinically active disease from asymptomatic carriage; (c) facilitating longitudinal monitoring of disease activity and progression through serial blood sampling; (d) serving as a pharmacodynamic biomarker for DUX4-targeted therapeutic interventions, including antisense oligonucleotides, small interfering RNAs, and small molecule inhibitors currently in clinical development; (e) enabling populationbased screening, including newborn screening; and (f) providing a research tool to advance understanding of FSHD biology, natural history, and the relationship between DUX4 transcriptional activity and clinical phenotype. This disclosure addresses these unmet needs by providing a cfDNA fragmentomics-based method and system for diagnosing FSHD.SUMMARY
[0022] The present disclosure provides methods, systems, biomarker panels, computer-implemented methods, and kits for non-invasive detection, diagnosis, classification, and monitoring of Facioscapulohumeral Muscular Dystrophy (FSHD) using cell-free DNA (cfDNA) fragmentomic features measured from biological samples such as plasma, serum, urine, or saliva. The disclosed approaches leverage characteristic cfDNA fragmentation patterns at genomic regions associated with DUX4-responsive genes and at D4Z4 macrosatellite repeat regions to distinguish FSHD-affected subjects from unaffected subjects.
[0023] In certain embodiments, the methods comprise obtaining cfDNA from a biological sample of a subject, sequencing the cfDNA to generate sequencing reads, mapping the sequencing reads to a reference genome, and determining one or more cfDNA fragmentomic features at genomic regions associated with DUX4-responsive genes.Fragmentomic features include, but are not limited to, cfDNA fragment-length distributions, nucleosome-positioning scores, promoter-subnucleosome enrichment (PSE) values, end-motif frequency profiles, and cfDNA coverage depth profiles. In some embodiments, the fragment-length distribution is quantified using fragment-length classes comprising subnucleosomal fragments (about 35-130 bp), nucleosomal fragments (about 130-200 bp), overlapping dinucleosomal fragments (about 200-330 bp), andAty ref. 2848-355-PCT CU2025-209dinucleosomal fragments (about 330-400 bp). In certain embodiments, the PSE value is computed as a ratio of subnucleosomal and overlapping dinucleosomal fragments to nucleosomal and dinucleosomal fragments within a promoter window or +1 nucleosome window of a DUX4-responsive gene.
[0024] The methods further comprise comparing one or more cfDNA fragmentomic features to a reference value derived from one or more control subjects or a reference population, and classifying the subject as having FSHD, being at risk for FSHD, or being unaffected based on a deviation of the fragmentomic features from the reference value. In some embodiments, the classification is performed using a trained machine-learning classifier, such as a random forest, gradient-boosted decision tree, support vector machine, logistic regression model, neural network, or combination thereof. In certain embodiments, a composite FSHD fragmentomic score is computed by aggregating two or more fragmentomic features across multiple DUX4-responsive gene loci.
[0025] The disclosure further provides biomarker panels comprising cfDNA fragmentomic biomarkers measured at genomic loci corresponding to DUX4-responsive genes. In some embodiments, the biomarker panel comprises at least two fragmentomic biomarkers selected from fragment-length distributions, PSE values, end-motif frequency profiles, and nucleosome-positioning scores. In certain embodiments, the biomarker panel discriminates between FSHD-affected and unaffected subjects with a sensitivity of at least about 70% and a specificity of at least about 70%.
[0026] Systems and computer-implemented methods are also provided for processing cfDNA sequencing data, computing fragmentomic features, comparing the features to reference values, and generating a classification output. In some embodiments, the system comprises a sequencer, one or more processors, and a non-transitory computer-readable storage medium storing instructions for performing the disclosed operations. In certain embodiments, the computer-implemented methods further comprise computing PSE values, nucleosome-positioning maps, end-motif diversity scores, or composite fragmentomic scores, and generating visual reports depicting fragmentomic features relative to reference values.
[0027] Kits for performing the disclosed methods are further provided. In certain embodiments, the kits comprise a cfDNA isolation reagent, a sequencing library preparation reagent set, capture probes or amplification primers directed to genomicAty ref. 2848-355-PCT CU2025-209regions corresponding to DUX4-responsive genes, and instructions for computing fragmentomic features and comparing the features to reference values to detect FSHD.
[0028] The disclosed methods, systems, biomarker panels, computer-implemented methods, and kits enable non-invasive detection and monitoring of FSHD using cfDNA fragmentomic signatures, providing a powerful alternative to invasive muscle biopsy and facilitating early diagnosis, disease stratification, and longitudinal assessment of disease progression or therapeutic response.BRIEF DESCRIPTION OF THE DRAWINGS
[0029] FIGS. 1A, IB, and 1C demonstrate cfDNA extraction and sequencing. FIG. 1A shows the total quantity of cfDNA extracted from approximately 4 ml of plasma from each donor displayed in a boxplot, categorized according to FSHD status. No significant difference in cfDNA quantity was observed concerning the presence of FSHD. FIG. IB is a representative trace from Agilent TapeStation illustrating cfDNA sequencing libraries prepared for Illumina sequencing. FIG. 1C shows the number of unique fragments sequenced from the cfDNA of each donor depicted as a boxplot, also categorized by FSHD status. No significant difference in sequencing depth was detected with respect to FSHD status.
[0030] FIG. 2 demonstrates PCA of promoter subnucleosome enrichment showing separation based on FSHD status when PC2 and PC4 loadings are plotted against each other.
[0031] FIGS. 3 A and 3B show the results of a fourfold cross-validation analysis to identify gene signatures predictive of the presence of FSHD. FIG. 3A shows the Ascore, representing the difference between the average PSE scores of positive features and the average PSE scores of negative features, calculated for the held-out test set in each fold. The positive and negative features were identified from the training set of the corresponding fold. FIG. 3B shows the positive and negative features (genes) that were common across all four folds utilized to determine the Ascore across the entire cohort.
[0032] FIGS. 4A and 4B show the distribution of promoter subnucleosome enrichment (PSE) score for individual genes across the study cohort categorized by clinical severity (0_None: unaffected individuals, l_Mild: CSS 0-3, 2_Mid: CSS 4-6; 3_Severe: CSS >6. FIG. 4A shows genes representing DUX4 signature for the whole cohort. FIG. 4A shows genes representing DUX4 signature for donors under 40.Aty ref. 2848-355-PCT CU2025-209
[0033] FIG. 5 Heatmap identification of genes that feature PSE that correlate with clinical severity. For each gene, the Z-score of the PSE was calculated comparing the mean and standard error of mean from each category of clinical severity to the unaffected category. Genes showing an increase with increasing severity or a decrease with increasing severity were chosen and plotted as a heatmap.DETAILED DESCRIPTIONA. Definitions
[0034] Unless otherwise defined herein, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. The following definitions are provided to facilitate understanding of the disclosure and are not intended to limit the scope of the claims.
[0035] As used herein, the term "about" means + / -10% of the recited value. Use of "about" is contemplated in reference to all ranges and values recited herein.
[0036] As used herein, the term “cell-free DNA” or “cfDNA” refers to extracellular DNA fragments present in a biological fluid, including but not limited to plasma, serum, urine, or saliva. cfDNA may be derived from apoptotic, necrotic, or actively secreting cells and may include fragments of varying lengths.
[0037] As used herein, the term “fragmentomic feature” refers to any quantitative or qualitative characteristic of cfDNA fragments derived from sequencing data, including but not limited to fragment-length distributions, nucleosome-positioning scores, promoter-subnucleosome enrichment (PSE) values, end-motif frequency profiles, cfDNA coverage depth profiles, jagged-end indices, orientation-aware fragmentation patterns (OAFP), and composite fragmentomic scores.
[0038] As used herein, the term “fragment-length distribution” refers to the distribution of cfDNA fragment lengths within a sample or within a defined genomic region, optionally categorized into fragment-length classes such as subnucleosomal fragments (about 35-130 bp), nucleosomal fragments (about 130-200 bp), overlapping dinucleosomal fragments (about 200-330 bp), and dinucleosomal fragments (about 330-400 bp).
[0039] As used herein, the term “promoter-subnucleosome enrichment” or “PSE” refers to a ratio quantifying the relative abundance of subnucleosomal and overlapping dinucleosomal cfDNA fragments compared to nucleosomal and dinucleosomal cfDNA fragments within a defined promoter or +1 nucleosome window. In certain embodiments,Aty ref. 2848-355-PCT CU2025-209the PSE value is computed as: (#(35-130 bp) + #(200-330 bp)) / (#(130-200 bp) + #(330-400 bp)).
[0040] As used herein, the term “promoter window” refers to a genomic interval extending upstream and downstream of a transcription start site (TSS) of a gene. In certain embodiments, the promoter window spans from about 2,000 base pairs upstream to about 2,000 base pairs downstream of the TSS. In other embodiments, the promoter window spans from about 1,000 base pairs upstream to about 1,000 base pairs downstream of the TSS.
[0041] As used herein, the term “+1 nucleosome window” refers to a genomic interval centered on the +1 nucleosome position downstream of a TSS, typically extending approximately 100 base pairs upstream and 100 base pairs downstream of the +1 nucleosome dyad.
[0042] As used herein, the term “nucleosome-positioning score” refers to a metric derived from cfDNA fragment endpoints or midpoints that reflects nucleosome occupancy or protection at a genomic locus, including but not limited to a windowed protection score (WPS) or a nucleosome-occupancy likelihood (NOL).
[0043] As used herein, the term “end-motif frequency profile” refers to the frequency of nucleotide motifs, including but not limited to 4-mer motifs, at the 5' ends and / or 3' ends of cfDNA fragments mapped to a genomic region.
[0044] As used herein, the term “DUX4-responsive gene” refers to a gene whose transcription is directly or indirectly activated by DUX4 expression, including but not limited to ZBTB17, ZSCAN4, TRIM43, MBD3L2, LEUTX, KDM4E, DUXA, PRAMEF1, KHDC1L, TRIM49, RFPL2, TRIM43B, SLC34A2, CCNA1, andH3Yl.
[0045] As used herein, the term “reference value” refers to a value, distribution, or threshold derived from one or more control subjects or a reference population, against which one or more cfDNA fragmentomic features from a subject are compared.
[0046] As used herein, the term “composite fragmentomic score” refers to a combined metric derived from two or more fragmentomic features, optionally comprising a weighted sum, a principal-component projection, or a machine-learning-derived score.
[0047] As used herein, the term “machine-learning classifier” refers to any computational model trained to distinguish between FSHD-affected and unaffected subjects based on cfDNA fragmentomic features, including but not limited to random forests,Aty ref. 2848-355-PCT CU2025-209gradient-boosted decision trees, support vector machines, logistic regression models, neural networks, or combinations thereof.
[0048] As used herein, the term “D4Z4 macrosatellite repeat region” refers to the tandem repeat array located on chromosome 4q35 or chromosome 10q26, including adjacent regulatory elements such as CpG sites, promoters, and polyadenylation signal (PAS) regions relevant to DUX4 expression.
[0049] As used herein, the term “clinical severity score” refers to any quantitative measure of FSHD disease severity, including but not limited to the Clinical Severity Score (CSS), Reachable Workspace (RWS), or a composite FSHD outcome measure.B. Mechanistic Biology Linking FSHD to cfDNA Signatures1. D4Z4 Contraction and Epigenetic De-Repression
[0050] The causal chain linking the genetic lesion in FSHD to a detectable cfDNA fragmentomic signature begins with the epigenetic de-repression of the D4Z4 macrosatellite repeat array on chromosome 4q35. In unaffected individuals, the D4Z4 array is maintained in a constitutively heterochromatic state characterized by dense DNA methylation at CpG dinucleotides within and flanking the D4Z4 repeats, enrichment of the repressive histone mark H3K9me3 (trimethylation of histone H3 at lysine 9), and recruitment of the heterochromatin-associated proteins HPly (heterochromatin protein 1 gamma, encoded by CBX3) and members of the cohesin complex. These epigenetic marks act cooperatively to prevent transcription from the D4Z4 units and maintain the array in a compact, transcriptionally silent chromatin configuration.
[0051] In FSHD1, contraction of the D4Z4 array to 1-10 repeat units on one allele reduces the total number of D4Z4 units available to nucleate and spread heterochromatin. Experimental studies have demonstrated that the contracted array exhibits reduced levels of CpG methylation, reduced H3K9me3, and decreased occupancy of HPly and cohesin relative to arrays of normal length. The threshold effect (the requirement that the array be contracted below a critical size (approximately 10 units)) is consistent with a model in which a minimum number of D4Z4 units is required to establish a stable heterochromatic domain through the spreading of repressive marks from one unit to the next.
[0052] In FSHD2, the D4Z4 array is of normal length (11 or more units), but the epigenetic regulators responsible for maintaining D4Z4 methylation are compromised. Heterozygous loss-of-function mutations in SMCHD1 impair the ability of the SMCHD1 protein to maintain and spread DNA methylation at D4Z4 and other repetitive elements.Aty ref. 2848-355-PCT CU2025-209SMCHD1 is a non-canonical member of the SMC protein family that plays a role in X chromosome inactivation and silencing of clustered gene families. Its loss of function results in D4Z4 hypomethylation on all alleles, but disease manifestation requires the cooccurrence of a permissive 4qA haplotype. Similarly, rare mutations in DNMT3B, a de novo DNA methyltransferase, can cause D4Z4 hypomethylation and FSHD2. In both FSHD1 and FSHD2, the net result of epigenetic de-repression is the transcriptional activation of DUX4 from the most distal D4Z4 unit, producing a DUX4 mRNA that is stabilized by the polyadenylation signal provided by the 4qA haplotype.2. DUX4 Transcription Factor Activity and Target Gene Activation
[0053] The DUX4 protein, upon translation in skeletal muscle cells, localizes to the nucleus and functions as a transcriptional activator. DUX4 contains two paired-type homeodomains that mediate sequence specific DNA binding. Genome-wide binding studies using chromatin immunoprecipitation followed by sequencing (ChlP-seq) have revealed that DUX4 binds to thousands of sites across the human genome, with a strong preference for the core motif TAATCTAATCA and related degenerate sequences. DUX4 binding sites are enriched at the promoters and enhancers of its target genes, and DUX4 binding is associated with the recruitment of transcriptional coactivators, including p300 and CBP (CREB-binding protein), which possess histone acetyltransferase activity. The recruitment of p300 / CBP to DUX4-bound regions results in the deposition of activating histone marks, particularly H3K27ac (acetylation of histone H3 at lysine 27) and H3K4me3 (trimethylation of histone H3 at lysine 4), converting the local chromatin environment from a repressed to an active state.
[0054] The transcriptional targets of DUX4 include a distinctive set of genes that are normally expressed exclusively in the germline, the pre-implantation embryo, or the cleavage- stage zygote. Representative members of this DUX4-activated gene program include: (a) ZSCAN4 (zinc finger and SCAN domain-containing 4), a transcription factor involved in telomere maintenance in embryonic stem cells; (b) TRIM43 (tripartite motifcontaining 43), an E3 ubiquitin ligase with roles in chromatin decondensation during zygotic genome activation; (c) MBD3L2 (methyl-CpG binding domain protein 3-like 2), a component of the NuRD complex variant expressed in the testis; (d) LEUTX (leucine twenty homeobox), a homeodomain transcription factor active in 4- to 8-cell embryos; (e) KHDC1L (KH domain-containing 1 -like), a gene involved in oocyte and early embryo function; (f) PRAMEF1, PRAMEF2, PRAMEF12, and other members of the PRAMEFAty ref. 2848-355-PCT CU2025-209gene family, which encode cancer-testis antigens; (g) DUXA (double homeobox A), a paralog of DUX4 that amplifies the DUX4 transcriptional cascade; and (h) KDM4E (lysine demethylase 4E), a histone demethylase that removes H3K9me3, potentially further de-repressing chromatin at other loci.
[0055] DUX4 also activates the transcription of endogenous retroelements, most notably members of the HERVL (Human Endogenous Retrovirus-L) family and their associated solo long terminal repeats (LTRs), including the MLT2A1 and MLT2A2 subfamilies. In normal somatic cells, these retroelements are silenced by DNA methylation and repressive histone modifications. DUX4-mediated activation of HERVL LTRs creates novel promoters that drive the transcription of chimeric transcripts, contributing to the aberrant transcriptome of FSHD muscle. Additionally, DUX4 activates genes involved in innate immune signaling, double-stranded RNA sensing, and interferon response pathways, which may contribute to the inflammatory component of FSHD pathology.3. Downstream Pathological Consequences of DUX4 Expression
[0056] The expression of DUX4 and its target gene program in differentiated skeletal muscle cells triggers a cascade of pathological events that ultimately lead to myofiber death and muscle tissue degradation. These events include, but are not limited to, the following. Oxidative stress is induced through disruption of mitochondrial function and upregulation of reactive oxygen species (ROS)-generating pathways, overwhelming the cellular antioxidant defenses and causing oxidative damage to proteins, lipids, and DNA. Apoptosis is triggered through both p53-dependent and p53-independent mechanisms, with activation of the intrinsic (mitochondrial) apoptotic pathway involving cytochrome c release, apoptosome formation, and activation of caspase-9 and executioner caspases (caspase-3 and caspase-7). Myogenic differentiation is impaired through DUX4-mediated suppression of PAX7, a transcription factor essential for muscle satellite cell identity and self-renewal, and MYODI, a master regulator of myogenic commitment and differentiation; this impairment compromises the regenerative capacity of the muscle tissue. Immune cell infiltration occurs as DUX4-activated muscle cells express chemokines and cytokines that recruit macrophages, CD8+ T cells, and other inflammatory cells to affected muscles, establishing a chronic inflammatory milieu.Finally, progressive loss of contractile myofibers is accompanied by their replacement with fibrotic connective tissue and adipose tissue, a process driven by the activity of fibroblasts and adipogenic progenitor cells within the muscle interstitium.Aty ref. 2848-355-PCT CU2025-209
[0057] The net result of these pathological processes is the progressive, often asymmetric, weakness and wasting of skeletal muscles that defines the clinical phenotype of FSHD. Importantly for this disclosure, the ongoing death of muscle cells through apoptosis and necrosis releases their genomic DNA content into the extracellular space and ultimately into the bloodstream, where it circulates as cfDNA.4. Muscle Cell Death and cfDNA Release
[0058] The release of genomic DNA from dying muscle cells into the circulation is a critical step in the mechanism by which this disclosure detects FSHD. During apoptosis, the activation of caspase-activated DNase (CAD, also known as DFF40, DNA Fragmentation Factor 40 kDa subunit) cleaves chromosomal DNA preferentially at the linker DNA between nucleosomes, generating DNA fragments corresponding to mono-, di-, and tri-nucleosomal units. These nucleosome-associated DNA fragments are released into the extracellular space and subsequently enter the bloodstream, where they circulate as cfDNA complexed with histone octamers. During necrosis, which may also contribute to cfDNA release in FSHD (particularly during episodes of active muscle damage), the release of DNA is less orderly and may produce a broader distribution of fragment sizes, including longer fragments; however, rapid degradation by plasma DNases (DNase I, DNASE1L3) reduces necrotic DNA to sizes comparable to apoptotic cfDNA within minutes to hours.
[0059] The critical insight underlying this disclosure is that the cfDNA fragments released from DUX4-expressing FSHD muscle cells retain the epigenomic imprint of those cells. Specifically, the nucleosome positioning, chromatin accessibility, and transcription factor binding patterns that characterize DUX4-activated muscle chromatin are preserved in the cfDNA fragments. At genomic loci where DUX4 has induced chromatin remodeling (opening the chromatin, displacing nucleosomes, and establishing active transcription), the cfDNA fragments exhibit patterns of coverage, fragment length, and fragment positioning that are distinct from the patterns at the same loci in cfDNA derived from cells in which those loci are transcriptionally silent.5. Nucleosome Positioning Reflects Transcriptional State
[0060] A fundamental principle of chromatin biology, and a foundational principle of this disclosure, is that nucleosome positioning at gene promoters is tightly linked to transcriptional state. At actively transcribed genes, the promoter region characteristically exhibits a nucleosome-free region (NFR), also referred to as a nucleosome depleted regionAty ref. 2848-355-PCT CU2025-209(NDR), spanning approximately 100 to 300 bp centered on or immediately upstream of the TSS. The NFR is essential for transcription factor binding, assembly of the pre-initiation complex (comprising RNA polymerase II and general transcription factors TFIIA, TFIIB, TFIID, TFIIE, TFIIF, and TFIIH), and transcription initiation. Flanking the NFR are two precisely positioned nucleosomes: the +1 nucleosome, located immediately downstream of the TSS (typically centered at approximately +135 bp relative to the TSS), and the -1 nucleosome, located immediately upstream (centered at approximately -215 bp relative to the TSS). Beyond these flanking nucleosomes, additional nucleosomes are positioned with decreasing precision in a periodic array extending in both directions.
[0061] At transcriptionally silent genes, the promoter may adopt one of several chromatin configurations: the TSS may be occupied by a nucleosome, occluding transcription factor binding sites and preventing transcription initiation; the NFR may be absent or narrowed; or the chromatin may be compacted into a heterochromatic state with high nucleosome density and repressive histone modifications. In each of these configurations, the nucleosome positioning pattern at the promoter is distinct from the pattern at an actively transcribed promoter.
[0062] These differences in nucleosome positioning at active versus silent promoters are directly reflected in the cfDNA coverage pattern at those loci. At an active promoter with a well-defined NFR, the cfDNA coverage profile shows a characteristic trough (reduced coverage) at the TSS, corresponding to the NFR, flanked by peaks of coverage at the +1 and -1 nucleosome positions, corresponding to the DNA protected by those nucleosomes. At a silent promoter where the TSS is nucleosome-occupied, the cfDNA coverage profile shows relatively uniform coverage across the promoter region without a pronounced trough. This relationship between promoter nucleosome architecture and cfDNA coverage pattern enables the inference of transcriptional state from cfDNA data; a methodology this disclosure applies specifically to the DUX4 target gene set for the purpose of FSHD diagnosis.6. DUX4-Specific Chromatin Remodeling Detectable in cfDNA
[0063] This disclosure is based on the recognition that DUX4-mediated transcriptional activation of its target genes in FSHD muscle cells is accompanied by characteristic chromatin remodeling at the promoters of those target genes. Specifically, DUX4 binding at a target gene promoter induces the following chromatin changes: (a) displacement of nucleosomes from the core promoter region, creating or deepening the NFR; (b)Aty ref. 2848-355-PCT CU2025-209recruitment of the histone acetyltransferases p300 and CBP, resulting in acetylation of histone H3 at lysine 27 (H3K27ac) on the flanking nucleosomes; (c) deposition of H3K4me3 at the +1 nucleosome by the COMPASS / MLL complexes; and (d) precise positioning of the +1 and -1 nucleosomes flanking the NFR. These chromatin changes are the hallmark of transcriptional activation and create a nucleosome architecture at DUX4 target gene promoters that is fundamentally different from the architecture at the same loci in non-FSHD cells, where these genes are not expressed and their promoters are either nucleosome-occupied or in a poised (bivalent) chromatin state.
[0064] When FSHD muscle cells undergo apoptosis, the cfDNA fragments released from these cells carry the nucleosome positioning imprint of DUX4-activated chromatin. At DUX4 target gene promoters, the cfDNA fragments from FSHD muscle cells exhibit the characteristic NFR signature: reduced fragment coverage at the TSS center, increased coverage at the +1 and -1 nucleosome positions, and an enrichment of short (sub-nucleosomal) fragments within the NFR region corresponding to DNA exposed by nucleosome displacement and at the +1 nucleosome position. In contrast, cfDNA fragments from non-FSHD cells, in which DUX4 target genes are silenced, exhibit a distinct pattern at the same promoters: uniform or flat coverage across the promoter region (reflecting nucleosome occupancy at the TSS) and a lower ratio of short to long fragments.
[0065] The cfDNA in a plasma sample from an FSHD patient is a mixture of fragments derived from multiple tissues, with hematopoietic cells contributing the majority of fragments and FSHD-affected muscle cells contributing a smaller but diagnostically significant fraction. The fragmentomic profile observed at any given DUX4 target gene promoter in the plasma cfDNA is therefore a weighted average of the profiles from all contributing cell types. The muscle-specific DUX4-activation signal is diluted by the predominant hematopoietic signal. However, because DUX4 target genes are not normally expressed in hematopoietic cells or in most other somatic cell types, the hematopoietic cfDNA contribution at these promoters reflects the silent (nucleosome-occupied) state. The small fraction of muscle derived cfDNA that carries the active (NFR and +1 nucleosome) signature introduces a measurable perturbation of the composite profile-a perturbation that is absent in healthy controls whose muscle cells do not express DUX4 target genes.Aty ref. 2848-355-PCT CU2025-2097. Aggregate Signal Across the DUX4 Target Gene Set
[0066] A key inventive concept of this disclosure is the recognition that while the DUX4-activation signal at any single promoter may be subtle (given the low fraction of muscle-derived cfDNA), the aggregate signal across dozens to hundreds of DUX4 target gene promoters provides sufficient statistical power for reliable diagnostic discrimination. This ensemble approach is analogous to Gene Set Enrichment Analysis (GSEA) in transcriptomics, where the coordinated expression changes across a defined gene set are more informative than expression changes at individual genes. By computing the PSE score across the entire DUX4 target gene set, the method of this disclosure converts a collection of individually weak signals into a robust composite biomarker.
[0067] The statistical power of the ensemble approach derives from the fact that, in FSHD, DUX4 coordinately activates a large number of target genes, all of which undergo the same type of chromatin remodeling (nucleosome depletion at their promoters). The probability that a random set of promoters would all show the same directional shift in cfDNA fragmentomic features by chance is exceedingly small. Thus, even when each individual promoter's signal is only marginally above the noise floor, the consistent directionality of the signal across many promoters provides strong evidence for the presence of DUX4-activated muscle cells contributing to the cfDNA pool.
[0068] The sensitivity of the ensemble approach is influenced by several factors, including: (a) the fraction of total cfDNA derived from FSHD-affected muscle cells, which is expected to be higher in patients with greater disease activity, more extensive muscle involvement, or following strenuous exercise; (b) the number of genes in the DUX4 target gene set, with larger gene sets providing greater statistical power but potentially introducing noise from genes with weaker DUX4 responsiveness; (c) the sequencing depth, with deeper sequencing providing more precise coverage profiles at each promoter; and (d) the specificity of the gene set, with genes that are uniquely and strongly activated by DUX4 (and not by other transcription factors or in other myopathies) providing the most discriminating features.
[0069] The contribution of skeletal muscle derived cfDNA to the total plasma cfDNA pool has been estimated at approximately 1 to 5% in healthy individuals, based on tissue-of-origin deconvolution studies using DNA methylation patterns. In patients with active myopathies characterized by ongoing muscle cell death, this fraction may be elevated to 5 to 15% or higher, depending on disease severity and activity. The methods of thisAty ref. 2848-355-PCT CU2025-209disclosure are designed to detect the DUX4-activation signature even when the muscle derived cfDNA fraction is at the lower end of this range, through the use of the ensemble approach, statistical enrichment methods, and, in certain embodiments, targeted sequencing to increase the depth of coverage at informative loci.
[0070] In certain embodiments, the signal-to-noise ratio of the assay is further enhanced by pre-analytical enrichment strategies. These strategies may include, but are not limited to: (a) size selection of cfDNA fragments to enrich for subnucleosomal fragments, which are preferentially generated from nucleosome-free regions at active promoters; (b) enrichment for muscle-derived cfDNA using tissue-specific methylation markers as selection criteria; and (c) analysis of cfDNA obtained after physical exercise or physiotherapy in the subject, which transiently increases the release of muscle derived cfDNA into the circulation.
[0071] In an additional embodiment of this disclosure, the specificity of the DUX4-activation signal for FSHD is established by comparison with other muscular dystrophies and myopathies. Duchenne muscular dystrophy (DMD), Becker muscular dystrophy (BMD), limb-girdle muscular dystrophies (LGMDs), and inflammatory myopathies (dermatomyositis, polymyositis, inclusion body myositis) are all associated with muscle cell death and elevated circulating cfDNA of muscle origin. However, in these conditions, DUX4 is not aberrantly expressed, and the DUX4 target gene set remains transcriptionally silent in the affected muscle cells. Therefore, while the overall fraction of muscle-derived cfDNA may be elevated in these conditions, the specific fragmentomic signature at DUX4 target gene promoters (the NFR signature of transcriptional activation) is absent. The PSE score in non-FSHD myopathy patients is expected to be comparable to that of healthy controls, providing specificity for FSHD.
[0072] It is recognized that a small number of DUX4 target genes may also be activated in certain cancers (e.g., some sarcomas and carcinomas have been reported to express DUX4 or PRAMEF family members as cancer-testis antigens) or in specific physiological states (e.g., the early cleavage- stage embryo). In the context of FSHD diagnosis, cancer related DUX4 target gene activation would represent a potential confound. In certain embodiments, the method includes a quality control step to assess cfDNA features associated with malignancy (e.g., copy-number alterations, elevated tumor mutation burden, or elevated total cfDNA concentration) and flag samples for further evaluation if such features are detected. In practice, the clinical context in which FSHD testing isAty ref. 2848-355-PCT CU2025-209ordered is typically a young to middle-aged patient with a suggestive pattern of muscle weakness. This makes co-occurrence with a DUX4-expressing malignancy an uncommon scenario.
[0073] This disclosure further recognizes that the DUX4 expression in FSHD muscle is stochastic, occurring in only a small percentage of myonuclei at any given time. Singlecell and single-nucleus RNA-seq studies of FSHD muscle biopsies have estimated that DUX4 is expressed in approximately 0.1% to 1% of myonuclei. Despite this low frequency of expression, the downstream consequences of DUX4 activity (including activation of target gene transcription), are propagated to neighboring cells through cell non-autonomous mechanisms, and the apoptosis triggered by DUX4 in individual myonuclei leads to the release of cfDNA from the entire myofiber, which is a multinucleated syncytium. Thus, the cfDNA released from a dying FSHD myofiber may carry the DUX4-activation chromatin signature from the affected nucleus as well as the non-activated signature from the other nuclei within the same fiber, yielding a composite signal at the single-fiber level. The ensemble approach across many DUX4 target gene promoters, as described herein, is designed to detect this composite signal.
[0074] In summary, the mechanistic chain linking the FSHD genetic lesion to a detectable cfDNA fragmentomic signature comprises the following sequential steps: (a) D4Z4 contraction (FSHD1) or SMCHD1 / DNMT3B mutation (FSHD2) causes epigenetic derepression of DUX4 on a permissive 4qA haplotype; (b) DUX4 is expressed in a fraction of skeletal muscle myonuclei; (c) DUX4 protein acts as a transcription factor, binding to and activating promoters of its target genes genome-wide; (d) transcriptional activation is accompanied by chromatin remodeling, including nucleosome displacement from target gene promoters; (e) DUX4 expression triggers myofiber apoptosis and necrosis; (f) dying muscle cells release cfDNA fragments that carry the nucleosome positioning imprint of DUX4-activated chromatin; (g) the muscle derived cfDNA circulates in plasma and contributes to the total cfDNA pool; and (h) analysis of the aggregate fragmentomic pattern at DUX4 target gene promoters in plasma cfDNA reveals the DUX4-activation signature, enabling FSHD diagnosis. This complete causal chain establishes the biological basis for the methods and systems of this disclosure.C. Biological Sample Collection and cfDNA Extraction
[0075] The methods of this disclosure begin with the collection of a biological sample from the subject. In preferred embodiments, the biological sample is peripheral venousAty ref. 2848-355-PCT CU2025-209blood. The blood is collected by standard phlebotomy techniques into blood collection tubes suitable for cfDNA analysis. Suitable collection tubes include, but are not limited to, cfDNA-stabilizing tubes such as Streck Cell-Free DNA BCT tubes, PAXgene Blood ccfDNA Tubes, or equivalent stabilizing tubes from other manufacturers. These tubes contain proprietary preservatives that stabilize white blood cells and prevent their lysis during transport and storage, thereby minimizing contamination of the cfDNA fraction with genomic DNA released from lysed leukocytes. Alternatively, standard K2-EDTA or K3-EDTA anticoagulant tubes may be used, provided that the blood is processed to plasma within 4 to 6 hours of collection to minimize leukocyte lysis.
[0076] The recommended volume of whole blood collected is 10 to 20 mL, although the method may be performed with as little as 3 to 5 mL in situations where blood volume is limited (e.g., pediatric patients, neonates). Larger volumes of blood yield proportionally greater amounts of cfDNA and may improve the sensitivity of the assay by increasing the total number of muscle-derived cfDNA fragments available for analysis.
[0077] Plasma is separated from whole blood by a two-step centrifugation protocol designed to remove cellular components while preserving the integrity and fragment-size distribution of cfDNA. In the first centrifugation step, the blood sample is centrifuged at l,600xg (relative centrifugal force) for 10 minutes at 4°C to pellet the cellular fraction (red blood cells, white blood cells, platelets). The plasma supernatant is carefully transferred to a new tube, avoiding disruption of the buffy coat layer. In the second centrifugation step, the plasma is centrifuged at 16,000*g for 10 minutes at 4°C to remove residual cellular debris, platelet fragments, and large-vesicle-associated DNA. The resulting clarified plasma is transferred to a clean tube for cfDNA extraction or stored at -80°C until extraction.
[0078] cfDNA is extracted from the clarified plasma using validated extraction methods designed for the isolation of short, fragmented DNA from biological fluids, such as the ultraprep method (Raymond CK, et al., PLoS One 15, e0231854 (2020)). This may include the use of nucleic acid extraction kits such as, but not limited to, the QIAamp Circulating Nucleic Acid Kit (Qiagen), the MagMAX Cell-Free DNA Isolation Kit (Thermo Fisher Scientific), the MagMAX Cell-Free Total Nucleic Acid Isolation Kit, the Quick-cfDNA Serum and Plasma Kit (Zymo Research), and equivalent kits from other manufacturers. The extraction is performed according to the manufacturer's recommended protocol, with attention to the following parameters: (a) the volume of plasma input, typically 1 to 10Aty ref. 2848-355-PCT CU2025-209mL; (b) the use of carrier RNA or DNA where recommended to improve recovery of low-abundance fragments; (c) the elution volume, typically 20 to 60 pL, optimized to balance cfDNA concentration and total recovery; and (d) the absence of size-selection steps that might bias the fragment-length distribution.
[0079] The typical yield of cfDNA extraction from the plasma of a healthy adult is approximately 5 to 30 nanograms (ng) per milliliter of plasma, corresponding to approximately 1,500 to 9,000 haploid genome equivalents per milliliter. The yield may be higher in subjects with elevated cfDNA concentrations due to active disease processes, including active myopathies. Conversely, the yield may be lower in neonates or in subjects with low disease activity.
[0080] Quality control of the extracted cfDNA is performed to confirm suitability for downstream sequencing and fragmentomic analysis. The fragment-size distribution of the extracted cfDNA is assessed by capillary electrophoresis using instruments such as the Agilent Bioanalyzer 2100 (using the High Sensitivity DNA Kit), the Agilent TapeStation (using the High Sensitivity D1000 or Cell-Free DNA ScreenTape), or equivalent platforms. The expected cfDNA size profile shows a predominant mononucleosomal peak at approximately 167 bp, with minor peaks at di-nucleosomal (-335 bp) and tri-nucleosomal (-500 bp) sizes. The absence of a high-molecular-weight DNA peak (>10,000 bp) confirms that the sample is not contaminated with genomic DNA from lysed cells. If significant genomic DNA contamination is detected, the sample is flagged and may require re-collection with improved pre-analytical handling.
[0081] The concentration of extracted cfDNA is quantified by a fluorometric assay, such as the Qubit dsDNA High Sensitivity Assay (Thermo Fisher Scientific) or equivalent assays, which provide accurate quantification of double-stranded DNA at low concentrations (0.01 to 100 ng / pL). Spectrophotometric methods (e.g., NanoDrop) are less preferred due to their lower accuracy at the low concentrations typical of cfDNA extracts and their susceptibility to interference from co-purified contaminants.
[0082] In alternative embodiments of this disclosure, the biological sample may be a sample type other than peripheral blood plasma. Suitable alternative samples include, but are not limited to: (a) serum, which contains cfDNA at somewhat higher concentrations than plasma due to the release of DNA from clotting cells, though with a potentially greater contribution from leukocyte-derived genomic DNA; (b) urine, which contains transrenal cfDNA that has been filtered through the kidneys-urinary cfDNA tends to beAty ref. 2848-355-PCT CU2025-209more fragmented than plasma cfDNA and may require adapted extraction protocols; (c) saliva, which contains cfDNA of both local (oral epithelial) and systemic origin; and (d) cerebrospinal fluid (CSF), which may be relevant in embodiments directed to the detection of CNS-derived cfDNA signatures.
[0083] In a particularly advantageous embodiment for newborn screening applications, the biological sample is a dried blood spot (DBS) collected on a standard newborn screening card (e.g., Guthrie card, Whatman 903 Protein Saver Card). The DBS is obtained by heel-prick capillary blood collection in the neonatal period, typically within 24 to 72 hours of birth. cfDNA extraction from DBS requires adapted protocols optimized for the small sample volume (typically equivalent to 50 to 100 pL of whole blood per spot), the drying and storage conditions of the DBS card, and the low total cfDNA yield. Specialized DBS cfDNA extraction protocols have been described and may involve rehydration of the DBS in an appropriate buffer, lysis of residual cells, and purification of cfDNA using magnetic bead-based or column-based methods.
[0084] In certain embodiments, the timing of blood collection relative to the subject's physical activity is controlled or documented, as physical exercise, particularly eccentric exercise involving muscle damage, is known to transiently increase the release of muscle-derived cfDNA into the circulation. In preferred diagnostic embodiments, the blood sample is collected under standardized conditions (e.g., after at least 12 hours of physical rest) to minimize variability. In monitoring embodiments, the effect of physical activity on the PSE score may itself be informative of muscle vulnerability and disease activity.
[0085] It is understood that the cfDNA in any given plasma sample is a complex mixture of fragments originating from multiple tissues and cell types. The relative contribution of each tissue to the total cfDNA pool depends on factors including the rate of cell turnover, the rate of apoptosis or necrosis in each tissue, the vascularization of the tissue, and the presence of pathological processes. In healthy individuals, hematopoietic cells (neutrophils, lymphocytes, monocytes) contribute the vast majority (approximately 55-75%) of plasma cfDNA, with smaller contributions from the vascular endothelium, hepatocytes, and other tissues. In FSHD patients with active disease, the contribution of skeletal muscle to the cfDNA pool is expected to be elevated relative to healthy individuals, providing the signal that this disclosure detects.
[0086] Once extracted and quality-controlled, the cfDNA is suitable for library preparation and sequencing. Extracted cfDNA may be stored at -20°C or -80°C forAty ref. 2848-355-PCT CU2025-209extended periods (weeks to months) prior to library preparation without significant degradation of fragment-length information, provided that repeated freeze-thaw cycles are avoided.
[0087] In certain embodiments, the extracted cfDNA may be subjected to additional pre-analytical processing steps. These steps may include, but are not limited to: (a) enzymatic treatment to repair damaged DNA ends (fill-in of 5' overhangs, removal of 3' overhangs, phosphorylation of 5' termini) prior to library preparation, thereby increasing the efficiency of adapter ligation; (b) treatment with bisulfite or enzymatic methylation conversion reagents (e.g., TET-assisted pyridine borane sequencing, TAPS; or enzymatic methyl-seq, EM-seq) to enable simultaneous analysis of DNA methylation and fragmentomic features; and (c) denaturation and single-strand library preparation to capture single-stranded cfDNA fragments, which may include a different population of fragments than double-stranded cfDNA and provide complementary information.
[0088] The pre-analytical workflow described in this section is designed to preserve the native fragment-length distribution and end characteristics of the cfDNA, as these features are the primary analytes of the fragmentomic analysis. It is critical that no enzymatic or mechanical fragmentation step be applied to the cfDNA during extraction or library preparation, as such fragmentation would destroy the naturally occurring fragment-length patterns that encode the nucleosome positioning information upon which this disclosure relies.
[0089] In summary, the sample collection and cfDNA extraction steps of this disclosure are compatible with standard clinical laboratory workflows and require only a routine blood draw from the subject. This represents a significant practical advantage over existing FSHD diagnostic methods that require specialized sample handling, high molecular weight DNA extraction, or invasive tissue biopsies.D. Sequencing Workflow1. Library Preparation
[0090] Following cfDNA extraction and quality control, a sequencing library is prepared from the cfDNA. The library preparation converts the cfDNA fragments into a form compatible with the sequencing platform to be used. In preferred embodiments, the library preparation employs a single-strand ligation strategy using splint adapters, which preserves nicked and partially degraded cfDNA molecules and enables efficient capture of short cfDNA fragments that are critical for accurate promoter subnucleosome enrichmentAty ref. 2848-355-PCT CU2025-209(PSE) analysis. In certain embodiments, the single-strand ligation workflow comprises denaturation of cfDNA fragments, hybridization of splint oligonucleotides, and ligation of sequencing adapters to the 5' and 3' ends of single-stranded cfDNA molecules. Standard double-stranded library preparation protocols relying on end repair, A-tailing, and adapter ligation may be used in some contexts but are generally disfavored for PSE computation because they preferentially retain intact nucleosomal fragments, lose nicked or damaged fragments, and do not enrich for the short cfDNA species that encode promoter-proximal chromatin accessibility.
[0091] In the end-repair step, the cfDNA fragments, which may have 5' overhangs, 3' overhangs, or damaged termini resulting from the in vivo nuclease digestion that generated them, are enzymatically processed to produce blunt-ended, 5'-phosphorylated doublestranded DNA fragments. This step typically employs a mixture of T4 DNA polymerase (for 3 '-to-5' exonuclease and 5'-to-3' polymerase activity to fill in or remove overhangs), Klenow fragment (for polymerase fill-in), and T4 polynucleotide kinase (for 5' phosphorylation). In the A-tailing step, a single deoxyadenosine (dA) residue is added to the 3' ends of the blunt-ended fragments by the Klenow fragment exo-minus (lacking 3'-to-5' exonuclease activity), creating a single-nucleotide 3' overhang that facilitates ligation of sequencing adapters bearing a complementary 3' deoxythymidine (dT) overhang.
[0092] In the adapter ligation step, platform-specific sequencing adapters are ligated to both ends of each cfDNA fragment using T4 DNA ligase or a proprietary ligation enzyme. The adapters contain: (a) sequences complementary to the oligonucleotides on the sequencing flow cell, enabling cluster generation; (b) sequencing primer binding sites for Read 1 and Read 2; and (c) index (barcode) sequences for sample multiplexing. Suitable library preparation kits include, but are not limited to, the KAPA HyperPrep Kit (Roche), the NEBNext Ultra II DNA Library Prep Kit (New England Biolabs), the Illumina DNA Prep Kit (Illumina), and equivalent kits from other manufacturers. The typical cfDNA input for library preparation is 5 to 30 ng, which is well within the range of cfDNA yields from a standard plasma extraction.
[0093] A critical feature of the library preparation protocol for cfDNA fragmentomic analysis is that no fragmentation step (enzymatic or mechanical) is applied. Unlike library preparation from high molecular weight genomic DNA, which requires shearing to generate fragments of the desired size range, cfDNA is naturally fragmented to mononucleosomal and sub-nucleosomal sizes by in vivo nuclease activity. The libraryAty ref. 2848-355-PCT CU2025-209preparation protocol preserves these natural fragment lengths, which encode the nucleosome positioning information that is central to the diagnostic method of this disclosure.
[0094] In certain embodiments, unique molecular identifiers (UMIs) are incorporated into the sequencing adapters during ligation. UMIs are short (typically 6 to 12 nucleotide) random sequences that uniquely tag each original cfDNA fragment prior to PCR amplification. After sequencing, reads sharing the same UMI and mapping to the same genomic coordinates can be identified as PCR duplicates originating from the same original fragment, enabling accurate deduplication without the biases introduced by coordinate-based duplicate removal. UMI-based deduplication is particularly valuable in cfDNA analysis, where the limited starting material necessitates PCR amplification and where accurate representation of the original fragment population is critical for quantitative fragmentomic analysis.
[0095] Following adapter ligation, the library is subjected to a minimal number of PCR amplification cycles (typically 4 to 10 cycles) to generate sufficient mass for sequencing. The number of PCR cycles is kept as low as possible to minimize amplification biases, particularly GC-content-dependent amplification bias, which can distort the coverage profile and fragment-length representation at specific genomic loci. In embodiments with sufficient cfDNA input, PCR-free library preparation protocols may be used, eliminating amplification bias entirely.
[0096] Size selection of the library is NOT performed in the preferred embodiments of this disclosure. Standard library preparation protocols for whole-genome sequencing often include a size-selection step (e.g., using AMPure XP beads at specific ratios) to remove adapter dimers and very small fragments. In cfDNA fragmentomic analysis, the full spectrum of fragment lengths must be retained, including sub-nucleosomal fragments (<150 data-id="211" bp) that are informative of nucleosome-free regions at active promoters. Therefore, size selection is limited to the removal of adapter dimers (which are typically <100 bp after adapter trimming) using a conservative bead ratio that retains the smallest cfDNA fragments of interest (down to approximately 35 bp insert size, corresponding to transcription factor footprints in certain embodiments).2. Sequencing Platforms
[0097] The prepared sequencing library is sequenced on a next generation sequencing (NGS) platform capable of generating short paired-end reads with high accuracy andAty ref. 2848-355-PCT CU2025-209throughput. In preferred embodiments, the sequencing is performed on Illumina short-read sequencing platforms, including but not limited to the NovaSeq 6000, NovaSeq X, NovaSeq X Plus, NextSeq 2000, and NextSeq 1000 instruments. These platforms generate paired-end reads of 2^150 bp (i.e., 150 bases from each end of the fragment), which is sufficient to completely sequence the majority of cfDNA fragments (most of which are shorter than 200 bp) and to precisely determine the fragment length from the paired-end alignment.
[0098] Paired-end sequencing is essential for cfDNA fragmentomic analysis because it enables the determination of the fragment length (insert size) of each sequenced cfDNA molecule. The fragment length is computed from the alignment coordinates of the two paired reads: fragment length equals the distance from the 5' end of Read 1 to the 3' end of Read 2 (or, equivalently, the absolute value of the template length, TLEN, field in the BAM alignment file). This fragment-level resolution is critical for computing fragmentlength distributions, WPS, and other fragmentomic features at genomic loci of interest.
[0099] Alternative sequencing platforms may be used in certain embodiments of this disclosure. These include, but are not limited to: (a) MGI / BGI DNBSEQ platforms (DNBSEQ-T7, DNBSEQ-G400), which use DNA nanoball (DNB) technology and combinatorial probe-anchor synthesis (cPAS) sequencing; (b) Element Biosciences AVITI platform; (c) Singular Genomics G4 platform; and (d) Ultima Genomics UG 100 platform. Each of these platforms is capable of generating high-accuracy, high-throughput shortread sequencing data suitable for cfDNA fragmentomic analysis, provided that paired-end sequencing is performed.
[0100] In alternative embodiments, long-read sequencing platforms may be used. These include Pacific Biosciences (PacBio) single molecule real time (SMRT) sequencing instruments (Sequel II, Sequel lie, Revio) and Oxford Nanopore Technologies (ONT) instruments (MinlON, GridlON, PromethlON, P2). Long-read platforms offer the advantage of sequencing entire cfDNA fragments in a single read without the need for paired-end sequencing, and they can provide native DNA methylation information (in the case of ONT and PacBio using kinetic analysis). However, the per-base error rates of long-read platforms are higher than those of short-read platforms, and the throughput and cost per genome are currently less favorable for the whole-genome cfDNA analysis contemplated by the primary embodiments of this disclosure. Long-read platforms may beAty ref. 2848-355-PCT CU2025-209particularly suitable for the targeted sequencing embodiments described below, or for the combined fragmentomic and methylation analysis embodiments.3. Sequencing Depth Requirements
[0101] For whole genome sequencing (WGS) embodiments, the minimum sequencing depth is approximately 5* genome-wide coverage. At 20* coverage, approximately 200 million read pairs are generated, providing an average of approximately 5 sequenced fragments covering each base position in the genome. This depth is sufficient for genomewide fragmentomic profiling and PSE score computation when combined with the ensemble approach across the DUX4 target gene set (which aggregates signal from multiple promoters). However, higher sequencing depths improve the precision of coverage profiles at individual promoters and increase the sensitivity of the assay for detecting low-level DUX4-activation signals.
[0102] The preferred sequencing depth for WGS-based cfDNA fragmentomic analysis is 10* to 30* genome-wide coverage, corresponding to approximately 400 million to 1.2 billion read pairs. At these depths, the coverage at individual promoter windows (each spanning 2 kb) is approximately 10 to 30 fragments per base position (before normalization), providing sufficient statistical power for reliable computation of NFR scores, WPS profiles, and fragment-length ratios at single-promoter resolution. The higher end of this range (20-30*) is preferred for applications requiring maximum sensitivity, such as detection of early-stage or mild FSHD with low disease activity and correspondingly low muscle-derived cfDNA fractions.
[0103] The trade-off between sequencing depth and cost is an important practical consideration. At the time of this disclosure, whole-genome sequencing at 30* coverage on current Illumina platforms costs approximately $200 to $600 per sample (reagent and consumable costs), making it feasible for clinical diagnostic applications. Cost reductions driven by ongoing improvements in sequencing technology are expected to make higher-depth WGS increasingly accessible.4. Targeted Panel Design (Optional Embodiment)
[0104] In an alternative embodiment of this disclosure, the sequencing is performed using a targeted sequencing approach rather than WGS. In this embodiment, a custom capture panel or amplicon panel is designed to selectively enrich for the genomic regions of interest (specifically, the promoter regions of DUX4 target genes) prior to sequencing. This targeted approach dramatically reduces the amount of sequencing required (andAty ref. 2848-355-PCT CU2025-209therefore the cost per sample) by concentrating the sequencing reads on the informative loci rather than distributing them uniformly across the genome.
[0105] The targeted capture panel comprises oligonucleotide probes (typically biotinylated RNA or DNA probes, 80 to 120 nucleotides in length) designed to tile the promoter windows (TSS ±1,000 bp) of genes in the DUX4 target gene set. In preferred embodiments, the panel targets the promoter regions of 50 to 500 DUX4 target genes, corresponding to a total target space of approximately 100 kb to 1 Mb. The panel may additionally include probes targeting: (a) promoter regions of a set of housekeeping genes (e g., GAPDH, ACTB, RPL13A, HPRT1, B2M, TBP, and other ubiquitously expressed genes) to serve as normalization controls; (b) promoter regions of genes specific to non-FSHD muscular dystrophies (e.g., DMD, CAPN3, DYSF, SGCA, FKRP) to enable differential diagnosis; (c) the D4Z4-adjacent genomic region on chromosome 4q35, including the pLAM and 4qA / 4qB haplotype-distinguishing sequences; (d) the SMCHD1 and DNMT3B gene loci, to enable detection of coding mutations associated with FSHD2; and (e) CpG-rich regions within and flanking D4Z4, to enable methylation analysis when combined with bisulfite or enzymatic conversion.
[0106] Following library preparation, the targeted capture is performed using standard hybrid-capture protocols. The library is hybridized with the biotinylated probe panel, the probe-target hybrids are captured using streptavidin-coated magnetic beads, non-target DNA is washed away, and the captured library is eluted and PCR amplified for sequencing. A critical consideration for cfDNA fragmentomic analysis is that the capture protocol must preserve the native fragment-length distribution to the greatest extent possible. Certain capture protocols are optimized for cfDNA applications and use gentle hybridization conditions and minimal post-capture amplification to achieve this goal.
[0107] The sequencing depth for targeted panel embodiments is 10 to 50 million read pairs, which is approximately 5 to 20-fold lower than the sequencing required for WGS at comparable coverage of the target regions. At the target regions, this corresponds to a mean coverage of approximately 100* to 1,000* or higher, providing significantly greater statistical power for fragmentomic feature extraction at individual promoters compared to WGS. The cost per sample for targeted panel sequencing is correspondingly lower, making this approach suitable for high-throughput screening applications and resource limited settings.Aty ref. 2848-355-PCT CU2025-209
[0108] In an alternative targeted sequencing embodiment, an amplicon based panel is used instead of a hybrid capture panel. In this embodiment, multiplex PCR amplification with specifically designed primer pairs is used to amplify the promoter regions of DUX4 target genes directly from the cfDNA library (or from the cfDNA itself after adapter ligation). Amplicon-based approaches offer simplicity, rapid turnaround, and low cost, but may introduce greater amplification bias and may less faithfully preserve the native fragment-length distribution compared to hybrid-capture approaches. In certain implementations, molecular inversion probes (MIPs) or single strand capture probes may be used as alternatives to standard hybrid-capture or amplicon approaches.5. Sequencing Data Generation
[0109] Regardless of the sequencing approach (WGS or targeted), the sequencing instrument generates raw sequencing data in the form of FASTQ files. Each FASTQ file contains the nucleotide sequences and per-base quality scores for each sequencing read. For paired-end sequencing, two FASTQ files are generated per sample: one for Read 1 and one for Read 2. The FASTQ files serve as the input to the computational pipeline.
[0110] In certain embodiments, the sequencing run includes internal controls to monitor sequencing quality and fragment-length accuracy. These controls may include: (a) a spikein of synthetic cfDNA fragments of known sequence and known length, serving as a size standard to verify that the sequencing and analysis pipeline accurately reports fragment lengths; (b) a negative control (water or buffer subjected to the entire extraction and library preparation workflow) to detect contamination; and (c) a positive control (cfDNA from a confirmed FSHD patient, or a synthetic mixture of cfDNA fragments designed to recapitulate the FSHD fragmentomic signature) to verify the sensitivity of the assay.
[0111] Sequencing data may be generated on a single sequencing run or across multiple runs. When samples are multiplexed (multiple samples sequenced on the same run), sample identity is maintained through index sequences incorporated during library preparation, and demultiplexing is performed as a standard step in the sequencing platform's primary analysis pipeline. The resulting per-sample FASTQ files are then processed individually through the computational pipeline.
[0112] In certain embodiments of this disclosure, the sequencing data may be augmented by the inclusion of additional data modalities. For example, in embodiments combining fragmentomic analysis with methylation analysis, the sequencing data may include methylation-converted reads (from bisulfite-seq or EM-seq), which provide bothAty ref. 2848-355-PCT CU2025-209fragment-length information and methylation status at CpG sites, enabling simultaneous fragmentomic and methylation-based analysis from a single sequencing library. In embodiments using Oxford Nanopore sequencing, native DNA methylation may be called directly from the raw signal without the need for chemical conversion.
[0113] The raw sequencing data, aligned reads, and derived fragmentomic features may be stored in standard bioinformatics file formats, including FASTQ (raw reads), BAM / CRAM (aligned reads), BED (genomic intervals), bigWig (continuous coverage tracks), and tab-delimited text or CSV formats (feature matrices and PSE scores). These files may be stored locally, on a laboratory information management system (LIMS), or on a cloud-based storage platform for downstream analysis and archival.
[0114] In summary, the sequencing workflow of this disclosure is compatible with standard NGS laboratory procedures and commercially available library preparation kits and sequencing instruments. The workflow produces paired-end sequencing data from which fragment lengths and genomic coordinates can be precisely determined, preserving the fragmentomic information necessary for downstream analysis. The choice between WGS and targeted sequencing embodiments depends on the desired balance between comprehensive genome-wide analysis and cost efficiency, and both approaches are encompassed by this disclosure.E. Computational Pipeline: From Raw Reads to Fragmentomic Features1. Read Quality Control and Preprocessing
[0115] The computational pipeline of this disclosure begins with quality control and preprocessing of the raw sequencing data. The raw FASTQ files are first assessed for overall quality metrics using tools such as FastQC (Babraham Bioinformatics) or equivalent quality assessment software. Metrics evaluated include per-base sequence quality, per-sequence quality scores, per-base sequence content, per-sequence GC content, sequence length distribution, sequence duplication levels, overrepresented sequences, and adapter content. These metrics provide an initial assessment of data quality and identify potential issues (e.g., adapter contamination, low-quality cycles, index hopping) that may require attention in subsequent processing steps.
[0116] Adapter trimming is performed to remove residual sequencing adapter sequences from the ends of reads. Even though the cfDNA insert sizes are generally shorter than the read length (meaning that Read 1 and Read 2 often extend into the adapter on the opposite end of the fragment), accurate adapter trimming is essential for correct alignment andAty ref. 2848-355-PCT CU2025-209fragment-length determination. Suitable adapter trimming tools include fastp, Trimmomatic, cutadapt, and BBDuk, among others. The trimming parameters are configured to remove the specific adapter sequences used in library preparation (e.g., Illumina TruSeq or Nextera adapters) while retaining the maximum length of insert sequence. A minimum read length after trimming (e.g., 35 bp) is enforced to exclude very short reads that may not align uniquely.
[0117] Quality filtering is applied to remove reads with low overall base quality. Reads with a mean base quality below Q20 (corresponding to a 1% base-call error rate) or with more than a specified number of low-quality bases (e.g., more than 5 bases with quality below QI 5) are discarded. This step ensures that only high-quality reads are used for alignment and downstream analysis, reducing the rate of misalignments and erroneous fragment-length assignments.
[0118] If UMIs were incorporated during library preparation (as described above), the UMI sequences are extracted from the reads and appended to the read headers (or stored in auxiliary BAM tags) at this stage. This is accomplished using specialized tools such as UMI-tools extract, fgbio ExtractUmisFromBam, or fastp (which supports inline UMI extraction). The UMI information is carried through the pipeline and used during the deduplication step to identify and collapse PCR duplicate read pairs that originated from the same original cfDNA fragment.2. Alignment to Reference Genome
[0119] The preprocessed paired-end reads are aligned to a human reference genome assembly. In preferred embodiments, the reference genome is GRCh38 / hg38 (Genome Reference Consortium Human Build 38) or the Telomere-to-Telomere (T2T-CHM13) assembly, which provides a more complete representation of repetitive regions, centromeres, and segmental duplications. The choice of reference genome may influence the alignment of reads at or near the D4Z4 region and other repetitive loci, and the use of T2T-CHM13 may provide improved alignment accuracy at these regions (T2T genome version CHM13v2.0 / hsl was used for the analyses presented in the Figures of this disclosure).
[0120] Alignment is performed using a short-read aligner optimized for speed and accuracy, such as BWA-MEM2 (Burrows- Wheel er Aligner, version 2), Bowtie2, or minimap2 (in short-read mode). BWA-MEM2 is preferred for its combination of high accuracy, fast execution, and wide adoption in clinical genomics pipelines. The aligner isAty ref. 2848-355-PCT CU2025-209configured for paired-end mode, and the output is a coordinate-sorted BAM (Binary Alignment Map) file containing the alignment coordinates, mapping quality scores, insert sizes, and other alignment-derived information for each read pair.
[0121] Post-alignment processing is performed to remove reads that are uninformative or potentially misleading for fragmentomic analysis. The following categories of reads are removed: (a) PCR and optical duplicates, identified by Picard MarkDuplicates (Broad Institute) or, if UMIs are available, by UMI-based deduplication tools (UMI-tools dedup, fgbio GroupReadsByUmi followed by CallMolecularConsensusReads, or Gencore); (b) reads with mapping quality below 20 (MAPQ < 20), which are likely to be ambiguously mapped or multi-mapped and may provide unreliable positional information; (c) improperly paired reads, including singletons (only one read of the pair aligned), discordant pairs (mates aligned to different chromosomes or in unexpected orientations), and pairs with insert sizes exceeding a maximum threshold (e.g., >1,000 bp), which are unlikely to represent genuine cfDNA fragments; (d) reads aligning to the mitochondrial genome (chrM), which represents a distinct cfDNA source (mitochondrial cfDNA) with different fragmentomic properties from nuclear cfDNA; (e) reads aligning to decoy sequences and alternate locus scaffolds in the reference genome; and (f) reads aligning to ENCODE blacklist regions, which are genomic regions known to produce anomalous, artifactually high coverage in sequencing experiments, typically corresponding to collapsed repeat sequences, centromeric satellites, and other problematic regions.
[0122] The filtered BAM file is then indexed (using samtools index or equivalent) to enable rapid random access to specific genomic regions. Summary alignment statistics, including total read count, alignment rate, duplicate rate, mean insert size, insert-size standard deviation, and genome-wide coverage, are computed (using samtools flagstat, Picard CollectlnsertSizeMetrics, and mosdepth or equivalent tools) and recorded for quality control purposes.3. Fragment-Length Extraction
[0123] From the filtered BAM file, the fragment length (insert size) of each properly paired read pair is extracted. The fragment length is defined as the distance from the leftmost aligned position of the forward read to the rightmost aligned position of the reverse read, which corresponds to the size of the original cfDNA fragment (inclusive of the nucleosome-protected core and any linker DNA). This value is stored in the TLENAty ref. 2848-355-PCT CU2025-209(template length) field of the BAM file and can be extracted using standard tools (samtools, pysam, or custom scripts).
[0124] A genome-wide fragment-length distribution histogram is generated by tabulating the frequency of each fragment length across all properly paired reads in the filtered BAM file. The expected distribution for cfDNA shows a major peak at approximately 167 bp (mononucleosomal), minor peaks at approximately 335 bp (dinucleosomal) and 500 bp (trinucleosomal), and a shoulder or secondary peak at approximately 143 bp (reflecting fragments corresponding to the nucleosome core particle without linker DNA).Additionally, a population of short fragments (<150 data-id="276" bp, sub-nucleosomal) is present, corresponding to DNA from nucleosome-free regions, promoter-proximal nucleosome positions, linker DNA, and transcription factor footprints.
[0125] For the purposes of fragmentomic analysis at promoter windows, fragments are categorized into size classes: (a) short fragments, defined as those with a length less than 130 bp, which are enriched at nucleosome-free regions, promoter-proximal nucleosome positions, and transcription-factor-bound regions; (b) mononucleosomal fragments, defined as those with a length of 130 to 200 bp, which correspond to DNA protected by a single nucleosome; and (c) overlapping dinucleosome fragments, defined as those with a length 200 to 330 bp, which correspond to DNA protected by a nucleosome and a subnucleosome, which are enriched at promoter-proximal nucleosome positions and generated due to active transcription, d) dinucleosomal fragments, defined as those with a length of 250 to 400 bp, corresponding to DNA protected by two adjacent nucleosomes. The boundaries of these categories may be adjusted in certain embodiments based on the specific fragment-length distribution of the sample and the species under study.4. Promoter Window Definition
[0126] For each gene in the analysis set (including the DUX4 target gene set, the housekeeping control gene set, and any additional gene sets under evaluation), a promoter window is defined centered on the annotated TSS. In preferred embodiments, the promoter window spans TSS ±1,000 bp, yielding a 2,000 bp (2 kb) analysis window. This window size is chosen to encompass the nucleosome-free region at the TSS and the flanking ±1 and -1 nucleosomes, as well as sufficient flanking sequence to establish the local background coverage level.
[0127] TSS annotations are derived from GENCODE (version 44 or later), Ensembl, or RefSeq gene annotation databases, using the TSS of the principal (canonical) transcriptAty ref. 2848-355-PCT CU2025-209isoform for each gene. Where multiple annotated TSSs exist for a single gene (reflecting alternative promoter usage or alternative first exons), the promoter window is defined for each annotated TSS separately, and the most informative TSS (as determined from training data) is selected for inclusion in the final analysis. In certain embodiments, all annotated TSSs for a gene are included, and the one yielding the strongest PSE signal in the training cohort is designated the primary TSS for that gene. In some embodiments, the promoter window spans TSS ± 2,000 base pairs to capture the nucleosome architecture surrounding the promoter region, and the fragmentomic analysis specifically evaluates cfDNA fragments that overlap the +1 nucleosome, defined as the first well-positioned nucleosome downstream of the TSS. Determining the position of the +1 nucleosome for each gene enables extraction of fragment-length distributions at this regulatory landmark, which forms the basis for accurate computation of promoter subnucleosome enrichment (PSE).
[0128] Promoter windows are subject to quality filtering to exclude regions that may yield unreliable fragmentomic data. Windows that overlap ENCODE blacklist regions are excluded. Windows with extreme GC content (below 20% or above 80% across the 2 kb window) are excluded, as such regions are prone to GC-dependent coverage biases that may confound fragmentomic analysis. Windows that overlap with other promoter windows from different genes (i.e., overlapping or closely spaced TSSs) are flagged and, in certain embodiments, merged or treated as a single combined window. Additionally, windows located in regions of low mappability (where short reads cannot be uniquely aligned due to sequence repetitiveness) are excluded or down-weighted in the PSE score computation.5. Coverage and Fragment-Count Matrices
[0129] For each promoter window, multiple coverage and fragment-count metrics are computed at single base or binned resolution. The primary metrics are as follows. Read depth coverage at each position within the promoter window is computed as the number of aligned fragments overlapping that position. This metric provides a direct measure of the number of cfDNA fragments contributing to the signal at each base and forms the basis for the coverage profile, which is the fundamental data representation for nucleosome footprint analysis.
[0130] Fragment midpoint coverage is computed as the count of cfDNA fragment midpoints (the center of the fragment, computed as the mean of the fragment's 5' and 3'Aty ref. 2848-355-PCT CU2025-209coordinates) falling at each base position within the promoter window. The fragment midpoint is expected to be located near the center (dyad) of the nucleosome that protects the fragment, and therefore the fragment midpoint coverage profile provides a more precise map of nucleosome dyad positions than the raw read-depth coverage profile.
[0131] Fragment-start coverage is computed as the count of 5' fragment ends at each base position within the promoter window. The 5' end of a cfDNA fragment corresponds to the position where the apoptotic nuclease cleaved the DNA, which is expected to occur preferentially at linker DNA regions between nucleosomes. Therefore, the fragment-start coverage profile provides complementary information about the boundaries of nucleosome-protected regions and the positions of linker DNA and NFRs.
[0132] Normalization of coverage profiles is performed to account for technical biases and enable quantitative comparison between samples sequenced at different depths. The normalization steps include: (a) library-size normalization, in which raw coverage counts are divided by the total number of aligned fragments in the sample and multiplied by a scaling factor (e.g., one million) to yield counts per million (CPM) or reads per million (RPM); (b) GC-bias correction, in which the coverage profile is adjusted for GC-content-dependent variation using LOESS (locally estimated scatterplot smoothing) regression-the GC content of each fragment is computed, the mean coverage is plotted as a function of GC content, a LOESS curve is fitted, and each fragment's contribution is scaled by the inverse of the LOESS -predicted coverage at that fragment's GC content; and (c) mappability correction, in which the coverage at each position is adjusted for the local mappability (the probability that a read of a given length originating from that position can be uniquely aligned), using precomputed mappability tracks derived from the reference genome.6. Windowed Protection Score (WPS)
[0133] The Windowed Protection Score (WPS) is computed across each promoter window to quantify nucleosome positioning and identify nucleosome-free regions. For each base position i within the promoter window, the WPS is defined as: WPS(i) = (number of fragments that completely span position i, with both 5' and 3' ends located outside a window of width W centered on i) minus (number of fragments that have at least one end located within the window of width W centered on i). The default window width W is 120 bp, which is approximately the length of DNA in contact with the histoneAty ref. 2848-355-PCT CU2025-209octamer surface and provides good discrimination between nucleosome-occupied and nucleosome-free regions.
[0134] The WPS profile reveals nucleosome positioning with the following interpretation: positive WPS values indicate positions that are protected by nucleosomes (where many long fragments span the position, consistent with intact nucleosomal DNA); negative WPS values indicate positions that are not protected by nucleosomes (where fragments preferentially begin or end at or near the position, consistent with linker DNA or NFRs that are accessible to nuclease cleavage). The WPS profile across a gene promoter typically shows a trough of negative WPS at the NFR (centered on or near the TSS of active genes) flanked by peaks of positive WPS at the +1 and -1 nucleosome positions.
[0135] In certain embodiments, the WPS is computed separately for different fragmentlength classes. Long-fragment WPS (computed using only fragments of 120 to 180 bp, corresponding to mononucleosomal fragments) provides a map of nucleosome positioning. Short-fragment WPS (computed using only fragments of 35 to 80 bp, corresponding to sub-nucleosomal fragments) can reveal the positions of transcription factor binding sites, as transcription factors protect shorter stretches of DNA from nuclease digestion than nucleosomes. The dual-WPS approach (long-fragment WPS and short-fragment WPS) provides complementary information about both nucleosome positioning and transcription factor occupancy at each promoter, which may be leveraged in certain embodiments of the classifier to improve diagnostic performance.7. Nucleosome-Free Region (NFR) Quantification
[0136] At each promoter window, the nucleosome-free region (NFR) may be inferred from the local chromatin architecture surrounding the transcription start site. In certain embodiments, the NFR is defined relative to the position of the +1 nucleosome, which is identified as the first well-positioned nucleosome downstream of the promoter. Because the NFR lies immediately upstream of the +1 nucleosome, its presence and depth are inferred from the depletion of cfDNA fragments of nucleosomal size in the region directly upstream of the +1 nucleosome rather than from just a minimum-coverage metric. In this framework, a more pronounced depletion of nucleosomal fragments upstream of the +1 nucleosome indicates a deeper nucleosome-free region and is consistent with active transcription. This +1 -nucleosome-anchored approach reflects the analyses performedAty ref. 2848-355-PCT CU2025-209herein and does not rely on minimum coverage of nucleosomal fragments calculations, which were not used in the present work.
[0137] The NFR score is defined as the ratio of the mean coverage at the flanking nucleosome peak positions (+1 and -1 nucleosomes, identified as the two highest peaks in the coverage or WPS profile within the promoter window) to the coverage at the TSS center (mean coverage within TSS ±50 bp). An NFR score greater than 1.0 indicates nucleosome depletion at the TSS relative to the flanking regions; higher values indicate more pronounced depletion. In the context of this disclosure, a high NFR score at a DUX4 target gene promoter in a test sample suggests active transcription of that gene, consistent with DUX4-driven transcriptional activation in FSHD.
[0138] An alternative NFR metric is the area under the WPS curve within the TSS ±150 bp window. A more negative area (larger magnitude of negative WPS) indicates stronger nucleosome depletion. Additional metrics may include the width of the NFR (defined as the distance between the -1 and +1 nucleosome peaks) and the symmetry of the NFR (comparing the depth of the trough on the upstream and downstream sides of the TSS).8. Fragment-Length Ratio at Promoters
[0139] At each promoter window, fragment-length distributions are quantified to capture the characteristic chromatin architecture associated with active versus inactive promoters. In certain embodiments, rather than computing a generic short-to-long fragment ratio, the analysis computes a Promoter Subnucleosome Enrichment (PSE) score specifically at the +1 nucleosome position, defined as the first nucleosome downstream of the transcription start site. The PSE score is calculated using cfDNA fragments that overlap the +1 nucleosome within a ±100 bp window, thereby capturing the fragment-length classes most informative of promoter accessibility. Using this framework, the PSE score for each gene is computed according to Formula I:PSE = [#(35-130 bp) ± #(200-330 bp)] / [#(130-200 bp) ± #(330-400 bp)] which is the number of 35-130 bp fragments added to the number of 200-330 bp fragments, divided by the number of 130-200 bp fragments added to the number of 330-400 bp fragments.
[0140] The numerator captures subnucleosomal and extended subnucleosomal fragments that arise from nucleosome destabilization and transcription-associated chromatin remodeling, while the denominator captures nucleosome-protected and dinucleosomal fragments characteristic of closed or inactive promoter states. Because DUX4-responsiveAty ref. 2848-355-PCT CU2025-209promoters exhibit increased chromatin accessibility in FSHD muscle, the PSE score is expected to be elevated at these loci in subjects with FSHD relative to unaffected controls. This PSE-based approach reflects the analyses performed herein and supersedes simpler short-to-long fragment ratios, which were not used in the present work.
[0141] In certain embodiments, finer fragment-length resolution is used, such as computing the ratio of fragments in the 80-140 bp range to fragments in the 150-200 bp range, or analyzing the full fragment-length distribution within each promoter window as a histogram feature. The Shannon entropy of the local fragment-length distribution may also be computed as a measure of the diversity of fragment sizes at each promoter, with higher entropy expected at active promoters where multiple chromatin configurations (nucleosome-free, partially occupied, fully occupied) coexist.
[0142] Collectively, the fragmentomic features: coverage profiles, WPS profiles, NFR scores, fragment-length ratios, and fragment midpoint and end distributions-constitute a rich multi-dimensional representation of the nucleosome architecture at each promoter window. These features are the inputs to the PSE score computation.
[0143] Quality control of the computational pipeline includes monitoring of aggregate metrics at each processing step: (a) alignment rate (percentage of reads successfully aligned; expected >95% for high-quality cfDNA data); (b) duplicate rate (percentage of reads marked as duplicates; expected <30% data id="322" for libraries with 4-8 PCR cycles); (c) mean insert size and insert size standard deviation (expected mean -167 bp for cfDNA); (d) on-target rate (for targeted panel embodiments, the percentage of reads mapping within the target regions; expected >60%); and (e) strand balance (the ratio of forward- to reverse-strand reads at each position; expected -1.0 for unbiased libraries). Samples failing one or more quality thresholds are flagged for review and may require resequencing or re-extraction.
[0144] The computational pipeline may be implemented in any suitable programming language or framework, including but not limited to Python, R, Java, C++, or Rust, and may be deployed as a standalone software package, a cloud-based analysis service, or an integrated module within a commercial bioinformatics platform. The pipeline may be orchestrated using workflow management systems such as Nextflow, Snakemake, Cromwell / WDL, or equivalent systems to ensure reproducibility, scalability, and auditability of the analysis. In preferred embodiments, the entire pipeline from FASTQ input to PSE score output is fully automated, requiring no manual intervention.Aty ref. 2848-355-PCT CU2025-209F. Promoter Architecture and cfDNA Signature Theory1. Nucleosome Organization at Gene Promoters
[0145] The theoretical foundation of this disclosure rests on the well-established relationship between nucleosome organization at gene promoters and transcriptional state. In eukaryotic cells, the genome is packaged into chromatin, the fundamental unit of which is the nucleosome: 147 bp of DNA wrapped in approximately 1.65 left handed superhelical turns around an octamer of histone proteins (two copies each of H2A, H2B, H3, and H4). Adjacent nucleosomes are connected by linker DNA of variable length (typically 20 to 80 bp in human cells), and the repeating pattern of nucleosomes along the DNA fiber creates a "beads-on-a-string" structure. The average nucleosome repeat length (NRL) in human cells is approximately 185 to 195 bp.
[0146] At gene promoters, nucleosome positioning is not random but follows a stereotypical architecture that is conserved across eukaryotic species. Most promoters of transcriptionally active genes exhibit a nucleosome-free region (NFR) of approximately 100 to 300 bp centered on or immediately upstream of the TSS. The NFR corresponds to the core promoter and is essential for the binding of general transcription factors (TFIID / TBP, TFIIA, TFIIB, etc.), mediator complex, and RNA polymerase II to form the preinitiation complex. The NFR is maintained by the combined action of ATP-dependent chromatin remodeling complexes (SWI / SNF, RSC, ISWI, CHD families), transcription factor binding (which competes with nucleosomes for DNA occupancy), and the intrinsic DNA sequence preferences of nucleosomes (certain sequences, such as poly(dA:dT) tracts, disfavor nucleosome assembly and are enriched in NFRs).
[0147] Immediately flanking the NFR are two precisely positioned nucleosomes, designated the +1 nucleosome (positioned downstream of the TSS, with its dyad typically at approximately +135 bp relative to the TSS) and the -1 nucleosome (positioned upstream, with its dyad at approximately -215 bp relative to the TSS). The +1 nucleosome is particularly well positioned and is often decorated with the histone modification H3K4me3, a hallmark of active or poised promoters. Beyond the +1 and -1 nucleosomes, additional nucleosomes are arranged in a periodic array (the nucleosome "phasing" pattern), with each successive nucleosome positioned at a distance approximating the NRL (-190 bp) from its neighbor. This phasing decays with distance from the TSS, reflecting the decreased influence of the TSS-proximal positioning cues.Aty ref. 2848-355-PCT CU2025-209
[0148] At transcriptionally silent genes, the promoter nucleosome architecture may differ in several ways from the active promoter pattern. In some cases, the TSS is occupied by a nucleosome (sometimes referred to as the "positioned nucleosome over TSS" or "+1 nucleosome shifted to the TSS"), occluding transcription factor binding sites and physically blocking pre-initiation complex assembly. In other cases, the NFR may be present but narrow, or the flanking nucleosomes may be less precisely positioned. In the case of genes that are silenced by Polycomb-mediated repression, the promoter may carry bivalent chromatin marks (H3K4me3 and H3K27me3) and maintain a narrow NFR in a poised state, ready for activation upon the appropriate signal. In the case of genes silenced by heterochromatin formation (as is the case for DUX4 target genes in non-FSHD cells), the promoter may be buried in compacted chromatin with high nucleosome density and repressive marks (H3K9me3, DNA methylation).2. Active vs. Silent Promoter Signatures in cfDNA
[0149] The relationship between promoter nucleosome architecture and cfDNA coverage patterns has been established through genome wide studies comparing cfDNA nucleosome maps with transcriptomic and epigenomic data from the tissues of origin. At promoters of genes that are highly expressed in the primary cfDNA source tissues (e.g., hematopoietic cells), the cfDNA coverage profile shows the characteristic active promoter pattern: a pronounced trough at the TSS (corresponding to the NFR) flanked by peaks at the +1 and -1 nucleosome positions, and an increased ratio of subnucleosomal to nucleosomal fragments at the +1 nucleosome position. At promoters of genes that are not expressed in the primary cfDNA source tissues, the coverage profile is relatively flat or shows a single peak centered on the TSS (corresponding to a nucleosome positioned over the TSS) and the ratio of subnucleosomal to nucleosomal fragments at the +1 nucleosome position is lower.
[0150] This relationship has been demonstrated to hold at a genome-wide scale, with the depth of the NFR trough in cfDNA correlating with the expression level of the gene in the tissue(s) contributing to the cfDNA pool. Moreover, when the relative contribution of a specific tissue to the cfDNA pool increases (as in organ specific pathology), the cfDNA coverage profiles at promoters of tissue-specific genes shift toward the active promoter pattern, reflecting the increased contribution of cfDNA from that tissue. This principle underlies the use of cfDNA fragmentomics for tissue-of-origin deconvolution and geneAty ref. 2848-355-PCT CU2025-209expression inference, and it is the principle upon which the FSHD diagnostic method of this disclosure is based.3. Mechanism of Nucleosome Imprinting in cfDNA
[0151] The mechanism by which nucleosome positioning information is transferred from the chromatin of living (or recently living) cells to the circulating cfDNA pool is as follows. During apoptosis, the activation of caspase-activated DNase (CAD / DFF40) initiates the digestion of chromosomal DNA. CAD cleaves DNA preferentially at the linker regions between nucleosomes, because linker DNA is accessible and unprotected, while DNA wrapped around nucleosome core particles is physically shielded from nuclease access. This nucleosome-guided cleavage produces DNA fragments that correspond in size to the DNA protected by one nucleosome (-147 bp core + -20 bp linker = -167 bp for mononucleosomal fragments), two nucleosomes (-335 bp for dinucleosomal fragments), and so on. Within the NFR, where nucleosomes are absent, the DNA is freely accessible to CAD and other nucleases, resulting in cleavage within the NFR and the generation of shorter fragments. At transcription factor binding sites within the NFR, the bound transcription factor may protect a short stretch of DNA (typically 8-30 bp of direct contact) from nuclease cleavage, producing very short protected fragments that leave a "footprint" in the sub-nucleosomal fragment size range (35-80 bp). At promoter-proximal nucleosome (+1 nucleosome), transcription produces subnucleosome particles containing less than an octamer of histones, resulting in shorter footprints as well (typically 35-130 bp).
[0152] After the initial intracellular nuclease digestion, the resulting DNA fragments are released into the extracellular space through membrane blebbing (in apoptosis) or membrane rupture (in necrosis). The fragments enter the bloodstream, where they circulate as cfDNA, either associated with histone proteins (as nucleosome-DNA complexes) or as free DNA. Additional processing by plasma nucleases (principally DNASE1 and DNASE1L3) further degrades cfDNA fragments, preferentially attacking unprotected DNA and further sharpening the nucleosome-guided fragmentation pattern. The net result is a cfDNA population in which each fragment is a "fossil record" of the nucleosome (or transcription factor or subnucleosome) that protected it from nuclease digestion in the cell of origin.Aty ref. 2848-355-PCT CU2025-2094. Tissue-of-Origin Deconvolution and the FSHD Signal
[0153] The aggregate cfDNA fragmentome in plasma is a composite of contributions from all cell types that are undergoing apoptosis or necrosis and releasing DNA into the circulation. The cfDNA coverage profile at any given gene promoter is therefore a weighted average of the profiles from all contributing cell types, where the weight of each cell type is proportional to its contribution to the total cfDNA pool. This principle is used in cfDNA-based tissue-of-origin deconvolution methods, which infer the fractional contribution of each tissue to the cfDNA pool by analyzing the coverage patterns at tissuespecific gene promoters.
[0154] In the context of FSHD, the relevant tissue-specific signal is the DUX4-activated transcriptional program in skeletal muscle cells. In healthy individuals and in patients with non-FSHD conditions, DUX4 target genes are not expressed in any tissue that contributes significantly to the cfDNA pool. Therefore, the promoters of DUX4 target genes are nucleosome-occupied (silent) in all contributing cell types, and the cfDNA coverage profile at these promoters reflects the silent pattern (uniform coverage, no NFR trough, decreased subnucleosome enrichment). In FSHD patients, a fraction of the cfDNA is derived from muscle cells in which DUX4 target genes are actively transcribed. This muscle-derived fraction contributes an active promoter pattern (NFR trough, flanking nucleosome peaks, increased subnucleosomes at +1 nucleosome) at DUX4 target gene promoters. The observed cfDNA coverage profile is the weighted average of the silent pattern (from hematopoietic and other non-muscle tissues) and the active pattern (from DUX4-expressing muscle cells). Even when the muscle-derived fraction is small (e.g., 1-5% of total cfDNA), the shift from the purely silent pattern is detectable at the statistical level provided by the ensemble analysis across many DUX4 target gene promoters.5. Signal-to-Noise Considerations
[0155] The sensitivity of the cfDNA fragmentomics-based FSHD diagnostic is fundamentally limited by the signal-to-noise ratio, which is determined by the fraction of muscle-derived cfDNA carrying the DUX4-activation signature relative to the total cfDNA pool. This disclosure overcomes the inherent challenges of detecting a small tissue-specific signal within a complex cfDNA mixture through several complementary strategies.
[0156] The first strategy is the ensemble approach, in which the fragmentomic signal is aggregated across dozens to hundreds of DUX4 target gene promoters. If the DUX4-Aty ref. 2848-355-PCT CU2025-209activation signal at each individual promoter has a signal-to-noise ratio of S, the aggregate signal-to-noise ratio across N independently informative promoters scales approximately as S x N, according to standard statistical principles. With N = 50 to 500 genes in the DUX4 target gene set, the aggregate signal-to-noise ratio is improved 7- to 22-fold relative to any single promoter.
[0157] The second strategy is the use of statistical enrichment methods (PSE score computation) that test for the coordinated directional shift of fragmentomic features across the gene set, rather than requiring a significant signal at any individual gene. This approach is analogous to GSEA and is robust to the presence of noise at individual genes.
[0158] The third strategy, employed in certain embodiments, is targeted sequencing, which concentrates sequencing reads at the informative loci (DUX4 target gene promoters) and achieves 10- to 100-fold higher coverage at those loci compared to WGS at the same total sequencing cost. The increased coverage provides more precise fragmentomic feature estimates and improves the detection of subtle signal shifts.
[0159] The fourth strategy is the use of machine learning classifiers trained on multidimensional feature vectors, which can capture complex, nonlinear relationships between fragmentomic features and disease status that may not be captured by simple summary statistics. These classifiers can learn to weight features optimally and to discount noisy or uninformative features, further improving diagnostic accuracy.G. Promoter Subnucleosome Enrichment (PSE) Score Computation1. Reference Profile Construction
[0160] The computation of the PSE score, which is the core diagnostic metric of this disclosure, begins with the construction of reference fragmentomic profiles from a training cohort. The training cohort comprises two groups: (a) an FSHD group consisting of subjects with genetically confirmed FSHD (FSHD1 with documented D4Z4 contraction to 1-10 units on a 4qA haplotype, and / or FSHD2 with documented SMCHD1 or DNMT3B mutation and D4Z4 hypomethylation on a 4qA haplotype), with a recommended minimum of 10 subjects and a preferred minimum of 30 subjects to adequately represent the variability within the FSHD population; and (b) a control group consisting of age- and sex-matched individuals without FSHD and without other active myopathies, with a recommended minimum of 10 and a preferred minimum of 30 subjects.
[0161] For each subject in the training cohort, cfDNA is collected, extracted, sequenced, and processed through the computational pipeline to generate fragmentomic featureAty ref. 2848-355-PCT CU2025-209profiles at each promoter window. The FSHD reference profile for each promoter window is then computed as the mean (or median, in alternative embodiments) of the normalized coverage profiles, WPS profiles, or other fragmentomic feature vectors across all FSHD subjects. Similarly, the control reference profile for each promoter window is computed as the mean of the same features across all control subjects.
[0162] In preferred embodiments, the reference profiles are computed not as position-based coverage vectors but as fragment-length distributions derived from cfDNA fragments mapping to defined promoter-proximal regions. In certain embodiments, fragment-length profiles are computed at 10-bp bin resolution across the range of approximately 35-405 bp using cfDNA fragments that overlap either (i) the +1 nucleosome position within a ±100 bp window or (ii) the broader promoter region spanning TSS ±2,000 bp. This approach yields a binned fragment-length vector that captures the relative abundance of subnucleosomal, nucleosomal, extended subnucleosomal, and dinucleosomal fragments at each promoter or regulatory locus. These fragment-length profiles serve as inputs to machine-learning or artificial-intelligence classification algorithms and are stored in the reference database component of the system for comparison with test samples.
[0163] The reference profiles may be updated as additional FSHD and control samples become available, improving the accuracy and representativeness of the reference. In certain embodiments, the reference profiles are stratified by demographic variables (age, sex, ancestry) to account for potential confounders. In other embodiments, batch correction methods (e.g., ComBat, Harmony, or linear regression-based methods) are applied to the reference profiles to remove technical variability between sequencing runs, platforms, or sites.2. Per-Promoter Scoring
[0164] In certain embodiments, the system computes a per-promoter quantitative score that reflects the promoter-proximal chromatin accessibility of each gene in the curated DUX4-responsive gene set. The method assigns a single promoter score per gene, derived from the fragment-length distribution of cfDNA fragments overlapping the +1 nucleosome position within a ±100 bp window. This promoter-specific score is sensitive to disease state and exhibits characteristic shifts in subjects with FSHD. For example, the promoter of ZBTB17, a canonical DUX4-responsive gene, displays a marked increase in promoterAty ref. 2848-355-PCT CU2025-209subnucleosome enrichment (PSE) in FSHD subjects relative to unaffected controls, demonstrating that even a single promoter score can discriminate disease status.
[0165] In preferred embodiments, the promoter score is computed using the PSE metric described herein, which quantifies the relative abundance of subnucleosomal (35-130 bp) and extended subnucleosomal (200-330 bp) fragments relative to nucleosomal (130-200 bp) and dinucleosomal (330-400 bp) fragments. The PSE score for each gene is computed according to Formula I:#(35-130 bp) + #(200-330 bp)PSE“ #(130-200 bp) + #(330-400 bp)This per-promoter PSE score (set forth above) captures the promoter-specific chromatin accessibility signature associated with DUX4 activation and forms the basis for gene-level and cohort-level analyses.
[0166] In another application, the fragment-length profiles described herein (computed at 10-bp resolution across the 35-405 bp range using cfDNA fragments mapping to either the +1 nucleosome ±100 bp window or the broader TSS ±2,000 bp region) are used as input features for a deep neural network (DNN) or other machine-learning model. In this embodiment, the model is trained directly on the binned fragment-length vectors for each promoter or genomic locus, enabling the classifier to learn complex, non-linear relationships between fragmentomic patterns and disease state. A new sample is classified by inputting its fragment-length profiles into the trained model, which outputs a predicted probability or categorical classification of FSHD status. This model-based approach leverages the full information content of the fragment-length distributions.3. Gene-Set-Level PSE Score
[0167] The aggregate PSE score across the DUX4 target gene set is computed using one of the following aggregation methods, each representing a distinct embodiment of this disclosure.
[0168] In a first embodiment, the mean PSE score is computed as the simple arithmetic mean of PSE i values across all N genes in the DUX4 target gene set: PSE_mean = (1 / N) x S PSE i. This method weights all genes equally and provides a straightforward summary of the aggregate signal. Its simplicity makes it interpretable and robust to outliers (which are averaged out over the gene set).Aty ref. 2848-355-PCT CU2025-209
[0169] In a second embodiment, the weighted PSE score is computed as a weighted average: PSE_weighted = X(w_i x PSE_i) / Xw_i, where w_i is a gene-specific weight reflecting the confidence or informativeness of gene i. The weights may be determined by: (a) the effect size (Cohen's d or standardized mean difference) of PSE i between the FSHD and control training groups, giving higher weight to genes with greater discriminatory power; (b) the inverse variance of PSE i in the training cohort, downweighting genes with high variability; (c) the strength of DUX4 binding at the promoter (as measured by ChlP-seq peak height), giving higher weight to genes with stronger DUX4 regulation; or (d) a combination of these factors.
[0170] In a third embodiment, a rank-based enrichment score is computed using a method analogous to Gene Set Enrichment Analysis (GSEA). In this embodiment, all gene promoters analyzed in the sample (genome-wide, or all promoters covered by the targeted panel) are ranked by their individual PSE i values, from highest (most FSHD-like) to lowest (most control-like). The positions of the DUX4 target genes within this ranked list are then evaluated using a running-sum statistic (Kolmogorov- Smirnov-like): as the ranked list is traversed from top to bottom, the running sum is incremented each time a DUX4 target gene is encountered and decremented at each non-target gene. The maximum deviation of the running sum from zero is the enrichment score (ES). The ES is then normalized by the number of permutations and the size of the gene set to yield a normalized enrichment score (NES). A positive NES indicates that DUX4 target genes are enriched at the top of the ranked list (i.e., among the genes whose cfDNA profiles most resemble the FSHD pattern), consistent with an FSHD diagnosis.
[0171] In a fourth embodiment, the vector of per-promoter PSE i values (or, more broadly, the multi-dimensional feature matrix encompassing NFR scores, S / L ratios, WPS features, and coverage profiles at all DUX4 target gene promoters) is used as the input to a machine learning classifier. Suitable classifiers include, but are not limited to: (a) logistic regression, which models the log odds of FSHD diagnosis as a linear function of the feature vector; (b) random forest, an ensemble of decision trees that captures nonlinear feature interactions; (c) gradient-boosted decision trees (e.g., XGBoost, LightGBM, CatBoost), which iteratively fit weak learners to the residuals of the previous iteration; (d) support vector machines (SVM) with linear or non-linear (radial basis function) kernels; and (e) deep neural networks (multi-layer perceptrons, convolutional neural networks, or attention-based architectures). The classifier is trained on the labeled training cohortAty ref. 2848-355-PCT CU2025-209(FSHD vs. control) and outputs a continuous probability score (between 0 and 1) representing the predicted probability that the test sample is from an FSHD-affected individual. This probability score is used as the PSE score in this embodiment.4. Statistical Significance Assessment
[0172] In certain embodiments, the robustness and statistical significance of the promoter subnucleosome enrichment (PSE) score are evaluated using a cross-validation framework rather than permutation testing. In preferred embodiments, the dataset is partitioned into four folds, with 25% of samples held out in each fold. The remaining 75% of samples (the training set) are used to identify genes exhibiting significantly increased or decreased PSE in FSHD relative to unaffected controls, thereby defining the positive (+) and negative (-) gene sets for that fold. These gene sets are then applied to the held-out test samples, and a composite PSE-based score is computed using only the genes identified from the corresponding training fold. This procedure is repeated across all four folds, ensuring that gene selection and score computation are performed on non-overlapping sample subsets. Consistent separation of FSHD and control samples across all held-out folds demonstrates that the observed PSE signal is not driven by overfitting and provides strong evidence that the promoter-level fragmentomic signature is reproducibly associated with disease state.
[0173] When multiple gene sets are tested (e.g., the DUX4 target gene set plus gene sets for other muscular dystrophies), multiple testing correction is applied using the Benjamini-Hochberg false discovery rate (FDR) procedure or the more conservative Bonferroni correction. The corrected p-value (q-value) is reported alongside the raw p-value.
[0174] Confidence intervals for the PSE score are computed by bootstrap resampling: the gene set is randomly resampled with replacement, the PSE score is recomputed for each bootstrap sample, and the 2.5th and 97.5th percentiles of the bootstrap distribution define the 95% confidence interval. The confidence interval provides a measure of the precision of the PSE score estimate and is reported in the diagnostic output.5. Threshold Determination and Classification
[0175] The diagnostic classification of a test sample as FSHD-positive or FSHD-negative is based on comparison of the sample's PSE score to a predetermined threshold. The optimal threshold is determined from the training cohort by analysis of the receiver operating characteristic (ROC) curve. The ROC curve plots sensitivity (proportion of true FSHD cases correctly classified as positive) against 1 -specificity (proportion of trueAty ref. 2848-355-PCT CU2025-209controls incorrectly classified as positive) across all possible threshold values. The AUC of the ROC curve provides an overall measure of diagnostic accuracy.
[0176] The optimal threshold may be selected by one of several criteria: (a) Youden's J index (J = sensitivity + specificity - 1), which maximizes the sum of sensitivity and specificity; (b) a minimum sensitivity target (e.g., sensitivity > 95%), with the threshold set at the value that achieves this sensitivity while maximizing specificity; (c) a minimum specificity target (e.g., specificity > 95%); or (d) a cost-based criterion that accounts for the relative costs of false-positive and false-negative diagnoses in the clinical context. The selected threshold is validated on an independent test cohort to confirm its generalizability.
[0177] The diagnostic output for each test sample includes the following elements: (a) the PSE score, as a continuous numerical value; (b) the 95% confidence interval of the PSE score; (c) the classification result (FSHD-positive or FSHD-negative); (d) the p-value from permutation testing; (e) quality metrics for the sample and assay (sequencing depth, duplicate rate, mean insert size, on-target rate for targeted panels); and (f) a visualization of the PSE score relative to the distribution of PSE scores in the training cohort (FSHD and control distributions), enabling the clinician to assess the confidence of the classification, a visualization of PSE scores of individual genes that correlate with clinical severity, relative to the distribution of PSE scores for each of those genes in the training cohort, stratified by the clinical severity score (FSHD and control distributions), enabling the clinician to assess severity of disease.
[0178] In certain embodiments, the classification output includes a three-category result: FSHD-positive (PSE score above a high-confidence positive threshold), FSHD-negative (PSE score below a high-confidence negative threshold), and Indeterminate (PSE score between the two thresholds). Samples classified as Indeterminate may be recommended for re-testing (with a new blood sample or at higher sequencing depth) or for confirmatory testing by an alternative method (Southern blot / PFGE, molecular combing, or OGM). H. DUX4 Target Gene-Set SignaturesI. Curation of the DUX4 Target Gene Set
[0179] The DUX4 target gene set used in the methods of this disclosure is curated from publicly available transcriptomic and epigenomic studies of DUX4 function. The primary sources for gene-set curation include: (a) RNA-seq studies of human myoblasts or myotubes transduced with DUX4 expression constructs, comparing transcriptomic profiles between DUX4-expressing and control cells at various time points post-transduction; (b)Aty ref. 2848-355-PCT CU2025-209RNA-seq studies of FSHD patient muscle biopsies compared with biopsies from unaffected control individuals or from muscles of the same patient that are clinically unaffected; and (c) ChlP-seq studies identifying genomic loci bound by DUX4 protein, with particular emphasis on binding at promoter regions within 2 kb of annotated TSSs.
[0180] Genes are included in the DUX4 target gene set if they satisfy one or more of the following criteria: (a) statistically significant upregulation (fold-change >2, adjusted p-value <0.05, data id="411" using Benjamini -Hochberg FDR correction) upon DUX4 expression in at least two independent transcriptomic studies; (b) the presence of a DUX4 CUT&RUN peak within the promoter window (TSS ±2,000 bp), as determined from high-resolution, CUT&RUN datasets generated by the inventors, which provide superior signal-to-noise and positional accuracy compared to ChlP-seq datasets; or (c) consistent identification as a DUX4 target in published curated gene lists from the FSHD research literature.
[0181] The DUX4 target gene set is organized into three tiers of confidence, reflecting the strength and consistency of evidence supporting each gene's identification as a DUX4 target. The Tier 1 (Core) gene set comprises approximately 50 genes that are the most robustly and consistently identified DUX4 targets across multiple independent studies and experimental systems. These include, but are not limited to: ZSCAN4, TRIM43, MBD3L2, LEUTX, KHDC1L, PRAMEF1, PRAMEF2, PRAMEF4, PRAMEF5, PRAMEF6, PRAMEF7, PRAMEF8, PRAMEF9, PRAMEF11, PRAMEF12, PRAMEF13, PRAMEF14, PRAMEF15, DUXA, KDM4E, RFPL2, RFPL4B, CCNA1, and ZNF679, among others.
[0182] The Tier 2 (Extended) gene set comprises approximately 200 genes that are identified as DUX4 targets with moderate confidence, including genes that are upregulated in some but not all DUX4 studies, genes activated via DUX4-responsive retroelement promoters (HERVL / MLT2A1 -driven transcripts), and genes encoding immune mediators and retroelement-encoded proteins. The Tier 3 (Full) gene set comprises approximately 500 genes that show statistically significant upregulation upon DUX4 expression, including genes with smaller fold-changes or less consistent identification across studies.
[0183] In preferred embodiments, the PSE score computation uses the Tier 1 (Core) gene set, which provides the highest signal-to-noise ratio due to the strong and consistent DUX4 responsiveness of these genes. In alternative embodiments, the Tier 2 or Tier 3 gene sets are used to capture a broader range of DUX4-responsive signals, which mayAty ref. 2848-355-PCT CU2025-209improve sensitivity at the cost of slightly reduced specificity. In certain embodiments, however, the gene set is not predetermined but is instead derived directly from the data. For example, in the analyses presented herein, genes were selected using across-validation framework in which 25% of samples were held out and the remaining 75% were used to identify genes exhibiting increased or decreased PSE in FSHD relative to controls (FIG. 3). In additional embodiments, genes were identified based on monotonic trends in PSE associated with clinical severity categories, yielding severity-correlated gene sets as shown in FIG. 5. These data-driven gene sets may be used independently or in combination for PSE-based scoring or machine-learning classification.2. Control Gene Sets
[0184] Several control gene sets are used in the methods of this disclosure for normalization, specificity assessment, and permutation testing. A housekeeping gene set comprising the promoters of 20 to 50 ubiquitously and stably expressed genes (e.g., GAPDH, ACTB, B2M, RPL13A, HPRT1, TBP, RPLPO, YWHAZ, SDHA, HMBS) is used to normalize the fragmentomic features of the test sample, accounting for global shifts in cfDNA quality, quantity, or composition that are unrelated to FSHD. The NFR scores and coverage profiles at housekeeping gene promoters are expected to be similar between FSHD patients and controls (since these genes are expressed in hematopoietic cells, the primary cfDNA source, in both groups), and any systematic differences observed at housekeeping promoters between samples or batches can be used as correction factors for the DUX4 target gene features.
[0185] Disease specific control gene sets are curated for other muscular dystrophies and neuromuscular diseases to assess the specificity of the DUX4 target gene PSE score for FSHD. For example, a dystrophin pathway gene set (DMD, UTRN, DAG1, DTNA, SGCA, SGCB, SGCD, SGCG, SNTB1, SNTB2) is used to evaluate whether the fragmentomic signal at these loci is elevated in Duchenne muscular dystrophy but not in FSHD. Similarly, gene sets for calpainopathy (CAPN3 and interacting partners), dysferlinopathy (DYSF and associated genes), and inflammatory myopathy (interferon stimulated genes, complement pathway genes) are used for differential diagnosis. Random gene sets of matched size (e.g., 50 randomly selected gene promoters) are generated for permutation testing.Aty ref. 2848-355-PCT CU2025-2093. Gene-Set Refinement
[0186] In certain embodiments, the DUX4 target gene set is refined empirically from the training cohort to optimize diagnostic performance. Feature selection is performed by ranking each gene in the initial gene set (Tier 1, 2, or 3) by its individual discriminatory power, measured by: (a) the AUC of a univariate ROC curve using PSE i as the predictor; (b) the absolute value of the t-test statistic comparing PSE i between FSHD and control groups; or (c) the mutual information between PSE i and the class label. Genes with individual AUC below a threshold (e.g., <0.55) data id="426" or with nonsignificant t-test results (p > 0.1) are removed from the gene set, as they contribute noise without meaningful signal. The resulting refined gene set typically comprises 30 to 100 genes and provides improved aggregate PSE score performance relative to the full, unrefined gene set.
[0187] Promoters that overlap other genes are discarded. This redundancy removal is performed to address the case of multiple genes sharing overlapping or adjacent promoter regions thus ensuring PSE scores are assignable to single genes.I. Optional Embodiments and Alternative Implementations1. Methylation Aware Fragmentomics
[0188] In a first alternative embodiment, the fragmentomic analysis of this disclosure is combined with DNA methylation analysis of the cfDNA. This is accomplished by subjecting the cfDNA to enzymatic methylation conversion (e.g., EM-seq using TET2 and APOBEC3 A enzymes, or TAPS using TET-mediated oxidation followed by pyridine borane reduction) or, alternatively, bisulfite conversion, prior to or during library preparation. The resulting sequencing data provides both the fragment-length and alignment information necessary for fragmentomic analysis and the methylation status of cytosine residues at CpG sites throughout the genome.
[0189] The dual analyte (fragmentomic + methylation) approach offers several advantages for FSHD diagnosis. D4Z4 methylation status is a direct indicator of the epigenetic state of the D4Z4 array, with hypomethylation (<25% data id="435" methylation at D4Z4-associated CpG sites) being a hallmark of both FSHD1 and FSHD2. By assessing D4Z4 methylation in cfDNA concurrently with the fragmentomic PSE score, the method can: (a) provide independent confirmation of the FSHD diagnosis; (b) distinguish FSHD1 from FSHD2, since FSHD1 patients typically have methylation in the 10-25% range while FSHD2 patients may have methylation levels below 10% at certainAty ref. 2848-355-PCT CU2025-209D4Z4-associated CpG sites; and (c) assess disease severity, as the degree of D4Z4 hypomethylation has been correlated with clinical severity and DUX4 expression levels. The methylation analysis at promoters of DUX4 target genes provides complementary information about the epigenomic state at these loci, independent of the nucleosome positioning based fragmentomic features.2. Fragment End Motif Analysis
[0190] In a second alternative embodiment, the fragmentomic analysis includes analysis of the 4-mer nucleotide sequences at the 5' and 3' ends of cfDNA fragments. These end motifs reflect the sequence preferences of the nucleases that generated the cfDNA fragments (principally CAD / DFF40 during apoptosis and DNASE 1 / DNASE 1L3 during extracellular processing). Different cell types and tissues have been shown to produce cfDNA with distinct end motif frequency profiles, reflecting differences in nuclease expression and activity, chromatin accessibility, and transcription factor binding. At DUX4 target gene promoters in FSHD muscle cells, the chromatin remodeling induced by DUX4 alters the local accessibility of DNA to nucleases and may produce characteristic end motif signatures that differ from the end-motif signatures at the same loci in non-FSHD cells. Analysis of end motif frequencies at DUX4 target gene promoters provides an additional, orthogonal feature for the FSHD classifier, complementing the coveragebased and fragment-length-based features.3. Multi Modal Integration
[0191] In a third alternative embodiment, the PSE score is combined with additional cfDNA-derived features in a multi modal diagnostic model. These additional features include: (a) total cfDNA concentration, which may be elevated in FSHD patients with active muscle damage; (b) global fragment-length distribution statistics, including the mean, variance, skewness, and kurtosis of the genome wide fragment-length distribution, as well as the proportion of sub-nucleosomal (<150 data id="442" bp), mononucleosomal (150-250 bp), and dinucleosomal (250-450 bp) fragments; (c) copy number variation (CNV) analysis at D4Z4-adjacent regions on chromosome 4q35, which may reveal contraction-associated dosage changes; and (d) plasma protein biomarkers (e.g., creatine kinase, aldolase, lactate dehydrogenase) measured from the same blood sample, providing complementary information about muscle damage. The multi-modal feature vector is input to a machine learning classifier (e.g., gradient boosted tree ensemble) for integrated FSHD classification.Aty ref. 2848-355-PCT CU2025-2094. Machine Learning Classifiers
[0192] In a fourth alternative embodiment, the diagnostic classification is performed by an ensemble machine learning classifier trained on the full multi-dimensional feature space. The classifier architecture may be: (a) a random forest comprising 100 to 1,000 decision trees, each trained on a bootstrap sample of the training data with a random subset of features; (b) a gradient-boosted decision tree ensemble (XGBoost, LightGBM, or CatBoost) with hyperparameters (learning rate, maximum depth, number of estimators, regularization parameters) optimized by cross-validation; (c) a deep neural network with multiple hidden layers, trained using backpropagation with dropout regularization; or (d) a stacked ensemble combining predictions from multiple base classifiers using a meta learner. The classifier is trained using k-fold cross-validation (k = 5 or 10) or “leave-one-out cross-validation” (LOOCV) on the training cohort, and performance is evaluated on a held-out independent validation cohort. Feature importance analysis (e.g., SHAP values, permutation importance) is performed to identify the most informative features and ensure model interpretability.5. Longitudinal Monitoring
[0193] In a fifth alternative embodiment, the method of this disclosure is applied to serial blood samples collected from an FSHD patient at multiple time points to monitor disease progression and therapeutic response. The PSE score is computed at each time point, and the trajectory of PSE scores over time is analyzed. A decrease in PSE score over time is interpreted as a reduction in DUX4 transcriptional activity in the muscle and may indicate: (a) response to DUX4-targeted therapy (e.g., antisense oligonucleotides, small interfering RNAs, gene therapy, or small molecule inhibitors such as losmapimod targeting p38a MAPK, which is involved in DUX4 expression regulation); (b) spontaneous fluctuation in disease activity; or (c) reduced muscle mass and correspondingly reduced muscle derived cfDNA contribution (in advanced disease with extensive muscle replacement by fibrotic / adipose tissue). Conversely, an increase in PSE score over time may indicate disease progression, therapeutic failure, or exercise induced muscle damage. The slope, inflection points, and variability of the PSE score trajectory over time provide clinically actionable information for disease management.6. Multiplexed Diagnostic Panels
[0194] In a sixth alternative embodiment, the method is extended to diagnose multiple neuromuscular diseases simultaneously from a single cfDNA sample. In this embodiment,Aty ref. 2848-355-PCT CU2025-209multiple disease-specific gene set signatures are defined (e.g., the DUX4 target gene set for FSHD, a dystrophin pathway gene set for DMD / BMD, a calpain-3 pathway gene set for LGMD2A, an interferon response gene set for inflammatory myopathies), and the PSE score (or an analogous enrichment score) is computed for each disease specific gene set. The pattern of enrichment scores across the disease specific panels provides a differential diagnostic profile, enabling classification of the subject into one of several disease categories or as unaffected. This multiplexed approach is particularly valuable in the clinical evaluation of patients presenting with nonspecific proximal muscle weakness, where the differential diagnosis includes multiple inherited and acquired myopathies.7. Point-of-Care and Rapid Implementations
[0195] In a seventh alternative embodiment, the method is adapted for rapid, low cost implementation suitable for point-of-care (POC) applications or resource-limited settings. In this embodiment, sequencing is performed on a portable, rapid turnaround sequencing platform, such as the Oxford Nanopore MinlON, which can generate sequencing data within hours of library preparation. A minimal DUX4 target gene set (10 to 20 genes) is used to reduce the sequencing depth required, and a simplified PSE score computation (e.g., mean NFR score across the minimal gene set) is implemented on a laptop computer or a cloud connected tablet device. The rapid POC embodiment sacrifices some diagnostic accuracy (due to the smaller gene set and lower sequencing depth) in exchange for speed, portability, and accessibility, and may be suitable for initial screening in settings where confirmatory testing is available.8. Computational System Embodiment
[0196] In another aspect, this disclosure provides a computational system for FSHD diagnosis comprising: (a) a bioinformatics pipeline module that receives sequencing data (FASTQ files or BAM files) as input and performs the read quality control, alignment, post-alignment filtering, fragment-length extraction, promoter window feature computation, and PSE score computation described herein; (b) a reference database module that stores the FSHD and control reference profiles, the curated DUX4 target gene set, the TSS annotations, the ENCODE blacklist, and other reference data; (c) a PSE computation engine that implements the calculation and aggregation methods; (d) a classification module that compares the PSE score to the predetermined threshold and generates the classification result; and (e) a reporting module that generates a structured diagnostic report including the PSE score, confidence interval, classification result, p-Aty ref. 2848-355-PCT CU2025-209value, quality metrics, and visualizations. The system may be deployed as a cloud-based software-as-a-service (SaaS) platform accessible via a web browser, as a standalone software package installable on a local server, or as an integrated component of a laboratory information management system (LIMS) or electronic health record (EHR) system.9. Kit Embodiment
[0197] In another aspect, this disclosure provides a diagnostic kit for FSHD, the kit comprising: (a) one or more cfDNA collection tubes suitable for blood collection and cfDNA stabilization (e.g., Streck Cell-Free DNA BCT tubes or equivalent); (b) cfDNA extraction reagents sufficient for the extraction of cfDNA from plasma, including lysis buffers, binding buffers, wash buffers, elution buffers, and solid phase extraction columns or magnetic beads; (c) sequencing library preparation reagents, including end repair enzymes, A-tailing enzymes, DNA ligase, sequencing adapters (optionally with UMIs), PCR primers, and PCR master mix; (d) optionally, a set of biotinylated capture probes (oligonucleotides) designed to hybridize with the promoter regions of the DUX4 target gene set and housekeeping control gene set, along with streptavidin coated magnetic beads and hybridization and wash buffers; and (e) a computer-readable medium (e.g., USB drive, SD card, or cloud access credentials) containing software implementing the bioinformatics pipeline, PSE computation engine, classification module, and reference database of this disclosure. The kit may further comprise quality control materials, including synthetic cfDNA fragments of known size, positive control cfDNA from an FSHD patient, and negative control cfDNA from a healthy individual.10. Newborn Screening
[0198] In another embodiment, the method of this disclosure is adapted for newborn screening for FSHD. In this embodiment, cfDNA is extracted from a dried blood spot (DBS) specimen collected from a neonate as part of standard newborn screening procedures. The DBS cfDNA extraction protocol is optimized for the small sample volume and the low total cfDNA yield from DBS, using magnetic bead-based extraction methods and extended incubation times. The extracted cfDNA is subjected to library preparation and targeted sequencing using the panel described above, and the PSE score is computed. Neonates with a PSE score above the screening threshold are classified as "at risk" for FSHD and are recommended for confirmatory genetic testing (D4Z4 sizing and haplotyping). Newborn screening for FSHD enables early identification of affectedAty ref. 2848-355-PCT CU2025-209individuals, facilitating genetic counseling for families, enrollment in clinical trials, and initiation of monitoring and supportive care before the onset of irreversible muscle damage.EXAMPLEScfDNA Extraction, Subnucleosome Profiling, and Identification of Gene Signatures Predictive of FSHD Presence and SeverityStudy Cohort and cfDNA Extraction
[0199] A cohort of 99 donors was assembled comprising 51 donors with genetically confirmed facioscapulohumeral muscular dystrophy (FSHD) and 48 unaffected control donors. Cell-free DNA (cfDNA) was extracted from approximately 2 ml of plasma obtained from each donor. The median yield across all extractions was 6.9 ng of cfDNA per 2 ml of plasma. No statistically significant differences in cfDNA quantity were observed between the FSHD-affected and unaffected donor groups (FIG. 1 A). These results confirm that cfDNA can be reliably extracted from this cohort without group-level technical bias, and that differences in downstream analytical outputs are not attributable to differential input material between groups.Sequencing Library Preparation and Quality Assessment
[0200] Sequencing libraries were prepared from the extracted cfDNA using a singlestrand library preparation method. High-quality libraries were obtained across all 99 donors, with the predominant population of library molecules migrating at approximately 300 bp, corresponding to a cfDNA insert size of approximately 170 bp. This fragment size distribution is consistent with mononucleosomal DNA protection in plasma, wherein the nucleosome core particle shields approximately 147 bp of genomic DNA from circulating nucleases. A characteristic ladder of approximately 170 bp multiples was observed in the fragment size profiles, confirming faithful capture of the nucleosomal fragmentation pattern characteristic of cfDNA (FIG. IB).
[0201] The prepared libraries were subjected to 150 bp paired-end sequencing on an Illumina sequencing platform. The median sequencing depth achieved was approximately 150 million fragment pairs per donor, corresponding to approximately 15x coverage of the human genome. No significant differences in sequencing depth were detected between FSHD-affected and unaffected donors (FIG. 1C). All 99 resulting cfDNA datasets were aligned to the Telomere-to-Telomere (T2T) human genome assembly (T2T-CHM13),Aty ref. 2848-355-PCT CU2025-209providing complete and contiguous genomic coverage, including regions not represented in prior reference assemblies.Promoter Subnucleosome Enrichment (PSE) Score Computation
[0202] For each donor, the cohort was stratified by FSHD status and age to empirically determine the position of the +1 nucleosome (defined as the first nucleosome positioned downstream of the transcription start site (TSS)) for approximately 15,000 genes. The Promoter Subnucleosome Enrichment (PSE) score was subsequently calculated for each gene in accordance with Formula I, as previously defined in the application. Briefly, the PSE score is defined as the ratio of the number of short cfDNA fragments (less than 130 bp) to the number of long cfDNA fragments (130-200 bp) mapping to the +1 nucleosome position of a given gene. This calculation yielded a PSE profile comprising approximately 15,000 values per donor across the full 99-donor cohort.
[0203] The biological rationale underlying the PSE score is as follows: genes that are transcriptionally active are associated with destabilized +1 nucleosomes at their promoter-proximal regions. Such destabilization results in the generation of shorter cfDNA fragments, as partially disassembled nucleosomes provide diminished protection against nuclease digestion. The PSE score thereby captures this chromatin accessibility signature on a gene-by-gene basis, enabling genome-wide inference of transcriptional activity from circulating cfDNA.Principal Component Analysis Reveals FSHD-Associated Separation
[0204] A principal component analysis (PCA) was performed on the complete matrix of PSE scores, comprising approximately 15,000 genes across all 99 donors. Plotting the second principal component (PC2) against the fourth principal component (PC4) revealed a visual separation between FSHD-affected donors and unaffected donors (FIG. 2). This preliminary, unsupervised observation demonstrated that the global landscape of promoter subnucleosome enrichment contains FSHD-discriminating information, and that this information was sufficiently structured to manifest in principal component space. These findings warranted further supervised analysis to identify specific genomic loci driving the observed separation.Four-Fold Cross-Validation Identifies a Robust FSHD Gene Signature
[0205] To identify a robust and generalizable gene signature capable of distinguishing FSHD-affected from unaffected individuals, a four-fold cross-validation analysis was conducted. The 99-donor cohort was partitioned into four subgroups, each balanced withAty ref. 2848-355-PCT CU2025-209respect to FSHD-affected and unaffected status. In each of the four cross-validation folds, three subgroups were combined to form the training set, while the remaining subgroup was held out as the independent test set.
[0206] During the training phase of each fold, genes whose PSE values exhibited an absolute Z-score greater than 2.5 when comparing FSHD-affected individuals to unaffected individuals within the training set were identified as candidate discriminating features. During the test phase, the FSHD score was calculated for each held-out test sample in accordance with Formula II, defined as the mean PSE at genes upregulated in FSHD-affected individuals ("up genes") minus the mean PSE at genes downregulated in FSHD-affected individuals ("down genes"). The resulting FSHD scores were then evaluated for their ability to separate affected from unaffected individuals within the held-out test set. In each of the four cross-validation rounds, statistically significant separation between affected and unaffected test samples was achieved, with an area under the receiver operating characteristic curve (AU-ROC) of 1.0 in each fold (FIG. 3 A).
[0207] Genes represented in the candidate feature sets of all four training folds were identified as a high-confidence, cross-validated gene signature. This common signature consisted of 98 genes whose PSE was increased in FSHD-affected individuals (up genes) and 147 genes whose PSE was decreased in FSHD-affected individuals (down genes), yielding a total of 245 genes. The FSHD score calculated using this common gene signature demonstrated clear and consistent separation between affected and unaffected individuals across the full 99-donor cohort (FIG. 3B). Notably, ZBTB17, a known target of the DUX4 transcription factor, exhibited one of the highest delta PSE scores across all four cross-validation folds. The prominence of ZB TB 17 in this analysis implicates DUX4-mediated transcriptional activation as a detectable and reproducible source of cfDNA signal in FSHD, consistent with the pathogenic role of DUX4 reactivation in disease. The ENSEMBL IDs of the 98 genes whose PSE was increased in FSHD-affected individuals (up genes) are: ENSG00000011243, ENSG00000014123, ENSG00000025796, ENSG00000033178, ENSG00000055211, ENSG00000064313, ENSG00000064995, ENSG00000071082, ENSG00000083814, ENSG00000087111, ENSG00000100227, ENSG00000100281, ENSG00000100359, ENSG00000100994, ENSG00000101152, ENSG00000101811, ENSG00000102699, ENSG00000102910, ENSG00000103248, ENSG00000104131, ENSG00000104365, ENSG00000105700, ENSG00000106993, ENSG00000107929, ENSG00000108671, ENSG00000115297, ENSG00000116685,Aty ref. 2848-355-PCT CU2025-209ENSGOOOOO 116809, ENSG00000116954, ENSG00000118518, ENSG00000119333, ENSG00000119844, ENSGOOOOO 121067, ENSG00000125520, ENSGOOOOO 127914, ENSG00000128581, ENSGOOOOO 130338, ENSGOOOOO 130713, ENSG00000131653, ENSGOOOOO 134480, ENSGOOOOO 134717, ENSGOOOOO 136243, ENSG00000137275, ENSGOOOOO 138035, ENSGOOOOO 138629, ENSGOOOOO 138778, ENSG00000140455, ENSG00000140575, ENSG00000143185, ENSG00000143207, ENSG00000143498, ENSG00000145555, ENSG00000147535, ENSG00000148334, ENSG00000148634, ENSG00000148719, ENSGOOOOO 151239, ENSGOOOOO 154222, ENSGOOOOO 154328, ENSG00000155287, ENSGOOOOO 156925, ENSGOOOOO 158158, ENSG00000159873, ENSGOOOOO 160752, ENSGOOOOO 160796, ENSGOOOOO 163558, ENSGOOOOO 163872, ENSGOOOOO 163900, ENSGOOOOO 164548, ENSGOOOOO 165678, ENSGOOOOO 167978, ENSG00000168653, ENSG00000171103, ENSG00000171806, ENSG00000173085, ENSG00000173273, ENSGOOOOO 174606, ENSGOOOOO 178074, ENSGOOOOO 178896, ENSGOOOOO 180776, ENSG00000181031, ENSGOOOOO 181220, ENSGOOOOO 184887, ENSGOOOOO 185619, ENSG00000186073, ENSGOOOOO 187118, ENSGOOOOO 188647, ENSGOOOOO 196313, ENSGOOOOO 196396, ENSGOOOOO 196782, ENSGOOOOO 196968, ENSGOOOOO 197147, ENSGOOOOO 198736, ENSG00000204130, ENSG00000214360, ENSG00000235750, ENSG00000262543, and ENSG00000275066. The ENSEMBL IDs of the 147 genes whose PSE was decreased in FSHD-affected individuals (down genes) are: ENSG00000004487, ENSG00000008988, ENSG00000012660, ENSG00000014164, ENSG00000021574, ENSG00000023697, ENSG00000054965, ENSG00000062822, ENSG00000065243, ENSG00000067955, ENSG00000071537, ENSG00000075461, ENSG00000075568, ENSG00000078668, ENSG00000089063, ENSG00000092199, ENSG00000094914, ENSG00000095203, ENSG00000095485, ENSGOOOOO 100150, ENSGOOOOO 100429, ENSGOOOOO 100526, ENSG00000101150, ENSG00000101363, ENSGOOOOO 103494, ENSG00000105443, ENSGOOOOO 106089, ENSGOOOOO 106682, ENSGOOOOO 107521, ENSGOOOOO 107959, ENSGOOOOO 108506, ENSGOOOOO 109685, ENSGOOOOO 110697, ENSGOOOOO 110925, ENSG00000111880, ENSGOOOOO 112062, ENSG00000112584, ENSG00000115137, ENSGOOOOO 115275, ENSGOOOOO 115760, ENSGOOOOO 118894, ENSG00000120805, ENSGOOOOO 121892, ENSGOOOOO 122417, ENSG00000122705, ENSG00000123575, ENSGOOOOO 124226, ENSGOOOOO 125124, ENSG00000125875, ENSG00000129351, ENSG00000130818, ENSG00000132155, ENSGOOOOO 132842, ENSG00000132912, ENSG00000133059, ENSGOOOOO 134200,Aty ref. 2848-355-PCT CU2025-209ENSG00000135919, ENSG00000135966, ENSG00000136895, ENSG00000138326, ENSG00000138382, ENSGOOOOO 138594, ENSG00000140386, ENSG00000141510, ENSGOOOOO 141551, ENSG00000142396, ENSG00000143390, ENSGOOOOO 143751, ENSGOOOOO 144747, ENSG00000144895, ENSG00000145741, ENSG00000145780, ENSG00000145916, ENSG00000147789, ENSG00000148572, ENSGOOOOO 150977, ENSG00000151304, ENSG00000152683, ENSG00000153037, ENSG00000153214, ENSG00000153786, ENSGOOOOO 155100, ENSG00000155115, ENSG00000156873, ENSGOOOOO 157426, ENSG00000157823, ENSGOOOOO 158402, ENSG00000158636, ENSGOOOOO 159202, ENSGOOOOO 159445, ENSGOOOOO 160633, ENSG00000161021, ENSGOOOOO 163291, ENSGOOOOO 164031, ENSGOOOOO 164167, ENSGOOOOO 164691, ENSGOOOOO 164815, ENSGOOOOO 164944, ENSGOOOOO 164975, ENSGOOOOO 165078, ENSGOOOOO 165417, ENSGOOOOO 166046, ENSGOOOOO 166167, ENSGOOOOO 166224, ENSGOOOOO 166801, ENSGOOOOO 166974, ENSGOOOOO 167065, ENSGOOOOO 167468, ENSGOOOOO 167614, ENSGOOOOO 168000, ENSGOOOOO 168994, ENSGOOOOO 169299, ENSGOOOOO 171451, ENSGOOOOO 172046, ENSGOOOOO 172062, ENSGOOOOO 172301, ENSGOOOOO 172500, ENSGOOOOO 172731, ENSGOOOOO 173064, ENSG00000175325, ENSG00000175595, ENSGOOOOO 175938, ENSGOOOOO 176092, ENSGOOOOO 176476, ENSG00000177151, ENSGOOOOO 177731, ENSGOOOOO 184216, ENSGOOOOO 184708, ENSG00000184983, ENSGOOOOO 185291, ENSG00000185880, ENSGOOOOO 186501, ENSGOOOOO 187514, ENSGOOOOO 189190, ENSG00000189339, ENSGOOOOO 197021, ENSGOOOOO 198960, ENSG00000206127, ENSG00000206527, ENSG00000213066, ENSG00000214595, ENSG00000216937, ENSG00000222028, ENSG00000229937, ENSG00000244038, ENSG00000253117, and ENSG00000267110.PSE Scores of DUX4-Induced Genes Correlate with Clinical Severity
[0208] To assess whether PSE scores at individual loci reflect the degree of clinical disease severity, the PSE scores of eight known DUX4-induced genes were evaluated across four clinical severity categories: unaffected; mild (Clinical Severity Score [CSS] 0-3); mid (CSS 4-6); and severe (CSS 7-9). The candidate genes examined were KHDC1L, LEUTX, MBD3L2, PRAMEF2, PRAMEF25, TRIM43, ZSCAN4, and ZBTB17.
[0209] In the analysis of the full cohort, most DUX4-induced genes showed limited correlation with CSS (FIG. 4A). This result is consistent with the biological observation that a high CSS does not necessarily reflect active, ongoing DUX4-mediated disease activity, particularly in older individuals who may be non-ambulatory and exhibit highAty ref. 2848-355-PCT CU2025-209cumulative clinical severity scores, yet have comparatively low levels of ongoing DUX4 transcriptional activity. ZBTB17 was an exception in the full-cohort analysis, demonstrating a positive trend in PSE scores across increasing CSS categories.
[0210] Restricting the analysis to donors under 40 years of age (a subgroup considered more likely to harbor active, ongoing disease with contemporaneous DUX4 activity) revealed positive correlations between CSS and PSE scores for at least two DUX4-induced genes: KHDC1L and LEUTX (FIG. 4B). These findings indicate that age is an important covariate when interpreting cfDNA-based disease severity correlations in FSHD, and that younger donors constitute a more informative subgroup for the identification of pharmacodynamic biomarkers reflective of active molecular disease.Genome-Wide Identification of PSE Signatures Tracking Clinical Severity
[0211] To systematically identify genes whose PSE profiles trend with clinical severity across the cohort, a genome-wide analysis was performed. For each of the approximately 15,000 genes assessed, a Z-score was calculated comparing the mean PSE value (and associated standard error of the mean) within each CSS category (mild, mid, and severe) to the mean PSE value in the unaffected category. Genes were selected that demonstrated either a monotonically positive delta Z-score with increasing CSS ("up-trending" genes) or a monotonically negative delta Z-score with increasing CSS ("down-trending" genes).
[0212] This analysis identified 110 genes exhibiting PSE values positively correlated with clinical severity, including ZBTB17, and 752 genes exhibiting PSE values negatively correlated with clinical severity (FIG. 5). The resulting two gene sets represent candidate pharmacodynamic biomarkers for monitoring disease activity and therapeutic response in future longitudinal and interventional studies. The identification of these severity-correlated gene sets demonstrates the potential of cfDNA subnucleosome profiling not only to discriminate FSHD-affected individuals from unaffected individuals, but also to provide a continuous, blood-based readout of disease progression across the clinical severity spectrum.Conclusion
[0213] Across a cohort of 99 donors comprising 51 FSHD-affected individuals and 48 unaffected controls, cfDNA subnucleosome profiling successfully distinguished FSHD-affected individuals from unaffected individuals with high diagnostic accuracy, achieving an AU-ROC of 1.0 in each fold of a four-fold cross-validation analysis. A 245-gene FSHD signature (comprising 98 genes with increased PSE in FSHD-affected individuals and 147Aty ref. 2848-355-PCT CU2025-209genes with decreased PSE) was identified and demonstrated reproducibility across independent sample subsets. Exploratory analysis further identified 110 PSE-increasing genes and 752 PSE-decreasing genes whose profiles correlate with clinical severity, providing a foundation for developing a blood-based, minimally invasive monitor of FSHD disease activity and therapeutic response in future clinical and interventional studies.K. Industrial Applicability
[0214] The methods, systems, kits, and compositions disclosed herein have direct industrial applicability in multiple domains. In the domain of clinical diagnostics, this disclosure provides a noninvasive, blood-based diagnostic test for FSHD that can be performed in any clinical laboratory equipped with standard NGS instrumentation and bioinformatics infrastructure. The test addresses the well documented limitations of current FSHD diagnostic methods (Southern blot / PFGE, molecular combing, optical genome mapping) by eliminating the need for high-molecular-weight DNA extraction, specialized gel electrophoresis equipment, and weeks-long turnaround times. The cfDNA-based test requires only a standard blood draw, utilizes commercially available reagents and sequencing platforms, and can deliver results within 3 to 5 business days from sample receipt, enabling faster clinical decision making and reduced patient anxiety.
[0215] In the domain of companion diagnostics and pharmacodynamics, this disclosure provides a biomarker (the PSE score) that reflects the activity of the DUX4-driven transcriptional program (the proximate cause of FSHD pathology) rather than merely the presence of the underlying genetic lesion. As DUX4-targeted therapies advance through clinical development (including antisense oligonucleotides, small interfering RNAs, gene therapy approaches, and small molecule inhibitors of pathways upstream of DUX4 expression such as p38a MAPK inhibitors), there is a critical need for pharmacodynamic biomarkers that can measure the biological effect of these therapies. The PSE score addresses this need by providing a quantitative, reproducible, noninvasive measure of DUX4 transcriptional activity that can be monitored longitudinally through serial blood draws. Regulatory agencies (FDA, EMA, PMDA) increasingly require companion diagnostic assays and pharmacodynamic biomarkers for the approval of targeted therapies, and the PSE score is well positioned to fulfill this regulatory requirement for DUX4-targeted FSHD therapies.Aty ref. 2848-355-PCT CU2025-209
[0216] In the domain of newborn screening, this disclosure provides a method for early identification of FSHD-affected neonates using dried blood spot specimens that are already routinely collected for other newborn screening tests. Early identification of FSHD enables genetic counseling for families, proactive monitoring for the onset of clinical symptoms, and early enrollment in clinical trials of preventive or disease modifying therapies. The DBS-based embodiment of this disclosure is compatible with the existing infrastructure of newborn screening programs and can be integrated into multiplexed newborn screening panels.
[0217] In the domain of genetic counseling, this disclosure provides a tool for assessing disease activity in genetically confirmed FSHD carriers who may be clinically asymptomatic. The penetrance of FSHD is incomplete, and many genetically affected individuals remain asymptomatic for years or decades. The PSE score may serve as an indicator of subclinical DUX4 activity, enabling risk stratification and personalized counseling regarding prognosis, family planning, and lifestyle recommendations.
[0218] In the domain of research, this disclosure provides a tool for advancing the understanding of FSHD biology, natural history, and the relationship between DUX4 transcriptional activity and clinical phenotype. The noninvasive nature of the cfDNA-based assay enables large scale longitudinal studies that would be impractical or unethical with repeated muscle biopsies. The PSE score can serve as a quantitative, objective endpoint in natural history studies and clinical trials, complementing subjective clinical outcome measures. Furthermore, the multiplexed diagnostic panel embodiment provides a research platform for the simultaneous investigation of multiple neuromuscular diseases.
[0219] The methods and systems of this disclosure are also applicable, with appropriate modification of the target gene sets, to the diagnosis and monitoring of other diseases characterized by aberrant activation of specific transcriptional programs. These include, but are not limited to: (a) other epigenetically-driven diseases in which loss of repressive chromatin marks leads to de-repression of specific genes or gene families; (b) cancers in which tumor-specific transcriptional programs (e.g., lineage-specific transcription factor targets) may be detected in ctDNA; and (c) autoimmune and inflammatory conditions in which tissue-specific gene expression programs are activated in the context of immune-mediated tissue damage. The general framework of computing an enrichment score for a disease-specific gene set based on cfDNA fragmentomic features at gene promotersAty ref. 2848-355-PCT CU2025-209represents a broadly applicable diagnostic paradigm that extends beyond the specific application to FSHD described in detail herein.
[0220] It will be understood by those of ordinary skill in the art that various modifications and variations of the methods, systems, kits, and compositions described herein can be made without departing from the spirit and scope of this disclosure. It is intended that this disclosure covers the modifications and variations of the invention provided they come within the scope of the appended claims and their equivalents. All references cited herein are incorporated by reference in their entirety.
Claims
Aty ref. 2848-355-PCT CU2025-209WHAT IS CLAIMED IS:
1. A method for detecting or diagnosing Facioscapulohumeral Muscular Dystrophy (FSHD) in a subject, the method comprising:a) obtaining a biological sample comprising cell-free DNA (cfDNA) from the subject;b) sequencing the cfDNA to generate a plurality of sequencing reads;c) mapping the sequencing reads to a reference genome;d) determining one or more cfDNA fragmentomic features at one or more genomic regions associated with DUX4-responsive genes, wherein the one or more cfDNA fragmentomic features are selected from the group consisting of:(i) a cfDNA fragment-length distribution,(ii) a nucleosome-positioning score,(iii) a promoter-subnucleosome enrichment (PSE) value,(iv) an end-motif frequency profile, and(v) a cfDNA coverage depth profile;e) comparing the one or more cfDNA fragmentomic features to a reference value derived from one or more control subjects; andf) classifying the subject as having FSHD, being at risk for FSHD, or being unaffected based on a deviation of the one or more cfDNA fragmentomic features from the reference value.
2. The method of claim 1, wherein the biological sample is selected from the group consisting of blood, plasma, serum, urine, and saliva.
3. The method of claim 1, wherein the cfDNA fragment-length distribution comprises determining a promoter subnucleosome enrichment (PSE) ratio defined as:#(35-130 bp) + #(200-330 bp)#(130-200 bp) + #(330-400 bp)’wherein the numerator comprises subnucleosomal and overlapping dinucleosomal cfDNA fragments and the denominator comprises nucleosomal and dinucleosomal cfDNA fragments.Aty ref. 2848-355-PCT CU2025-2094. The method of claim 3, wherein an increased PSE at the genomic regions associated with DUX4-responsive genes relative to the reference value is indicative of FSHD.
5. The method of claim 1, wherein determining the cfDNA fragment-length distribution comprises quantifying cfDNA fragments within fragment-length classes comprising:(a) subnucleosomal fragments of about 35-130 base pairs;(b) nucleosomal fragments of about 130-200 base pairs;(c) overlapping dinucleosomal fragments of about 200-330 base pairs; and (d) dinucleosomal fragments of about 330-400 base pairs,and computing a promoter-subnucleosome enrichment (PSE) ratio based on the relative abundance of these fragment-length classes.
6. The method of claim 1, wherein the promoter-subnucleosome enrichment (PSE) value is computed for each of the one or more genomic regions by:(i) defining a promoter window extending from about 2,000 base pairs upstream to about 2,000 base pairs downstream of a transcription start site (TSS) of a DUX4-responsive gene, and defining a +1 nucleosome window extending approximately 100 base pairs upstream and 100 base pairs downstream of a +1 nucleosome position downstream of the TSS; and(ii) calculating the PSE value as a ratio of fragment-length classes comprising:#(35-130 bp) + #(200-330 bp)#(130-200 bp) + #(330-400 bp)’wherein the numerator comprises subnucleosomal and overlapping dinucleosomal cfDNA fragments and the denominator comprises nucleosomal and dinucleosomal cfDNA fragments.
7. The method of claim 6, wherein the promoter window extends from about 1,000 base pairs upstream to about 1,000 base pairs downstream of the TSS.
8. The method of claim 1, wherein the one or more genomic regions associated with DUX4-responsive genes comprise loci of at least two genes selected from the group consisting of: ZBTB17, ZSCAN4, TRIM43, MBD3L2, LEUTX, KDM4E, DUXA, PRAMEF1, KHDC1L, TRIM49, RFPL2, TRIM43B, SLC34A2, CCNA1, andH3Yl.Aty ref. 2848-355-PCT CU2025-2099. The method of claim 8, wherein the one or more genomic regions comprise loci of at least three genes selected from the group consisting of ZBTB17, ZSCAN4, TRIM43, MBD3L2, and LEUTX.
10. The method of claim 8, wherein the one or more genomic regions comprise loci of at least five genes selected from the group consisting of ZBTB17, ZSCAN4, TRIM43, MBD3L2, LEUTX, KDM4E, DUXA, and PRAMEF1.
11. The method of claim 1, further comprising determining a cfDNA fragmentomic feature at a genomic region comprising a D4Z4 macrosatellite repeat region on chromosome 4q35.
12. The method of claim 11, wherein the cfDNA fragmentomic feature at the D4Z4 macrosatellite repeat region comprises a methylation-correlated cfDNA fragment-length signature.
13. The method of claim 1, wherein the nucleosome-positioning score comprises a windowed protection score (WPS) or a nucleosome-occupancy likelihood (NOL) computed from cfDNA fragment midpoints or endpoints mapped within the one or more genomic regions.
14. The method of claim 1, wherein the end-motif frequency profile comprises determining the frequency of 4-mer nucleotide motifs at 5' ends and / or 3' ends of cfDNA fragments mapped to the one or more genomic regions.
15. The method of claim 1, wherein the sequencing comprises whole-genome sequencing (WGS) at a depth of at least about 1 x to about 30* .
16. The method of claim 1, wherein the sequencing comprises targeted sequencing using a capture panel comprising probes or primers directed to the one or more genomic regions associated with DUX4-responsive genes.
17. The method of claim 16, wherein the capture panel further comprises probes directed to a D4Z4 macrosatellite repeat region on chromosome 4q35 and / or chromosome 10q26.
18. The method of claim 1, wherein the sequencing comprises whole-genome bisulfite sequencing (WGBS) or enzymatic methyl sequencing (EM-seq), and wherein the method further comprises determining a cfDNA methylation level at one or more CpG sites within a D4Z4 repeat, a DUX4 promoter, or a DUX4-PAS region.Aty ref. 2848-355-PCT CU2025-20919. The method of claim 1, wherein classifying the subject comprises applying a trained machine-learning classifier to the one or more cfDNA fragmentomic features, the trained machine-learning classifier selected from the group consisting of: a random forest, a gradient-boosted decision tree, a support vector machine, a logistic regression model, a neural network, and a combination thereof.
20. The method of claim 19, wherein the trained machine-learning classifier was trained on fragmentomic feature data from a training cohort comprising subjects with genetically confirmed FSHD type 1 (FSHD1), subjects with genetically confirmed FSHD type 2 (FSHD2), and unaffected control subjects.
21. The method of claim 1, further comprising determining a composite FSHD fragmentomic score by aggregating two or more of the one or more cfDNA fragmentomic features across the one or more genomic regions.
22. The method of claim 21, wherein the composite FSHD fragmentomic score comprises a weighted sum or a principal-component projection of PSE values and fragment-length ratios across at least four DUX4-responsive gene loci.
23. The method of claim 1, further comprising determining a disease-severity metric for FSHD based on a magnitude of deviation of the one or more cfDNA fragmentomic features from the reference value.
24. The method of claim 1, further comprising monitoring FSHD disease progression or response to a therapeutic intervention by comparing the one or more cfDNA fragmentomic features from the biological sample to cfDNA fragmentomic features from a prior biological sample obtained from the same subject at an earlier time point.
25. The method of claim 1, further comprising distinguishing between FSHD type 1 (FSHD1) and FSHD type 2 (FSHD2) based on a difference in cfDNA fragmentomic features at a D4Z4 macrosatellite repeat region combined with cfDNA methylation levels at D4Z4 CpG sites.
26. A biomarker panel for non -invasive detection of Facioscapulohumeral Muscular Dystrophy (FSHD), the biomarker panel comprising: a set of cell-free DNA (cfDNA) fragmentomic biomarkers measured at genomic loci corresponding to a panel of DUX4-responsive genes, the panel of DUX4-responsive genes comprising at least three genes selected from the group consisting of ZBTB17, ZSCAN4, TRIM43, MBD3L2,Aty ref. 2848-355-PCT CU2025-209LEUTX, KDM4E, DUXA, PRAMEF1, KHDC1L, TRIM49, RFPL2, TRIM43B, SLC34A2, CCNA1, and H3Y1;wherein the set of cfDNA fragmentomic biomarkers comprises at least two biomarkers selected from the group consisting of:(i) a cfDNA fragment-length distribution at each locus;(ii) a promoter-subnucleosome enrichment (PSE) value at each locus;(iii) an end-motif frequency profile of cfDNA fragments at each locus; and (iv) a nucleosome-positioning score at each locus; and wherein the biomarker panel discriminates between FSHD-affected subjects and unaffected subjects with a sensitivity of at least about 70% and a specificity of at least about 70%.
27. The biomarker panel of claim 26, wherein the panel of DUX4-responsive genes comprises ZBTB17, KHDC1L, ZSCAN4, TRIM43, MBD3L2, and LEUTX.
28. The biomarker panel of claim 26, further comprising a cfDNA fragmentomic biomarker measured at a genomic locus corresponding to a D4Z4 macrosatellite repeat region on chromosome 4q35.
29. The biomarker panel of claim 26, wherein the biomarker panel further comprises a cfDNA methylation biomarker at one or more CpG sites within a D4Z4 repeat, a DUX4-PAS region, or a DR1 region.
30. The biomarker panel of claim 26, wherein the sensitivity is at least about 85% and the specificity is at least about 85%.
31. A system for detecting Facioscapulohumeral Muscular Dystrophy (FSHD), the system comprising:a) a sequencer configured to sequence cell-free DNA (cfDNA) from a biological sample obtained from a subject and to generate a plurality of sequencing reads; b) one or more processors operably coupled to the sequencer or to a data store containing the sequencing reads; andc) a non-transitory computer-readable storage medium storing instructions that, when executed by the one or more processors, cause the one or more processors to:(i) map the sequencing reads to a reference genome;Aty ref. 2848-355-PCT CU2025-209(ii) determine one or more cfDNA fragmentomic features at one or more genomic regions associated with DUX4-responsive genes, the one or more cfDNA fragmentomic features selected from the group consisting of a cfDNA fragment-length distribution, a nucleosome-positioning score, a promoter-subnucleosome enrichment (PSE) value, an end-motif frequency profile, and a cfDNA coverage depth profile;(iii) compare the one or more cfDNA fragmentomic features to a reference value derived from one or more control subjects; and(iv) generate an output classifying the subject as having FSHD, being at risk for FSHD, or being unaffected based on a deviation of the one or more cfDNA fragmentomic features from the reference value.
32. The system of claim 31, wherein the sequencer is selected from the group consisting of an Illumina sequencing platform, an Oxford Nanopore sequencing platform, a PacBio sequencing platform, an Ion Torrent sequencing platform, and a combination thereof.
33. The system of claim 31, wherein the instructions further cause the one or more processors to compute a composite FSHD fragmentomic score by aggregating cfDNA fragmentomic features across at least four genomic regions associated withDUX4-responsive genes.
34. The system of claim 31, wherein the instructions further cause the one or more processors to apply a trained machine-learning classifier to the one or more cfDNA fragmentomic features.
35. The system of claim 31, wherein the output comprises a report indicating the classification of the subject and a confidence score associated with the classification.
36. The system of claim 31, further comprising a sample-preparation module configured to isolate cfDNA from the biological sample and to prepare a sequencing library from the isolated cfDNA.
37. A computer-implemented method for detecting Facioscapulohumeral Muscular Dystrophy (FSHD), the method comprising:a) receiving, by one or more processors, a plurality of sequencing reads derived from cell-free DNA (cfDNA) obtained from a biological sample of a subject;Aty ref. 2848-355-PCT CU2025-209b) mapping, by the one or more processors, the sequencing reads to a reference genome;c) computing, by the one or more processors, one or more cfDNA fragmentomic features at one or more genomic regions associated with DUX4-responsive genes, the one or more cfDNA fragmentomic features selected from the group consisting of:(i) a cfDNA fragment-length distribution,(ii) a nucleosome-positioning score,(iii) a promoter subnucleosome enrichment (PSE) value,(iv) an end-motif frequency profile, and(v) a cfDNA coverage depth profile;d) comparing, by the one or more processors, the one or more cfDNA fragmentomic features to a reference value derived from a reference population; ande) outputting, by the one or more processors, a classification indicating the subject as having FSHD, being at risk for FSHD, or being unaffected.
38. The computer-implemented method of claim 37, further comprising training a machine-learning classifier on fragmentomic feature data from a training cohort comprising subjects with genetically confirmed FSHD type 1 (FSHD1), subjects with genetically confirmed FSHD type 2 (FSHD2), and unaffected control subjects, and wherein the comparing comprises applying the trained machine-learning classifier to the one or more cfDNA fragmentomic features.
39. The computer-implemented method of claim 37, wherein computing the PSE value comprises:(i) defining a promoter window spanning about 2,000 base pairs upstream to about 2,000 base pairs downstream of a transcription start site (TSS) of aDUX4-responsive gene, and defining a +1 nucleosome window spanning about 100 base pairs upstream to about 100 base pairs downstream of a +1 nucleosome position downstream of the TSS; and(ii) calculating the PSE value as a ratio of fragment-length classes comprising: (#(35-130 bp) + #(200-330 bp)) / (#(130-200 bp) + #(330-400 bp)),Aty ref. 2848-355-PCT CU2025-209wherein the numerator comprises subnucleosomal and overlapping dinucleosomal cfDNA fragments and the denominator comprises nucleosomal and dinucleosomal cfDNA fragments.
40. The computer-implemented method of claim 37, wherein the one or more genomic regions associated with DUX4-responsive genes comprise loci of at least two genes selected from the group consisting of ZBTB17, ZSCAN4, TRIM43, MBD3L2, LEUTX, KDM4E, DUXA, PRAMEF1, KHDC1L, TRIM49, RFPL2, and H3Y1.
41. The computer-implemented method of claim 37, further comprising computing a nucleosome-positioning map by determining a windowed protection score (WPS) from cfDNA fragment midpoints at the one or more genomic regions.
42. The computer-implemented method of claim 37, further comprising computing an end-motif diversity score from 4-mer nucleotide motifs at 5' ends and / or 3' ends of cfDNA fragments mapped to the one or more genomic regions.
43. The computer-implemented method of claim 37, further comprising normalizing the one or more cfDNA fragmentomic features for GC-content bias, mappability, and / or cfDNA input amount.
44. The computer-implemented method of claim 37, further comprising computing a composite FSHD fragmentomic score from a weighted combination of PSE values, fragment-length ratios, end-motif diversity scores, and nucleosome-positioning scores across the one or more genomic regions.
45. The computer-implemented method of claim 37, further comprising generating a visual report depicting the one or more cfDNA fragmentomic features at each of the one or more genomic regions relative to the reference value.
46. A non-transitory computer-readable storage medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:a) receiving a plurality of sequencing reads derived from cell-free DNA (cfDNA) obtained from a biological sample of a subject;b) mapping the sequencing reads to a reference genome;c) computing one or more cfDNA fragmentomic features at one or more genomic regions associated with DUX4-responsive genes, the one or more cfDNAAty ref. 2848-355-PCT CU2025-209fragmentomic features selected from the group consisting of a cfDNAfragment-length distribution, a nucleosome-positioning score, apromoter-subnucleosome enrichment (PSE) value, an end-motif frequency profile, and a cfDNA coverage depth profile;d) comparing the one or more cfDNA fragmentomic features to a reference value derived from one or more control subjects; ande) outputting a classification indicating the subject as having FSHD, being at risk for FSHD, or being unaffected based on a deviation of the one or more cfDNA fragmentomic features from the reference value.
47. The method of claim 1, wherein the one or more cfDNA fragmentomic features further comprise a jagged-end index measuring a length of single-stranded overhangs at cfDNA fragment termini at the one or more genomic regions.
48. The method of claim 1, wherein the one or more cfDNA fragmentomic features further comprise an orientation-aware fragmentation pattern (OAFP) measuring strand-specific fragment-end preferences at the one or more genomic regions.
49. The method of claim 1, wherein the one or more genomic regions further comprise genomic loci of genes encoding proteins involved in muscle-specific chromatin remodeling selected from the group consisting of PAX7, MYODI, MYF5, MYOG, MYH3, and DES.
50. The method of claim 1, further comprising determining cfDNA fragmentomic features at a plurality of tissue-of-origin marker loci to estimate a skeletal-muscle-derived cfDNA fraction.
51. The method of claim 1, wherein the sequencing comprises long-read sequencing using an Oxford Nanopore or PacBio platform, and the method further comprises phasing cfDNA fragments to a chromosome 4q35 D4Z4 allele versus a chromosome 10q26 D4Z4 allele.
52. The method of claim 1, further comprising performing size selection on the cfDNA prior to the sequencing to enrich for cfDNA fragments having a length within a range of about 35 to about 400 base pairs.
53. The method of claim 1, wherein the one or more cfDNA fragmentomic features at the one or more genomic regions are determined within a plurality of non-overlappingAty ref. 2848-355-PCT CU2025-209genomic bins spanning each genomic region, each bin having a width of about 50 to about 500 base pairs.
54. The method of claim 1, further comprising correlating the one or more cfDNA fragmentomic features with a clinical severity score for FSHD selected from the group consisting of a Clinical Severity Score (CSS), a Reachable Workspace (RWS) metric, and a FSHD composite outcome measure.
55. A kit for performing the method of claim 1, the kit comprising:a) a cfDNA isolation reagent;b) a sequencing library preparation reagent set;c) a set of capture probes or amplification primers directed to genomic regions corresponding to at least three DUX4-responsive genes selected from the group consisting of ZBTB17, ZSCAN4, TRIM43, MBD3L2, LEUTX, KDM4E, DUXA, PRAMEF1, KHDC1L, TRIM49, RFPL2, and H3Y1; andd) instructions for use comprising a protocol for computing one or more cfDNA fragmentomic features and comparing the one or more cfDNA fragmentomic features to a reference value to detect FSHD.
