Discovery of SBMA proteomic signature from patient cohorts. The outline of our study is shown in Figure 1. Briefly, we investigated plasma and serum proteomes from patients with SBMA and age-matched healthy controls. We examined reproducibility across multiple cohorts, conducted gene set enrichment analyses (GSEA) to characterize findings, investigated correlations between protein abundance and clinical features, and characterized protein abundance in a SBMA mouse model.

Study design. Illustration outlining the experimental design and analyses. Multiplexed proximity extension assay was used to discover protein biomarkers in the plasma/serum from patients with SBMA. These proteins were then measured against clinical outcomes from these patients and in the AR113Q mouse model of SBMA.

For a summary of the data used in our study, see Table 1. We leveraged 2 independent discovery cohorts to investigate protein differences between SBMA and control samples, with all samples from male donors and patients with SBMA having a confirmed genetic diagnosis. One discovery cohort used samples obtained from research conducted at the NIH, with average age of patients with SBMA of 57.6 years (y) (±7.8 y) and of control donors 58.7 y (±7.3 y) (17). The other discovery cohort came from the UCL, from a longitudinal nonintervention study (UCL Discovery). The average ages from this cohort were 60.8 y (±9.3 y) and 59.4 y (±10.7 y) from patients with SBMA and control donors, respectively. To determine the proteomic profile of patients with SBMA, we utilized the Olink Explore 3072 platform, which assesses the abundance of 2,925 unique proteins. Using principal component analyses (PCA) on all assessed proteins, we found distinct separation of patients with SBMA from matched control samples across cohorts (Figure 2, A and C). We identified 185 and 280 proteins that exhibited differential abundance between SBMA versus controls in NIH and UCL cohorts, respectively (Padj < 0.05 and limit of detection > 50%; Figure 2, B and D). Applying a more conservative log2 fold change > |0.5| cut-off did not substantially change the number of differentially abundant proteins, with 95% (NIH cohort) and 97.5% (UCL cohort) of proteins retaining their differential abundance. Across cohorts, 46 proteins were significantly associated with SBMA (Padj <0.05 and limit of detection > 50%). The differential abundance of these proteins was highly correlated across cohorts (r = 0.78, P < 0.0001) with 40 of 46 of these proteins (87%) being estimated in the same direction in SBMA across cohorts (Figure 2E and Supplemental Table 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.176383DS1). Herein, these 40 proteins are referred to as the “SBMA-associated proteins” (Table 2).

Discovery of protein biomarker signature in SBMA. (A) PCA of control and patients with SBMA from the NIH cohort (n = 15 SBMA; n = 15 control). Differential abundance in protein expression was calculated using a linear regression framework, adjusting for age. (B) Volcano plot of all measured proteins from the NIH cohort. Proteins in red were increased, and proteins in blue were decreased. Proteins in green were not significantly different (Padj < 0.05 cut-off, dotted red line). Top proteins are labeled, and clinical labs measured in those patients are in bold and italicized. (C) PCA of samples from the UCL cohort (n = 19 SBMA; n = 12 control). (D) Volcano plot as described in B. (E) Correlation of proteins consistent across both cohorts; red indicates increased in both, blue indicates decreased in both, and green indicates inconsistent across cohorts. (F) Gene set enrichment analysis (GSEA) of SBMA proteomic associations showing the tissues and biological functions that were significantly enriched across cohorts.

Metadata of fluid samples for analysis

Listing of SBMA-associated proteins

The NIH cohort included clinical laboratory tests from both patients and controls, revealing differences in 6 of 14 measures between SBMA and controls (including creatinine, CK, glucose, insulin, alanine transaminase, and aspartate transaminase; Padj < 0.05). CK is widely used across several neuromuscular disorders, including prior SBMA interventional trials (18, 19). In the NIH cohort, half of the SBMA-associated proteins (20 of 40) exhibited a more significant association with SBMA than CK (Padj < 4.6 × 10–5) and 77.5% of these proteins (31 of 40) showed less variability than CK in patients with SBMA (coefficient of variation < 49.8%). This suggests that these proteins may offer enhanced sensitivity and reliability to currently available biomarkers.

To gain more insight into the biological systems in which these proteomes were enriched, we performed GSEA on thousands of gene sets using all detected proteins in each cohort. We found that 29 gene sets were significantly enriched across cohorts (Padj < 0.05). The most significant enrichment was observed in gene sets associated with skeletal muscle expression and function, but significant enrichment was also detected in RNA Binding, Mitochondrion, and Calcium Signaling, among other gene sets (Figure 2F and Supplemental Table 2).

Replication of SBMA-associated proteins in independent patient cohorts. Plasma samples were also collected in a separate, multisite observational study composed of only patients with SBMA (Table 1). Plasma proteomes from these patients were analyzed on the same Olink Explore 3072 platform and compared with the control samples from the 2 discovery cohorts. Once again, PCA indicated that the SBMA samples clustered separately from the control samples, suggesting proteome-level differences between cases and controls (Figure 3A). In this data set, the abundance of 126 proteins were significantly different in SBMA than controls (P < 0.05; Figure 3B). Furthermore, all 40 SBMA-associated proteins demonstrated differential abundance (P < 0.05) in this independent SBMA cohort, building further evidence for the robustness of these biomarkers (Figure 3C).

Replication of SBMA proteomic signature. (A) PCA of Nido Biosciences SBMA samples (n = 9) compared with control samples from NIH and UCL. (B) Volcano plot of increased (red), decreased (blue), or unchanged (green) proteins. (C) Log2 fold change of 40 SBMA-associated proteins from Nido Bio compared with control samples from NIH (n = 15) and UCL cohorts (n = 12). Data are plotted as mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Most of the SBMA-associated proteins had increased abundance in patients, while Repulsive Guidance Molecule A (RGMA), MSTN, and ART3 were the only 3 proteins that were decreased in patients as compared with healthy controls. Some of the proteins with the highest fold change have also been identified in samples from patients with other skeletal muscle–related disorders such as Duchenne muscular dystrophy (DMD) or facioscapulohumeral muscular dystrophy (FSHD) (20–25) (Table 2). It is noteworthy that CK is also elevated in cases of DMD and FSHD, but the creatine kinase – muscle isoform (CKM) is not measured in the 3072 platform. By comparison, biomarker papers from more neural-related disorders such as amyotrophic lateral sclerosis, Parkinson’s disease, or Huntington’s disease did not report differential abundance of the 40 SBMA-associated proteins (26–28).

A separate replication cohort of patients with SBMA and healthy controls from UCL (UCL Replication) was analyzed on a more limited panel of 1,536 antibody pairs. Similarly in this cohort, we found that patients with SBMA showed modest clustering from healthy controls and that 113 proteins showed differential abundance between patients with SBMA and healthy controls (P < 0.05; Supplemental Figure 1, A and B). However, within this limited panel, only 13 of the 40 SBMA-associated proteins were assessed, and 10 of the 13 assessed proteins (76.9%) exhibited differential abundance between SBMA and controls (Supplemental Figure 1C).

To further establish that these proteins could reliably distinguish patients with SBMA from healthy controls, we generated ROC curves for each of the 40 proteins and confirmed substantial consistency across cohorts with an average AUC of 0.94 (SD = 0.06) (Supplemental Figure 2). While these data illustrate the utility of these markers in separating patients from healthy controls, they are not intended to be diagnostic, which is done by a genetic test.

Using the Human Protein Atlas, we mapped the human tissue expression of these 40 SBMA-associated proteins and found that many of them were enriched in skeletal muscle and tongue and slightly less in cardiac muscle (Supplemental Figure 3). These expression profiles align with our GSEA and further support that muscle tissue is a major site of pathology of SBMA. A few of the SBMA-associated proteins have ubiquitous expression patterns across tissues, such as NDUFS6 and CHCHD10, which are localized to mitochondria, with the latter being associated with a subtype of ALS (29). Others are expressed in tissues relevant to other sites of SBMA pathology, such as ART3 and PHOSPHO1 (testis), MEGF10 and NGRN (spinal cord), and GOT1 (liver).

Overlap of SBMA-associated proteins and AR gene regulation data sets. Given that these proteins were detected in circulating biofluids, identifying their tissue expression patterns is insufficient to fully explain their relevance to SBMA pathogenesis. To gain further insight into potential associations with AR dysfunction, we intersected the 40 SBMA-associated proteins with a published data set that used primary human skeletal muscle myoblasts to perform AR ChIP coupled with microarrays (30). Proteins whose genes contain androgen response elements (AREs) and are bound by AR may be more likely to have altered levels in the context of SBMA. Using this data set, we found 19 proteins whose encoding genes have evidence of being regulated by AR (Supplemental Figure 4A). Dystrobrevin β (DTNB) followed by CAPN3 and DMD have the highest peak values, suggesting that their genes contain AREs that are more likely to be occupied by AR.

To further investigate whether these 40 proteins may be regulated by AR, we also interrogated a data set derived from RNA-Seq on skeletal muscle from WT and Ar-KO mice (31). Genes that were differentially expressed between Ar-KO and WT muscle after DHT administration include Mstn, Art3, Mb, Dmd, and Nos1 (Supplemental Figure 4B). In total, 12 genes had altered expression with the 5 most affected genes being downregulated. Although these data sets have some limitations, their information is complementary, and therefore proteins that were identified in both analyses — MSTN, ART3, MB, DMD, TTN, and MYL1 — suggest an increased likelihood of being directly regulated by AR. It should be noted that gene regulation by AR is one of many mechanisms by which fluid biomarkers can be altered. Proteolysis of extracellular domains or passive release from tissue damage are also likely mechanisms.

Correlations between patient functional assessments and SBMA-associated proteins. For some SBMA cohort blood measures, muscle MRI and clinical functional assessments were also performed at the time of sample collection. Muscle MRI has recently been shown to be an informative biomarker, as it quantifies muscle atrophy and fat infiltration into the muscles, 2 major features of SBMA (32). Furthermore, muscle MRI measurements were shown to correlate with clinical assessments. These additional measurements allowed us to correlate the abundance of the 40 SBMA-associated proteins with other meaningful aspects of the disease within patients and investigate the connection of these biomarkers with relevant metrics of disease. In the NIH cohort, we found that nearly every protein correlated with at least 1 clinical parameter measured in those patients. Only 3 proteins, including RGMA, MSTN, and ART3, and 1 metabolite (creatinine) were found to have decreased abundances in plasma, all of which significantly correlated with higher thigh MRI muscle fat fraction (MFF) (Figure 4). In the UCL cohort, the SBMA Functional Rating Scale (SBMAFRS; a rater-based measurement of 5 domains and multiple subdomains relevant to SBMA phenotypes) (33) and the 6-minute walk test (6MWT) functional assessments were collected. Many proteins that were increased in plasma had an inverse correlation with “walking” and “stairs” subscales of the SBMAFRS scores, indicating that higher protein levels correlated with lower scores and increased disease severity. Many of the proteins that correlated with lower-limb items on the SBMAFRS were also associated with the 6MWT in the same direction, suggesting consistency across functional assessments. Examples of individual proteins significantly associated with lower limb impairment are reported in Supplemental Figure 5.

A heatmap showing the correlations (r) of the 40 SBMA-associated proteins with clinical labs and functional readouts in the same patients using linear regression. The different cohorts are labeled on the left of the heatmap, proteins and clinical labs (italicized) are labeled on the bottom, and traits are labeled on the right. The dendrogram at the top clusters the proteins via a hierarchical clustering algorithm. #P < 0.1; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 as assessed by linear regression. NA, not applicable; MFF, muscle fat fraction.

Some patients in the UCL cohort also had longitudinal follow-up assessments and blood collections every 13 months on average (Table 1), and this allowed us to assess the dynamics of the SBMA-associated proteins over time. Results show that, of the 40 SBMA-associated proteins, only RGMA and MEGF10 change over time (Supplemental Figure 6) while most proteins in the SBMA proteomic signature were stable over approximately 1 y.

Correlations between SBMA-associated proteins and phenotypes in SBMA preclinical mouse model. Finally, we utilized the AR113Q preclinical mouse model of SBMA, which was generated by knocking in a human AR exon 1 with 113 CAG repeats to the mouse Ar locus (34). These mice mimic the genetic cause of SBMA and have been shown to recapitulate the pathological features of the disease such as muscle atrophy and polyQ AR protein accumulation. Consistent with previous studies, our data confirm that AR113Q mice exhibit body weight loss of approximately 15% from WT and exhibit decreased survival to 50% by 30 weeks of age (Figure 5, A and B). Currently, the cause of either of these phenotypes is not well known. To determine if these 2 phenotypes were related to each other, we used time-varying Cox proportional hazard ratio modeling and found that decreases in body weight were significantly associated with premature death in AR113Q mice (P = 0.0374; Figure 5C).

Phenotypes and protein signature in AR113Q mice. (A) Weekly body weight of WT and AR113Q mice from 5–30 weeks of age (2-way ANOVA). At study start, n = 47 for WT and n = 45 for AR113Q, with the number of mice that died each week during the study subtracted. (B) Kaplan-Meier survival plot comparing WT and AR113Q mice. The significance was measured at 30 weeks (Mantel-Cox). (C) Relationship between body weight and survival, categorized by AR113Q mice that died or survived. (D) Volcano plot of 92 proteins, as described previously. The red dashed line represents a P < 0.05 from multilevel metaregression models. (E and F) Volcano plots showing the correlation of protein levels with body weight or rate of death, respectively, across time in AR113Q mice. Eda2r: hazard ratio = 2.63, P = 0.003; Rgma: hazard ratio = 0.019, P = 0.0185 via time-varying Cox proportional hazard ratio models. All proteins below the red dashed line (P < 0.05) were not associated with weight or survival. The β represents the regression coefficient/slope of the model. (G) Differential abundance (log2 fold change) of Eda2r and Rgma between AR113Q and WT at different ages. n = 12 for WT and n = 15–30 for AR113Q mice, depending on each week. Week 18 samples were not available for analysis. ****P < 0.0001. (H) Change in levels of Eda2r and Rgma and association with body weight in AR113Q mice. (I) Change in levels of Eda2r and Rgma over time in AR113Q mice that died.

To measure the plasma proteomic profile of AR113Q mice, we again used the Olink platform; however, this rodent panel was limited and only measured 92 proteins. Of these 92 proteins, only 2 of the 40 SBMA-associated proteins were measured. We found that 36 proteins exhibited differential abundance between AR113Q versus WT (P < 0.05; Figure 5D and Supplemental Table 3). Interestingly, Ectodysplasin A2 receptor (Eda2r; Xedar, Tnfrsf27) which was one of the 2 SBMA-associated proteins measured in this panel, was significantly increased in AR113Q mice (P = 5.47 × 10–7). Eda2r is a single-pass transmembrane receptor of the TNF family, located near AR on the X-chromosome, and is linked to several overlapping features of patients with SBMA or AR biology, including aging, metabolic dysfunction, dysregulation of hair growth (androgenic alopecia), and muscle atrophy (35, 36).

We also tested whether plasma protein levels in AR113Q mice correlated with body weight or survival phenotypes. Our analyses identified several proteins that were significantly associated with body weight (Figure 5E) and survival (Figure 5F) using mixed-effects regression and time-varying Cox proportional hazard ratios models. Increases in Eda2r were associated with decreases in body weight and premature death in AR113Q mice (Figure 5, G–I). Decreases in Rgma, the other SBMA-associated protein measured in this panel, was associated with decreased body weight and premature death in AR113Q mice (Figure 5, G–I). Rgma is a GPI-anchored glycoprotein that functions as a guidance molecule, with tissue-dependent roles of cellular adhesion or repulsion (37–39). Rgma has an important role in nervous system development and maintenance as well as myoblast fusion and myotube hypertrophy. Together, both Eda2r and Rgma were constituents of the SBMA proteomic signature, correlated with other SBMA-relevant endpoints, and are reported to function in biology relevant to the pathophysiology of SBMA, making them interesting targets for further investigation in the context of SBMA.