Longitudinal motor function and biomarker correlates in treated adult spinal muscular atrophy: a single-center cohort study

Figures

• 

  Figure 1. Genomic structures of variant SMN2 genes. Five SMN2-related genomic structures were identified in this cohort, comprising combinations of standard SMN2 alleles and hybrid alleles. In the schematic diagram, blue boxes represent exon 8 derived from the SMN1, while orange boxes represent exons from SMN2. Hybrid alleles contain exon 7 from SMN2 and exon 8 from SMN1, and are illustrated accordingly. These structural variations were used to classify genotypes and examine genotype–phenotype correlations.

• 

  Figure 2. Genotype-specific distributions of clinical, laboratory, and physiological/electrophysiological parameters. Boxplots show the distribution of clinical scores (age at onset, RULM, HFMSE), laboratory markers (CK, Cr, creatine), and physiological/electrophysiological parameters (%VC, ulnar nerve CMAP amplitude, ulnar nerve decremental response), stratified by SMN2 genotype. Each group represents a distinct SMN2 copy number and hybrid configuration: three SMN2 copies without hybrids (n = 4); three copies with one hybrid (n = 4); four copies without hybrids (n = 11); four copies with one hybrid (n = 3), and four copies with two hybrids (n = 1). RULM, Revised Upper Limb Module; HFMSE, Hammersmith Functional Motor Scale Expanded; CK, creatine kinase; Cr, serum creatinine; %VC, vital capacity; CMAP, compound muscle action potential.

• 

  Figure 3. Treatment and follow-up. Nusinersen was administered with loading doses at 0, 1, and 3 months, followed by maintenance dosing every 6 months with clinical evaluations at each injection. For patients who were switched to risdiplam, evaluations were performed every 6 months. Orange bars indicate nusinersen treatment, and green bars indicate risdiplam. Five patients transitioned to risdiplam during the study period, and two of these later reverted to nusinersen. The mean follow-up duration was 57 months, with a maximum of 81 months.

• 

  Figure 4. RULM and HFMSE scores after treatment initiation. (A,B) Individual trajectories of RULM and HFMSE scores over time. (C,D) Mean changes from baseline (ΔRULM and ΔHFMSE) at each follow-up period. RULM, Revised Upper Limb Module; HFMSE, Hammersmith Functional Motor Scale Expanded.

• 

  Figure 5. SHAP analyses for random forest models predicting RULM and HFMSE. (A) Mean absolute SHAP values (feature importance) for the RULM model. (B) Mean absolute SHAP values for the HFMSE model. (C,D) SHAP summary plots showing the distribution of SHAP values for each feature; colors denote feature values from low (blue) to high (pink). Ulnar CMAP amplitude ranked highest for RULM, whereas CK ranked highest for HFMSE. Higher ulnar CMAP values were generally associated with higher predicted RULM, and higher CK often increased predicted HFMSE, although contributions of CK varied widely across observations, reflecting between-patient heterogeneity. Models represent the best-performing random forests selected under leave-one-subject-out (LOSO) cross-validation with fold-internal preprocessing. SHAP values reflect contributions to model predictions and do not imply causality. SHAP, SHapley additive explanations; RULM, revised upper limb module; HFMSE, Hammersmith functional motor scale expanded; CK, creatine kinase; Cr, serum creatinine; %VC, vital capacity; CMAP, compound muscle action potential.