A retrospective, exploratory analysis identifies biomarkers predictive of spinal muscular atrophy (SMA) response to nusinersen and suggests the efficacy of using machine learning algorithms to anticipate patient outcomes.
Researchers integrating biomarker and clinical data into a machine learning algorithm found success in identifying predictors of motor functionality improvement following nusinersen treatment in patients with spinal muscular atrophy (SMA), according to a study published in the Journal of Clinical Medicine.
Although the availability of FDA-approved, disease-modifying treatments have recently altered the natural disease course of SMA, the cellular pathways affected by restoration of the survival motor neuron (SMN) protein are still unexplained. Furthermore, research has been unable to use biomarkers to adequately predict patient responses to treatment. However, the authors note that cerebrospinal fluid (CSF) offers a good opportunity to identify biomarkers associated with SMA and the effects of nusinersen because it is not in direct contact with the degenerating ventral horn motor neurons that SMA influences. Additionally, the authors write that the neurofilaments that leak into CSF could be promising biomarkers in SMA as they are not disease-specific and can signal disease activity throughout many neurological disorders. To address this gap and build on this knowledge, the researchers conducted a CSF proteomic study in patients with SMA to investigate treatment pathways linked to nusinersen treatment and predict motor improvement.
The analysis included 49 patients who began nusinersen treatment between 2017 and 2018. Each patient submitted CSF samples at baseline (T0) and 6 months (T6) to the Stanford Neuromuscular Repository. CSF was sampled immediately before their first intrathecal injection of nusinersen, before injections on day 15, 29, and 64, and subsequently every 4 months. The first 0.5 mL of CSF was sent for an analysis of basic biochemistry and cell count while the rest was centrifuged. These patients’ proteomic samples were gathered through the Olink platform to detect protein biomarkers and characterize 1113 total peptides.
At baseline and throughout follow-up, motor function was also assessed with the Children’s Hospital of Philadelphia Infant Test of Neuromuscular Disorders (CHOP-INTEND), Hammersmith Functional Motor Scale Expanded (HFMSE), Revised Upper Limb Module (RULM), and/or the 6-Minute Walk Test (6MWT). Patient outcomes were evaluated after 2 years of nusinersen treatment or the last clinical follow-up.
Of the 49 patients, 10 had SMA type 1, 18 had type 2, and 21 had type 3. At a median of 746 days, final motor assessments were conducted. Thirty-one patients exhibited improvements in motor function, with the highest proportion occurring in younger patients who were the most severely affected by SMA (type 1) compared with older, less severely affected patients (types 2 and 3), respectively.
The CHOP-INTEND assessment registered the most patient improvements (n = 31) with median scores increasing by 5 points and the highest improvements reaching over 10. For the HFMSE, RULM, and 6MWT assessments, those that improved did so with median measurements increasing by 3 points, 1 point, and 54 meters.
At T6, a total of 595 peptides were reliably identified across the entire cohort. CSF proteomic changes revealed that 43 proteins had statistically significant reductions in CSF. Through the use of a random forest machine learning algorithm, researchers identified 4 key proteins that were viewed as the most predictive of 2-year motor improvement: arylsulfatase B (ARSB), ectonucleoside triphosphate diphosphohydrolase 2 (ENTPD2), neurofilament light chain (NEFL), and interferon-gamma-inducible protein 30 (IFI30). The authors note that this algorithm was able to use baseline clinical data and 6-month CSF proteomic changes to predict patient motor improvement at a rate of nearly 80%.
The findings here demonstrate that most proteins underwent CSF concentration reductions following 6 months of nusinersen intervention. Furthermore, the machine learning algorithm used was able to adequately identify proteins indicative of a patient’s treatment response. Authors express the need to validate these findings in larger cohorts utilizes patients with various types of SMA to not only confirm their results but also better their model’s predictive capacity. Overall, they believe the use of a machine learning predictive model can provide great benefits to informing treatment approaches and improving outcomes in patients with SMA.
Beaudin M, Kamali T, Tang W, et al. Cerebrospinal fluid proteomic changes after nusinersen in patients with spinal muscular atrophy. J Clin Med. 2023;12(20):6696. doi:10.3390/jcm12206696