# Surrogate Endpoint Lessons from Tofersen (Qalsody): Evidence Since Approval and Conditions for Generalization Prepared: 2026-05-04 --- ## 1. Executive Summary Tofersen (brand name Qalsody) received accelerated approval from the FDA in April 2023 for SOD1-associated amyotrophic lateral sclerosis (SOD1-ALS), based primarily on reduction of plasma neurofilament light chain (NfL) as a surrogate endpoint. Since approval, incremental real-world evidence and long-term extension data have accumulated, but fundamental questions remain about whether the surrogate-endpoint model can translate to the broader ALS landscape. This note reviews (a) post-approval functional clinical data, (b) efforts to extend the surrogate approach to non-SOD1 subtypes, and (c) emerging targets where similar antisense oligonucleotide (ASO) or gene-therapy strategies are in early development. We then identify three to five critical gaps that must close before the surrogate-endpoint framework can be generalized across ALS. --- ## 2. Post-Approval Evidence on Functional Clinical Outcomes (a) ### 2.1 Phase-3 VALOR Trial and Its Biomarker Foundation The VALOR phase-3 trial established that intrathecal tofersen reduced cerebrospinal fluid (CSF) SOD1 protein, plasma NfL, and CSF NfL in SOD1-ALS patients over 28 weeks. The primary analysis did not achieve statistical significance on the ALS Functional Rating Scale–Revised (ALSFRS-R) slope at the interim analysis, but the FDA granted accelerated approval based on the well-established mechanistic relationship between NfL reduction and neuronal injury, combined with the unmet need in a uniformly fatal disease. ### 2.2 Long-Term Pharmacodynamic Biomarkers from VALOR CSF A 2026 proteomic analysis of longitudinal VALOR CSF samples identified 56 proteins significantly modulated by tofersen versus placebo (Guise et al., PMID 41850233). Notably, **CSF GPNMB (glycoprotein nmb)** was significantly and continuously elevated across all post-baseline timepoints, detectable as early as 4 weeks. This provides a second pharmacodynamic biomarker beyond NfL and suggests that unbiased proteomic screening can surface treatment-response signatures that could be leveraged in future trials. Whether GPNMB elevation predicts functional stabilization—or merely reflects on-target biology—remains to be validated against long-term clinical outcomes. ### 2.3 Real-World Functional Data: Genetically Homogeneous Cohorts Two instructive real-world cohorts have been published: | Cohort | Variant | N | Key Functional Finding | Citation | |---------|---------|---|------------------------|----------| | Croatia (homogeneous) | SOD1 p.Leu145Phe | 8 | Median ALSFRS-R slope -0.28 pts/month; 2 stable, 6 gradual decline. Patterns consistent with known slow natural history of this variant. | Bilić et al. 2026 (PMID 41821425) | | Iceland (homogeneous) | SOD1 p.Gly94Ser | 4 | All achieved "nonprogressive chronic ALS" on monthly tofersen | Iceland case series 2025 (PMID 41670738) | These observations highlight a central interpretive challenge: **slow-progressing founder variants (e.g., p.Leu145Phe) may confound functional outcome assessment**, because their natural history already extends well beyond typical ALS survival. In the Croatian cohort, the authors explicitly note that trajectories "reflected the mutation's slow-progressing phenotype" and cautioned against inferring treatment efficacy. The Icelandic cohort's dramatic phenotype (nonprogressive chronic ALS) is intriguing but based on a very small sample with the same founder variant. ### 2.4 Real-World Safety and Access A systematic review/meta-analysis on tofersen safety is in press (Tarazona-Santabalbina et al., PMID 41746412). The Croatian and Icelandic cohorts both report tofersen as well tolerated, with no serious treatment-related adverse events. The Dutch orphan-drug access protocol (PMID 41974046) documents practical experience in rapid, controlled access for SOD1-ALS patients. ### 2.5 Synthesis: The Functional-Outcome Picture Is Still Incomplete - **No randomized, placebo-controlled functional outcome data** have been published post-approval that would satisfy the traditional confirmatory-evidentiary standard. - Real-world series are small, single-center, and dominated by specific founder variants with atypical natural history. - Surrogate biomarkers (NfL, GPNMB) continue to track pharmacodynamic activity robustly, but the magnitude of NfL reduction required for clinical benefit, and the time lag before functional effects become detectable, remain undefined. - The Icelandic experience raises the tantalizing possibility that **very early, presymptomatic or minimally symptomatic intervention** could fundamentally alter disease course—a hypothesis currently untested in controlled trials. --- ## 3. Translation of Surrogate Endpoint Approach to Non-SOD1 ALS Subtypes (b) ### 3.1 The SOD1 Seeding Paradox: Relevance to Sporadic ALS A 2026 study demonstrated SOD1 seeding activity in antemortem CSF from sporadic ALS patients—i.e., individuals without SOD1 mutations (PMID 41929296). This finding implies that wild-type SOD1 can adopt aggregate-prone conformations in the broader ALS population and that SOD1 seeding activity correlates with ALS progression (ALSFRS-R decline and CSF NfL). Conceptually, this opens the door to using SOD1-targeting strategies beyond the strict SOD1-mutation-positive population, although tofersen itself is allele-specific and would not lower wild-type SOD1 mRNA. ### 3.2 What Would It Take to Apply the NfL-Surrogate Model Elsewhere? The FDA's accelerated approval of tofersen rested on a specific evidentiary chain: 1. **Monogenic disease with a single well-defined toxic gain/loss mechanism** 2. **Biomarker (NfL) that is** (i) measurable in blood/CSF, (ii) tied mechanistically to axonal injury, (iii) responsive to therapy, and (iv) correlated with disease progression in natural-history studies 3. **Target engagement biomarker** (CSF SOD1 reduction) confirming on-target activity Applying this template to non-SOD1 subtypes encounters several hurdles: | Subtype | Prevalence (familial) | Molecular Target | ASO/Gene Therapy Development Stage | Surrogate Validity Challenge | |---------|----------------------|----------------|-----------------------------------|------------------------------| | **C9orf72** | ~40% of fALS | Hexanucleotide (GGGGCC) repeat expansion → toxic RNA, dipeptide repeats | Multiple ASOs in preclinical/early clinical | Repeat length heterogeneity; gain-of-function vs. haploinsufficiency debate; no validated blood surrogate with same pedigree as NfL | | **TARDBP (TDP-43)** | ~4% of fALS; TDP-43 in >95% sALS | Nuclear loss-of-function + cytotoxic aggregation | Conceptual; TDP-43 is essential for survival, making knockdown risky | No single blood biomarker captures TDP-43 pathology; NfL is nonspecific | | **FUS** | ~1% of fALS | Cytoplasmic aggregation + loss of nuclear function | Early preclinical | Rare; natural history data extremely limited | | **ATXN2** | Risk modifier (intermediate repeats) | PolyQ expansion enhances TDP-43 toxicity | Preclinical ASO concepts | Not a monogenic ALS cause; modifier status complicates trial design | | **STMN2** | Cryptic exon inclusion downstream of TDP-43 loss | Stathmin-2 depletion → axonal regeneration failure | Preclinical ASO splice correction | TDP-43-dependent; no human clinical data | ### 3.3 C9orf72: The Most Important Non-SOD1 Test Case C9orf72 repeat expansions cause both ALS and frontotemporal dementia (FTD), representing the single largest genetic ALS subtype. Multiple academic and industry groups are developing ASOs targeting C9orf72 repeat-containing RNAs to block toxic RNA foci and dipeptide repeat protein production. However, unlike SOD1-ALS, C9orf72 disease involves: - **Pleiotropic molecular pathology**: sense and antisense RNA foci, five dipeptide repeat species (poly-GA, poly-GR, poly-GP, poly-PA, poly-PR), and possible C9orf72 protein haploinsufficiency affecting autophagy. - **Repeat-length variability** ranging from <100 to >2,000 repeats, potentially modulating phenotype and therapeutic window. - **No plasma biomarker** with the same extensive natural-history-to-trial correlation as NfL. Poly-GP dipeptide repeat proteins in CSF have been explored as target-engagement biomarkers, but their correlation with clinical progression is less well established than NfL. A hypothetical C9orf72-ASO trial could use NfL as a secondary/surrogate endpoint, but regulators would likely require stronger evidence that NfL dynamics in C9orf72 disease mirror those in SOD1 disease. ### 3.4 Sporadic ALS and the NfL Problem In sporadic ALS, TDP-43 proteinopathy dominates. NfL is elevated and prognostic, but it is a downstream measure of axonal degeneration, not a readout of a specific molecular pathway. If an ASO targeting, for example, STMN2 were developed for sporadic ALS, NfL reduction could plausibly serve as a surrogate, but the evidentiary bridge would need to demonstrate that rescuing STMN2 directly reduces axonal injury (measured by NfL) and that this reduction is reasonably likely to predict clinical benefit. This requires substantially more mechanistic validation than was needed for SOD1-ALS, where the direct causal link between mutant SOD1 and motor-neuron death was already well established. --- ## 4. Emerging Targets in Early Development (c) Beyond SOD1, several targets are in preclinical or very early clinical stages where antisense, RNA-targeting, or gene-therapy strategies are being explored: ### 4.1 C9orf72 - **Approaches**: ASOs targeting sense/antisense repeat transcripts (Ionis, Wave Life Sciences, and academic labs), small molecules blocking RAN translation. - **Stage**: Preclinical to early-phase clinical. - **Key biologic challenge**: Balancing suppression of toxic RNA/dipeptides against preserving residual C9orf72 protein function in autophagy. ### 4.2 STMN2 (Stathmin-2) - **Approach**: ASO-mediated splice correction to prevent TDP-43–dependent cryptic exon inclusion, restoring full-length STMN2 required for axonal regeneration. - **Stage**: Preclinical. - **Relevance**: STMN2 loss is a downstream consequence of TDP-43 pathology, making it a potentially broad target for sporadic (and C9orf72) ALS rather than a genotype-specific therapy. - **Surrogate considerations**: NfL could serve as an axonal-injury readout, but target-engagement biomarkers (splice correction in CSF) would need to be developed. ### 4.3 ATXN2 - **Approach**: ASO lowering of ataxin-2 protein, which modifies TDP-43 toxicity when present in intermediate polyQ expansions. - **Stage**: Preclinical. - **Relevance**: Not a monogenic ALS cause; instead a genetic risk modifier. Trial design would require stratification by ATXN2 repeat length, complicating the straightforward monogenic model used for tofersen. ### 4.4 FUS - **Approach**: ASOs or gene therapies targeting mutant FUS; challenging because FUS is essential for DNA repair and RNA processing. - **Stage**: Conceptual/early preclinical. - **Relevance**: Very rare subtype; limited natural history data. ### 4.5 TARDBP (TDP-43) – The "Holy Grail" - **Approach**: Direct TDP-43 lowering is considered risky because TDP-43 autoregulates its own expression and is essential for neuronal survival. Alternative strategies include enhancing nuclear import or blocking pathological phosphorylation/aggregation. - **Stage**: Preclinical target identification. - **Relevance**: Would address >95% of ALS if successful, but the biology is far more complex than SOD1. --- ## 5. Gaps That Must Close Before Generalizing the Surrogate-Endpoint Approach We identify **five critical gaps** standing between the SOD1-tofersen precedent and a broadly generalizable surrogate-endpoint framework for ALS: ### Gap 1: Establishing That NfL Reduction Is Clinically Meaningful Across Genotypes In SOD1-ALS, NfL reduction was accepted as reasonably likely to predict clinical benefit based on natural-history correlations and the known centrality of SOD1 mutation to disease pathogenesis. For C9orf72, TARDBP, FUS, and sporadic ALS, the same evidentiary standard has not been met. It is biologically plausible that NfL lowering reflects reduced axonal injury regardless of upstream cause, but **regulators and the field need genotype-specific natural-history studies correlating baseline NfL, on-treatment NfL trajectories, and long-term functional outcomes**. Without these, each non-SOD1 application of NfL as a surrogate restarts the evidentiary argument from first principles. ### Gap 2: Absence of Target-Engagement Biomarkers for Most Non-SOD1 Targets Tofersen's regulatory package included two parallel biomarker signals: CSF SOD1 (proving target engagement) and plasma/CSF NfL (proving downstream benefit). For C9orf72, candidate target-engagement markers (e.g., poly-GP in CSF) exist but have weaker clinical-correlative validation. For STMN2, splice-correction biomarkers would need de novo development. For TDP-43, there is no validated CSF target-engagement marker at all. **A surrogate-endpoint framework without a target-engagement counterpart is substantially weaker**, because it cannot distinguish pharmacologically active drug from placebo-level noise. ### Gap 3: Unknown Therapeutic Window and Heterogeneity of Genetic Subtypes The Croatian real-world cohort (PMID 41821425) illustrates that SOD1-ALS itself is not homogeneous: the p.Leu145Phe variant progresses slowly, making functional outcomes ambiguous over typical trial durations. C9orf72 repeat lengths vary enormously, and phenotype can range from pure ALS to mixed ALS-FTD to pure FTD. FUS mutations cause juvenile, rapidly progressive disease in some families. **A one-size-fits-all surrogate threshold (e.g., "NfL must fall by X%") is unlikely to be valid across this heterogeneity.** Tailored surrogate thresholds, or composite endpoints combining NfL with genotype-specific markers, may be needed—and these do not yet exist. ### Gap 4: Delivery and Distribution Uncertainties for CNS-Targeted ASOs Tofersen is delivered via intrathecal injection every 4 weeks, achieving broad CSF distribution but uncertain deep parenchymal penetration. For non-SOD1 targets—particularly those affecting cortical motor neurons more diffusely (e.g., C9orf72, sporadic TDP-43)—the same delivery modality may be insufficient. **Gene therapies delivered via intrathecal or intraparenchymal AAV** (e.g., SOD1-targeting AAV5-miRNA approaches in development) could in principle achieve more durable target suppression but carry unique immunogenicity, dosing, and distribution uncertainties. The surrogate-endpoint framework assumes reliable target engagement; if delivery is inconsistent, biomarker signals become uninterpretable. ### Gap 5: Regulatory Precedent vs. Scientific Rigor The FDA's accelerated approval of tofersen was enabled by an exceptional circumstance: a monogenic disease with a single causal mutation, a druggable mRNA target, a well-understood biomarker (NfL), and a uniformly fatal prognosis with no alternatives. Each deviation from this ideal erodes the applicability of the precedent. For sporadic ALS, where etiology is multifactorial and TDP-43 pathology may be irreversible by the time NfL becomes elevated, regulators may demand **higher evidentiary hurdles before accepting NfL reduction alone as a surrogate**. The field needs formal qualifying studies—ideally conducted by consortia such as the Critical Path for ALS (CP-ALS) or NEALS biomarker efforts—to establish NfL (or composite biomarker panels) as a drug-development tool across ALS subtypes. --- ## 6. Concluding Perspective Tofersen's approval represents a landmark in ALS therapeutics, but it is best understood as **a successful N-of-1 proof of principle** rather than a template ready for immediate replication across the disease spectrum. Post-approval data have reinforced the safety and pharmacodynamic consistency of SOD1-lowering ASOs while exposing the difficulty of interpreting functional outcomes in slow-progressing variants. Translation to C9orf72, the next most logical genetic target, is plausible but blocked by the lack of a comparably mature target-engagement biomarker and by the molecular complexity of repeat-expansion disease. For sporadic ALS, the surrogate-endpoint path is conceptually viable (axonal injury is common to all subtypes, and NfL tracks it), but the evidentiary bridge from biomarker to clinical benefit is longer and more fragile than it was for SOD1. Closing the five gaps above—with particular emphasis on genotype-specific natural-history biomarker studies, target-engagement marker development, and formal qualification of NfL (or next-generation composite panels) as a drug-development tool—will determine whether the tofersen precedent opens the door to a broader platform of precision ASO and gene therapies for ALS, or whether it remains an isolated milestone. --- ## References 1. Guise AJ, et al. Identification of tofersen PD-response biomarkers in VALOR clinical trial CSF via multiplexed quantitative proteomics. *Cell Reports Medicine*. 2026. doi:10.1016/j.xcrm.2026.102648 (PMID 41850233) 2. Bilić H, et al. Tofersen treatment in SOD1 p.Leu145Phe ALS: real-world outcomes in a genetically homogeneous Croatian cohort. *Amyotroph Lateral Scler Frontotemporal Degener*. 2026. doi:10.1080/21678421.2026.2638589 (PMID 41821425) 3. Iceland case series. Treating SOD1-ALS with tofersen results in nonprogressive chronic ALS. 2025. (PMID 41670738) 4. Sebogo MA, et al. Longitudinal Analysis of Superoxide Dismutase 1 Seeding Activity in Amyotrophic Lateral Sclerosis Cerebrospinal Fluid. *medRxiv*. 2026. doi:10.64898/2026.03.20.26348753 (PMID 41929296) 5. Tarazona-Santabalbina FJ, et al. Safety profile of tofersen in amyotrophic lateral sclerosis: a systematic review and meta-analysis. *Neurol Sci*. 2026. doi:10.1007/s10072-025-08793-5 (PMID 41746412) 6. Willemse SW, et al. Fast and controlled access to tofersen for SOD1-related ALS in The Netherlands. *Amyotroph Lateral Scler Frontotemporal Degener*. 2026. doi:10.1080/21678421.2026.2655738 (PMID 41974046) 7. Yesbek Kaymaz A, et al. Gene Targeted Therapies for Neurodegenerative Disorders: Strategies and Implications in ALS and SMA. *Genes*. 2026. doi:10.3390/genes17040419 (PMID 42074537) 8. Kurdi MA, et al. Amyotrophic Lateral Sclerosis (ALS) Genetics and Microbiota: A Comprehensive Review. *Int J Mol Sci*. 2026. doi:10.3390/ijms27041978 (PMID 41752118)