Peripheral frataxin levels govern long-term clinical progression in Friedreich ataxia

Abstract

Background

Novel therapeutics for Friedreich ataxia employ diverse strategies to increase frataxin protein levels, and a better understanding of the relation to clinical outcomes could strengthen their use as pharmacodynamic markers, and potentially as surrogate endpoint in therapeutic development. An elaborate modelling framework was developed to evaluate the suitability of frataxin as a biomarker across assays, tissues and disease stages.

Methods

Frataxin levels generated previously through two distinct assay platforms and from two separate clinical cohorts: whole blood frataxin was measured by a lateral-flow immunoassay (LF cohort), and a triple-quadrupole LC-MS/MS method (TQ cohort), which enables separate quantification of mature frataxin (FXN-M) and erythrocyte-specific frataxin (FXN-E). Results were compared descriptively with control and heterozygous carriers, and several distinct modelling strategies were employed to correlate them with clinical function.

Results

Both cohorts represented the relevant disease spectrum, with minor differences in both genetic and clinical severity, which correlated with frataxin levels. Heterozygous carriers showed intermediate levels. Modelling confirmed the predictive value of frataxin across multiple clinical assessments, such as age of symptom onset, age at loss of ambulation and long-term progression. GAA1, the shorter repeat expansion, was confirmed as the dominant predictor of frataxin itself, and, in most situations, clinical function.

Discussion and conclusion

Although isoform biology and tissue-specific expression remain important considerations, peripheral frataxin quantification provides biologically grounded measure of the pathophysiology and disease progression, with strong potential for application in therapeutic trials. Frataxin is a valid clinical biomarker, and our findings support advancing its candidacy as a surrogate endpoint in Friedreich ataxia.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Emerging therapies in Friedreich ataxia aim to restore frataxin deficiency, the central pathogenic mechanism of the disease. Small studies suggest that peripheral frataxin levels reflect disease severity, but their prognostic relevance across assays, tissues and disease stages has not been systematically established.

WHAT THIS STUDY ADDS

• Using an integrated prognostic modelling framework across two independent cohorts and distinct biochemical assay platforms, this study shows that peripheral frataxin levels consistently predict key clinical outcomes, including age at onset, age at loss of ambulation and long-term progression measured by clinical rating scales.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• These findings support peripheral frataxin as a clinically meaningful prognostic and pharmacodynamic biomarker, strengthening its candidacy for disease stratification and as a surrogate endpoint in therapeutic development and regulatory decision-making in Friedreich ataxia.

Introduction

Friedreich ataxia (FRDA) is an autosomal recessive neurodegenerative disorder caused by deficiency of frataxin, a mitochondrial protein involved in iron sulphur cluster synthesis.¹,³ Clinically, FRDA manifests as progressive gait ataxia, loss of hand coordination, dysarthria and cardiomyopathy, leading to an average age of death in the late 30s.^{4 5} Approximately 96% of mutant alleles contain an expanded GAA repeat in intron 1 of the FXN gene, leading to transcriptional silencing and reduced frataxin levels; the length of the shorter GAA repeat predicts age of onset and other markers of clinical severity.¹,³ Among the remaining 4% of alleles, most cause reduced or absent frataxin, and some missense variants (eg, G130V) produce structurally abnormal protein at normal levels. Despite these differences, the shared consequence is frataxin deficiency leading to downstream mitochondrial abnormalities as the fundamental disease mechanism.⁶ Frataxin levels in tissue provide a biomarker of clinical severity, as they correlate with GAA repeat length and age of onset, and predict results from clinical measures accounting for disease duration or age.⁷,¹⁸

Several clinical trials have addressed the mitochondrial dysfunction in FRDA, with one agent, the NRF2 activator (omaveloxolone), being approved.^{19 20} However, ideal therapies would address the cause of disorder, deficiency of functional frataxin, through protein replacement, gene therapy, gene editing or epigenetic approaches.²¹,²⁸ The development of such therapies would be markedly aided by the ability to assess frataxin levels even in surrogate tissues, enabling them to serve as a pharmacodynamic biomarker in early trials to establish the appropriate dose and potentially as a surrogate endpoint in efficacy trials. A variety of approaches has been used to measure frataxin, including immunoassays and, more recently, mass spectrometry-based methods.⁷,¹⁸ While such measures predict clinical outcomes to a moderate degree, their value in predicting future progression or change in disease status is unknown. In the present study, we have used serial analysis of clinical measures from the large FACOMS and FACHILD natural history studies in conjunction with expanded cohorts in which frataxin has been measured to assess the ability of frataxin measurement to predict long-term progression in FRDA.²⁹,³⁴

Methods

Frataxin assays

Frataxin levels were generated previously through two distinct assay platforms and from two separate clinical cohorts: in the first (LF cohort), frataxin levels from whole blood samples were measured via a lateral-flow (LF) immunoassay and expressed as per cent of normal based on control samples run on the same plate.⁷ This assay does not differentiate between frataxin isoforms, but consists of at least 80% of erythrocyte-specific frataxin (FXN-E).⁸ The second (TQ cohort) employed a recently developed method based on a triple quadrupole LC-MS/MS platform,¹⁰ a method that allows separate quantification of mature frataxin (FXN-M), present in most cell types, and FXN-E; results are reported in ng/mL. Both cohorts included healthy control and heterozygous GAA repeat expansion carriers. Samples were obtained during enrolment in the natural history cohorts of FACOMS and FACHILD; controls were neurologically normal individuals evaluated at the institutions. Clinical data of the individuals affected by FRDA are now stored in the Friedreich Ataxia Global Clinical Consortium UNIFIED Natural History study (UNIFAI).³⁵

Statistical considerations and methodology

Demographic characteristics were summarised. Patient frataxin levels were compared descriptively with those of controls and heterozygous carriers. To assess the relationship between frataxin and clinical outcomes, we fit multivariable models adjusted for sex, age, age at symptom onset (AOO), and both GAA repeat lengths (GAA1, GAA2). Backward elimination yielded parsimonious models, balancing explanatory value (R²) with model fit (Akaike Information Criterion, AIC; Bayesian Information Criterion, BIC) and preserving comparability across the three data sets. Model-specific specifications are described below. We hypothesised that frataxin predicts long-term progression in FRDA without a prespecified model hierarchy. We emphasise estimation and report effect sizes with 95% CIs and, where applicable, means with SEs. All p-values are two-sided, nominal, unadjusted for multiplicity and interpreted descriptively.

After cross-sectional, multivariate modelling, long-term clinical progression was evaluated using two distinct analyses. First, age at loss of ambulation (LoA) was assessed with Cox proportional hazards models, based on item E7 of the Modified Friedreich Ataxia Rating Scale (mFARS; score of 5=non-ambulatory). For time-to-event regression analyses, we report the Cox-Snell pseudo-R², derived from the log-likelihood ratio between the fitted and null models. Unlike linear regression, Cox-Snell pseudo-R² reflects a relative measure of model fit and was interpreted alongside information criteria (AIC and BIC) and likelihood ratio tests. Individuals who had not reached LoA at their last visit were censored at that time, while those who were already non-ambulant at baseline were excluded. This introduces truncation bias, rendering the resulting LoA estimates not directly interpretable.³³ The enrolment age was omitted from the stepwise procedure, as it is not a meaningful predictor in this context.

Second, the effect of frataxin levels on progression was evaluated with random-coefficient regression models, implemented as mixed models for repeated measures, using long-term slopes of mFARS and its axial subscore, the Upright Stability Score (USS). Analyses were restricted to ambulant participants in the decline phase (baseline mFARS 10–50; USS 5–30) to mirror typical clinical trial inclusion criteria and avoid ceiling effects (online supplemental Figure 1). Slope effects were estimated through predictor–time interactions, with baseline differences captured by non-interaction covariates. To limit collinearity, predictors (frataxin, AOO and shorter GAA repeat length (GAA1)) were first assessed individually; baseline covariates were then added to the frataxin model to adjust for intercept differences and reduce residual variance. Model performance was summarised using marginal and conditional R² from mixed-effects models,³⁶ reflecting variance explained by fixed effects alone and by fixed plus random effects, respectively.

Results

In both the LF and TQ cohorts, individuals were excluded from further analysis if they were heterozygous for the GAA repeat expansion (n=24 and 16, respectively) or if no clinical follow-up data were available (n=26 and 18). The resulting frataxin data sets had 391 affected individuals in the LF cohort and 245 in the TQ cohort (online supplemental Table 1). The two cohorts differed in their genetic severity, both by average and by distribution. This is best expressed by differential proportions of individuals by onset group: the LF cohort had more individuals with intermediate and late-onset, while the TQ cohort was more skewed towards early and typical FRDA onset (p=0.025). This was also expressed through the lower GAA1 value (655 vs 690, p=0.045) and a higher age of onset (median 10 years vs 11 years, p=0.008). With regard to their longitudinal status, the LF cohort was older (19.8 years vs 15.8 years, p<0.001), and had more advanced disease (mFARS 46.7 vs 39.5, p<0.001) with fewer ambulant individuals (74.9% vs 87.8%, p<0.001). The number of follow-up visits was similar (p=0.167), though follow-up time was slightly longer in the LF cohort (10 years vs 8 years, p=0.003). Despite these differences, both cohorts covered the full spectrum of FRDA severities and follow-up times in UNIFAI.

Frataxin levels differed markedly by onset group (table 1), with early (AOO<8 years) onset showing only 9.9 to 19% of control values across assays. Levels rise progressively with onset, reaching up to 50% in late onset FRDA (AOO>25 years), while carriers show values intermediate between later onset and controls (boxplot: online supplemental Figure 2). This confirms that residual frataxin is not a strong discriminator between those with later onset of symptoms and heterozygous individuals but strongly correlates with genetic disease severity.

Table 1. Frataxin results by onset and control groups.

Cohort, frataxin type	Onset and control groups	N	Frataxin level (ng/mL)	Percentage of control (%)	Multiple of 0–7 years group
LF cohort, frataxin	0–7 years	108		13.6
	8–14 years	171		21.6	1.6
	15–24 years	97		34.6	2.6
	>24 years	43		49.5	3.7
	Carrier	170		64.3	4.7
	Control	109			7.4
TQ cohort, frataxin E	0–7 years	84	2.9	9.9
	8–14 years	117	4.3	14.6	1.5
	15–24 years	51	7.1	24.2	2.4
	>24 years	12	7.4	25.2	2.6
	Carrier	207	15.7	53.3	5.4
	Control	215	29.4		10.1
TQ cohort, frataxin M	0–7 years	84	2.4	19.9
	8–14 years	117	3.5	28.7	1.4
	15–24 years	51	5.1	42.0	2.1
	>24 years	12	5.9	48.5	2.4
	Carrier	207	9.5	77.6	3.9
	Control	215	12.2		5.0

E, erythrocyte; LF, lateral-flow; M, mature; TQ, triple-quadrupole.

Multivariate modelling

In the multiple linear regression analyses, frataxin levels were modelled as a function of GAA1 and GAA2 repeat lengths (table 2). Model fits were comparable, though slightly weaker than previously reported,¹⁰ and age at sampling was not informative.

Table 2. Frataxin levels predicted through multiple linear regression analysis.

Cohort	TQ, FXN-E (n=347)	P value	TQ, FXN-M (n=347)	P value	LF, FXN-M (n=581)	P value
Model (adj.R², AIC, BIC)	0.28 (111, 130)	<0.0001	0.25 (91, 110)	<0.0001	0.37 (197, 219)	<0.0001
AOO	0.007 (0.001, 0.013)(2.0–43.0))	0.0206	0.011 (0.005, 0.017)(2.0–43.0))	0.0001	0.010 (0.006, 0.014)(1.0–63.0))	<0.0001
GAA1	−0.056 (−0.075, −0.036)(55 – 1275)	<0.0001	−0.027 (−0.046, −0.008)(55 - 1275)	0.0055	−0.053 (−0.068, −0.038)(41 – 1150)	<0.0001
GAA2	−0.016 (−0.030, −0.001)(100 – 1800)	0.0338	−0.028 (−0.043, −0.014)(100 – 1800)	0.0001	−0.020 (−0.032, −0.009)(165 – 1500)	0.0007

Values show estimates with 95% CIs and range. Shaded by significance levels: p<0.001 (dark green), p<0.01 (medium green), p<0.05 (light green).

AIC, Akaike Information Criterion; AOO, age at symptom onset; BIC, Bayesian Information Criterion; FXN-E, erythrocyte-specific frataxin; FXN-M, mature frataxin; LF, lateral-flow; TQ, triple-quadrupole.

In parallel, AOO was most strongly explained by frataxin levels and independently by GAA1, with R² values of 0.44–0.55 (table 3). Neither sex nor GAA2 contributed independent explanatory value.

Table 3. Prediction of age of onset through multiple linear regression analysis.

Cohort	TQ, FXN−E (n=347)	P value	TQ, FXN−M (n=347)	P value	LF, FXN−M (n=583)	P value
Model (adj.R², AIC, BIC)	0.44 (2120, 2135)	<0.0001	0.45 (2112, 2128)	<0.0001	0.55 (3741, 3759)	<0.0001
Frataxin	2.105 (0.207, 4.003)	0.0299	3.381 (1.499, 5.263)	0.0005	4.682 (3.034, 6.331)	<0.0001
GAA1	−1.867 (−2.153, −1.582)	<0.0001	−1.818 (−2.087, −1.548)	<0.0001	−2.394 (−2.652, −2.135)	<0.0001

Values show estimates with 95% CIs. Shaded by significance levels: p<0.001 (dark green), p<0.01 (medium green), p<0.05 (light green).

AIC, Akaike Information Criterion; AOO, age at symptom onset; BIC, Bayesian Information Criterion; FXN−E, erythrocyte−specific frataxin; FXN−M, mature frataxin; LF, lateral−flow ; TQ, triple−quadrupole.

At the functional level, we examined cross-sectional rating scale results. Baseline mFARS and USS were modelled, with the latter restricted to ambulant individuals (table 4). R² values of ~0.57 for mFARS and ~0.31 for USS were obtained, consistent across both frataxin assays. Stepwise procedures eliminated sex and GAA2, while both age and AOO provided independent contributions; removing substantially reduced R². Frataxin levels, though associated with severity, consistently emerged as a weaker predictor compared with GAA1 and AOO.

Table 4. Prediction of mFARS and Upright Stability Scores.

	Cohort	TQ, FXN−E (n=250)	P value	TQ, FXN−M (n=250)	P value	LF, FXN−M (n=405)	P value
	Model (adj.R², AIC, BIC)	0.56 (1877, 1898)	<0.0001	0.57 (1871, 1892)	<0.0001	0.57 (3141, 3165)	<0.0001
mFARS	Frataxin	−6.258 (−10.819, −1.698)	0.0074	−8.384 (−12.919, −3.849)	0.0003	−7.707 (−11.715, −3.699)	0.0002
mFARS	Age	1.456 (1.280, 1.632)	<0.0001	1.429 (1.255, 1.603)	<0.0001	1.349 (1.223, 1.475)	<0.0001
mFARS	AOO	−1.537 (−1.850, −1.223)	<0.0001	−1.477 (−1.789, −1.165)	<0.0001	−1.212 (−1.442, −0.981)	<0.0001
mFARS	GAA1	2.315 (1.499, 3.131)	<0.0001	2.345 (1.568, 3.121)	<0.0001	2.917 (2.159, 3.674)	<0.0001
	Cohort	TQ, FXN−E (n=220)	P value	TQ, FXN−M (n=220)	P value	LF, FXN−M (n=305)	P value
	Model (adj.R², AIC, BIC)	0.31 (1360, 1380)	<0.0001	0.31 (1359, 1379)	<0.0001	0.32 1926, 1948	<0.0001
USS	Frataxin	−1.925 (−4.558, 0.707)	0.1509	−2.248 (−4.945, 0.450)	0.1020	−3.586 (−5.820, −1.352)	0.0017
USS	Age	0.543 (0.398, 0.687)	<0.0001	0.546 (0.402, 0.691)	<0.0001	0.527 (0.429, 0.624)	<0.0001
USS	AOO	−0.520 (−0.723, −0.317)	<0.0001	−0.512 (−0.715, −0.309)	<0.0001	−0.464 (−0.607, −0.322)	<0.0001
USS	GAA1	1.365 (0.914, 1.817)	<0.0001	1.414 (0.985, 1.842)	<0.0001	1.212 (0.776, 1.649)	<0.0001

Values show estimates with 95% CIs. Shaded by significance levels: p<0.001 (dark green), p<0.01 (medium green), p<0.05 (light green).

AIC, Akaike Information Criterion; AOO, age at symptom onset; BIC, Bayesian Information Criterion; FXN−E, erythrocyte−specific frataxin ; FXN−M, mature frataxin; LF, lateral−flow ; mFARS, Modified Friedreich Ataxia Rating Scale; TQ, triple−quadrupole; USS, Upright Stability Score.

Overall, multivariate modelling confirmed that repeat lengths strongly predict frataxin levels, while AOO adds only minor impact. This pattern extends to functional predictions: GAA1 remains the dominant predictor across all models, with no contribution from sex in any model.

Time to event analysis: loss of ambulation

Cox proportional hazards models were used to evaluate predictors of age at LoA (table 5). Neither sex nor GAA2 contributed meaningfully to model fits. Among the remaining predictors, stepwise procedures identified the combination of frataxin and AOO as the strongest explanatory set. This analysis was unique in that GAA1 emerged as the weakest predictor among the three covariates, in contrast to cross-sectional models where GAA1 had been consistently robust.

Table 5. Age at loss of ambulation: Cox models, including Snellen partial R².

Cohort	TQ, FXN-E (n=189)	P value	TQ, FXN-M (n=189)	P value	LF, FXN-M (n=346)	P value
Model (Snell R², AIC, BIC)	0.63 (1527, 1534)	<0.0001	0.66 (1493, 1500)	<0.0001	0.62 (3192, 3200)	<0.0001
Frataxin	0.364 (0.216, 0.616)	0.0002	0.147 (0.087, 0.247)	<0.0001	0.210 (0.154, 0.286)	<0.0001
AOO	0.744 (0.708, 0.781)	<0.0001	0.741 (0.707, 0.778)	<0.0001	0.868 (0.851, 0.885)	<0.0001

Values show HRs with 95% CIs. Shaded by significance levels: p<0.001 (dark green), p<0.01 (medium green), p<0.05 (light green).

AIC, Akaike Information Criterion; AOO, age at symptom onset; BIC, Bayesian Information Criterion; FXN-E, erythrocyte-specific frataxin ; FXN-M, mature frataxin; LF, lateral-flow ; TQ, triple-quadrupole.

Random coefficient regression models of long-term clinical progression

We first tested null models with only time as a fixed factor, yielding marginal R² values of 0.19/0.24 for mFARS and 0.41/0.44 for USS (online supplemental Table 2). Higher values for USS suggest a cleaner, more linear slope with more linear change and fewer floor/ceiling effects. Notably, this difference disappeared when baseline values were added as covariates, as baseline scores strongly predict future values and therefore absorb most of the explained variance. Next, slope interactions were assessed separately for frataxin, GAA1 and AOO, with fixed predictors capturing baseline differences. This only modestly improved model fit: R² increased from 0.19/0.24 to 0.21/0.29 (mFARS/frataxin) and from 0.41/0.44 to 0.43/0.48 (USS/frataxin). Similar improvements were seen for GAA1 and AOO. Predictor quality was consistent: GAA1 >AOO ≈ frataxin (online supplemental Tables 3-5).

The following stepwise procedure eliminated baseline fixed factors GAA2 as well as sex and age, while AOO added only marginally to model fits (online supplemental Table 6). The final model to best express the predictive effect of frataxin on progression included only baseline factors frataxin and GAA1, in addition to frataxin*time interaction (table 6).

Table 6. Frataxin levels predict the speed of progression in mFARS and Upright Stability Scores.

Variable		Tq, FXN−E		Tq, FXN−M		Lf, FXN−M
		Estimate (95% CI)	P value	Estimate (95% CI)	P value	Estimate (95% CI)	P value
	Cohort	subjects (slopes) 203		subjects (slopes) 203		subjects (slopes) 271
	Model (R²m/R²c; AIC, BIC)	0.26/0.89 (8084, 8130)		0.27/0.89 (8062, 8108)		0.30/0.87 (11657, 11 707)
mFARS	Frataxin	−2.35 (−7.28, 2.58)	0.3492	−4.96 (−9.91, −0.00)	0.0499	−6.66 (−10.29, −3.04)	0.0004
mFARS	GAA1	1.31 (0.63, 2.00)	0.0002	1.25 (0.61, 1.88)	0.0001	0.45 (−0.09, 0.99)	0.1025
mFARS	Time	1.82 (1.64, 1.99)	<0.0001	1.80 (1.63, 1.96)	<0.0001	1.80 (1.66, 1.93)	<0.0001
mFARS	Frataxin×time	−0.86 (−1.41, −0.31)	0.0025	−1.56 (−2.10, −1.01)	<0.0001	−0.97 (−1.37, −0.57)	<0.0001
	Cohort	subjects (slopes) 191		subjects (slopes) 191		subjects (slopes) 256
	Model (R²m/R²c; AIC, BIC)	0.48/0.88 (6365, 6411)		0.49/0.88 (6356, 6402)		0.50/0.87 (9315, 9364)
USS	FXN	−1.31 (−3.61, 0.98)	0.2595	−2.69 (−5.04, −0.35)	0.0245	−2.50 (−4.31, −0.69)	0.0069
USS	GAA1	0.75 (0.43, 1.08)	<0.0001	0.70 (0.40, 1.01)	<0.0001	0.50 (0.23, 0.77)	0.0003
USS	Time	1.54 (1.43, 1.65)	<0.0001	1.52 (1.42, 1.63)	<0.0001	1.37 (1.30, 1.44)	<0.0001
USS	Frataxin×time	−0.56 (−0.89, −0.23)	0.0009	−0.73 (−1.06, −0.39)	<0.0001	−0.53 (−0.74, −0.32)	<0.0001

Values show estimates with 95% CIs. Shaded by significance levels: p<0.001 (dark green), p<0.01 (medium green), p<0.05 (light green).

AIC, Akaike Information Criterion; AOO, age at symptom onset; BIC, Bayesian Information Criterion; FXN−E, erythrocyte−specific frataxin; FXN−M, mature frataxin; LF, lateral−flow; mFARS, Modified Friedreich Ataxia Rating Scale; TQ, triple−quadrupole; USS, Upright Stability Score.

Discussion

The present study demonstrates that frataxin levels directly correlate with all major clinical outcomes in FRDA, including not only cross-sectional markers such as AOO and disease severity, but also LoA as a disease milestone and long-term progression slopes of mFARS and USS. The associations were observed consistently across two independent cohorts and two distinct assay platforms, providing robust support for peripheral frataxin level as a clinically meaningful biomarker in both natural history studies and interventional trials.

The findings align with the established pathophysiology of FRDA, in which expanded GAA repeats in the FXN gene lead to transcriptional silencing of frataxin, reduced tissue levels of the protein and the resulting clinical and biochemical features of the disorder.¹,³ In this context, the observed correlations are not surprising (or even new), but their consistent replication across cohorts, assay platforms and clinical outcomes reinforces the coherence of frataxin as disease-relevant biomarker. While the shorter GAA repeat length in the FXN gene remains the primary determinant of genetic severity, it has important limitations as a biomarker. Repeat length is largely genetically fixed, usually unresponsive to therapy, and often confounded by mosaicism and interruptions of unclear significance.¹,^{337 38} Frataxin levels, on the other hand, are dynamic, vary across tissues and may be directly modulated by therapeutic interventions. This distinction underscores the unique potential of frataxin as a pharmacodynamic marker of treatment response, offering advantages over static genetic predictors in contexts where frataxin restoration is targeted.

At the same time, frataxin, GAA1 and AOO are highly intercorrelated. Their apparent contributions in regression models largely reflect statistical partitioning of shared variance rather than independent biological effects. Each provides a complementary perspective on disease severity—genetic, biochemical or clinical—but the degree of overlap limits the ability to disentangle distinct mechanistic pathways using current datasets. As a result, explanatory power gains are incremental when adding predictors together. For now, their greatest utility lies in providing parallel stratification handles that reflect disease biology from different vantage points, rather than in serving as wholly independent predictors.

Assay methodology plays a critical role in how frataxin levels are measured and interpreted, but results were consistent across cohorts and assay techniques. The more recently developed mass spectrometric method (TQ) achieved comparable statistical significance with far fewer subjects, suggesting superior analytical precision. Nevertheless, methodological factors such as tissue isolation remain critical in determinants of assay fidelity.³⁹ The TQ assay quantifies two frataxin isoforms (E and M), whereas the LF assay does not distinguish isoforms and is dominated by FXN-E. Consequently, whole blood results in LF assays are broadly analogous to FXN-E. However, whole blood heterogeneity introduces substantial variability, particularly when values are normalised per blood volume: cell counts alone contribute ~25% variability for erythrocytes, ~33% for individual leucocyte subtypes and up to 50% for platelets, independent of assay performance.¹⁰ While FXN-M detected in whole blood by LC-MS/MS is largely platelet-derived, normalisation beyond erythrocytes remains challenging. Prior work showed that normalisation to haemoglobin—an erythrocyte-specific marker—improves consistency for both FXN-E and FXN-M by mitigating volume-related variability, whereas no stable, disease-independent normalisation marker is available for leukocytes or platelets.¹⁰ Accordingly, apparent separation between heterozygotes and late-onset FRDA assay precision, biological overlap and normalisation strategy underscores that absolute peripheral frataxin thresholds are poorly suited for therapeutic decision-making; within-individual change relative to baseline is likely the more robust readout in interventional trials. Isoforms E and M show similar statistical associations, and the slightly stronger associations in the LF cohort are explained by its larger sample size.

The biology of frataxin isoforms further complicates interpretation. Frataxin E is not directly involved in disease pathogenesis and is not produced in large amounts in affected tissues, such as neurons or the myocardium. Nevertheless, FXN-E levels strongly correlate with those of FXN-M, as both are tightly controlled by GAA repeat lengths. A paradoxical finding is that retention of FXN-E in patients with point mutations near the N-terminus is associated with more severe phenotypes,^{8 9} likely reflecting the uncoupling of FXN-E stability from the functional levels of mature frataxin. FXN-E, a soluble protein found almost exclusively in red blood cells, is readily extracted, whereas FXN-M is intramitochondrial and requires more complex extraction procedures. Because of its erythrocytic origin, one also should consider whether variation in haemoglobin or other haematologic parameters contributes to observed differences and adjust analyses accordingly. This may well lead to variability in assessment and then less significant statistical associations. Consequently, careful consideration of tissue source and extraction methodology is essential when interpreting frataxin as a biomarker across studies. Importantly, available evidence indicates that peripheral frataxin levels in untreated individuals are largely stable over time,¹⁰ with variability driven mainly by genetic severity rather than disease duration. Therefore, our analyses reflect primarily between-individual differences rather than longitudinal within-individual change. To confirm temporal stability across tissues and disease stages, further longitudinal sampling will be required.

In clinical trials in particular, understanding the mode of frataxin restoration will be crucial. Protein replacement, epigenetic activation or gene-based delivery is designed to normalise cellular frataxin, but a rise in peripheral tissue does not necessarily reflect restoration in the nervous system, where disease impact is greatest. Moreover, postnatal change in frataxin levels may not mirror the effects of the frataxin level an individual was born with. Consequently, measurement of frataxin restoration must consider not only the overall level (as measured by the assays described here), but also cellular and tissue distribution to understand whether sufficient restoration has been achieved where it matters most. Still, the present data show that peripheral tissue frataxin levels provide a marker of clinical status in FRDA, particularly when interpreted in conjunction with immunohistochemistry or in situ hybridisation that assesses cellular distribution. Using measurements of frataxin levels in therapeutic trials should therefore provide a meaningful, though not complete surrogate marker of treatment efficacy.

Current frataxin assays have limitations such as technical challenges with tissue-specific quantification, effects of somatic expansion or repeat interruptions, and uncertain correspondence between peripheral and target-tissue frataxin levels. Yet, our integration of genetic, biochemical and clinical data establishes a compelling framework for frataxin as a biomarker. Peripheral frataxin levels align with core aspects of FRDA pathophysiology, correlate with genetic severity and track disease progression by governing the rate of functional decline over time, while (like GAA repeats) remaining unchanged over time.⁴⁰ They offer a dynamic pharmacodynamic indicator of therapeutic response. With assay refinement, deeper understanding of isoform biology and cross-tissue validation, frataxin quantification could become a foundational marker in FRDA research and clinical care. Crucially, these findings bring the field a meaningful step closer to positioning frataxin as a surrogate endpoint under the FDA’s accelerated approval pathway.

Data availability statement

Data are available upon reasonable request.