The mitochondrial disease biomarker GDF15 is dynamic, quantifiable in saliva, and correlates with disease severity

Abstract

Circulating growth differentiation factor 15 (GDF15) is a biomarker of mitochondrial diseases and aging, but its natural dynamics and response to acute stress in blood and other biofluids have not been well defined. Using extensive samples from MiSBIE participants with rare mitochondrial diseases (MitoD), we examined GDF15 biology in 290 plasma and 860 saliva aliquots from 40 subjects with the m.3243A>G mutation (n=25) or with single, large-scale mtDNA deletions (n=15). Compared to healthy controls, both MitoD groups exhibited significantly elevated blood and saliva GDF15 (p<0.0001). To examine the origin of GDF15 protein in saliva, we profiled GDF15 expression in 48 tissues from the GTEx dataset and identified high GDF15 expression in salivary gland secretory cells. Despite being chronically elevated in MitoD, saliva GDF15 further increased in response to experimental laboratory mental stress alone (without physical exertion), whereas the stress-induced plasma GDF15 reactivity was blunted in MitoD compared to controls. Using a home-based saliva collection protocol, we show that similar to other stress-related metabolic hormones, saliva GDF15 is highest upon awakening and declines rapidly by 61.2% within 45 minutes. Elevated saliva GDF15 levels persisted throughout the day in MitoD. Clinically, saliva GDF15 correlated with neurological symptoms, fatigue, and functional capacity. Importantly, stress-evoked changes in GDF15 did not amplify noisy disease severity associations, but rather consistently increased the effects sizes for GDF15-symptoms connections, pointing to converging psychobiology underlying the responses to mitochondrial OxPhos defects and acute mental stress. These results open the door to exploring saliva GDF15 as a non-invasive monitoring approach for mitochondrial diseases and call for further studies examining the psychobiological processes linking mitochondria, mental stress, and GDF15 dynamics.

1. Introduction

Growth differentiation factor 15 (GDF15) is the leading biomarker candidate for primary mitochondrial diseases (MitoD) caused by mitochondrial DNA (mtDNA) pathogenic variants ^1–6. Elevated levels of GDF15 have been observed in children with confirmed mitochondrial disease³, particularly in children with mtDNA single deletions, translation defects such as tRNA mutations, and SURF1 mutations⁷. In adult patients with the 3243A>G mutation, GDF15 is associated with disease severity⁵. A comprehensive meta-analysis involving 845 MitoD patients quantified GDF15’s diagnostic accuracy⁸. Notably, GDF15 has been shown to outperform conventional biomarkers in both sensitivity and specificity, particularly in patients with muscle-manifesting mitochondrial diseases⁹. Based on these data, the Mayo Clinic recommends a diagnostic threshold of 750 pg/mL for plasma GDF15¹⁰; however, there is no universally accepted cutoff. Importantly, GDF15 is also the most elevated circulating protein with aging^11,12, making age an important factor to consider when interpreting the clinical significance and diagnostic thresholds. Beyond MitoD, blood GDF15 has emerged as the leading plasma protein for distinguishing individuals with various current¹³ and future clinical illnesses¹⁴, highlighting its broad induction in various psychopathological conditions. Because circulating GDF15 is one of the strongest prognostic indicators of mortality and disability in clinical populations and older adults^11,15, fully characterizing GDF15 biology and dynamics in MitoD can inform future research and clinical work.

Mechanistically, elevated levels of GDF15 have been shown in various conditions associated with impaired mitochondrial biology. Systemic GDF15 release results from the activation of the integrated stress response (ISR), which is triggered by mtDNA defects that impair energy transformation via oxidative phosphorylation (OxPhos) in mitochondria, leading to intracellular reductive stress (i.e., elevated NADH/NAD⁺ ratio)^16–19. A marked induction of GDF15 has been shown in skeletal muscles of patients with thymidine kinase 2 (TK2) defects, which causes mtDNA depletion/deletions and OxPhos dysfunction²⁰. Plasma GDF15 is elevated in adults with mtDNA pathogenic point variants and single deletions, distinguishing it from other neuromuscular conditions^2,21. Mitochondrial translation defects and polymerase gamma mutator (POLG) deficiency in mouse models also induce GDF15²². These findings underscore the specific relationship between mitochondrial diseases and GDF15, highlighting its potential utility in diagnosis and monitoring.

In preclinical studies of mitochondrial disorders, we previously reported that OxPhos defects and other mitochondrial defects altered transcriptional, metabolic, and endocrine responses to evoked stress²³, pointing to the psychobiological interaction of OxPhos biology and stress responses. In humans, the release of neuroendocrine hormones like glucocorticoids and catecholamines that alter cellular metabolism and immune function can be triggered by psychosocial and mental stress²⁴. As a result, stress responses impose additional metabolic demands²⁵ which may contribute to the burden and embedding of chronic stress²⁶. Prolonged exposure to psychological stress compromises immunity, increases susceptibility to infections²⁷, disrupts neuroendocrine balances²⁸, affecting metabolism²⁹, mood³⁰, and cardiovascular function³¹, among other systems³². A recent survey study found that MitoD patients reported worse symptoms including fatigue, muscle pain, and neurological issues when they experience more stress and emotional difficulties³³, highlighting the potential intersection of stress systems and MitoD pathophysiology. However, there is a need for robust, objective biomarkers of mitochondrial disorders to examine the stress➔disease cascade in MitoD. Understanding whether psychosocial stressors influence the MitoD biomarker GDF15 could help understand how the full spectrum of physiological factors interact with the pathophysiology of mitochondrial diseases.

Although most GDF15 research has focused on circulating blood levels, preliminary evidence suggests that it may also be detectable in human saliva^34,35. Determining if saliva GDF15 is physiologically regulated together with, or independently from, circulating blood levels would expand our understanding of GDF15 biology. It would also establish whether easily accessible saliva can be used to monitor human GDF15 dynamics in health and disease, particularly in MitoD under different clinical situations (morning fasting vs afternoon) and in free-living conditions (at home).

Here, we use repeated-measures data from the Mitochondria Stress, Brain Imaging, and Epigenetics (MiSBIE)³² study to define the dynamic properties of blood and saliva GDF15, including its psychobiological regulation under stress and its natural diurnal pattern, among individuals with rare genetic OxPhos defects causing MitoD.

2. Results

2.1. Saliva GDF15 in health controls and patients mitochondrial diseases

In this study, we collected plasma and saliva simultaneously from 40 patients with mitochondrial diseases aged 20-60 (70% females, Figures 1A, see Supplemental Table 1 for detailed clinical symptoms characteristics), compared to age- and physical activity-matched healthy controls (n=70) who do not have a diagnosis of MitoD³⁶. The MitoD subject population includes three groups: (1) individuals with single, large-scale mtDNA deletions (n=15)³⁷, (2) mtDNA pathogenic variant m.3243A>G carriers who do not present stroke-like episodes (n=20)³⁸, and (3) subjects with m.3243A>G who present stroke-like episodes and more severe overall disease severity ([MELAS]: mitochondrial encephalomyopathy lactic acidosis and stroke-like episodes; n=5)³⁹. Saliva and plasma GDF15 were quantified for up to 26 timepoints per individual, over the course of a 2-day hospital visit and a 3-day at-home saliva protocol in MiSBIE (Figure 1B, see Materials and Methods for details; total 3,226 samples).

Figure 1. Mitochondrial OxPhos defects increase plasma and saliva GDF15 and abolishes age correlation.

(A) MiSBIE control and MitoD groups including the m.3243A>G point mutation without (Mutation) or with MELAS, and single, large-scale mitochondrial DNA deletion (Deletion). (B) MiSBIE experimental design. GDF15 was measured from human plasma and saliva samples over up to 26 timepoints during a 2-day hospital visit and a 3-day home-based saliva. (C) Mean plasma and saliva GDF15 concentrations in MitoD participants (n=40, 70% female) aged 20-60 years old. See Supplemental Figure S1A for log 10 transformed distribution with individual datapoints. (D) Associations between age and plasma (left) or saliva (right) GDF15 levels in controls and MitoD patients. Data from morning fasting timepoints. (E) Change in plasma (left) and saliva (right) GDF15 levels from morning to afternoon for MitoD group patients. Data shown as mean ± SEM in (C). Effect sizes and P-values from (D) Spearman’s rank correlation (E) Wilcoxon-paried t-test. *p<0.05, **p<0.01, **p<0.001, ****p<0.0001.

Compared to the control group (average 404.1 pg/mL, median 378.6 pg/mL, range 148.5-1,411pg/mL), plasma GDF15 levels were 2.5-folder higher in subjects with MitoD (average 1,418 pg/mL, median 1,265 pg/mL, range 130.7-5,295 pg/mL), confirming that plasma GDF15 is elevated in primary OxPhos defects^1–4. Specifically, plasma GDF15 levels were 4.3-fold higher in the mtDNA deletion group (average 2,155 pg/mL, median 2,242 pg/mL, range 583-5,292 pg/mL), 1.5-fold higher in individuals with the m.3243A>G mutation without MELAS (average 1,006 pg/mL, median 1,040 pg/mL, range 130-3,725 pg/mL), and 1.7-fold higher in those with m.3243A>G and MELAS (average 1080 pg/mL, range 685-1,837 pg/mL) (Figure 1C, left, see Figure S1A left for log distribution).

We then tested whether OxPhos defects also increased saliva GDF15. As in blood, compared to healthy controls (average 32.25 /mL, median 17 pg/mL, range 2.02-683.3 pg/mL), saliva GDF15 levels were 1.4-fold higher in patients with MitoD (average 77.8 pg/mL, median 43.8 pg/mL, range 2.0-1,677 pg/mL). For each disease group, saliva GDF15 levels were 1.7-fold higher in deletion subjects (average 87.2 pg/mL, median 42.2 pg/mL, range 2.03-1,677 pg/mL), 0.97-fold higher in the m.3243A>G group (average 63.5 pg/mL, median 38.8 pg/mL, range 2.05-873 pg/mL), and 2.6-fold higher in the MELAS group (average 116.2 pg/mL, median 90.68 pg/mL, range 10.8-330.9 pg/mL) (Figure 1C, right, see Figure S1A right for log distribution).

In contrast to healthy controls where plasma GDF15 correlated with age (r=0.61, p<0.001), there were no associations between MitoD plasma GDF15 and age (Figure 1D, left), suggesting that mitochondrial OxPhos defects disproportionally increase GDF15 beyond the more modest, progressive effect of aging. Saliva GDF15 did not correlate with age in healthy controls nor MitoD (Figure 1D, right). The AM-PM (within day) and Day 1-Day 2 (day-to-day) stability of blood and saliva GDF15 in MitoD is shown in Figures 1E reveal substantial variation. Between Day 1 and Day 2, morning fasting saliva GDF15 in MitoD did not change significantly (Figure S1B).

2.2. Potential origin of saliva GDF15

To investigate the presence and potential origin of GDF15 in human saliva (Figure 2A) and explore its relationship with circulating GDF15 measured in plasma, we first leveraged the Gene Tissue Expression database⁴⁰ to systematically quantify GDF15 gene expression across the human body. Figure 2B shows GDF15 transcript abundance across 48 organs and tissues (GTEx, n=948 individuals, tissues with >20 participants; see Materials and Methods for details), including the minor salivary gland which produces the bulk volume of saliva in the mouth. This analysis revealed that GDF15 is widely expressed among most human tissues: highest in the kidney and other glandular and secretory tissues, and lowest in the brain (where its receptor GFRAL is expressed⁴¹). This anatomical expression pattern is consistent with its systemic extracellular secretion and role as a body-to-brain signaling protein⁴². The minor salivary gland ranked 17 out of 48 tissues based on GDF15 expression (Supplemental Table 2).

Figure 2. GDF15 is expressed in human salivary gland and GDF15 protein is detectable in saliva, where it does not correlate with age.

(A) Study design where GDF15 is concurrently measured in human plasma and saliva. (B) GDF15 gene expression (RNA transcript levels) across human tissues from the GTEx dataset. (C) GDF15 gene expression in single cells from the minor salivary gland, retrieved from the Human Protein Atlas. (D) Immunohistochemistry for GDF15 protein in the minor salivary gland of a 62-year-old male (HPA subject ID 2367). For additional mitochondrial proteins, see Supplemental Figure S1. (E) GDF15 measured in plasma and saliva from healthy controls (N=70, 69% female. All timepoints, plasma: n=571 observations, 59 missing due to failed blood draw; saliva: n=1,476 observations, 315 missing due to low saliva volume, 29 below the minimum detection limit).(F) Correlation between morning fasting plasma and saliva GDF15 levels in healthy controls (n=47, 66% female). GTEx data (B,G,H) excludes cultured cells and tissues with n<20; see Supplemental Table 2 for full dataset used; neither placenta nor isolated immune cell populations such as macrophages were available in the GTEx database (placenta, not available in GTEx, exhibits high GDF15 expression). Single cell sequencing (C) and histological staining (D) data retrieved from Human Protein Atlas; see Supplemental Table 3 for single cell RNA sequencing based GDF15 expression in minor salivary gland. Effect sizes and p-values from (E) Mann-Whitney t-test, (F,H,I) Spearman’s rank correlation. Data below minimum detectable concentration excluded for graphs and analyses (E,F,I). *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

In the salivary gland analyzed by single-cell transcriptomics, GDF15 was most highly expressed in salivary duct cells and mucus glandular cells (Figure 2C, Supplemental Table 3; data retrieved from GTEx, n=24 cell types, see Materials and Methods for details). Histological analysis of GDF15 protein abundance by immunohistochemistry⁴³ confirmed its preferential expression in saliva-producing glandular epithelial cells (Figure 2D), providing a basis for its secretion in human saliva. Histologically, salivary duct cells were also enriched for most of the examined mitochondrial proteins (TOMM20 [membrane-associated], TFAM [mtDNA-associated], NDUFS4 [OxPhos]), MCU [calcium import]), although not all (CPT1A [fatty acid oxidation]) (Figure S2). This provided further confidence that mitochondria-rich duct epithelial cells express high levels of the GDF15 gene and may produce the GDF15 protein detected in human saliva.

In healthy controls from our study, saliva GFD15 levels were, on average, 8.0% of plasma levels (Figure 2E). Interestingly, plasma and saliva GDF15 levels were not correlated in healthy controls (Spearman r=0.034, p=0.82, Figure 2F) or MitoD patients (Figure S1C). This suggested that unlike systemic stress hormones that are correlated between plasma and saliva, such as cortisol⁴⁴, saliva GDF15 may not reflect passive “spillage” of blood into saliva. Saliva GDF15 may therefore be regulated separately from systemic blood levels, calling for a systematic effort to map its association with age and its response to psychobiological stressors.

2.3. Saliva GDF15 does not correlate with age

Consistent with its status as an age-related circulating biomarker^11,12, across hundreds of bulk GTEx tissue transcriptomes from individuals aged 20-70 years, GDF15 expression was positively correlated with age in most human tissues (Figures S3A, Supplemental Table 1). As expected, in healthy controls, circulating plasma GDF15 was also positively correlated with age (Figure S3C left, r=0.61, p<0.0001), increasing on average 101.8pg/mL per decade of life. However, GDF15 expression in some tissues, including the salivary gland, showed no association with age; in fact, the age-to-GDF15 correlation was second lowest and null in the salivary gland (Figure S3B, r=−0.040, p=0.61). Accordingly, in the present study, we observed no significant association between age and saliva GDF15 levels (Figure S3C, right). However, in a separate recent study from our group measuring saliva GDF15 from 198 healthy controls, we did detect a modest but statistically significant correlation between saliva GDF15 and age (r=0.21, p=0.003)⁴⁵, though this association was weaker than that observed in plasma.

Together, i) the selective expression of GDF15 in salivary glandular cells, ii) the lack of correlation between GDF15 in plasma and saliva, and iii) the absence of an age-related pattern in GDF15 gene expression in salivary glands—along with the lack of, or weaker, age associations in saliva GDF15 protein levels—converge to indicate that saliva GDF15 is regulated independently from circulating blood levels.

2.4. Plasma and saliva GDF15 dynamics in OxPhos defects

The presence of the GDF15 protein in saliva and its elevated levels in MitoD offered the opportunity to examine whether the dynamic properties and psychobiological regulation of saliva GDF15 presented previously³⁶ are influenced or preserved among individuals with OxPhos defects (Figures 3A).

Figure 3. Psychological stress elicited differential GDF15 dynamics in patients with mitochondrial OxPhos defects.

(A) Schematic of MiSBIE experimental design measuring psychological response in GDF 15 after a modified socio-evaluative stress task. Blood and saliva samples were collected simultaneously at 8 timepoints. Plasma (B) and saliva (C) GDF15 levels before and after socio-evaluative stress task. For individual trajectories in percent change separated within the MitoD group see Figure S4A–B. Data shown as mean ± SEM. P-values for effect of time from mixed-effect modeling corrected for multiple testing, p-values for interaction effect from two-way mixed-effect modeling corrected for multiple testing. See Supplemental Table 3 for sample size at each timepoint.

Stress-evoked regulation of GDF15 in MitoD.

Compared to controls, plasma GDF15 levels were consistently elevated throughout the speech task session in OxPhos-deficient MitoD individuals (Figure S4A, left). When each participant’s trajectory was assessed relative to their baseline levels, on average, MitoD groups showed a relatively distinct reactivity pattern in response to stress when each participant’s trajectory is projected relative to their baseline levels (Figures 3B). In the deletion group, the already elevated plasma GDF15 did not increase further; it decreased by 3.6% at 30 min. In the m.3243A>G non-MELAS group, plasma GDF15 peaked at 11.6% at 120 min, and peaked 2.1% at 5 min in the MELAS group (Figure S4A, right).

As in blood, saliva GDF15 levels were elevated in most MitoD subjects compared to controls throughout the speech task (Figure S4B, left). On average, MitoD individuals showed a trajectory similar to that of controls (Figure 3C). In MitoD subgroups, acute socio-evaluative stress increased the already elevated saliva GDF15 at the 5 min timepoint by 46% in deletion, 23% in non-MELAS m.3243A>G, and 55% in MELAS. However, the effect was only statistically significant in the more robustly powered non-MELAS m.3243A>G group. In the non-MELAS m.3243A>G group, saliva GDF15 declined from 5 to 120 minutes in a pattern similar to controls, whereas levels in the deletion group remained marginally elevated at 120 minutes (Figure S4B, right).

The GDF15 concentration is already elevated at baseline in MitoD groups, therefore, the stress-induced relative % change from baseline likely underestimates the absolute release of GDF15. Examined as absolute change in concentrations (delta, pg/ml), socio-evaluative stress produced plasma and saliva GDF15 elevations that were either similar or exaggerated by up to 6.7-fold (5min post stress, MELAS group vs control in saliva, p=0.021) in MitoD groups compared to controls (Figures S4C–D).

MitoD patients exhibit a normal awakening response.

In addition to socio-evaluative stress, we also investigated saliva GDF15 awakening response in MitoD (Figure 4A). In healthy controls, saliva GDF15 levels were highest at awakening (Figures 4B) and exhibited a negative awakening response³⁶. Here, we observed that awakening saliva GDF15 levels were markedly higher in all three MitoD groups compared to controls: 2.3-fold higher for deletion (p=0.0041), 1.0-fold higher for m.3243A>G non-MELAS (p=0.0021), 1.6-fold higher in MELAS (p=0.0035) (Figure S5B). All MitoD groups exhibited robust awakening responses, whereby saliva GDF15 levels decreased by 71.5% in deletion, 56.5% in mutation, and 38.8% in MELAS at 45min (p<0.05 for deletion and mutation). In all disease groups, bedtime saliva GDF15 levels were lower compared to the awakening levels, validating the results observed in the control group and suggesting a preserved circadian rhythm of saliva GDF15 in two subgroups of MitoD patients.

Figure 4. Awakening response in patients with mitochondrial diseases.

(A) Schematic of experimental design measuring GDF15 awakening response from human saliva samples, including the classical expected cortisol awakening response that peaks 30-45 minutes after waking up. Saliva was sampled at awakening, 30 and 45 min after awakening, and at bedtime on Monday, Wednesday, and Friday. (B) Saliva GDF15 diurnal variation in mitochondrial diseases at awakening, 30 and 45 min after awakening, and bedtime shown as and group average. See Figure S5A for individual trajectory. Data shown as mean ± SEM. P-values for effect of time from mixed-effect modeling corrected for multiple testing, p-values for interaction effect from two-way mixed-effect modeling corrected for multiple testing. See Supplemental Table 3 for sample size at each timepoint.

2.5. Blood and saliva GDF15 correlate with MitoD disease severity

Sensitivity and specificity to MitoD diagnosis.

Using all plasma and saliva timepoints, we systematically computed effect sizes at each timepoint to compare MitoD subjects and controls (Figures S6A–D). The standardized effect sizes in plasma and saliva GDF15 tended to be lowest across the three MitoD groups when sampled in the morning fasting state – when physiology is most “unperturbed”. In contrast, the stress task moderately increased group differences (Figures 5A and 5B). In the afternoon of Day 2, participants performed cognitively demanding tasks during a 2-hour magnetic resonance imaging (MRI) of the brain, during which saliva was collected at 3 timepoints. Compared to the morning fasting sampling, the control-MitoD differences in saliva GDF15 were largest during the afternoon MRI (Figure 5B), indicating that stressful/demanding challenges may reveal group differences otherwise minimized at rest.

Figure 5. Plasma and post-stress saliva GDF15 classify individuals with mitochondrial OxPhos defects.

(A) Standardized effect sizes (Hedges’ g) for the difference in plasma GDF15 between control and MitoD participants in MiSBIE, using measurements performed at different timepoints: Day 1 morning fasting, Day 2 morning fasting, stress reactivity quantified by area under the curve relative to ground, pre-MRI session, mid-2-hour MRI session, awakening sample collected at home. Bars highlight the higher effect size relative to the morning fasting timepoint for 5 of the 6 group comparisons. Hedges’ g>0.8 is considered a large effect size. Control n=48-69, MitoD n=31-37. (B) Same as (A) for saliva GDF15. (C,D) Receiver Operating Curve (ROC) curve analysis for plasma (C) or saliva (D) GDF15 in distinguishing Controls (n=48-68) from MitoD groups combined (n=30-33). For ROC curve analysis separately for Deletion and Mutation groups, see Supplemental Figures S5J–K. Total sample sizes: Patients with MitoD plasma timepoints: Deletion n=15, 73% female, 102 observations; 3243A>G mutation n=20, 75% female, 160 observations; MELAS n=5, 40% female, 28 observations. Patients with MitoD saliva timepoints: Deletion n=15, 73% female, 302 observations; 3243A>G n=20, 74% female, 453 observations; MELAS n=5, 940% female, 5 observations. Only participants with pre-stress baseline and more than five stress timepoints were included in the AUC analysis. See Supplemental Table 3 for sample size at each timepoint. (E,F) Mean fasting plasma (E) and (F) saliva GDF15 levels by clinical symptoms in MitoD participants. Bars represent the mean ± SEM of GDF15 levels for MitoD participants grouped by the presence (striped bars) or absence (solid bars) of each clinical symptom. Sample sizes for each group are indicated at the base of the bars. See Supplemental Table 5 for full analysis including other timepoints, *p<0.05, **p<0.01, **p<0.001, ****p<0.0001.

We then examined the ability of blood and saliva GDF15 levels to distinguish or diagnose MitoD groups (all three groups combined) from controls. In a receiver operating characteristic (ROC) curve analysis for morning fasting plasma GDF15, the area under the curve combining sensitivity and specificity was 0.85 (95% C.I. 0.75-0.96, cut-off value at 689.6pg/mL, Figure 5C), similar to previous studies⁴. The ROC curve was similar at the 20min post-TSST timepoint with an AUC of 0.86 (95% C.I. 0.75-0.97, cut-off value at 621.3pg/mL). Using saliva GDF15 at the same time as the morning fasting blood timepoint (Figure 5D), the AUC was 0.75 (95% C.I. 0.63-0.86, cut-off value at 20.9pg/mL), indicating moderately lower sensitivity/specificity characteristics in saliva compared to plasma. However, when saliva GDF15 was sampled in the afternoon after the MRI, the AUC was 0.81 (95% C.I. 0.71-0.91, cut-off value at 65.66pg/mL), comparable to plasma. Thus, in this MitoD sample, saliva GDF15, particularly when measured in the afternoon in the context of a relatively stressful medical assessment (MRI), provides similar sensitivity/specificity characteristics as plasma GDF15.

Using the morning fasting cut-off values, out of the 40 MISBIE patients, 24 had both fasting plasma and saliva samples. Among these, 63% (15/24) showed concordance and 37% (9/24) showed discordance between the biofluids. In the 9 patients that showed discordance between biofluids, 5 patients showed elevated plasma but not saliva GDF15 levels.

Blood and stress-induced saliva GDF15 correlate with MitoD disease severity.

Finally, to examine whether GDF15 levels across biofluids reflect the severity of disease, we correlated the independently regulated plasma and saliva GDF15 levels with disease severity indices. Disease severity was measures using five independent approaches⁴⁶ (Figures 6A and 6B).

Figure 6. Plasma and post-stress saliva GDF15 correlate with disease severity.

(A,B) Correlations between disease severity indices (rows) and plasma or saliva GDF15 (columns) measured from at all MiSBIE timepoints. NMDAS: Newcastle Mitochondrial Diseases Assessment Score; COMPASS: Composite Autonomic Symptom Score-31, CNS: Columbia Neurological Score, FIS: total Fatigue Impact Scale. For the CNS and 30 seconds sit-stand test, higher scores indicate fewer symptoms and better functional capacity. (C) Spearman’s rank correlations between plasma GDF15 measured from Day 1 morning fasting (top), 10 min post stress (bottom). n=28-32 MitoD. (D) Same as (C) for Saliva GDF15 measured from Day 1 morning fasting (top) and sample taken at the end of 2-hour MRI session (bottom). n=26-30 MitoD. Total sample sizes: Patients with MitoD plasma timepoints: Deletion n=15, 73% female, 102 observations; 3243A>G mutation n=20, 75% female, 160 observations; MELAS n=5, 40% female, 28 observations. Patients with MitoD saliva timepoints: Deletion n=15, 73% female, 302 observations; 3243A>G n=20, 74% female, 453 observations; MELAS n=5, 940% female, 5 observations. Only participants with pre-stress baseline and more than five stress timepoints were included in the AUC analysis. See Supplemental Table 3 for sample size at each timepoint. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

Plasma GDF15 from the fasting morning blood draw was expectedly correlated with multi-system disease severity index by the NMDAS score (r=0.71, p<0.0001), as in previous studies⁵. Notably, this GDF15-symptom correlation was moderately stronger when measured on blood collected in the afternoon at the 10 min post stress task timepoint (r=0.75, p<0.0001) (Figure 6C). Similar patterns were observed for other disease severity indices: the GDF15-symptoms correlations in MitoD patients were stronger for plasma sampled at the 20 min poststress, suggesting that the mental stress challenge may elicit functionally and clinically significant GDF15 release that is more strongly indicative of disease severity. These analyses repeated with age-adjusted GDF15 values yielded similar results for all symptoms.

Interestingly, although blood and saliva GDF15 were not correlated, MitoD severity correlated with saliva GDF15 under certain sampling conditions. Morning fasting saliva GDF15 levels were only moderately associated with symptoms, albeit all in the same direction as plasma (rs=0.06-0.27, ps=0.16-0.96). However, afternoon MRI saliva GDF15 (after patients experience the stress and discomfort of the study/clinical procedures and report more fatigue) was more strongly and significantly correlated with clinician-rated MitoD disease severity, objective neurological signs and symptoms, functional capacity, and fatigue (rs=0.34-0.63, p-values=0.23-0.00018, Figure 6D).

Finally, given the clinical heterogeneity of mitochondrial disease⁴⁷ and the potential for organ-specific symptoms to differentially influence biomarker levels⁴⁸, we examined the plasma and saliva GDF15 concentrations according to the presence or absence of all clinical syndromes and symptoms, at two separate timepoints (AM-fasting and PM-post-stress) (Supplemental Table 4).

For symptom-level comparisons (Supplemental Table 5), fasting plasma GDF15 levels were significantly elevated in individuals with myopathy (g=1.37, p=0.01), weakness (g=1.71, p<0.0001), ophthalmoparesis (g=1.67, p<0.0001), ptosis (g=1.53, p<0.0001), and chronic fatigue (g=0.80, p=0.008) (Figure 5E, Figure S7A left). Following stress exposure, several symptom-related differences became newly significant, including depression (g=0.93, p=0.01) and chronic shortness of breath (g=1.65, p=0.004) (Figure S7A right). In contrast, saliva GDF15 comparisons (Figure 5F, limited by smaller sample sizes at the morning fasting timepoint related to overnight fasting) showed stronger associations with ataxia (AM-fasting, g=2.59, p=0.01, Figure 5F, Figure S7B left) and chronic fatigue (PM, mid-MRI, g=0.97, p=0.002, FigureS7B right). These findings highlight some meaningful differences in the sensitivity of GDF15 to specific symptoms based on the time of day when the collection is made, and differences in symptoms sensitivity between blood and saliva GDF15.

Together, our findings highlight the compounded influence of stress and OxPhos impairments on GDF15’s associations with disease severity.

3. Discussion

By documenting the presence and elevation of saliva GDF15 in patients with the m.3243A>G pathogenic variant or with single, large-scale mtDNA deletions, our work extends prior knowledge regarding the utility and sensitivity of GDF15 as a biomarker of MitoD ^1–4,6. In our small sample of MitoD participants with rare homogenous genetic lesions causing OxPhos defects, we show for the first time that saliva GDF15 is robustly elevated by 1.0-3.5-fold. However, the effect sizes for saliva are smaller than in blood. Two key findings should be highlighted. First, that GDF15 is elevated by mental stress. And second, that saliva and plasma GDF15 measured after stress exposure were not noisier, but rather more robustly associated with symptoms and disease severity than baseline, fasting samples.

A strength of this work is the repeated saliva sampling under different conditions. Combining both standardized laboratory stressors and ecological sampling at home was possible because of the non-invasive nature of saliva collection. Using this approach, we document the correlation of GDF15 with multiple indices of disease severity and symptomatology, including clinical signs and symptoms, functional capacity, and the subjective experience of fatigue. We show that saliva and plasma GDF15 correlate robustly with multi-system symptoms (up to r=0.75, meaning that GDF15 accounts for >56% of variance in symptoms), particularly when sampled in the afternoon after moderately challenging or psychologically stressful medical procedures (MRI). These non-resting timepoints likely capture the compounded effect of mental stress and metabolic stress, which could explain the heightened association between GDF15 and symptomatology when the “system” is perturbed. When we further stratified the patient population based on clinical symptoms, we observed that patients with muscle related symptoms, such as weakness, myopathy, ptosis, and ophthalmoparesis, showed higher plasma GDF15 levels (Figure 5E). Notably, blood and saliva samples collected in the afternoon after stress revealed additional symptom-related GDF15 differences, including elevated plasma GDF15 in individuals with depression and chronic shortness of breath (Figure S7A, right) and elevated saliva GDF15 in individuals with chronic fatigue (Figure S7B, right), which was not apparent in morning-collected samples. These findings align with prior research on the interaction of psychosocial stress and mitochondrial biology in mouse models of mitochondrial disorders^23,32. This also reinforces the idea that symptom context and sampling condition are critical in capturing physiologically relevant GDF15 levels in mitochondrial diseases. One implication of this work that deserves further research is that stress-related processes may exacerbate the cellular and/or systemic consequences of OxPhos defects³³, and that GDF15 may represent a potential mediator of this interaction.

Furthermore, our home-based sampling protocol where participants easily collect saliva on their own, store salivettes in their freezer, and subsequently ship samples by regular mail, provides a portable opportunity to perform intensive longitudinal, repeated-measures studies or to explore the clinical utility of saliva GDF15 in hard-to-reach study populations. Our results support the feasibility of integrating saliva GDF15 monitoring into MitoD care and potentially into broader clinical and research applications for mitochondrial health assessment. As mitochondrial diseases often require continuous monitoring⁴⁹, the ease of at-home saliva sampling offers a practical, scalable solution that may improve patient quality of life through non-invasive examination of health status, disease onset, and progression⁵⁰.

This study has several limitations. First, the sample size for our MitoD group was relatively small. Future larger cohorts are needed to validate the sensitivity and specificity of saliva GDF15 as a MitoD biomarker, or biomarker for specific disease subtypes or symptoms involvement. Second, although the MiSBIE cohort includes two homogenous group of genetically-defined MitoD patient groups (m.3243A>G, and single, large-scale mtDNA deletions), those groups are in themselves fairly heterogenous. Larger samples of well phenotyped patients will be needed to examine the dynamics and stability of GDF15-symptoms associations. Third, in relation to age, while we did not observe an association between age and saliva GDF15, nor between saliva and plasma GDF15 levels, a recent larger study reported that saliva GDF15 is higher in older individuals⁴⁵, suggesting that our null findings here may be due to the limited sample size. To enhance clinical utility of plasma and saliva GDF15, future work with larger sample sizes will be necessary to establish meaningful cutoffs by age groups. Finally, this study is cross-sectional and not informative about the prognostic value of neither plasma or saliva GDF15 levels. Now that we know that saliva GDF15 can be measured in MitoD, future studies could aim to determine whether blood and saliva GDF15 measured in the clinic (AM or PM, resting or after a stressor), or at home at awakening with saliva, predict disease progression or track with symptoms trajectories.

One technical point worth noting is that MiSBIE saliva samples collected in lab were transported on ice (4°C) to be processed the same day, while home-collected samples were frozen immediately in a home freezer (typically −20°C) followed by mail shipment at room temperature (thawing) to be processes within 48 hours. Given the novelty of saliva GDF15 measurements, we include here some preliminary analyses that may provide useful information on the potential impact of sampling method, timing, and processing (Supplemental Figure S8). What these analyses reveal is that in both controls and MitoD groups, GDF15 levels were significantly higher in home-collected samples, suggesting either a potential effect of timing or sampling method (Figure S8A). We also compared MiSBIE saliva GDF15 results to another study (SGA study⁴⁵, n=194) where saliva was also collected in the lab, but by passive drool (vs cotton swab salivettes in MiSBIE). Saliva GDF15 levels in the SGA study were 161% higher than those in MiSBIE controls across the stress reactivity timepoints, and baseline levels in SGA were 123% higher than MiSBIE baseline levels (Figure S8B). The ages of both cohorts were 37.5 (MiSBIE) and 40.3 years (SGA). These differences reflect either an effect of sampling methodology, population differences, and/or processing conditions. These initial findings should be taken into considerations when designing future studies.

In summary, this work identifies saliva GDF15 as a dynamic marker of mitochondrial OxPhos defects and reveals its rapid regulation by psychosocial stress. Our strategic repeated-measures sampling across laboratory and home-based studies demonstrates general dynamic principles for the diurnal and psychobiological regulation of both blood and saliva GDF15 in individuals with mtDNA mutations and single, large-scale mtDNA deletions, establishing its dynamic “state” properties beyond its stability as a MitoD biomarker. This broadens our understanding of human GDF15 biology beyond disease. Our results also suggest that OxPhos defects and psychobiological processes underlying the response to mental stress share converging biology, possibly converging on energy needed to both adapt to OxPhos deficiency⁵¹ and to power the stress response²⁵. This novel finding points to converging biology for mental stress and OxPhos defects, calling for further studies to examine the clinical utility of highly accessible saliva GDF15 measurement as a prognostic marker of mitochondrial diseases.

4. Methods

4.1. MiSBIE cohort

4.1.1. Participants

Participants were enrolled in the Mitochondrial Stress, Brain Imaging, and Epigenetics (MiSBIE) study in adherence to the directives outlined by the New York State Psychiatric Institute IRB protocol #7424 and the Columbia University Medical Center IRB protocol #AAAU9470. The study was registered in ClinicalTrials.gov under #NCT04831424. Recruitment was conducted both within our local clinic at the Columbia University Irving Medical Center, and nationally throughout the United States and Canada. All enrolled participants provided written informed consent, authorizing their participation in the investigative procedures and the potential dissemination of data. Recruitment occurred from June 2018 to May 2023. A total of 110 participants ages 18 to 60 years were enrolled, including 70 healthy controls (females n=48, males n=22) and 40 individuals with genetically defined mitochondrial diseases (MitoD, females n=28, males n=12). This manuscript specifically reports on MiSBIE GDF15 dynamics in blood and saliva. Sex was determined by self-report. This single-step method is limited, does not distinguish between gender and sex, and can exclude both transgender and intersex people; current best practices include a two-step⁵² measurement and broadening the available answer options.

Our inclusion criteria for controls included healthy adults willing to provide saliva samples and have blood collected using an intravenous catheter during the hospital visits. Exclusion criteria included the presence of pronounced cognitive impairment precluding informed consent, concurrent neoplastic disease, recent occurrences of flu or other temporally pertinent infections within the four-week window preceding study participation, Raynaud’s syndrome, engagement in ongoing therapeutic or exercise trials registered on ClinicalTrials.gov, and the existence of metallic elements within or on the body, alongside claustrophobia posing an impediment to magnetic resonance imaging (MRI). In-person laboratory procedures were performed over 2-days (Tuesday and Wednesday), followed by a 1-week home-based saliva collection protocol.

For the MitoD groups, eligibility criteria for inclusion included the presence of a genetically ascertained diagnosis of symptomatic MitoD, which encompassed either (i) the m.3243A>G pathogenic variant, with³⁹ or without³⁸ the clinical manifestations of mitochondrial encephalopathy, lactic acidosis, and stroke-like episodes (MELAS), or (ii) a single, large-scale mitochondrial DNA deletion-associated presentation of chronic progressive external ophthalmoplegia (CPEO), Kearns-Sayre syndrome (KSS)³⁷, or other syndromic variant (Supplemental Table 1). A standardized questionnaire was administered to all participants, collecting essential information including demographic attributes, age, gender, ethnicity, prevailing health status, socioeconomic status, and ongoing medication regimens. Exclusion criteria were similar to those of healthy controls, with the additional criteria excluding patients who used steroid therapy (including oral dexamethasone, prednisone, or analogous agents).

Disease severity and symptomatic profiles were quantified using five different approaches. 1) The Newcastle Mitochondrial Disease Adult Scale (NMDAS)⁵³, a validated metric administered by a qualified clinician, which assesses symptomatic and functional limitations across several domains of life; 2) Autonomic dysregulation (Composite Autonomic Symptom Score – COMPASS 31)⁵⁴, which assesses autonomic symptoms in six domains; 3) neurological symptoms (Columbia Neurological Scale – CNS)⁵⁵, which assesses central nervous system symptoms; 4) Functional capacity (30-second sit-to-stand test)⁵⁶, 5) Perceived fatigue and its impact on physical, cognitive, and psychosocial functioning (Fatigue Impact Scale – FIS)⁴⁶.

Non-stress plasma and saliva samples were collected both in the morning, under fasting conditions around 9:30 AM, or after lunch around 1:00 PM. Both the breakfast and lunch meals were selected from a restricted MiSBIE study menu to avoid large differences in meal types between participants.

4.1.3. Plasma and saliva collection

The blood and saliva collection protocols are illustrated in Supplemental Figure S2A. Plasma was collected via a central venous catheter at 9 time points: 1) day 1 morning fasting sample, 2) 8 afternoon samples collected during the stress reactivity protocol. A total of 16-20mL of whole blood was collected at the morning fasting timepoint in two 10 mL K₂EDTA blood collection tubes (BD #366643). A total of 5mL of whole blood was collected at each afternoon time point both before and after stress in 6mL K₂EDTA blood collection tubes (BD #367899).

Saliva was collected at 26 timepoints: 1) day 1 morning fasting sample, 2) 8 afternoon samples collected during the stress reactivity protocol, 3) 1 sample after a cold pressor test, 4) day 2 morning fasting sample, 5) 3 afternoon stress samples before and during the MRI, 6) 12 at-home samples collected across 3 days: (i) immediately upon awakening, (ii) 30 min after waking up, (iii) 45 min after waking up, and (iv) at bedtime. A total of 1-2 mL of saliva was collected at each sample timepoint using a salivette (Starstedt # 51.1534.500) following the Biomarker Network recommended procedure⁵⁷. Participants were directed to position the cotton swab at the center of their tongue within their mouth for 2 to 5 min. It was emphasized that they should refrain from biting down on the cotton and ensure the swab did not come into contact with their cheeks. Afterward, participants reinserted the cotton swab into the salivette recipient tube. Samples were collected on visit days 1 and 2, the salivettes were placed on ice (4°C) in a styrofoam box for transportation and further processing at the laboratory. For home-based sample collection, salivettes were frozen immediately after collection in a home freezer (typically −20°C). During the at-home morning collections, individuals were advised to ideally delay tooth brushing and eating until after the third sample (+45 minutes), while the nighttime sample was to be taken prior to bedtime tooth brushing. If participants needed to, they were advised to eat breakfast within the 30 min break of awakening sampling and note the time and content. Participants were instructed to avoid consuming water or any other liquids within 10 minutes of each saliva sample collection. This was monitored and controlled for the laboratory samples. Within 2 weeks of collection, participants either transported (~10%) or shipped them (~90%) to the laboratory in a non-temperature controlled, pre-stamped USPS priority shipping.

4.1.4. Biofluids (Plasma and Saliva) processing and storage

Whole blood tubes for morning fasting plasma samples were immediately inverted 10-12 times and centrifuged at 1000xg for 5 minutes at room temperature after collection. Samples were placed on ice (4°C) in a styrofoam box and transported to the laboratory for further processing. When samples reached the laboratory, they were immediately centrifuged at 2000xg for 10 minutes at 4°C. To avoid platelets and other cellular debris, approximately 80% of the plasma was collected from the upper portion of each tube, transferred and pooled into a fresh 15 mL conical tube. To further eliminate cellular components, pooled plasma tube was centrifuged at 2000xg for 10 minutes at 4°C. Around 90% of the resulting plasma supernatant was transferred to fresh 15 mL conical tube and mixed through inversion. The resulting cell-free plasma was aliquoted into 0.5-1.5 mL aliquots and stored at −80°C before being used for stress reactivity GDF15 measurements by ELISA.

Stress reactivity time points include 8 tubes for each timepoint, starting at 5 minutes to 120 minutes. Whole blood tubes for afternoon stress plasma samples were immediately inverted 10-12 times and centrifuged at 2000xg for 3.5 minutes at room temperature after collection. Samples were placed on ice (4°C) in a styrofoam box and transported to the laboratory together for further processing. When samples reached the laboratory, they were immediately centrifuged at 2000xg for 10 minutes at 4°C. To avoid platelets and other cellular debris, approximately 80% of the plasma was collected from the upper portion of each tube and transferred into fresh 8*15 mL conical tubes. To further eliminate cellular components, plasma tubes were centrifuged at 2000xg for 10 minutes at 4°C. Around 90% of the resulting plasma supernatant was transferred to fresh 8*15 mL conical tubes and mixed through inversion. The resulting cell-free plasma was aliquoted into 0.5-1.5 mL aliquots for each timepoint and stored at −80°C before being used for stress reactivity GDF15 measurements by ELISA.

When saliva samples reached the laboratory, salivettes were centrifuged at 1000xg for 5 min in a refrigerated centrifuge at 4°C. To avoid cellular contamination with leukocytes or epithelial buccal cells⁵⁸, supernatant was carefully removed from the top of the tube, transferred to cryovials, and stored immediately at −80°C. Before use, saliva was thawed and centrifuged at 5000xg for 10 min to further eliminate cellular components. The resulting supernatant was collected as cell-free saliva and was used for subsequent GDF15 assessment.

4.5. GDF15 assays

Plasma, serum and saliva GDF15 levels were quantified using a high-sensitivity ELISA kit (R&D Systems, DGD150, SGD150) following the manufacturer’s instructions. Different lot numbers were used, and the average coefficient of variation (C.V.) between lot numbers was determined by reference samples for quality control. Plasma and serum samples were diluted with assay diluent (1:4 ratio) to maximize the number of samples within the dynamic range of the assay, including MitoD participants. Saliva samples were not diluted. Absorbance was gauged at 450nm, and concentrations were computed utilizing the Four Parameter Logistic Curve (4PL) model. Samples were run in duplicates, on separate plates when possible, and the concentration for each sample was computed from the average of the duplicates. Standard curve (5 samples per plate) and plasma reference samples (2-3 samples per plate, same sample per batch) were run with each individual assay and the inter-assay C.V. was monitored. All standard curves and references were overlaid on top of each other to monitor failed runs. Samples were run in duplicate plates when possible and those with C.V. larger than 15% were re-ran. When it was not possible to rerun (e.g., no sample left), sample sets with a C.V. >15% between the duplicates were included. We performed sensitivity analyses excluding these samples, confirming that the results were unchanged by the presence of absence of these samples with lower reliability. Values below the mean minimum detectable dose (2.0pg/ml) were considered as non-detectable (reported as NA) and excluded in the graphs or statistical analyses. For the MiSBIE study, samples were run in 3 batches over 2 years, multiple quality control measures were applied to monitor batch-to-batch and within batch variability. For a full experimental setup, see Supplementary Table 4. Data-preprocessing and quality control measures was done using R Software (version 4.2.2) and are available as supplemental files.

Based on the 1000 Genomes Project, around 15-30% of the population assessed in the project carried the common H202D variant in GDF15⁵⁹. Compared to the Roche Elecsys assay, the R&D ELISA kits was found to report GDF15 concentrations that are 36% lower in individuals carrying one D allele or 61% lower in those that carry two D alleles at position 202 of the pro-peptide (or position 6 of the mature peptide)⁶⁰. If an unequal number of individuals with this variant were included in our control and MitoD groups, this limitation could affect our group comparisons. However, our within-person findings related to psychosocial stress reactivity, diurnal variation, and awakening responses would be unaffected by the underestimation of GDF15 concentration in some participants with the H202D variant.

To analyze GDF15 gene expression, we utilized RNA sequencing data from the Genotype-Tissue Expression (GTEx) Project v8 dataset encompassing 948 donors and a total of 17,382 samples. GTEx v8 is available for download at https://gtexportal.org/home/downloads/adult-gtex/bulk_tissue_expression. Gene read counts were normalized using the trimmed mean of M-values (TMM) normalization method with the edgeR package⁶¹. The age of donors is part of the protected access data and was obtained through the dbGAP accession #phs000424.v8.p2 under the project #27813 (Defining conserved age-related gene expression trajectories). We extracted donor’s age and GDF15 gene expression data was sequenced in all 54 human tissues. Single cell data sequenced in the minor salivary gland and immunohistochemistry analysis of GDF15 protein abundance in various tissues were extracted from the Human Protein Atlas and presented in this paper. The Human Protein Atlas data is available for download at https://www.proteinatlas.org/about/download. Histology graphs used in this paper are: GDF15 (https://www.proteinatlas.org/ENSG00000130513-GDF15/tissue/salivary+gland), TOMM20 (https://www.proteinatlas.org/ENSG00000173726-TOMM20/tissue/salivary+gland), CPT1A (https://www.proteinatlas.org/ENSG00000110090-CPT1A/tissue/salivary+gland), TFAM (https://www.proteinatlas.org/ENSG00000108064-TFAM/tissue/salivary+gland), NDUFS4 (https://www.proteinatlas.org/ENSG00000164258-NDUFS4/tissue/salivary+gland), MCU (https://www.proteinatlas.org/ENSG00000156026-MCU/tissue/salivary+gland).

4.6. Statistical analyses

The association between age and plasma/saliva GDF15 level at morning, fasting sample (baseline) was determined using Spearman rank correlation. Group differences (e.g. sex difference, difference between the controls, 3243 and deletion groups) were assessed using non-parametric Mann Whitney test and Tukey multiple comparisons post hoc test where appropriate. Differences across two timepoints in the same individual (e.g. Morning to afternoon, day 1 to day 2 fasting) were assessed using paired non-parametric signed-rank Wilcoxon test. Diurnal pattern was assessed using Tukey multiple comparison post hoc test. Stress reactivity was operationalized as % change from baseline at each time point, and a mixed effect models were used to test whether acute stress affected plasma and saliva GDF15 levels over time. The magnitude of GDF15 stress reactivity and comparisons of GDF15 levels between MitoD groups and the control group were quantified as effect sizes (Hedges’ g) which calculates differences of the means divided by the pooled standard deviations. To capture longitudinal stress reactivity and awakening response over the time range of 8 time points spanning from 5 minutes before speech task to 120 min after speech task, we used area under the curve (AUC). Specifically, for stress reactivity, AUC of GDF15 levels over the 8 time points was calculated as the area under the longitudinal GDP15 measures, where missing GDF15 values were imputed using mean GDF15 of all available time points if missing happened at last few consecutive time points, or using weighted average of two adjacent available GDF15 measures weighted by fraction of time intervals, with bigger weights for closer measures, if the missing value happened elsewhere. Only participants with pre-stress baseline and at least 4 stress timepoints were included in the AUC analysis. For awakening response analysis from immediate upon awakening to 45 min, we averaged GDF15 measures collected at the same timepoint during a day from three days (Mon, Weds, and Fri). Awakening response AUC was calculated similarly using GDF15 at 3 timepoints. We also examined the ability of blood and saliva GDF15 levels to distinguish (i.e., diagnose) all three MitoD groups combined from controls using simple logistic models, where results are presented using receiver operating characteristic (ROC) curves. The optimum GDF15 cutoff values were determined using Youden index. Statistical analyses were conducted using GraphPad Prism (version 9.4.1) and R Software (version 4.2.2 and 4.3.0)

5. Code and data availability statement

All data is available and submitted along with his manuscript. The code for data analysis is available on GitHub at https://github.com/mitopsychobio/2024_GDF15_Dynamics_Huang