Topical Steroid Withdrawal is a Targetable Excess of Mitochondrial NAD+

Abstract

Topical Steroid Withdrawal (TSW) is a controversial diagnosis advocated by patients but often confused for atopic dermatitis (AD). We conducted a multi-modal pilot study of 16 patients fitting the TSW diagnostic profile: contrasting them against patients with AD (N=10) and healthy controls (N=11). Our clinical evaluations established objective diagnostic criteria which distinguish TSW from AD; metabolomics and transcriptomics of skin biopsies suggested that neuro-inflammatory pathways are associated with complex I mediated oxidation of NAD+; cellular and mouse models demonstrated that NAD+ metabolism was proinflammatory and glucocorticoid responsive; while functional assays demonstrated that the metabolic effects of glucocorticoids on the only cell type which aligns with the distribution and duration of TSW pathology could be mitigated by complex I blockade. These results informed a successful open-label trial using complex I inhibiting interventions, metformin and berberine. Although this work represents a pilot study, to our knowledge, this work offers previously unreported mechanistic insights into TSW.

Graphical Abstract

Introduction

Topical steroids are first-line therapies for atopic dermatitis (AD) and other inflammatory dermatological conditions (Eichenfield et al., 2014). Although more accurately described as glucocorticoids, topical corticosteroids (TCS) is the more common clinical term. In recent decades, there has been a concerning rise in reports of severe systemic adverse reactions due to long-term use and abrupt cessation of TCS (commonly referred to as Topical Steroid Withdrawal; TSW) (Hwang and Lio, 2022). Although no formal diagnostic criteria for TSW exist, reports of patients experiencing TSW are very common online (Bowe et al., 2022). Initial reports concluded that TSW was limited to exposed areas for patients who applied high-potency TCS to the face or genitals (Hajar et al., 2015). However, subsequent case series have better identified the ways in which TSW may be different from AD, contact dermatitis, or other dermatoses (Ahuja and Lio, 2024, Barta et al., 2023, Brookes et al., 2023, Hwang and Lio, 2022, Marshall et al., 2024, Sheary and Harris, 2020), including disease manifestations on regions of the body where TCS were never applied. Although resolution may take months to years, improvement is seen through avoidance of TCS therapy, further suggesting that TSW is a distinct clinical entity rather than a flare of the underlying dermatoses (Feschuk and Pratt, 2023).

Although some have labeled the patient concerns as “steroid phobia” (Finnegan et al., 2023), there have been no systematic clinical or mechanistic studies to either refute or support TSW as molecularly distinct from other dermatopathies. Herein, we report a re-analysis of a previous survey of 1,889 adults with eczematous skin disorders (Barta et al., 2023) to clinically distinguish those that self-diagnosed with TSW from those that denied having TSW. We then report the results of a pilot study of 16 patients with symptoms consistent with TSW. Using multi-omics analysis of serum and skin samples we identified increased vitamin B3 (nicotinic acid; NAD+) oxidation stemming from overexpression of known NAD+ producers (Qin et al., 2018, Vial et al., 2019) such as mitochondrial complex I. Open-label treatment with anti-complex I medications (metformin or the “herbal metformin” berberine) led to notable improvements in symptoms. Our work represents an investigation into the pathogenesis of TSW resulting in a promising targeted therapy.

Results

Patients with TSW have symptoms distinct from atopic dermatitis

Reanalysis of a previously published survey of patients with eczematous skin disease (Barta et al., 2023) identified worsening skin symptoms despite TCS use as most predictive of self-diagnosed TSW compared to those with eczematous skin disease who denied having TSW (Fig. 1A; Supplementary Fig. S1A). Worsening signs may include reduced TCS effectiveness, a need for higher potency, spreading rash despite treatment, or new-onset symptoms of temperature dysregulation and full body redness/burning (Fig. 1A). Peak itch intensity and sleep disturbance were significantly higher in those affirming TSW (Fig. 1B). However, there were no consistent, significant differences in body site for either severity (Fig. S1B) or application site (Fig. 1A). The dryness and oozing scores typically collected for the Patient Oriented SCORing Atopic Dermatitis (PO-SCORAD (Faye et al., 2020)) severity metric were inversely associated with TSW (Fig. 1A), however a greater degree of redness, scratching, swelling, and thickening were seen in TSW (Fig. S1C). Skin specific symptoms with a greater than 80% prevalence that are distinct from AD were deemed as major criteria while the remainder were utilized as minor criteria (Fig. 1C). Twenty-five percent of patients reported all related symptoms, while the combination of ≥1 major and ≥3 minor offered >90% sensitivity for capturing those reporting TSW in the survey (Fig. 1C). Nearly 25% of patients reported symptoms lasting longer than 3 years after discontinuation of TCS (Fig. 1D).

Figure 1: Topical Steroid Withdrawal (TSW) is distinguishable from other eczematous skin diseases.

(A) Lasso regression model of previously reported survey results from 1,889 individuals with eczematous skin disease (Barta et al., 2023) establishing the responses that are most predictive of self-identified TSW. Results separated by over-the-counter hydrocortisone (HC) versus prescription topical steroids (TS). (B) Intensity of itch and sleep disruption for the respondents affirming TSW versus those denying such diagnosis. (C) Proposed diagnostic criteria derivation evaluating the sensitivity (Sens.) for number of reported Major and Minor criteria extracted from survey from respondents reporting suffering from TSW (N=1,486). (D) Time to symptom of TSW symptoms as reported by respondents (N=1,486). (E-I) Representative images of TSW associated flushing and erythema (E-F), ‘red sleeves’ (G), “elephant wrinkles” (H), hair loss (I), and pronounced desquamation (J). All images used with written consent of participants or their legal guardian (1F).

Although many reports imply that TSW is limited to the application of high-potency glucocorticoids to the face or genitals (Hajar et al., 2015, Spergel and Leung, 2023), after written, informed consent was obtained we enrolled a cohort (N = 16) who each experienced full-body disease, including areas that had never been directly treated with TCS (Supplementary Fig. S1D; Supplementary Table S1). Patients described full body redness, flushing, and anhidrosis (Fig. 1E) occurring 4–6 weeks after discontinuation of TCS, which: tended to spare the nose (Fig. 1F; termed “headlamp sign” (Sheary, 2018)), palms, and soles (Fig. 1F); and often formed “red sleeves” on the arms (Fig. 1G). Additional symptoms included: subjective temperature changes; stabbing neuropathic pain (termed “zingers” by patients; Supplementary Fig. S1D); loose skin over the flexor surfaces (“elephant wrinkles”; Fig. 1H); hair loss (Fig. 1I); and substantial skin shedding (Fig. 1J; Supplementary Fig. S1D). The shortest time since last steroid exposure in the cohort was 4 months, with a mean time since last exposure of over 47 months (Table S1); this excluded concerns for possible contact dermatitis from TCS exposure or TCS-specific excipients.

Serum metabolomics suggest systemic lipid dysregulation

Serum metabolomics for patients with TSW was compared to healthy volunteers (HV; N = 11), and patients with AD who did not report symptoms of TSW (N = 10). Although matched for sex and ethnicity, the TSW cohort was slightly older (mean 37.1 years) than AD (26 years) and HV (25 years; Table S1). Significant differences were seen under global similarity analysis but not for any specific identifiable metabolites (Supplementary Fig. S2A–D). However, collating the results into MetaboAnalyst pathway analysis (Pang et al., 2021) suggested patients with TSW had significant deficiencies in the sphingolipid and urea cycle amino acid pathways (Supplementary Fig. S2E) compared to both HV and AD. Patients with TSW had significant alterations of metabolites in several pathways, such as glycerophospholipids and C21-steroid hormones including cortisol (Supplementary Fig. S2F; Supplementary Fig. S3).

Markers of dysbiosis and inflammation in TSW were similar to those previously described in atopic dermatitis

Evaluation of the microbiome signatures in TSW revealed significant differences from HV in bacterial speciation at the genus level (Fig. 2A–B; Supplementary Fig. S4A). Similar to AD (Gough et al., 2022), there was a predominance of Staphylococcal species, particularly S. aureus, burden (Fig. 2C) and reduced alpha bacterial diversity (Fig. 2D). Less robust differences with respect to HV were seen in fungal speciation, with results again similar to those described for AD (Fig. 2E; Supplementary Fig. S4B) (Tao et al., 2022). Virome signatures demonstrated an increased burden of viruses associated with fungal and bacterial taxa (Supplementary Fig. S4C) while functional metagenomics revealed global differences (Fig. 2F) that were dominated by transcripts associated with Staphylococcus infection (Supplementary Fig. S5A–B).

Figure 2: Microbiome and Th2 cytokine signatures in TSW differ from healthy controls.

Shotgun metagenomic sequencing was performed on skin swabs from patients with TSW (N=16) versus healthy volunteers (N=7). (A) Speciation at genre level with most abundant taxa indicated (full list in Supplementary Fig. S4A). (B) Similarity plot as measured by NMDS (non-metric multidimensional scaling) for genre-level bacterial analysis. (C) Normalized counts for Staphylococcus aureus. (D) Chao1 alpha-diversity index. (E-F) Speciation and NMDS for fungal genre. (G) IgE levels (mean ± SD) for TSW cohort (dotted line indicates upper limit of normal for our hospital). (H) Serum cytokine levels for HV, TSW and patients with atopic dermatitis without history of TSW (AD) for interleukin (IL-) 13, IL-6, CCL13, and IL-4. Significance adjusted for three-way comparison, but not the full number of cytokines analyzed. * = p value <0.05, ns = not significant, other p values indicated.

Like AD (Sugita and Akdis, 2020), preliminary characterization of cell infiltrates were primarily CD4+ and CD8+ T-cells, as well as CD68+ monocytes (Supplementary Fig. S5C); these immune cells were detectable in thin sections of skin biopsies for patients with TSW but required 3-dimensional volume analysis to detect in HV samples (Supplementary Fig. S5D; Movie S1). For 11 of the patients in the cohort, IgE levels were elevated (Fig. 2G). Serum cytokine analysis indicated significantly increased levels of interleukin (IL-) 4 and IL-13, as well as proinflammatory IL-6 and CCL13 (Fig. 2H; Table S2).

Metabolomics implicates vitamin B3 metabolism

Skin metabolomics was performed using MALDI-imaging mass spectrometry (Yadav et al., 2022), which identifies metabolites in space (example shown in Fig. 3A). Ranking the abundance of metabolites that could be identified using m/z and collisional cross section (CCS) revealed distinctions between TSW and HV (Fig. 3B). Using MetaboAnalyst (Pang et al., 2021) to contrast TSW with both HV and AD identified a significant increase in vitamin B3 (nicotinate and nicotinamide) metabolism (Fig. 3C). Several pathways were downregulated compared to the HV and AD cohorts, most prominently tryptophan metabolism (Fig. 3D). Of note, nicotinic acid is a derivative of tryptophan through a process that generates the known neuropathologic, and anticholinergic intermediate kynurenine via the enzymes indoleamine 2,3-dixygenase (IDO) and tryptophan 2,3-dioxygenase (TDO) (Qin et al., 2018). Consistent with the known interaction between NAD+ and the itch regulating transient receptor potential (TRP) family ion channels (Arendt-Nielsen et al., 2022, Ma et al., 2014), staining a subset of patients and controls for expression of TRPA1 revealed enhanced dermal, but not total, expression (Supplementary Fig. S5E–G).

Figure 3: Skin from patients with TSW demonstrates disruptions in Vitamin B3 and Tryptophan metabolism.

Skin biopsy samples from patients with topical steroid withdrawal (TSW; N=16), Atopic Dermatitis (AD; N=10), and healthy volunteers (HV; N=11) underwent MALDI-imaging mass spec. (A) Representative image for distribution of metabolite at 772.5213 m/z. for HV, AD, and TSW; scale bar represent 8mm (B) Intensity values for metabolites identifiable through combination of m/z and collisional cross-section for HV and TSW. (C-D) MetaboAnalyst pathway analysis for metabolites ranked by how predictive of TSW they were compared with both HV and AD cohorts (C) and metabolites ranked by ROC value (receiver operating characteristic) for the HV and AD cohorts compared to TSW (D). Bolding indicates pathways not identified as distinctive for AD relative to HV.

RNAseq suggests mitochondrial dysfunction is present in TSW

Biopsies subjected to RNAseq identified cell cycle and neurodegeneration as the most distinguishing pathways when comparing TSW to HV (Supplementary Fig. S6A). The neurodegeneration pathway includes Wnt signaling (Fig. 4A) which has been linked to AD previously and would be expected to produce increased cell proliferation (Myles et al., 2018). Several mitochondrial complexes were also upregulated in patients with TSW, particularly complex I (Fig. 4B) which produces NAD+ as a byproduct. Several other aspects of nicotinate metabolism were upregulated in TSW (Supplementary Fig. S6B). Opposing expression of IDO2 (EC 1.13.11.11) and TDO2 (EC 1.13.11.52) was seen between TSW and controls (Fig. 4C; full image in Supplementary Fig. S6C). STRING analysis (Szklarczyk et al., 2023) demonstrated a strong signal for histone modifying and DNA processing pathways but were most strongly linked to proton motive force in ATP synthesis (Supplementary Fig. S6D). These results were inconsistent with the reports of distinguishing transcriptomics in HV versus AD (Guo et al., 2023, Mitamura et al., 2023, Schabitz et al., 2022, Yu and Li, 2022). Activation of inflammatory pathways known to respond to upregulations in the oxidative stress pathways (Supplementary Fig. S6A) was evident for MAP kinase (Supplementary Fig. S7A), JAK-STAT (Supplementary Fig. S7B), NFκB (Supplementary Fig. S8A), and T-cell receptor signaling (Supplementary Supplementary 8B).

Figure 4: Skin from patients with TSW demonstrates disruptions in Niacin related metabolic pathways.

Skin biopsy samples from patients with topical steroid withdrawal (TSW; N=16) and healthy volunteers (HV; N=11) underwent RNAseq analysis. (A-C) KEGG pathway analysis for Wnt signaling (A), Oxidative phosphorylation (B), and Nicotinate and Nicotinamide metabolism (C; full image in Supplementary Fig. S6B). Superimposed log10 fold change, green indicates product is more abundant in HV, red indicates product is more abundant in patients with TSW.

Nicotinic acid and clobetasol exposure impact stem cell metabolism

Although niacin and nicotinic acid are known to induce skin flushing (Kamanna et al., 2009), most research into the skin effects of niacin has focused on dietary intake. To assess the effects of topical application of nicotinic acid, we applied various concentrations to mouse ears. Nicotinic acid yielded a dose-dependent increase in swelling (Fig. 6A; Fig. S9A), but with redness that was below the mouse hairline (Fig. 6B). This finding is distinct from the standard murine models of AD such as MC903, which induce dermatitis on the entire ear (Supplementary Fig. S9B). Combining the prior reports of nasal, palmar, and plantar sparing, the protracted resolution (Sheary, 2018), and murine disease limited to areas with hair (Fig. 6B) suggested a possible role for stem cells in the pathology of TSW given that these cells reside at the base of each hair follicle and serve as progenitors for replenishing other skin cells (Diaz-Garcia et al., 2021).

Figure 6: Steroid withdrawal induces mitochondrial associated niacin overproduction in human skin cells.

(A-B) Mice were treated for 8 days with daily topical application of nicotinic acid. Resultant ear thickness (A) and representative image of redness below shaved hairline (B) are shown; box and mouse image to separate murine studies from cell culture work. (C) Human follicular stem cells (HFSC) exposed to increasing doses of nicotinic acid, scale bars indicate 2mm. (D) Total cell count for HFSC incubated in clobetasol for 48 hours (Steroid), clobetasol for 24 hours followed by normal media for 24 hours (Steroid withdrawal), or under steroid conditions with 10uM for metformin or berberine. Dots indicate replicate wells. (E-F) HFSC under mitochondrial stress test for Seahorse assay (as assessment of metabolic activity which contrasts oxidative phosphorylation and glycolysis); indicating basal extracellular acidification rate (ECAR; a measurement of glycolysis) and basal oxygen consumption rate (OCR; a proxy for oxidative phosphorylation) for cells incubated as in panel D. (G) MetaboAnalyst pathway derive index of pathway significance for HFSC, keratinocytes (KC), fibroblasts (FB), or Schwann cells (Sch.) under steroid (S) and steroid withdrawal (W) conditions as measured by imaging mass spec. (H) Basal OCR and ECAR for HFSC incubated with metformin or berberine (but no glucocorticoids) for 24 hours before Seahorse analysis; 0uM represents the diluent value. Data represent 3 independent experiments (A-I) and are indicated by mean ± SEM. * = p <0.05, ** = p <0.01, *** = p <0.001, **** = p <0.0001.

Similar to prior reports (Tan et al., 2019), incubation of human follicular stem cells (HFSC) with nicotinic acid led to enhanced proliferation (Fig. 6C). Exposure to the commonly prescribed, high-potency glucocorticoid clobetasol also significantly increased HFSC proliferation (Fig. 6D). Functional metabolic analysis of HFSC indicated clobetasol exposure increased maximal extracellular acidification rate (ECAR; a measurement of glycolysis; Fig. 6E, Supplementary Fig. S9C) but not resting oxygen consumption rate (OCR; a proxy for oxidative phosphorylation; Fig. 6F, Supplementary Fig. S9D). Incubation of HFSC with clobetasol did not impact vitamin B3 metabolism, however, removing steroids from the media after clobetasol exposure impacted vitamin B3 and other metabolic pathways including the TCA cycle (Fig. 6G; Supplementary Fig. S9E). Differences between continuous glucocorticoid exposure and steroid withdrawal were also evident for human primary keratinocytes, fibroblasts, and Schwann nerve cells (Fig. 6G; Supplementary Fig. S9F–H).

These results suggested that TSW pathology may represent an overactivation of complex I causing an increase in NAD+ oxidation, either through increased complex I expression or increased NADH availability via tryptophan metabolism (Supplementary Fig. S10A). Therefore, inhibition of complex I activity would be a hypothetical therapeutic target. Metformin is a known blocker of complex I (Vial et al., 2019), has pre-clinical efficacy in models of AD (Choi et al., 2020), and reduces the conversion of tryptophan to kynurenine (Muzik et al., 2017). Similarly, berberine is the main ingredient in the traditional Chinese medicine (Wu Mei Wan) which was previously shown to improve TSW (Wang et al., 2021); berberine is also a known complex I inhibitor with similar uses as metformin (Fang et al., 2022). Metformin treatment of HFSC prevented hyperproliferation (Fig. 6D) as well as the alterations in basal OCR and ECAR (Fig. 6E–F, Fig. 6H–I, Supplementary Fig. S9B–C). Berberine did not alter proliferation (Fig. 6D) or basal ECAR (Fig. 6I, Supplementary Fig. S9B) but was a more potent inhibitor of OCR (Fig. 6F, Fig. 6H, Supplementary Fig. S9C). Diluting berberine to as low as 0.6uM had inhibitory activity without a strong coloration (Fig. 6H, Supplementary Fig. S10B).

Topical glucocorticoids alter mitochondrial transcription pathways

Nineteen healthy controls aged 18–64 were treated with the topical glucocorticoid methylprednisolone aceponate. Skin biopsies were taken prior to exposure, then again 4 hours after topical application. RNAseq identified numerous significant differences (Fig. 5A) concentrated in the keratinocyte differentiation pathways (Fig. 5B). Mitochondrial complexes, especially the complex I, NAD+ producing enzyme, ubiquinone reductase (EC 7.1.1.2) were significantly upregulated by topical glucocorticoid exposure (Fig. 5C; Supplementary Fig. S11). Impacts on Wnt and tryptophan metabolism were evident, but less robust than for TSW (Supplementary Fig. S12A–B).

Figure 5: Short term exposure to topical glucocorticoids induces mitochondrial complex I.

19 healthy controls underwent baseline skin biopsy followed by topical exposure to a glucocorticoid (see methods). 4 hours later, a repeat biopsy was performed in the area of glucocorticoid exposure. RNA-seq was performed, followed by differential expression analysis contrasting post- and pre-treatment skin transcriptomes. Volcano plot (A) and derived pathways (B) are shown. (C) KEGG pathway for oxidative phosphorylation with superimposed log₁₀ fold change. Green indicates product is more abundant before glucocorticoid exposure and red indicates product is more abundant after glucocorticoid exposure.

Therapies Selected for Complex I Inhibition Improve TSW Symptoms

Given the data demonstrating the blockade of complex I by metformin and berberine (Fig. 6H), we theorized that patients with TSW may receive benefit from these treatments. Patients in the cohort were informed about the potential utility of metformin and berberine and provided the analysis by the commercial outlet ConsumerLabs identifying that Natural Factors WellBetX and Solaray supplement brands contained the indicated amount of berberine without harmful contaminants. Three patients elected to take metformin through their primary care providers, 9 elected to take berberine purchased online, 2 made no changes in treatment, while 1 opted for alternate treatments (Table S1). No serious complications were reported (Table S1).

After 3–5 months of use, subjective overall improvement between 5% and 80% was reported for both interventions (Fig. 7A), with non-significantly greater improvement reported for those on berberine. These results were significantly greater than what would be expected (p = 0.012) when considering the time since symptom onset for each participant (Table S1) against the expected time to resolution reported in the larger survey (Fig. 1D) (Fig. 7B). Patients initially noted their skin becoming “softer” and/or “smoother” followed by an “inside out” resolution of itch in which the bone-deep itch resolved before the superficial itch. Rashes improved in both groups (Fig. 7C). Patients still reported flares related to known triggers such as infection, stress, or temperature change; however, these flares were described as of lower severity or shorter duration.

Figure 7: Use of complex I targeting medications correlated with improved outcomes in TSW.

(A) Response from patients for global assessment of overall improvement after 3–5 months of treatment with indicated intervention. All participants took 500mg of the indicated treatment daily for 5–7 days; then each escalated to twice daily thereafter. Open diamond indicates patient who started on metformin for 2 months, then switched to berberine for 2 months. Open circle indicates patient that discontinued berberine after 4 weeks when she discovered she was pregnant. Open triangle indicates a patient who used both berberine and ‘activated plasma’. Closed square indicate patient with a multitude of various interventions (See Table S1 for more details). One participant was loss to follow up. (B) Time since symptom onset for each patient using either metformin or berberine (detailed in Table S1) was contrasted against the odds of resolution by month as displayed in Fig. 1D; the expected (Exp.) rate of improvement was calculated and contrasted against the observed (Obs.) rate. (C) Representative images of improvement in patients who used berberine. All images used with written consent of participants.

DNA variants with differentially expressed transcripts in TSW included PAX6 and FMO2

Genome sequencing only identified the highly variable TTN as the gene with variants of uncertain significance (VUS) across the cohort (Supplementary Fig. S13A–B). Expanded analysis of the genome identified 206 genes with DNA variants present in at least 12 of the 16 participants with combined annotation dependent depletion (CADD) scores greater than 20 (a predictor of how deleterious to protein function a variant might be; Supplementary Fig. S13C). STRING analysis of these variants suggested the corresponding genes were enriched for involvement in cell adhesion, neuron development, and mitochondrial depletion (Supplementary Fig. S13D–F). Comparing the variant list with the RNAseq results indicated that genes which contained DNA variants and differentially expressed transcripts were led by the known modulator of NADPH metabolism FMO2 (Bailleul et al., 2021) and PAX6, which may mediate AD-relevant pathways in epithelial-to-mesenchymal transition (Anderson et al., 2020, Jin et al., 2020) (Supplementary Fig. S13G–I). Sequence analysis of the mitochondrial genomes did not identify any common non-synonymous variants within our cohort (Table S3).

Discussion

Our results indicate that TSW is a distinct, iatrogenic dermatopathy deserving of further investigation. The strength of this investigation is the shared pathogenic signals spanning multiple modalities. Our clinical evaluations identified that protracted neuro-inflammatory symptoms distinguish TSW from AD; metabolomics and transcriptomics of skin biopsies each suggested that neuro-inflammatory pathways are associated with complex I mediated oxidation of NAD+; cellular and mouse models demonstrated that NAD+ metabolism was proinflammatory and glucocorticoid responsive; while functional assays demonstrated that the metabolic effects of glucocorticoids on the only cell type which aligns with the distribution and duration of TSW pathology could be mitigated by complex I blockade. Ultimately, these results informed a successful open-label trial using complex I inhibiting medication.

Our working hypothesis of TCS induced complex I hyperactivity in follicular stem cells is supported by the established effects of glucocorticoids on mitochondria (Choi and Han, 2021) and the clinical data detailing the distribution and duration of TSW symptoms (Barta et al., 2023, Brookes et al., 2023, Hwang and Lio, 2022, Sheary and Harris, 2020). In some individuals, prolonged use may result in a form of endogenous chemical irritation due to overproduction of oxidized NADH (NAD+). The enhanced NADH production generated by tryptophan degradation results in an increase in neuro-toxic kynurenine metabolites via the corticosteroid-responsive enzymes IDO and TDO (Qin et al., 2018). We hypothesize that epigenetic changes may underly the protracted nature of TSW recovery, however cell specific analysis will be needed to further elucidate our findings beyond the stem cell pathology identified.

A hypothesis predicated on NAD+ is consistent with each of the proposed phases of TSW (Barta et al., 2023): the initial rise in NAD+ combined with the resultant anticholinergic effects of kynurenic mediators may lead to the early signs of flushing, anhidrosis, palpitations, neuralgia, and vision changes (Lopez-Alvarez et al., 2019). Subsequent Wnt overactivity, mitochondrial inflammatory reactive oxygen species, and TRP receptor activation would be expected to induce desquamation, pruritus, skin proliferation, changes in temperature sensation, and may make those with TSW more susceptible to effects of chemical triggers associated with pruritic skin diseases (Yadav et al., 2023).

Our work suggests clinical insights for practitioners treating patients experiencing worsening eczematous skin disease despite >4 months of TCS treatment. Specifically, our data indicates that the diagnosis of TSW should be made for patients with eczematous skin disease in the presence of at least 1 of our identified major criteria and 3 of the minor criteria. The prior claims of disease limited to the face or genitals (Hajar et al., 2015) are not supported by our data and should not guide diagnosis. Patients with the identified TSW predictive symptoms should be evaluated for the diagnosis and might be considered for treatment using mitochondrial complex I inhibitors; however, overall our data supports only the progression to placebo trials to evaluate the safety and efficacy of these treatments. Although topical berberine was not attempted in this study, the efficacious concentrations in vitro are akin to dissolving one 500 mg capsule in a liter of water and then adding ¼ - 1 cup of the resultant solution to a standard 60-gallon bathtub as is done with bleach baths in AD (Panel et al., 2024). Exposure to heat may also be a speculative intervention given higher temperatures preferentially inhibit mitochondrial complex I (Chretien et al., 2018). However, TSW shares several biochemical features with ‘routine’ AD, including microbiome signatures, the types of cellular infiltrates, and an increase in Th2 cytokines, the latter is consistent with reported utility of dupilumab in TSW (Arnold et al., 2018).

Our work is limited in several ways. First, querying TSW-specific symptoms from eczematous patients who deny having TSW will be required to validate the diagnostic criteria through calculating specificity and accuracy. The small pilot enrollment and open label case series may not fully capture either the pathology or treatment response for a disorder with a reported diversity of presentation and large population of TSW. Our global subjective assessment of TSW improvement should be replaced by prospective severity indices specific to TSW which incorporate scores for thermodysregulation, burning, and flushing as well as validated metrics for quality of life. Additionally, larger studies will be needed to test whether the proposed DNA variants are specifically associated with TSW and, if so, what the biologic consequences of those variants might be and to help distinguish the variability in clinical response seen in our cohort. Similarly, the statistical association with Protopic (tacrolimus; Fig. 1A) will require targeted evaluation. Furthermore, our clinical symptoms, metabolomics, and microbiome were contrasted against both HV and patients with AD, however our transcriptomic data was contrasted against HV alone; however, the transcriptomic findings for our TSW cohort are distinct from the vast literature for AD, which does not implicate mitochondrial metabolism (Guo et al., 2023, Mitamura et al., 2023, Schabitz et al., 2022, Yu and Li, 2022). Similarly, although the severity of skin disease was greater in our TSW cohort than their AD comparators, our findings are distinct from the numerous reports of lipid-centric defects seen in severe AD (Zhang et al., 2023). Although the medications used in out pilot case series were targeted in their selection, additional mechanistic insights could come from reassessment of metformin’s complex I dependance (Wheaton et al., 2014); use of HFSC and co-assessment with berberine may help clarify if mechanisms beyond mitochondrial inhibition are involved.

Importantly, while our case series suggests expanded clinical assessments are warranted, no patient reported complete resolution of symptoms. While prolonged treatment may yield continued improvement, preemptive caution may be warranted for exposures longer than 4 consecutive months and identifying preventive strategies and predictive factors should remain paramount.

Materials and Methods

Survey re-analysis

Survey data was collected and processed as previously described (Barta et al., 2023). Statistical re-analysis was conducted in R, using the software packages: tidyverse (https://joss.theoj.org/papers/10.21105/joss.01686) and glmnet (https://www.jstatsoft.org/article/view/v033i01). Only complete surveys were included. Least absolute shrinkage and selection operator (Lasso) was used to select a subset of 154 demographic, treatment, and disease progression variables correlated with TSW development. Briefly, Lasso is similar to linear regressions but uses a penalization term that reduces the coefficients of some variables to zero. This method emphasizes variables with strong relationships to the outcome and decreases the risk of overfitting a model. During the model fitting, the penalization parameter was determined using 10-fold cross-validation.

Only participants who reported having TSW were asked about their experience with TSW symptoms. Therefore, we were not able to calculate specificity of diagnostic criteria. Instead, major criteria were selected by a clinician as symptoms which were present in at least 80% of the participants with TSW and were distinct from atopic dermatitis (AD). Of this, we excluded the symptoms of Fatigue and Suicidal Ideation from the diagnostic criteria as these symptoms were more general. The remaining symptoms were classified as minor criteria. From the minor criteria we again eliminated insomnia, emotional distress, appetite changes, and reactions to new triggers because we felt that these symptoms were nonspecific, secondary to skin disease, and/or could be a manifestation of AD. The prevalence of these symptoms across TSW participants was estimated and the optimum combination was selected, having a sensitivity of 92%.

Patients and Controls

Participants with TSW and heathy controls were recruited on clinical protocol NCT04864886 which was approved by the institutional review board (IRB) of the National Institutes of Health (NIH). Patients were recruited in collaboration with the International Topical Steroid Awareness Network. Written, informed consent for use of the images presented was provided by the participants (or the guardian in the case of the image with a minor). Powering was performed using an anticipated difference in alpha diversity of the microbiome. A Chao2 was predicted as 250 in controls with a standard deviation of 25; by comparison, patients with TSW were expected to have a Chao2 of 215. For a continuous endpoint, independent sample study with an alpha error of 0.05, 11 patients per group would provide 90% power; while NCT04864886 continues to enroll patients with AD or primary immune deficiency, after enrolling the subpopulation for the TSW specific evaluations, enrollment for the TSW subanalysis was closed. After informed consent was obtained, participants underwent a complete history and physical.

Inclusion criteria were:

1. Has documentation of symptoms meeting the previous TSW case reports.

2. Aged 18 to 75 years.

3. Willing to allow storage of blood, biopsy tissue, bacterial and fungal cultures, and any other samples collected for future research.

4. Able to provide informed consent.

Exclusion criteria were:

1. Current or prior (within 3 months) anticoagulant or anti-platelet therapy (other than aspirin or non-steroidal anti-inflammatory drugs).

2. Current or prior (within 3 months) use of immunomodulatory drugs (eg, chemotherapy, steroids), except if approved by the principal investigator.

3. History of keloid formation.

4. Pregnancy, lactating, or breastfeeding.

5. Any condition that, in the opinion of the investigator, contraindicates participation in the study.

Blood work up included complete blood count with differential, chemistry panels, and IgE levels. Two 2mm skin punch biopsies were obtained. One biopsy was embedded in OCT freezing media and flash frozen using liquid nitrogen. Both biopsies were from lesional skin in patients with TSW and AD, but (definitionally) uninvolved skin for controls. Samples were sent to Histoserv (Germantown, MD) for sectioning on intelislides (Bruker Daltonik) and stored in a −70°C freezer for later use. The second biopsy was stored in RNALater (Fisher Scientific), flash frozen in liquid nitrogen, and stored at −70°C until ready for RNA extraction. Participants were instructed to avoid any topical products for 48 hours prior to the clinical visit and sample collection. Each participant was also assured to have been free from antibiotic exposure for the proceeding 2 months.

Slides for MALDI analysis were placed in a desiccator for 1 hr and brought up to room temperature. Matrix deposition of 2,5-dihydroxybenzoic acid matrix (Sigma Aldrich) 20 mg/mL dissolved in 70% acetone with 0.1% TFA was applied using the same robotic sprayer resulting in a matrix density of 0.003111 mg/mm². Mass spectrometry was carried out in positive ion mode with an m/z range of 400 to 2,000 (high molecular weight; HMW) and then again at 20–450 m/z (low molecular weight; LMW) with a spatial resolution of 100 μm. The feature list and intensities for both LMW and HWM were combined prior to analysis.

Cell culture

Human follicular stem cells (CellProgen, Torrance, CA), 3T3 fibroblasts, HaCaT keratinocytes, and Schwann nerve cells (ATCC, Manassas, VA) were purchased and cultured per manufacturer’s instructions. Nicotinic acid (Sigma) in water or 10–25uM clobetasol (Sigma Aldrich) in DMSO were added to the respective culture media for select experiments. DMSO was used at a final concentration of 0.1% v/v. Cells were cultured in T25 Flacon flasks (Fisher), Seahorse plates (Agilent, Santa Clara, CA), or imbidi 8 well removable chamber slides (Fitchburg, WI). Mitochondrial stress test was performed per manufacturer instructions using Seahorse XF Pro (Agilent). For MALDI, the cells were seeded at a density of 50,000 cells/well in 8 well chamber labtek culture slides (Lab-Tek II, CC2. ThermoFisher Scientific). Three trials of the same experiment were carried out for reproducibility. Cells were washed with 1x BPS, fixed with 4% PFA for 15 minutes and washed a second time with PBS and placed in a low-pressure desiccator for 1 hour. Optical images of each well were taken prior to matrix deposition of 2,5-dihydroxybenzoic acid matrix (Sigma Aldrich) 15 mg/mL dissolved in 70% ACN with 0.1% TFA and was applied using a TM-Sprayer robotic sprayer (HTX Technologies). The estimated matrix density was 0.001611 mg/mm². Mass spectrometry data was collected on a MALDI timsTOF Fleximager (Bruker Daltonik, Bremen, Germany) operated in TIMS qTOF positive ion mode from m/z 20 to 1,100 with a spatial resolution of 30 μm, mass measurement error <5 ppm, and 40,000 resolving power. All MALDI imaging data were visualized using SCiLS Lab Version 2021 (Bruker Daltonics). Denoising was carried out by moving the sliding window feature in SCiLS for the maximum number of peeks for each experiment. The average ion signals up-regulated in the experimental group vs the average signals found in the control were ordered based on threshold and important into Metaboanalyst for pathway identification using the functional analysis tool.

Metabolic pathway analysis

Pathway identification was performed using MetaboAnalyst (https://www.metaboanalyst.ca/) which determines the probability of a pathway being precent based on MS1 features associated with multiple metabolites within the reaction hierarchy of a specific pathway and adjusts the reported significance for the number of pathways analyzed. The output of MetaboAnalyst (functional analysis), index of pathway significance (IPS) values was calculated as previously described (Ratley et al., 2024, Yadav et al., 2023). IPS was calculated by ranking the metabolites most indicative of the TSW cohort versus both AD and HV using the ROC feature of SCiLS (Bruker); then repeated after ranking metabolites by those most predictive of the HV and AD cohort compared to TSW.

Healthy control glucocorticoid exposure

For controlled glucocorticoid skin exposure, healthy participants were enrolled on the NIAID IRB approved NCT02798523. 19 healthy volunteers ages 18–64 were recruited and, after informed consent was obtained, were treated with topical methylprednisolone aceponate (Advantan emulsion 0.1%; Bayer). A baseline 3mm skin punch biopsy was obtained from one arm, then topical methylprednisolone was applied to a limited area of skin in the contralateral arm. An additional 3mm skin biopsy was obtained 4 hours after drug administration, from the area where topical methylprednisolone was applied. With each biopsy, the epidermis layer was collected in tissue-grinding CKMix50-R 2ml tubes containing beads (Bertin Corp) in 500 uL TRIzol (ThermoFisher Scientific) and homogenized immediately using a Precellys 24 machine (Bertin Technologies) for 2 cycles of 5000 rpm for 20sec with a 90 sec break on ice between cycles. Samples were stored without the beads in TRIZol at −70°C until the time of RNA purification.

Serum Metabolomics

Serum was diluted 1:10 in 70% methanol (Fisher Chemical; Waltham, MA) with 0.1% trifluoroacetic acid (Thermo Fisher; Waltham, MA) and 30mcg/mL of dihydroxybenzoic acid (Sigma; St. Louis, MO). 1mcL of the matrix and serum mixture was plated on a spot plate (Bruker, Billerica, MA). Serum was run once for m/z range of 20–650, then replated and run again at 650–1700m/z. The data were analyzed separately except for analysis of ROC (performed using SCiLS; Bruker) in which the combined data set was ranked by ROC before pathway analysis by MetabAnalyst(Pang et al., 2021). Both spot plate serum and skin biopsies were analyzed by MALDI-TOF imaging mass spec as previously described on positive ion mode(Yadav et al., 2022). Collisional cross section identification was performed using MetaboScape (Bruker). Pathway disruption was visualized using Pathview(Luo and Brouwer, 2013, Luo et al., 2017).

Microbiome

Skin swabs for microbiome was performed using Floq swabs (Copan; Murrieta, CA) moistened with normal saline (Sigma). DNA was extracted and sequenced by COSMOSID (Germantown, MD) then analyzed as previously described(Chaudhary et al., 2023).

Serum cytokines

Serum cytokines were analyzed using the O-Link system (Uppsala, Sweden) performed under contract with Vanderbilt University (Nashville, TN).

Tissue Section Preparation and Immunostaining

Fresh frozen tissue sections of patient skin biopsy samples were fixed with BD Cytofix/Cytoperm solution (BD Bioscience, Cat#: 554722) diluted 1:4 in PBS for 30 minutes at 4°C. Healthy human skin samples were kindly provided by Dr. Jonathan M. Hernandez, Surgical Oncology Section, National Cancer Institute. Sections were washed in 1X PBS for 30 minutes before blocked in 1% BSA and 1% mouse serum in 0.1 M pH 7.4 Tris buffer containing 0.3% Triton X-100 for 30 minutes at room temperature. Then, tissue sections were incubated with directly conjugated primary antibodies (listed below) diluted in blocking solution for 10–15 h at 4°C, washed in 0.1 M pH 7.4 Tris buffer and mounted with SlowFade Gold Antifade Mounting solution (ThermoFisher Scientific Cat#S36937) using #1.5 cover glass (VWR, Cat#: 48393–241). A combination of the following fluorescent chromophores were used for immunostaining: Brilliant Violet 421, Brilliant Violet 510, Violet 450, Brilliant Violet 570, Brilliant Violet 650, PE, Alexa Fluor 488, Alexa Fluor 647, Alexa Fluor 700, Alexa Fluor 790, eFluor 570 and eFluor 615. Antibodies included from BioLegend: mouse anti-human CD1c-BV421 clone: L161, Cat# 331526; Mouse anti-human HLA-DR-BV570 clone: L243 Cat# 307638; Mouse anti-human CD69-AF700 clone: FN50 Cat# 310922; Mouse anti-human CD11c-BV510 clone:S-HCL-3 Cat# 371514. From eBioscience: Mouse anti-human FoxP3-eF570 clone: 236A/E7 Cat# 41-4777-82; Mouse anti-human CD20-eF615 clone: L26 Cat# 42-0202-82; Mouse anti-human CD103-PE clone: B-Ly7 Cat# 12-1038-42. From R&D Systems: Goat anti-human CD4-AF488 poly clone Cat# FAB8165G. From BD Biosciences: Mouse anti-human CD3-BV650 clone: SP34–2 Cat# 563916; Mouse anti-human CD8-V450 clone: RPA-T8 Cat# 560347; Mouse anti-human CD56-AF647 clone: B159 Cat#557711. From Santa Cruz Biotechnology Mouse anti-human CD68-AF790 clone: KP1 Cat# sc-20060 AF790.

Whole Skin Preparation, Immunostaining and Clearing

Fresh healthy skin tissue blocks were fixed in BD Cytofix/Cytoperm solution (BD Bioscience, Cat#: 554722) diluted 1:4 in PBS for 5 days at 4°C, washed in PBS 3X and left in PBS for 10 hours at 4°C. Then, the tissue blocks were transferred to PBS containing 30% glucose and left for 2 days. The tissues were then embedded in OCT and stored in −80 °C before use. A 300 micron thick skin cross sections were cut by use a Leica CM1950 cryostat. Immunostaining and clearing of the thick skin sections were performed using the protocol previously described(Li et al., 2019). The following fluorescent dye directly conjugated primary antibodies were used for immunostaining: CD8-V450, HLADR-AF647, CD4-AF488, CD3-iF594 and CD69-AF700.

Confocal Microscopy Imaging

Digital scan of both the thin and healthy thick skin sections was performed using an inverted Leica TCS SP8 X confocal system equipped with an 80 MHz pulsed white light laser, 4 Gallium-Arsenide (GaAs) Hybrid Detectors (HyDs) and 1 multialkali photomultiplier tube (PMT) with spectral detection capability. A 40X (NA 1.3) oil emersion objective lens was used for scanning the thin skin sections with a pixel size of 568.74X568.74 mm² and pixel dwell time of 1.2 ms. A 20X (NA 0.75) multi-emersion objective lens was used for volume scanning of the thick skin sections with a voxel size of 1.139X1.139X2.0 mm³. Pixel dwell time was maintained at 1.2 ms.

Image Processing and Visualization

Digital images acquired by the Leica confocal system were tile stitched and then processed to correct signal spillovers from nearby channels and autofluorescence signal within the skin by using the manual unmixing method of the Automatic Dye Separation function within the Leica Application Suite X (LAS X) software package (version 4.4.0.24861). The output images (.lif files, Leica file format) were converted into Imaris version 5.5 files by Imaris software (Version 9.5.0, Bitplane) and imported into Imaris. Images of individual channels were passing through a Gaussian filter to remove random noise before pseudo-color assignment, visualization and animation creation. In all image processing steps, image size and aspect ratio were maintained identical by keeping the pixel or voxel sizes unchanged.

IF staining TRPA1

Human skin biopsies were freshly frozen after embedding in the OCT medium and sent out for sectioning (10µm) (Histoserv Inc., Germantown, USA). Followed by placing the slides in the desiccator to dry them and placing them into the PBS solution to remove the OCT. Skin biopsy we fixed using 4%PFA for 30 minutes, followed by washing with PBS and permeabilized with 0.5% tritone X100 for 20 minutes. Blocking was performed using with 5% Normal Goat Serum (NGS) for 1 hour. Primary antibody (#ACC-037 Alomone labs) solution was added according to the manufacturer’s instruction at 4°C overnight. Slides were washed 3 times for 15 minutes each in PBS. A secondary antibody (Alexa flour 488 or 568) solution was added according to the manufacturer’s instructions for 1 hour at room temperature. Slides were washed 3 times for 15 minutes each in PBS. DAPI solution was added and incubated for 15 minutes at room temperature. Slides were washed 3 times with PBS and were mounted with mounting medium.

Genome sequencing

Genome sequencing and analysis was performed as previously described(Ghosh et al., 2022). VCF files were filtered using bcftools(Danecek et al., 2021). CADD scores were calculated for all variants(Schubach et al., 2024). Genes for which at least 12 of 16 patients had at least one heterozygous or homozygous variant with a PHRED-scaled CADD score of 20 were submitted to STRING for functional enrichment analysis(Szklarczyk et al., 2023). These genes were also cross-referenced to RNA-seq data to identify variants which were associated with changes in gene expression. Gene maps of the identified genes, PAX6 and FMO2, as well as TTN were generated using trackViewer in R(Ou and Zhu, 2019).

RNAseq

For biopsies of healthy controls exposed to glucocorticoids in a controlled setting, punch biopsies were homogenized for 40s in lysing matrix D tubes (MP Biomedicals, Santa Ana, CA) containing 1,000μl Trizol (Thermofisher Scientific, Waltham, MA) in a FastPrep^® FP 120 instrument (MP Biomedicals) at speed 6.0 meters per second. Homogenized Trizol lysate was combined 1-Bromo-3-chloropropane (MilliporeSigma, St. Louis, MO), mixed, and centrifuged at 16,000 x g for 15 min at 4°C. RNA containing aqueous phase was collected from each sample and passed through QIAshredder column (Qiagen, Valencia, CA) at 21,000 x g for 2 minutes to homogenize any remaining genomic DNA in the aqueous phase. Aqueous phase was combined with equal amount of RLT lysis buffer (Qiagen, Valencia, CA) with 1% beta mercaptoethanol (MilliporeSigma, St. Louis, MO) and RNA was extracted using Qiagen AllPrep DNA/RNA mini columns (Valencia, CA). RNA integrity was assessed using the Agilent 2100 Bioanalyzer using RNA 6000 Pico kit (Agilent Technologies, Santa Clara, CA). RNA was quantitated using a fluorescence assay (Quantit RiboGreen RNA, Thermofisher Scientific, Waltham, MA) on a Tecan Spark multiplate reader (Tecan, Switzerland).

Sequencing libraries were generated using the SMARTer Stranded Total RNA-Seq Kit v3-Pico Input Mammalian (Takara Bio USA, Inc., San Jose, CA) following the manufacturer’s protocol without modification. RNA input for NGS library preparation was 10 ng for the TSW samples and 3.75 ng for the punch biopsy samples. The TSW samples were amplified for 12 cycles and the punch biopsy samples were amplified for 13 given the amount of RNA in the preparation. Microfluidic electrophoresis for library quality control was performed on a TapeStation 4200 (Agilent Technologies, Santa Clara, CA) with D1000 screen tapes. Two microliters of each library were combined and sequenced on the MiSeq (Illumina, San Diego, CA) using a Micro v2 300 cycle chemistry kit to acquire the reads/uL for each library. The MiSeq data was used to create a normalized library pool for sequencing on the NovaSeq X Plus (Illumina, San Diego, CA) with 200 cycle chemistry.

Using the CogentAP NGS Analysis v2.0 pipeline (https://www.takarabio.com/products/next-generation-sequencing/bioinformatics-tools/cogent-ngs-analysis-pipeline ), we first processed raw fastq files to trim UMIs and append UMI sequence to the read ID. The pre-processed fastq files were then trimmed for quality and adapter contamination using Cutadapt v4.0(Martin, 2011). Trimmed reads were mapped to the hg38 reference genome and Gencode GRCh38 v.42 transcriptome using STAR v2.7.6a(Dobin et al., 2013) in two-pass mode, and PCR duplicates were flagged using Picard MarkDuplicates from GATK4 v4.2.6.0(Auwera and O’Connor, 2020). Gene-level expression quantification was performed using RSEM v.1.3.0 (Li and Dewey, 2011). Differential gene expression was carried by using DESeq2 package in R. Functional enrichment analysis of significantly different transcripts was performed using STRING. Pathway disruption was visualized using Pathview.

Mice

Animal experiments were approved and monitored by the Animal Safe Practices committee of the National Institutes of Health. Male and female BALBc/J mice aged 6–11 weeks (age and sex matched within each experiment) were purchased from Jackson Labs (Bar Harbor, MN). Nicotinic acid (Sigma) was diluted in water to final concentrations of 5M, 500mM, and 5mM and applied topically to each mouse ear at 10mcL per daily for 8–10 days. Thickness was measured using calipers (Matsui; Shiga, Japan) on day 9–11.

Berberine dilution

1 capsule of berberine (Solaray; Jim Beck, Utah) was dissolved in 1L of sterile water (Sigma). Based on the molecular weight, the resultant concentration was calculated at 1.5M. Dilutions were performed to achieve 0.6uM to align with the cell culture lowest effective dose. Images were taken after 15min of dilution.

Data availability statement

Shotgun metagenomics and RNA seq datasets generated during the current study are available in the NCBI under the BioProject ID: PRJNA1098044. Metabolomic data will be accessible via MetaboLights accession number MTBL9913.