Wu et al. Genome Medicine (2020) 12:73 https://doi.org/10.1186/s13073-020-00768-9 RESEARCH Open Access A population-based gene expression signature of molecular clock phase from a single epidermal sample Gang Wu1, Marc D. Ruben1, Lauren J. Francey1, David F. Smith2,3, Joseph D. Sherrill4, John E. Oblong4, Kevin J. Mills4 and John B. Hogenesch1* Abstract Background: For circadian medicine to influence health, such as when to take a drug or undergo a procedure, a biomarker of molecular clock phase is required––one that is easily measured and generalizable across a broad population. It is not clear that any circadian biomarker yet satisfies these criteria. Methods: We analyzed 24-h molecular rhythms in human dermis and epidermis at three distinct body sites, leveraging both longitudinal (n = 20) and population (n = 154) data. We applied cyclic ordering by periodic structure (CYCLOPS) to order the population samples where biopsy time was not recorded. With CYCLOPS- predicted phases, we used ZeitZeiger to discover potential biomarkers of clock phase. Results: Circadian clock function was strongest in the epidermis, regardless of body site. We identified a 12- gene expression signature that reported molecular clock phase to within 3 h (mean error = 2.5 h) from a single sample of epidermis––the skin’s most superficial layer. This set performed well across body sites, ages, sexes, and detection platforms. Conclusions: This research shows that the clock in epidermis is more robust than dermis regardless of body site. To encourage ongoing validation of this putative biomarker in diverse populations, diseases, and experimental designs, we developed SkinPhaser––a user-friendly app to test biomarker performance in datasets (https://github.com/gangwug/SkinPhaser). Keywords: Skin, Dermis, Epidermis, Circadian medicine, Population rhythm, Circadian biomarkers Background aims to incorporate this knowledge to improve treat- In the last 50 years, dozens of clinical studies showed ment outcomes. This requires a reliable measure of a that dosing time-of-day can impact the efficacy and patient’s molecular clock phase as wall time does not safety of many different types of medical treatments [1, necessarily equate to clock gene phase. Accumulating 2]. We know from multi-tissue studies in mice [3]and evidence shows interpersonal variation in the timing of humans [4] that thousands of rhythmically expressed physiology and behavior due to genetics, age, sex, and genes encode known drug targets. Circadian medicine lifestyle [5–8]. Robust markers of the molecular clock phase are therefore in high demand. Previous research focused on dim-light melatonin- * Correspondence: [email protected] onset (DLMO) as a marker of suprachiasmatic nucleus 1Divisions of Human Genetics and Immunobiology, Center for Circadian Medicine, Department of Pediatrics, Cincinnati Children’s Hospital Medical (SCN) phase [9, 10]. However, DLMO is inconvenient, Center, 240 Albert Sabin Way, Cincinnati, OH 45229, USA costly, difficult to standardize, and thus rarely used Full list of author information is available at the end of the article © The Author(s). 2020 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Wu et al. Genome Medicine (2020) 12:73 Page 2 of 12 clinically. With the development of high-throughput the associated protocol was approved by an Institutional molecular detection platforms and computational tech- Review Board (Aspire; http://aspire-irb.com/). Participants niques, recent efforts shifted to machine learning predic- donated one skin punch biopsy at each of four time points tions based on the transcriptome or metabolome from (6 AM, 12 PM, 6 PM, and 12 AM) over a 24-h period. one or two samples of whole blood [11–16], or a specific Biopsies were separated into epidermal and dermal layers blood cell type (e.g., CD14+ monocytes) [17]. This by LCM. The mRNA extraction, target labeling, and approach can predict DLMO phase to within 3 h from a hybridization to microarrays were described previously single blood sample [14, 17]. However, studies were [21]. small (≤ 74 participants) and limited to younger partici- pants (18–41 years of age). Developing and validating MetaCycle analysis of longitudinal data blood-based biomarkers of DLMO phase in the broader The RMA algorithm from the Affy R package [22] population remain a challenge. Additionally, it is not was used to extract the expression profile from the clear what the melatonin phase tells us about clock gene raw CEL files of 79 human longitudinal arm dermis phase in other tissues. samples (Additional file 1: Table S1). Expression pro- Previously, we showed human skin has a more robust files were analyzed with the meta3d function of Meta- circadian clock than blood, and developed a 29-gene Cycle (Additional file 1: Table S2) R package [23] expression signature from ~ 300 human forearm epider- using default settings, except “cycMethodOne” = “ARS,” mis samples. From a single biopsy, this set estimated “minper” = 24, and “maxper” = 24. ARSER [24] was used molecular clock phase in epidermis to within 3 h [18]. to analyze the time-series data individual by individual. However, for clinical use, critical questions remain such Then, meta3d was used to integrate the analysis results as whether this expression signature performs across from 19 participants (participant 115 had one missing different skin locations, in different populations, and on time point and was excluded from the longitudinal different platforms. analysis). Circadian genes were defined by P <0.05 and Using an experimental design that captures the ad- relative amplitude (rAMP) > 0.1. We applied a less strict vantages of both longitudinal and population-based cutoff (P < 0.1 and rAMP > 0.1) for evaluating overlap studies, we analyzed 24-h molecular rhythms in human between human epidermis and dermis, and phase set dermis and epidermis at three distinct body sites. enrichment analysis (PSEA) [25]. Salivary melatonin and Participants in the longitudinal group (n = 20, aged cortisol were also measured every 3 h over 24 h for 16 of 21–49 years) each donated one skin punch biopsy 20 participants (participants 101, 102, 103, and 104 only every 6 h across a day. Participants in the population have 3 or 4 measures) [18]. MetaCycle’smeta2dfunction group (n = 154, aged 20–74 years) each donated one predicted melatonin and cortisol phases (Additional file 2: punch biopsy from the forearm, buttock, and cheek Table S3) for these 16 participants using default settings, without recording time. For all biopsies, the dermis except “minper” =24and“maxper” = 24. We note that the was separated from the epidermis by laser capture relatively low sampling resolutions of skin biopsy and microdissection (LCM) and then profiled on gene ex- salivary collections reduce the accuracy of MetaCycle pression arrays. We applied cyclic ordering by periodic phase predictions. structure (CYCLOPS) [19]toorderthepopulation samples where biopsy time was not recorded. Comparing clock robustness between human epidermis We found that circadian clock function was strongest and dermis in the epidermis, regardless of body site. Based on this, We used the microarray dataset from a previous we applied ZeitZeiger [20] and identified a potential bio- population study of human epidermis and dermis marker set from a single epidermal sample that reported (Additional file 1:TableS1)[21]. This study recruited molecular clock phase in the skin to within 3 h. This set 154 Caucasian females (aged 20–74 years) in the performed well across body sites, ages, sexes, and detec- USA. Three punch biopsies were collected from each tion platforms and represents a forward path to clinical participant, with one biopsy per body site (arm, but- application of circadian medicine. tock, and cheek). Sample collection times were not recorded. Epidermal and dermal layers were separated Methods by LCM. This yielded six groups of skin samples: arm Collection of human longitudinal dermis samples epidermis, buttock epidermis, cheek epidermis, arm Twenty healthy Caucasian male participants from the dermis, buttock dermis, and cheek dermis. We mea- USA were recruited for the longitudinal dermis sample sured pairwise clock gene correlations [26] between collection. These were the same 20 participants in our epidermis and dermis at each body site. A reference previous collection of longitudinal epidermis samples correlation matrix of 17 mouse clock and clock-associated [18]. All participants provided informed consent, and genes was computed from 12 mouse tissues [3]. The Wu et al. Genome Medicine (2020) 12:73 Page 3 of 12 correlation matrix of 17 human homolog genes was com- seed set #2. Circadian genes were selected with FDR < puted for each skin group. To evaluate clock robustness in 0.05, fitmean > 16, rAMP > 0.1, and goodness-of-fit each skin group, we computed a Mantel test statistic (rsq) > 0.1 from the cosinor regression analysis results between the reference correlation matrix and the correl- of matched epidermis and dermis samples with ation matrix from each skin group.