Abstract
Expression levels of circadian clock genes, which regulate 24-hour rhythms of behavior and physiology, have been shown to change with age. However, a study holistically linking aging and circadian gene expression is missing. Using the colonial chordate Botryllus schlosseri, we combined transcriptome sequencing and stem cell-mediated aging phenomena to test how circadian gene expression changes with age. This revealed that B. schlosseri clock and clock-controlled genes oscillate organism-wide, with daily, age-specific amplitudes and frequencies. These age-related, circadian patterns persist at the tissue level, where dramatic variations in cyclic gene expression of tissue profiles link to morphological and cellular aging phenotypes. Similar cyclical expression differences were found in hundreds of pathways associated with known hallmarks of aging, as well as pathways that were not previously linked to aging. The atlas we developed points to alterations in circadian gene expression as a key regulator of aging.
One Sentence Summary The Ticking Clock: Systemic changes in circadian gene expression correlates with wide-ranging phenotypes of aging
Circadian clocks are systems that oscillate with a consistent phase and control daily functions at the cellular, tissue, and organismal levels. Consisting of a small number of core clock genes operating within transcription-translation feedback loops, they activate and repress the expression of hundreds of downstream genes and regulate the response of cells to daily and seasonal fluctuations in light and temperature (1–2). Proper functioning of the circadian rhythm system is essential for maintaining cellular and tissue homeostasis. The robustness of this system is reduced with age, resulting in disturbed circadian rhythms in old organisms relative to young adults (3). This disrupted circadian control leads to disparities in the rhythms of waking activity, core body temperature, suprachiasmatic nucleus (SCN) firing, release of hormones such as melatonin and cortisol, and more (4). Growing molecular evidence indicates that the circadian clock and aging process are closely associated (3–8). The reduction in the expression levels of the core clock genes Bmal1 and Per2 in the mammalian SCN has been found to be age-associated (5). The expression patterns of the other core clock genes in the aged mammalian SCN is less clear. Reports have indicated unchanged expression levels, reduced expression levels, or period lengthening (7, 9). Changes in the expression patterns of the core clock in both aged and peripheral tissues have been inconsistent as well, with different species and even different tissues showing varying expression profiles (4, 7, 8). Despite these inconsistencies, both aging and circadian rhythms are embedded in every tissue and organ of the body. Therefore, they must operate on the organismal level at some capacity. For this reason, we expect the molecular changes throughout the day due to age to be detectable on a systemic level. However, limitations of sampling and processing whole body tissues over time and age in conventional aging models make it difficult to directly monitor and study the important effect of changes in circadian rhythms on aging phenotypes.
Colonial chordates like Botryllus schlosseri provide a key to bridge this gap. This marine organism belongs to the chordate subphylum Urochordata, a taxonomic group considered to be the closest living invertebrate relative of vertebrates (10). This group shares homology with at least 75% of the human gene repertoire (11).
B. schlosseri exhibits several characteristics that render it a valuable model for the study of the impact of changes in circadian rhythms on aging. These include: (i) A long life span, with colonies in our lab living up to 20 years; (ii) Clonal asexual reproduction (Fig. S1), which allows separation of one genotype (colony) into several clonal replicates; (iii) Weekly stem cell-mediated asexual tissue regeneration of all body organs (12–16; Fig.S1), allowing for the study of stem cell aging’s effect at the organism level; (iv) Easy, serial collection of whole-body samples along their entire lifespan, which offers a unique approach to characterize genetic changes across time and age. Using B. schlosseri, this study sets out to holistically characterize the link between aging and circadian rhythms at the system, tissue, and organism level. Aiming to link organism-wide aging phenotypes and specific changes in circadian gene expression, we first characterizedB. schlosseri aging and circadian behaviors. Morphological phenotypic changes observed in young versus old B. schlosseri colonies include changes in pigmentation level (transparent vs. highly pigmented), terminating blood vessel (ampullae) shape (oval vs. elongated), significantly reduced zooid size as measured by average zooid area, and decreased regeneration capacity which wanes with age (measured by number of individuals (zooids)) in a colony (Fig. 1A-C; S1D). To further identify potential cellular changes that characterize the biology of aging in B. schlosseri, we compared the proportion of 24 cell populations (as described in 15) between 3-5 days old (n=190) and 4-13 years old colonies (n=37) (Fig. 1E). This comparison revealed age-associated alterations in the cellular composition. Juveniles showed a higher level of enriched hematopoietic and germline stem cell populations compared to the higher level of engulfing phagocytic, cytotoxic, and pigment cells in aged colonies (Fig. 1E). We also found diurnal, circadian behaviors that show reduced nocturnal activity. Slower heart rate and siphon activity measured at night suggest that B. schlosseri is a diurnal organism. Young and mid-aged colonies demonstrated slower heartbeats at night (Fig. 1D) as well as infrequent siphon contractions (Movie S1) when compared to the day. Circadian changes in heart rate and siphon activity were not observed in aged colonies, demonstrating diurnal phenotypes diminishing with age. (Fig. 1D; Movies S1).
Next, we investigated the B. schlosseri genome (11), to identify its circadian photoreceptors and core clock genes(Table S1A; Fig. 2, S2). Three homologs of the vertebrate photoreceptor pinopsin/melanopsin were found. Homologs for core clock genes including multiple genes for βHLH - PAS domain transcription factors (clock/bmal/arnt) and multiple genes for clock regulators like casein kinase (ε and δ), Rev-erb, RorB, Glycogen synthase kinase 3 (GSK3; also known as Shaggy (sgg)), Timeless (tim), Timeout, Sirt, and Mtor, were also identified. These represent a mix of genes resembling either vertebrate or invertebrate sequences (based on sequence homology; Table S1A). However, orthologs for the core clock suppressors period and cryptochrome were not found in the B. schlosseri genome assembly, as well as in 14 other tunicates genomes sequenced to date (aniseed database; 17).
A comprehensive alignment of raw RNAseq B. schlosseri reads to period and cryptochrome alleles from diverse species (see methods for details), and a search for domains associated with both genes in the B. schlosseri genome also did not reveal putative homologous for these genes, suggesting that other clock/bmal regulators exist in tunicates.
After identifying key clock genes in the clock gene network, we characterized their cyclical expression and found it varied widely with age. We also identified genes with light-induced (entrained) expression (Table S1D), an important characteristic in circadian gene regulation. This was achieved through free-run and light pulse experiments which generated comprehensive, whole-organism transcriptional profiles from a total of 88 samples (multiple biological replicates, see annotations and gene counts in Table S1B-C and Methods). We identified clear, circadian expression of core-clock and clock-controlled genes using bioinformatics methods developed to find circadian (18) and chronologically differentially expressed genes (16 and Methods) (Fig. 2, S2; Tables S2-3). The amplitudes and frequencies of these oscillations differed consistently between ages (Fig. 2G-I, S2; Tables S2-3). The number of genes that cycle throughout the day is higher in the mid-age group compared to old and young age groups (Mid>Old>Young). Each age group has unique sets of genes that cycle (Fig. 2C; Table S3). Both elderly and young colonies show a reduced amplitude of change and higher expression levels of clock-controlled genes. In middle-aged colonies, the amplitudes and frequencies of the gene oscillations are greater than their young and old counterparts (Fig. 2C).
We studied cyclic gene dynamics at the tissue level by looking at the number of genes expressed at different time points in detailed tissue-specific gene profiles/enrichments(15–16; Table S4). This analysis revealed that the observed morphological and cellular aging phenotypes (Fig. 1) were accompanied by dramatic changes in the cycles of organ-specific gene expression (Fig. 3, Table S4).
Significant age-associated changes were found in the cycling dynamics of genes associated with the hematopoietic system, the endostyle, the germline stem cells, and the central nervous system (Fig. 3 and Table S4). This is exemplified by the clear age-associated, cyclical enrichment pattern in the hematopoietic stem cell (HSC) gene profile. A higher percentage of HSC associated genes are dynamically expressed with larger, opposite amplitudes in young colonies compared to mid and old age groups (Fig. 3A). The endostyle associated genes behave differently, where throughout the day, the highest percentages of enrichment expression are observed in old colonies (Fig. 3B). This raised expression could be due to the old colonies attempting to compensate for reduced enriched HSC’s numbers in their endostyle niches (13, 15) (Fig. 1E). Young colonies also showed a diminished cycling pattern in the endostyle gene profile. The dynamic cycling of the gene enrichment associated with the B. schlosseri ampullae/blood vessels was similar to that of the endostyle. More genes were consistently expressed in old and mid-age colonies compared to the young. However, young colonies showed an undiminished cycling pattern that was the opposite of the pattern seen in mid and old colonies (Fig. 3C). The percent of genes associated with putative germline stem cell populations was higher in old and mid-age colonies. This resembled the ampullae enrichment patterns, where young colonies showed the near exact opposite cyclical patterns to mid and old colonies (Fig. 3D). The central nervous system (CNS) associated genes behaved differently than previous tissues discussed. In this system, the mid-age group showed the highest enrichment throughout the day. Unique to the CNS gene profile, all three age groups’ cycling dynamics were relatively similar (Fig. 3E). This is particularly intriguing considering the role of the CNS as the center of clock activity in mammals (specifically the suprachiasmatic nuclei)(19). Investigating the specific genes associated with these patterns can point to CNS aging mechanisms by looking for deviations in the cycling dynamics of the mid-age group (Table S4).
This analysis revealed that tissue-specific gene expression oscillates in a daily cycle with different amplitudes and frequencies between ages. The number of genes that are expressed cyclically is generally higher in old and mid-age groups compared to the young. The age-associated tissue-specific oscillation patterns observed correlate with the age-specific oscillation patterns detected in genes associated with the circadian entrainment pathway (Fig. 3F). This demonstrates a strong link between clock-controlled genes and tissue activities.
After finding a link between aging and cyclic expression at the tissue level, we studied the relationship between the circadian clock and aging in hundreds of known pathways. This was made possible by having a comprehensive set of genes for each age and timepoint (Table S2A-C). These sets of genes were analyzed to find any associations to known pathways. We found over 450 pathways, all with some level of age-associated changes in their cycling dynamics (Table S5). Substantial circadian variation between age groups was discovered in the expression patterns of pathways linked to hallmarks of aging (20–21), cellular processes, and tissue functions. (Table S5; Fig. 4). The age-specific patterns of these pathways (Fig. 4) correlated with the age-specific patterns of the circadian entrainment pathway (Fig. 3F). These cyclic patterns also correlated with the age-specific trends observed in the longevity regulation path (Fig.4P). These correlations indicate a strong link between the circadian clock and aging, suggesting that clock genes are key regulators of aging.
Through building an aging-clock molecular atlas, this study generated the most complete transcriptomic database that links systemic changes in the molecular clock to aging to date. Using a model organism that is closely related to vertebrates with an intrinsic connection to stem cell aging (22), this atlas emphasizes the extensive cycling variation in gene expression across ages, showcasing the importance that clocks must play in regulating stem cell aging.
Funding
This study was supported by NIH grants R01AG037968 and RO1GM100315 (to I.L.W., S.R.Q., and A.V.), and by R21AG062948 to I.L.W and A.V. the Chan Zuckerberg investigator program (to S.R.Q., I.L.W and A.V), the Virginia and D. K. Ludwig Fund for Cancer Research, a grant from the Siebel Stem Cell Institute and a Stinehart-Reed grant (to I.L.W.). C.A was supported by a Postdoctoral Fellowship of the Larry L Hillblom foundation, by Stanford School of Medicine Dean’s Postdoctoral Fellowship. B.R. was supported by the Human Frontier Science Program Organization LT000591/2014-L and NIH Hematology training grant T32 HL120824-03, and by ISF grant 1416/19, and HFSP Research Grant RGY0085/2019.
Author contributions
Conception and design: A.V., Y.V., A.G., M.K., R.B., D.S.; mariculture, observation and sample collection: Y.V., A.G., K.J.I., K.J.P., A.V., C.A., B.R., T.G.,; RNA isolation and library preparation: K.J.P., A.V.; sequencing: N.F.N.; sequencing analysis and development of analytical tools: M.K., Y.V., A.G., D.S.,; flow cytometry: B.R.; writing of manuscript: A.V., A.G.,Y.V., M.K., R.B., I.L.W.; technical support and conceptual advice: A.V., R.B., D.S., N.F.N., S.R.Q., I.L.W.;
Competing interests
Authors declare no competing interests.
Data and materials availability
RNA-Seq data are available on the Sequence Read Archive (SRA) database: BioProject: PRJNA682759
Materials and Methods
Mariculture
Mariculture procedures have been described previously (23). Briefly, wild type Botryllus schlosseri colonies collected in Monterey bay, were tied to 3×5-cm glass slides and placed 5 cm opposite another glass slide in a slide rack. The slide rack was placed into an aquarium, and within a few days the tadpoles hatched, swam to the settlement slide, and metamorphosed into the adult body plan (oozooid). Single oozooids are then transferred to individual slides and grown at 18-20°C and under 14h/10h light/dark regimen (6am-8pm light/ 8pm-6am dark). Colonies were fed daily using a marine invertebrate diet prepared in the lab as described in Kowarsky et al. 2020).
Identification of B. schlosseri aging phenotypes
Pictures of young and old colonies were taken. Morphological changes were recorded (e.g. pigmentation level, blood vessels shape). Zooid sizes were measured using image J (mm2). T-test was used to find significant changes.
Recording diurnal behavior
We monitored heartbeat (per 30 sec) and siphon openings in zooids on developmental stage A1 during day (6pm light) and night (12am dark). Three ages were observed: 40 days old, (light n=12, dark n=10), 1515 days old, (light n= 9, dark n=6), and 6905 days old (light n=9 dark n=9). T-tests were used to find significant changes.
Flow Cytometry
All flow cytometry analyses were performed using BD ACCURI-C6. Colonies were taken at two different ages: juvenile (3-5 days old; N=190) and old (4-13 years old, N=37). Cell suspension was isolated as described in 15; briefly, B. schlosseri systems were meshed and filtered through a 40 μm mesh using a sterile 1 ml syringe pump. Cells were washed and collected in staining media: 3.3x PBS, 2% FCS and 10 mM Hepes. Cells were stained with 1) Propidium Iodide (PI), to differentiate live vs dead cells 2) alkaline phosphatase substrate measuring alkaline phosphatase activity in green fluorescence, 3) Cd49d-PE-Cy7, 4) CD57-Pacific Blue, 5) ConcavilinA-AF633, and 6) anti BHF (25) labeled with mouse serum APC-Cy7. Live cells were gated on a forward scatter (FSC) and side scatter (SSC) panels using log scale, and then gated further into different populations. Analysis of flow cytometry data was accomplished using FlowJo V10 (FlowJo).
Clock experiments
Free-run experiment
Colonies from 3 different age groups (36-140 days old days; 2142-2146 days; 5869-5871 days) were placed in 1.5 liter tubs, 2-3 slides per tub and kept in dark. Following a minimum of 3 days under dark, whole systems (Fig. S1A) were sampled every 3 hours along the clock under red light: 12am (n=12), 3am (n=9) 6am (n=9), 9am (n=10), 12pm (n=10), 3pm (n=14), 6pm (n=11), 9pm (n=8). All samples were taken at either developmental stage A1 or A2 (Figure S1A).
Light pulse
Following 4 days under dark, samples (whole systems) were collected at midnight. 3 samples were exposed to light for half an hour (light pulse) and sampled afterwards. 5 colonies (100-135 days old; developmental stage A1-A2) were sampled in this experiment.
Library preparation
All samples collected for RNAseq library prep (Detailed annotation in Table S1) were frozen in liquid nitrogen, and held at −80. RNA was prepared from frozen samples using Zymo Research Quick RNA MIcro Prep Kit #R1050, and cleaned using Zymo Research RNA Clean and Concentrator, #R1015. Samples were analyzed on an Agilent QC 2100 Bioanalyzer to determine quality prior to library preparation. cDNA was prepared using the Nugen Ovation RNA Sequencing System V2, #7102 and cleaned using the Qiagen QIAquick PCR purification kit, #28104, and then analyzed on the Agilent QC Bioanalyzer. If needed, samples were size-selected using Zymo Research Select-A-Size DNA Clean and Concentrator #D4080 prior to barcoding. Final library was prepared using NEB NEBNext Ultra II DNA Library Prep Kit #27645 and barcoded using NEBNext Multiplex Oligos for Illumina #E6609S. All magnetic bead purification was accomplished using BullDogBio CleanNGS RNA and DNA Spri Beads #CNGS005. Samples were then analyzed on the Agilent QC 2100 Bioanalyzer to determine the concentration of each sample prior to determining dilution prior to sequencing. On average, 12 million 2×150 bp reads (Illumina Nextseq 500) were sequenced for each library.
Gene counts
Following sequencing, reads were processed using a Snakemake (24) pipeline: they were trimmed to remove low quality bases and primers, merged if the reads from both ends overlapped, and aligned to a database of B. schlosseri transcripts using bwa (mem algorithm), with likely PCR duplicates removed and then read counts determined for each transcript, resulting in a count matrix (Table S2).
Gene orthology and methods used to search for cry and per
Gene orthology was based on sequence similarities between the B. schlosseri gene models and human and mouse gene annotations (BLAST score smaller than 10−10 as described in the B. schlosseri genome papers 11,25). All discussions regarding putative clock genes are based on sequence and domains similarities alone. To search for the missing period and cryptochrome genes all alleles for both genes were taken from fly (flybase.org), fish (zfin.org), and frog (xenbase.org) databases and blasted against the Botryllus assembly using NCBI BLAST. Then all the alleles were combined to make a mini-assembly of these missing genes and aligned to the B. schlosseri RNA-seq data using Krypto and BWA-mem and the alignment was visualized with Jbrowse. A consensus sequence of the aligned RNA was looked for using Samtools, but none was found. Finally, we identified all the conserved domains in the Botryllus genome using NCBI domain search and focused on the ones that contained conserved domains that appear in Per and Cry. We generated a Maximum Likelihood Bootstrap Phylogenetic Tree with alleles and genes we thought were possible candidates using Clustal Omega Multiple Sequence Alignment and MegaX but did not identify putative cry or per.
Identification of genes entrained by light
Differentially expressed genes between control colonies (kept in dark) and colonies that were exposed to light pulse at midnight were found using edgeR ((FDR < 0.05); 26 Table S1D).
Identification of cycling genes
To identify circadian cycling genes we used MetaCycle (18), a package combining 3 different methods, (JTK_CYCLE, ARSER, Lomb-Scargle) developed to identify cycling genes. Each age group was tested by itself and all cycling genes identified by these programs are presented in Table S3A-C.
Identification of time dependent gene signature and formation of binary tables
Identification of chronologically differential expressed genes and formation of binary tables have been described previously (16). Briefly, we used DEseq2 (27; FDR < 0.05) to identify differentially expressed genes between all possible combinations of contiguous and individual time points, resulting in a hierarchy of expressed time points for each gene for each age group. Based on these analyses, a binary gene-time expression matrix for every expressed gene recorded along the time points was produced for each age group individually, with the binary pattern chosen being the one with the highest correlation to the time hierarchy for that gene. With 1 indicating dynamically “high” expression and 0 indicating “low” or zero expression (Table S2A-C).
Using the binary matrix we identified pathways and Go terms associated with each age group for every time point (GeneAnalytics; 28; Fig 4; Table S5).
Gene enrichment plots
Gene enrichment plots have been described in 16. Briefly, at each time point the proportion of genes in a gene set that are active (indicated by a 1 in the gene-time expression binary matrix defined above) is calculated. This gives a value between 0% (no genes in common) and 100% (all genes in the gene set are active at that time). A baseline expectation of the proportion of overlapping genes is calculated using a hypergeometric model that gives the likelihood that the same number of genes as in the selected gene set would be randomly selected from the matrix. In addition, the 68% confidence interval (1 standard deviation) of proportion of shared genes (‘enrichment’) from the hypergeometric is calculated and plotted, presented as a shaded region in the plot. Then the baseline is subtracted from the values calculated, with the confidence interval also subtracted, to show the expected range of values and how far the actual enrichment result differs from a null result. If the baseline expectation is greater than the actual enrichment (where the subtracted value would be negative) a value of 0% is used (as a negative percent was considered meaningless).
Tissue enriched and specific pathways gene set used
For the gene enrichment plots the following gene sets were used:
For tissue specific nervous system, endostyle, ampullae, enriched hematopoietic stem cells (HSCs) and putative germ cells, we used existing Botryllus tissue enriched gene lists (15, 16; Table S4).
For the pathway and go terms identified by GeneAnalytics within our time data, we focused on the gene sets of the same names from PathCards. From each of these gene sets we removed any gene that did not have a putative homolog to a known Botryllus gene. As such the percentages within the enrichment plots refer to this curated Botryllus specific gene sets. If a gene name appeared more than once in the Botryllus gene model annotation (11, 25) all matching Botryllus gene ids were included in the gene set.
Supplementary Tables
S1A B. schlosseri candidate clock core and regulators genes.
S1B Annotation of the 88 samples sequenced.
S1C Transcript counts for all 88 samples sequenced.
S1D Differentially expressed genes between control colonies (kept in dark) and colonies that were exposed to light pulse at midnight.
S2A-C A binary gene-time expression matrix for the 3 age groups.
S3A-C Meta cycle results for the 3 age groups.
S4 Tissue specific gene enrichment at every time point across the age groups (nervous (A-C), HSCs (D-G), endostyle (H-J), GSC (Q-T)).
S5 Pathway enrichment expression at every time point across ages based on GeneAnalytics (28) analysis superpaths enriched in Young (A), Mid (B) and Old (C) and the gene associated with them per every time point. (D) Color coded comparison between pathways enriched in all three ages.
Movie S1-Siphons opening during dark (night) versus light (day) of a young colony.
Siphon activity recorded at night versus day in young colonies, suggesting that B. schlosseri is a diurnal organism.
Acknowledgments
We thank C. Lowe, L. Crowder. C. Patton, J. Thompson, J. Lee, B. Compton, T. Naik L. Quinn for technical advice and help.