Longitudinal Comparative Transcriptomics Reveals Unique Mechanisms Underlying Extended Healthspan in Bats

Home , AKIRIN2, BTAF1, CCDC130, CHD1L, COMMD4, DRG2

Articles https://doi.org/10.1038/s41559-019-0913-3

Longitudinal comparative transcriptomics reveals unique mechanisms underlying extended healthspan in bats

Zixia Huang 1, Conor V. Whelan1, Nicole M. Foley1, David Jebb1, Frédéric Touzalin1,2, Eric J. Petit 3, Sébastien J. Puechmaille 1,4,5 and Emma C. Teeling 1*

Bats are the longest-lived mammals, given their body size. However, the underlying molecular mechanisms of their extended healthspans are poorly understood. To address this question we carried out an eight-year longitudinal study of ageing in long- lived bats (Myotis myotis). We deep-sequenced ~1.7 trillion base pairs of RNA from 150 blood samples collected from known aged bats to ascertain the age-related transcriptomic shifts and potential microRNA-directed regulation that occurred. We also compared ageing transcriptomic profiles between bats and other mammals by analysis of 298 longitudinal RNA sequencing datasets. Bats did not show the same transcriptomic changes with age as commonly observed in humans and other mammals, but rather exhibited a unique, age-related gene expression pattern associated with DNA repair, autophagy, immunity and tumour suppression that may drive their extended healthspans. We show that bats have naturally evolved transcriptomic signatures that are known to extend lifespan in model organisms, and identify novel genes not yet implicated in healthy ageing. We further show that bats’ longevity profiles are partially regulated by microRNA, thus providing novel regulatory targets and pathways for future ageing intervention studies. These results further disentangle the ageing process by highlighting which ageing pathways contribute most to healthy ageing in mammals.

geing is the leading risk factor for major life-threatening (Myotis) maintains the length of their telomeres with age without conditions such as cancer, neurodegeneration and cardio- developing cancer12 and do not show an increased level of mito- Avascular disorders1. Despite centuries of study, the com- chondrial damage as expected given their metabolic rate13. This plexity of the ageing process has hampered our understanding of is potentially due to adaptations in their DNA repair and mainte- what drives ageing, with multiple theories of how and why we age nance mechanisms13. However, to date, longitudinal ageing studies but little consensus2,3. Given that the predicted increase in human in bats have focused on only one ageing pathway or process at a lifespan (50% increase in people aged >60 years, 340% increase in time—for example, telomeres12, mitochondria13, microbiome14 or people aged >80 years by 2050)4 is not associated with a similar metabolome11. As ageing is an inherently complex process, a better increase in healthspan, we urgently need to understand ageing to approach is to study the interplay of multiple biological pathways relieve its maladies5. Most ageing studies have been carried out in to uncover the molecular mechanisms of healthy ageing evolved laboratory model species, as these are easier to manipulate and in bats. Comparative, population-level transcriptomics provides an house and have shorter lifespans6,7. Although substantial prog- alternative means that can potentially capture a full repertoire of ress has been made in extending lifespan and healthspan in these RNA species and reveal subtle age-related transcriptional variation short-lived model organisms, there is limited evidence that these across individuals. approaches will be effective in more long-lived species such as Recent transcriptome studies have catalogued signatures of age- humans5. An alternative approach is to explore ageing in species ing in model organisms and reveal a remarkable conservation of that are even more ‘ageing-resistant’ than humans and have natu- age-related pathways15–17. These molecular signatures illustrate a rally evolved longer healthspans6–9. combination of degenerative phenotypes, notably reflected by low- By far the most successful mammals in this regard are bats. Bats grade, chronic inflammation, dysregulation of metabolic activities account for 20% of all living mammals, have evolved self-powered and insufficient cellular maintenance16,18. By contrast, new evidence flight and have exceptional longevity10. Nineteen species of mam- has shown that long-lived mammals, including bats, may harbour mals live longer than humans, given their body size, of which 18 species-specific transcriptional signatures that are beneficial for the are bats6. Specifically, some wild individuals of the ~7-g Brandt’s maintenance of cellular functions into old age19–22. Here, focusing bat (Myotis brandtii) can live >41 years, over ten times longer on a unique, longitudinal capture–mark–recapture study of a wild than expected for their body size (humans live only five times lon- population of long-lived Myotis myotis bats (maximum lifespan, ger than expected), showing little signs of ageing6,7. Logistically 37.1 years)12, we performed a systems-level comparative analysis it is difficult to study bats in an ageing context, as most are only (Fig. 1a) to ascertain the age-related transcriptional changes and found in the wild and not easily maintained in captivity11. Initial microRNA (miRNA)-directed regulation that may underlie bats’ longitudinal ageing studies suggest that the longest-lived genus exceptional longevity.

1School of Biology and Environmental Science, Science Centre West, University College Dublin, Dublin, Ireland. 2Bretagne Vivante-SEPNB, Brest, France. 3Ecology and Ecosystem Health, Agrocampus Ouest, INRA, Rennes, France. 4Zoological Institute and Museum, University of Greifswald, Greifswald, Germany. 5ISEM, University of Montpellier, Montpellier, France. *e-mail: [email protected]

1110 Nature Ecology & Evolution | VOL 3 | JULY 2019 | 1110–1120 | www.nature.com/natecolevol Nature Ecology & Evolution Articles abComparative M. myotis blood M. myotis blood transcriptomics transcriptomics miRNomes 100 Age (n = 100) (n = 50) 0 years 1 years 2 years 3 years Enriched GO terms of 100 4 years highly expressed transcripts 75 iptome output 5 years Viral transcription 6 years Translational initiation

7 + years rRNA processing

Mitochondrial protein catabolic process

Oxygen transporter activity 50 Haem biosynthetic process Integrative analyses Removal of superoxide radicals Hydrogen peroxide catabolic process

Virion assembly

Transcriptome Expression Comparative miRNA DNA damage response

overview modules analyses regulation Positive regulation of proteolysis e percentage of total transcr 25 Defense response to virus Autophagy

Global genome nucleotide-excision repair

010 20 Cumulativ –log (FDR) Transcriptomic signatures of bat longevity miRNA-directed regulatory mechanisms

0 101 102 103 104 c 100 Transcripts (n) n = 12,263 d 0 years 1 years 2 years 3 years 4 years 5 years 6 years 7 + years 75

plaine d 0 years 251 1,296 1,027 944 1,650 379 1,470

1 years –142 117 280 111 104 9 175 ance ex ri 50 2 years –1,536 –90 318 4 79 49 298 3 years –1,617 –517 –256 117 17 21 186

4 years –891 –40 –2 –100 59 39 436 25 5 years –2,248 –423 –61 –72 –136 2 26 ercentage of va P 6 years –427 –5 –23 –20 –21 –1 5

0 7 + years –2,130 –577 –134 –311 –545 –1 0

y Age Batch Colon Individual Residual ear of capture Y

Fig. 1 | Overview of M. myotis blood transcriptome. a, Schematic drawing of the data analyses. b, Evaluation of the transcriptome complexity. Cumulative percentage of total transcriptome output was contributed by transcripts that were sorted from most to least based on expression values (n = 48,749). Lines represent average values across samples of the same age cohort, and the more lightly shaded adjacent regions reflect 95% confidence intervals. The intersection between the cumulative lines and dashed line indicates the overall contribution of the top 100 highly expressed transcripts across cohorts. c, Evaluation of gene expression variation. Residual variance represents the contribution from uncharacterized variables. d, Pairwise differential gene expression analyses across age cohorts. The values in each square represesnt the number of differentially expressed genes. Red (upper triangular matrix) indicates up-regulation while blue (lower triangular matrix) indicates down-regulation. GO, gene ontology.

Results and discussion scriptome, we developed a comprehensive pipeline (see Methods) Overview of M. myotis ageing transcriptomes. Using a non-lethal to assemble and annotate full-length transcripts from 100 RNA-Seq sampling process developed to maximize transcript representation samples, which resulted in 31,460 protein-coding transcripts (cor- from bat blood (>60% of all protein-coding genes represented)23, we responding to 12,263 protein-coding genes), 10,775 long non-cod- deep-sequenced ~1.7 trillion base pairs of RNA from 100 bat blood ing RNAs (lncRNAs) and 6,514 miscellaneous RNAs (miscRNAs) samples (69.6 ± 9.0 s.d. million reads per sample) using Illumina (Supplementary Fig. 2). More details regarding samples, tran- RNA-sequencing (RNA-Seq). These blood samples (~50–200 μl) scriptome assembly and annotation are described in Methods and were collected from 70 individual bats ranging in age from 0 to Supplementary information. >7 years (for example, first caught as an adult 6 years before subse- Similar to humans24, the M. myotis blood transcriptome was quent recapture) at five colonies in Brittany, France (Supplementary dominated by a few highly abundant transcripts, with ~75% of the Tables 1–3 and Supplementary Fig. 1). The majority of the raw reads overall transcriptional output derived from the top 100 most highly (98.5%) showed high quality (>Q30). On average, 77% of clean expressed transcripts (Fig. 1b). Interestingly, these dominant tran- reads were successfully mapped to the reference genome (Myotis scripts were enriched in several cellular maintenance bioprocesses, lucifugus), 64.2% of which had unique mapping coordinates. Of such as autophagy and DNA repair (Fig. 1b). This high level of cel- all mapped reads, 88.1% were concordantly aligned. Only those lular maintenance expression profile was not observed in human reads that were uniquely and concordantly mapped were used for or mouse blood, based on a comparative cross-sectional analysis downstream analyses. To gain an overview of the bat blood tran- (Supplementary Fig. 3; see also Methods). From a global perspective,

Nature Ecology & Evolution | VOL 3 | JULY 2019 | 1110–1120 | www.nature.com/natecolevol 1111 Articles Nature Ecology & Evolution

a DNA replication Cell cycle DNA replication Microtubule organizing centre organization M1 M2 3 M3 3 Chromosome segregation 2 2 Regulation of organelle assembly 2 Microtubule-based process Autophagy 1 Modification-dependent macromolecular 1 1 catabolic process 0 Organelle localization 0 0 Chromatin organization –1 –1 –1 n = 556 P < 0.001 n = 501 P < 0.001 n = 943 P < 0.001

ears ears ears ears ears ears ears 0 y 1 y 2 y 3 y 4 y 5 y 6 y 7 + years

DNA replair c Actin cytoskeleton organization DNA duplex unwinding Nucleobase-containing compound catabolic process Regulation of chromosome organization DNA synthesis involved in DNA repair Viral process Establishment of protein localization to membrane Negative regulation of cell cycle Positive regulation of establishment of protein localization Cell cycle phase transition Nuclear envelope organization Endomembrane system organization Positive regulation of cell cycle process Protein modification by small protein removal b Ribonucleoprotein complex assembly Antigen receptor-mediated Ribosomal large subunit biogenesis signalling pathway Leukocyte migration RNA splicing Maturation of SSUrRNA Cytokine-mediated signalling pathway Regualtion of defense response rRNA modification 2 M4 Inflammatory response M6 Regulated exocytosis 1 1

xpressio n 0 0 e (z-score)

–1 –1 Relati ve n = 882 P < 0.001 n = 734 P = 0.08 Lymphocyte activation Cytokine production T cell selection 2 M7 Regulation of interleukin-2 2 M5 production Cytokine production 1 RNA catabolic process 1 Cellular response to biotic stimulus Nucleocytoplasmic transport Viral process Interleukin-6 production Establishment of protein localization to organelle 0 Regulation of leukocyte mediated 0 Ribonucleoprotein complex biogenesis immunity Covalent chromatin modification Leukocyte differentiation –1 –1 Negative regulation of protein modification process Negative regulation of immune n = 547 P = 0.4 n = 1,415 P < 0.001 system process Myeloid cell differentiation ears ears ears ears ears ears ears ears ears ears ears ears ears ears Platelet activation 0 y 1 y 2 y 3 y 4 y 5 y 6 y Mitochondrial gene expression 0 y 1 y 2 y 3 y 4 y 5 y 6 y 7 + years 7 + years Small GTPase mediated signalling transduction Reactive oxygen species metabolic process Cellular response to nitrogen compounds Negative regulation of transcription by RNA polymerase

Fig. 2 | Co-expression network analysis based on 6,692 age-associated candidate genes. a–c, Gene expression modules showed positive (M1–M3; a), negative (M4, M5; b) and no correlation (M6, M7; c) with age. The networks indicate the enriched gene ontology terms and their connectivity in the modules under a–c, respectively. In each module, n indicates the number of genes in the module. The expression regression line was generated using the Loess curve-fitting method, and P values indicate the significance of Spearman’s rank correlation coefficient between the eigengene and age. The eigengene is a central gene whose expression pattern can represent the whole module. SSUrRNA, small subunit ribosomal RNA. transcript correlation analyses showed high concordance across a network that could maintain genomic stability in M. myotis samples (average Spearman’s rank correlation coefficient = 0.89, (Fig. 2a). In particular, we conducted a network analysis using all Supplementary Fig. 4), and gene expression variation analyses indi- the genes (n = 107) enriched in DNA repair (GO: 0006281) that cated that ‘age’ accounted for only 8.7% of total variance (Fig. 1c), exhibited a positive correlation with age in M. myotis. Based on suggesting that the age-related transcriptomic changes in M. myotis various sources of evidence (see Methods), the majority of genes are subtle. This percentage of age-related variance is comparable to (95 out of 107) were functionally related (Fig. 3a). Remarkably, we that of humans (7.8) but lower than that of short-lived mice (12.8; found that a number of genes directly involved in several mam- see Methods). Consistent with the long-lived naked mole rat20, only malian DNA repair pathways (for example, nucleotide excision a small proportion of genes showed differential expression (false repair, mismatch repair, base excision repair and double-strand discovery rate (FDR) < 0.05) with increasing age (adults aged 1 to break repair) were strongly enriched and connected in the network >7 years) (Fig. 1d), suggesting that maintenance of transcriptomic (Fig. 3b). This suggests that M. myotis bats may have evolved an profiles may be crucial for longevity in mammals. elaborate DNA repair machinery to maintain genome stability. This age-related increase in DNA maintenance could underlie the low A transcriptomic signature of ageing in bats. Next, we identified level of cancer incidence reported in bats12. An alternative expla- 6,692 age-associated genes that have been tagged at least once as dif- nation is that the enhancement of DNA repair pathways could be ferentially expressed between any two age cohorts (Supplementary a response to cumulative DNA damage over time. However, our Table 4). We performed weighted gene co-expression network anal- earlier longitudinal studies suggest that bats do not experience yses (WGCNA) on this dataset and identified seven gene expres- increased levels of DNA damage with age12,13 as compared to other sion modules that exhibited distinct ageing patterns in M. myotis mammals. Further estimation of the level of DNA damage experi- (Fig. 2a–c, Supplementary Table 5 and Supplementary information). enced with ageing by bats is required to test these hypotheses. M. myotis bats showed canonical age-related transcriptomic signa- In our previous study we speculated that telomeres, which did tures shared with other mammals16, such as age-related declines not significantly shorten with age, may be maintained through in the expression of genes associated with the adaptive immune the mechanism of alternative lengthening of telomeres (ALT) in response and mitochondrial activity (M4 and M5; Fig. 2b). In con- M. myotis12. This hypothesis is strongly supported by up-regulation trast, M. myotis showed an up-regulation of pathways involved of large proportions of genes directly involved in ALT (for example, in DNA damage signalling and DNA repair with age (M1–M3; MRE11A, PCNA, BLM, DNA2, MND1 and WRN) and inactivation Fig. 2a). These pathways were interconnected, suggesting an interplay of TERT over age, as observed in this study (Supplementary Data 1). between DNA damage sensing, signalling, repair and replication, An age-related increase in the expression of genes involved in cell

1112 Nature Ecology & Evolution | VOL 3 | JULY 2019 | 1110–1120 | www.nature.com/natecolevol Nature Ecology & Evolution Articles

a TFIP11 DNA repair RBM17 MSH4 HELQ ISY1 RMI1 NSMCE2 RAD51B RFWD3 MCM9 RAD51C SPIDR RMI2

MND1 POLA1 ALKBH1 MCM8 FANCL FANCB TOP3A

POLQ REV1 RAD51AP1 TERF2

BARD1 RAD51 DCLRE1A NEIL3 POLE MLH3 RIF1 RPA2 RFC3 OGG1 ATR FANCI POLB WHSC1 DDX11 PMS2 NPLOC4 FANCD2 LIG1 XRCC6 RFC4 KIAA0101

ERCC6 ERCC6L2 DOT1L DCLRE1C DTL

PRKDC ERCC8 GTF2H1 CDK7 KDM2A UBE2V2 CHD1L XRCC6BP1 XPC GTF2H4 UBE2W RAD23B SUPT16H EP300 DDB2 ATXN3 SIRT1 UVRAG

TICRR FOXM1 CD40LG JMY ACTL6A ITCH UCHL5 TFRC MAP2K7 PSME4 EPC2 NUPL1 NUP188 NDFIP1 MAP3K7 USP28

DUSP19 POLG2 MAP2K4 NUP54 MAP4K5 NUP93

RAE1 MAP3K4 SPAG9

b DNA repair mechanisms (KEGG) No. of FDR genes

Nucleotide excision repair 13 1.31 × 10–17

Mismatch repair 6 1.6 × 10–7

Homologous recombination 5 1.51 × 10–5

Base excision repair 5 3.6 × 10–5

Non-homologous end joining 3 1.57 × 10–3

Fig. 3 | Network analysis of the genes enriched in DNA repair. a, The gene product network. The network included 107 genes (nodes) enriched in DNA repair that exhibited a positive correlation with age in M. myotis. The thickness of edges indicates the strength of data support (see Methods). Coloured nodes indicate their engagement in different DNA repair pathways. b, The numbers of genes enriched in different mammalian DNA repair pathways. KEGG, Kyoto Encyclopedia of Genes and Genomes. cycle arrest was also observed in M. myotis. Induction of cell cycle We did not detect up-regulation of any enriched gene ontology control may represent a universal tumour-suppressant mecha- terms associated with chronic inflammatory responses with advanc- nism in mammals, but one that results in cellular senescence25,26. ing age (M6 and M7; Fig. 2c). This is unusual given that increased Interestingly, M. myotis exhibited enhanced autophagy activity with inflammation is a typical hallmark of ageing in mammals1,27. advancing age (Fig. 2a), which may potentially eliminate cumulative However, this supports the hypothesis that bats have evolved unique damaged organelles and protein aggregates and thereby promote immune systems that enable them to dampen the constant sterile cell survival. inflammation they experience from their high metabolic rates25,26.

Nature Ecology & Evolution | VOL 3 | JULY 2019 | 1110–1120 | www.nature.com/natecolevol 1113 Articles Nature Ecology & Evolution

a b 2 GO :0000226 microtube cytoskeleton organization Gene Correlation Gene Correlation GO :0000075 cell cycle checkpoint LRRC37A2 0.601 WDR12 –0.633 1.5 GO :0006281 DNA repair IPP2 0.579 CD19 –0.619 GO :0031124 mRNA 3′-end processing DZIP3 0.566 MAGED2 –0.603

pper 100 gene s PLK4 0.561 CLEC17A –0.598 GO :0007062 sister chromatid cohesion 1 ZEB1 0.56 VPS16 –0.59 GO :0006882 cellular zinc ion homeostasis WHAMM 0.559 KLHL6 –0.583 Lower 100 Upper 100 0.5 GO :0050853 B cell receptor signalling pathway AKIRIN2 0.556 TGIF2 –0.58 genes genes GO :0046649 lymphocyte activation CENPK 0.55 NEIL1 –0.578 GO :0032611 interleukin-1 beta production UBP25 0.542 SFT2D2 –0.571 0 Lower 100 gene sU GO :0010921 regulation of phosphatase activity PAF15 0.536 FRMD6 –0.571 BTAF1 0.534 WRAP53 –0.568 3 GO :0070734 histone H3-K27 methylation RUFY1 0.532 COMMD4 –0.566 GO :0072757 protein localization to membrane CXorf58 0.531 EXOSC7 –0.565 GO :0002228 natural killer cell-mediated immunity LRR1 0.531 AFF3 –0.562 2 GO :0006303 double-strand break repair via CSGALNACT1 0.53 DRG2 –0.561 non-homologous end joining KIAA1328 0.529 KLHDC3 –0.551 GO :0002250 adaptive immune response MED4 0.528 NOP56 –0.55 MRE11A 0.527 MPZL1 –0.55 1 GO :0001510 RNA methylation UCHL5 0.524 CCDC130 –0.549 GO :0006359 regulation of transcription by RNA CENPQ 0.521 CD72 –0.548 polymerase III c 0 GO :0043627 response to ostrogen z-score 1 GO :0070269 pyroptosis GO :0006954 inflammatory response –2 02 GO :0030162 regulation of proteolysis GO :0006281 DNA repair 37 0.75 GO :0048638 regulation of developmental growth GO :2000779 regulation of double-strand break repair8 GO :1990391 DNA repair complex8 GO :0030282 bone mineralization GO :0006301 post-replication repair9 0.5 GO :0072331 signal transduction by p53 class mediator 19 GO :0046649 lymphocyte activation GO :0000784 nuclear chromosome, telomeric region 14 Genomi c maintenance GO :0030111 regulation of Wnt signalling pathway GO :0007062 sister chromatid cohesion 15 0.25 GO :0007059 chromosome segregation 40 GO :0001568 blood vessel development GO :0044772 mitotic cell cycle phase transition 51 GO :0031347 regulation of defence response GO :0000086 G2/M transition of mitotic cell cycle27 0 GO :0000302 response to reactive oxygen species GO :0007093 mitotic cell cycle checkpoint 19 Cell cycle GO :0007095 mitotic G2 DNA damage checkpoint 7 regulatio n GO :0000082 G1/S transition of mitotic cell cycle20 GO :0007271 synaptic transmission cholinergic GO :0070125 mitochondrial translational elongation 27 GO :0007129 synapsis GO :0005761 mitochondrial ribosome 19 GO :0009636 response to toxic substance GO :0032774 RNA biosynthetic process 317

1 GO :0044212 transcription regulatory region DNA binding 48 Metabolism GO :0009620 response to fungus GO :0050852 T cell receptor signalling pathway 26 GO :0015844 monoamine transport GO :0050853 B cell receptor signalling pathway 13 GO :0030217 T cell differentiation 30 Adaptive GO :0002253 activation of immune response GO :0002250 adaptive immune response39 immunit y 0.5 GO :0030593 neutrophil chemotaxis 14 Density GO :0045058 T cell selection GO :0002523 leukocyte migration involved in inflammatory response6 GO :0060759 regulation of response to GO :0001819 positive regulation of cytokine production 35 cytokine stimulus GO :0050707 regulation of cytokine secretion 17 GO :0042742 defence response to bacterium17 response

0 GO :0015748 organophosphate ester transport GO :0051092 positive regulation of NF-kappa-B transcription factor activity 14 Inflammator y GO :0050729 positive regulation of inflammatory response10 –1 –0.5 0 0.5 1 GO :0016049 cell growth 34 GO :0070851 growth factor receptor binding 14

Spearman’s correlation coefficient GO :0001676 long-chain fatty acid metabolic process11 Nutrient sensing

Fig. 4 | Comparative transcriptomic analyses between bat and human, mouse and wolf. a, Distribution of Spearman’s correlation coefficient between gene expression and age across the four species. Gene ontology terms were enriched for the top 100 genes (both upper and lower). b, Top 20 genes that exhibited the strongest positive and negative correlation with age in M. myotis. c, Comparison of the pathway expression pattern with age across the four species. Within each species, the median z-scores of all genes under each of enriched age-associated gene ontology terms are used to represent their overall expression pattern with age. The z-scores were converted from Spearman’s correlation coefficients. The values following the gene ontology terms indicate the number of genes enriched.

This may in turn limit inflammation-driven ageing and also under- immuno-inflammatory gene expression profiles during ageing, lie their apparent tolerance of pathogens (for example, those respon- which have been shown to accelerate the ageing process16. Uniquely sible for Ebola, rabies and severe acute respiratory syndrome)28,29. seen in bats, our results indicate that increased genome maintenance and cell cycle regulation may represent an evolutionary Comparison of longitudinal age-related signatures across adaptation that enables bats to achieve exceptional longevity. mammals. To ascertain whether these age-related transcriptomic changes are unique to bats, we conducted comparative transcrip- Longevity signatures evolved naturally in long-lived bats. tomic analyses between M. myotis and three other mammals, Homo Research on the biology of ageing has discovered a wealth of genes sapiens (human, aged ~25–75 years, n = 147), Mus musculus (mouse, that can be transcriptionally modified to prevent age-related dis- aged ~0.2–2.5 years, n = 25) and Canis lupus (wolf, aged ~1–9 years, eases and promote longevity in model organisms30. We therefore n = 26) (see Methods). When we focused on the top 200 genes that investigated 207 human ageing-associated genes that are curated exhibited the strongest correlation with age, each species displayed in the GenAge database31 (see Methods) and compared their cor- unique age-correlated pathways (Fig. 4a). These top age-corre- relation with age across humans, mice, bats and wolves to ascer- lated genes found in M. myotis are mainly enriched in the maintain whether bats have naturally evolved known ‘life-extension’ tenance of genome stability (for example, DZIP3 and PLK4) and expression profiles. A large body of these candidate genes showed anti-cancer activity (for example, WDR12 and WRAP53) (Fig. 4b contrasting expression changes in M. myotis compared with the and Supplementary Data 2). The majority of these have not been other mammals (Supplementary Data 4). In total, 23 genes had the associated directly with ageing in humans or model species (Fig. 4b). opposite direction of expression with age in M. myotis compared However, many of these genes do interact upstream or downstream to each of the other three mammals (Fig. 5a), suggesting a role in with known human ageing-associated genes, as curated in the driving bat longevity. Intriguingly, the expression changes of certain GenAge database (Supplementary Table 6). Although age-related candidate genes occurring naturally in M. myotis during ageing immunosenescence and reduced mitochondrial activity did occur were observed to extend lifespan in model organisms when modu- in the blood transcriptome of all four mammal species, the path- lated. For example, PTEN (Spearman’s rs = 0.395, P = 0.002; Fig. 5b) ways associated with genome maintenance and cell cycle regula- is a well-established tumour suppressor involved in DNA damage tion had higher positive correlations with age in bats (Fig. 4c and repair and cell cycle arrest32. Transgenic mice carrying an extra Supplementary Data 3). These pathways were down-regulated in PTEN dosage displayed enhanced protection from cancer and pre- the other three mammals (Fig. 4c). Despite being of similar size sented a modest (16%) extension of lifespan33. Also, overexpression to a bat, mice uniquely exhibited increased nutrient sensing and of SIRT1, GCLM and BUB1B (Fig. 5b) was verified as prolonging

1114 Nature Ecology & Evolution | VOL 3 | JULY 2019 | 1110–1120 | www.nature.com/natecolevol Nature Ecology & Evolution Articles

ab 2 PTEN GCLM

1 2 MRE11A 0.527 –0.108 –0.244 –0.011 0.395 0.284 PCNA 0.394 –0.123 –0.282 –0.131 0 –0.003 –0.122 –0.242 0 0.041 PDPK1 0.38 –0.13 –0.126 –0.113 z-score –1 –0.493 –0.199 PRKDC 0.319 –0.04 –0.227 –0.498 BRCA1 0.291 –0.376 –0.057 –0.369 –2 –2 P = 0.002 P = 0.027 CREBBP 0.249 –0.1 –0.348 –0.197 SIRT1 0.243 –0.091 –0.074 –0.574 01234567+ 012 34567+ RAE1 0.217 –0.25 –0.126 –0.556 4 3 SIRT1 BUB1B ZMPSTE24 0.192 –0.307 –0.145 –0.039 2 PPM1D 0.151 –0.286 –0.007 –0.514 2 0.243 0.234 TERF2 0.115 –0.066 –0.248 –0.375 1 –0.091 0.004 MED1 0.103 –0.236 –0.331 –0.739 –0.074 0.03 –0.574 0 TOP1 0.058 –0.009 –0.023 –0.41 0 0.044

INSR –0.009 0.075 0.05 0.057 –1 STUB1 –0.019 0.018 0.212 0.11 –2 P = 0.05 P = 0.05 –2 FLT1 –0.048 0.221 0.286 0.531 01234567+ 01234567+ CLU –0.068 0.079 0.464 0.286 HIF1A –0.086 0.189 0.195 0.493 3 IKKB 4 MYC FGFR1 –0.122 0.21 0.11 0.443 2 APP –0.157 0.11 0.201 0.732 –0.31 –0.305 2 IL7 –0.159 0.109 0.121 0.673 1 –0.098 –0.229 –0.019 –0.093 PPARG –0.163 0.027 0.329 0.363 0 0.233 0.519 VEGFA –0.278 0.282 0.249 0.05 0 –1

P = 0.015 P = 0.017 –2 01234567+ 0 1234567+

Fig. 5 | Comparison of expression patterns of human ageing-associated genes between bat and human, mouse and wolf. a, Of 207 human ageing- associated genes, those that exhibited the opposite direction of expression changes with age in bat compared to human, mouse and wolf are shown. b, Examples of gene expression patterns during ageing in M. myotis that were observed to extend lifespan in model species. Spearman’s correlation coefficients with age are presented for all four species.

mean or maximum lifespan in fruit flies and mice via in vivo func- signatures of longevity, we sequenced 50 blood miRNomes from the tional assays34–36. same population (Supplementary Table 7 and Supplementary infor- Conversely, repression of IKKB and MYC, as seen in bats, mation); 98% of the miRNA sequences were obtained from the same increased longevity and enhanced healthspan in mice37,38. In partic- blood samples used for messenger RNA-Seq (Supplementary Table 1). ular, MYC has been shown to be evolving under divergent selection Principle component analysis (PCA) of the miRNA expression pro- in bat lineages12. MYC is a highly conserved proto-oncogene that file (Fig. 6a) revealed a striking similarity to that of the mRNA tran- is critically engaged in many essential cellular processes39. Despite scriptome (Supplementary Fig. 5), with a noticeable shift between its importance in maintaining cell function, its overexpression has juvenile and adult. However, expression analyses demonstrated that been documented in a variety of cancers40. In humans, the product a major change also occurred between 4- and 5-year-old cohorts, of MYC directly activates telomerase by inducing transcription of with most up-regulated miRNAs acting as tumour suppressors (for telomerase reverse transcriptase (TERT)41, whose overexpression is example, miR-146a/b, miR-30a) and down-regulated miRNAs pro- detected in ~90% of malignant cancer types40. This co-expression moting cell cycle or carcinogenesis (for example, miR-18a, miR-29c; probably occurred also in M. myotis, as supported by the high expres- Fig. 6b and Supplementary Fig. 6). miR-146a, which plays a sup- sion correlation between MYC and TERT (Spearman’s rs = 0.305, pressive role in prostate and lung cancer by inhibiting cell prolifera- P = 0.017) according to our analysis. Being under selective pres- tion and migration42, was up-regulated 2.27-fold (P < 0.001) while sure exclusively along the bat ancestral branch, down-regulation of miR-18a, used as a sensitive screening biomarker in a wide vari- MYC, coupled with its synergistic effect on TERT inactivation, may ety of human cancers43, was down-regulated 2.08-fold (P < 0.001). underlie the long lifespan and low cancer incidence observed in Although currently no direct evidence of lifespan extension was M. myotis, despite its small size and high metabolic rate. observed in model organisms by genetic modification of any of These results suggest that study of the ageing transcriptome in these differentially expressed miRNAs, some (for example, miR- wild, long-lived bats can uncover molecular mechanisms that underlie 146a and miR-18a) did show enhanced protection from various their increased healthspans, and that the novel candidates identified cancer types and improved health in humans42,43. No miRNAs were (for example, TGIF2, WRAP53 and IPP2; Fig. 4b and Supplementary differentially expressed amongst the older age cohorts (5, 6 and Data 2) are suitable targets for future ageing intervention studies. 7+ years; Supplementary Fig. 6), suggesting that this anti-cancer miRNA expression profile is maintained into old age in M. myotis. MiRNA regulation of longevity signatures in bats. To ascertain Next, we established a gene regulatory network to gauge the impact the molecular mechanisms that may underlie these transcriptional of miRNA on gene expression. The miRNA–mRNA interactions

Nature Ecology & Evolution | VOL 3 | JULY 2019 | 1110–1120 | www.nature.com/natecolevol 1115 Articles Nature Ecology & Evolution ab z-score 0.4

–3 03

Age miR-151a 0.2 0 years miR-18a 1 years 2 years miR-146a 3 years miR-146a 0 4 years miR-24 5 years PC2 (15.05%) miR-28 6 years miR-30a 7 + years miR-542

–0.2

miR-499a miR-29c –0.25 0 0.25 miR-331 PC1 (28.15%) 0 years 1 years 2 years 3 years 4 years 5 years 6 years 7 + years cd

6 8

wth 8 4 7 Viral process CK 2 APL Cell gro W RB11A erentiation RO 6 2 6 Histone modification

ocyte diff P = 0.05 5 P = 0.049 P = 0.02 0 4 Leuk 5 6.5 7 7.58 8.5 99.5 10 10.5 11 12345 miR-502b miR-532 miR-671

Response to ROS 8 P = 0.011 8 9.75 6 7 9.5 Cell division 4 6 9.25 PHC3 POGZ 5 CDK19 Autophagy 2 9 4 P = 0.021 P = 0.023 Regulation of lipid 0 8.75 metabolic process 56789 10 8.5 99.5 10 10.5 11 78910 Response to insulin miR-339 miR-342 miR-130a Chromosome segregatio 7.5 miR-423 10.5 10

miR-185 5 miR-330 10 EP300 CDC2 7

5 WDR3 3 miR-24 n miR-140 2.5 miR-1388 miR-138 miR-34b miR-532 miR-542 9.5 P < 0.001 P = 0.013 P = 0.05 miRNA 0 0 12.5 13 13.5 9.2 9.610 55.5 6 6.57 miR-15b miR-148b miR-185

Fig. 6 | miRNA analyses and their regulatory network. a, PCA based on 117 miRNA expression data. b, Cluster analysis of 117 miRNAs. The miRNAs in red indicate their differential expression between 4 and 5 years of age (red, up-regulation; black, down-regulation). c, The regulatory network between miRNAs and their top targeted biological pathways. The connectivity indicates that certain miRNAs regulate genes that are grouped under particular gene ontology terms. The top ten miRNAs with the highest numbers of targets are highlighted. d, Examples of predicted miRNA–mRNA pairs that were negatively correlated. These mRNAs are mainly involved in DNA repair and cell cycle regulation. Both miRNA and mRNA expression values were normalized using the TMM method and were further log2-transformed. P values indicate the significance of negative correlation. The grey shading reflects the 95% confidence interval. ROS, reactive oxygen species.

were determined by in silico miRNA target prediction44 and were Supplementary Table 8). For example, the up-regulation of EP300, further validated by confirming a negative correlation between a gene involved in DNA repair, correlated with the down-regula- expression of the predicted miRNA–mRNA pair (see Methods and tion of miR-148b; on the other hand, SHC1 and MTOR, genes that Supplementary information). Collectively, 10,108 miRNA–mRNA induce cell growth and proliferation, appeared to be progressively interactions were established between 117 known miRNAs and suppressed by increased expression of miR-30a and miR-330 with 7,051 mRNAs, only 9.45% of which (955 out of 10,108) exhibited sig- advancing age. These results are suggestive of the prominent roles of nificant negative correlation (FDR < 0.1). The enrichment analysis miRNA in the regulation of ageing pathways that may underpin the illustrated that miRNA mediated a wide spectrum of bioprocesses, extreme longevity observed in M. myotis. As these candidates are encompassing nutrient sensing, immunity and cellular mainte- suggested on the basis of in silico analyses, future functional assays nance pathways (Fig. 6c and Supplementary Fig. 7). In particular, are necessary to confirm their roles in bat longevity. we noticed that a large number of genes associated with cell cycle This unique longitudinal study of ageing in a wild population regulation and DNA repair were mediated by miRNA (Fig. 6d and of long-lived bats revealed that M. myotis did not show the same

1116 Nature Ecology & Evolution | VOL 3 | JULY 2019 | 1110–1120 | www.nature.com/natecolevol Nature Ecology & Evolution Articles transcriptomic changes with age commonly observed in other Summary of sequenced samples. We deep-sequenced 100 transcriptomes and mammals. Their unique ageing-transcriptomic shifts suggest that 50 miRNomes from 150 blood samples which were collected from 71 female M. myotis bats (Supplementary Table 1). These individuals were caught in five the regulation of genes associated with DNA repair, autophagy, colonies in Brittany between 2013 and 2016 (Supplementary Table 2). Among these immunity and tumour suppression underlies their extraordinary individuals, two were caught four times in consecutive years, two were caught three longevity. Our results show that bats have naturally evolved tran- times and 20 were captured twice, with the remainder being caught only once. Of scriptomic signatures that are known to extend lifespan in model all samples, 49 had the transcriptome and corresponding miRNome sequenced organisms, and also identify potential targets for future ageing simultaneously (Supplementary Table 1). The statistics derived from these samples are summarized in Supplementary Tables 3 and 7. intervention studies. The regulatory network uncovered suggests that these pathways are partially mediated by miRNA and provide Individual transcriptome assembly. The quality control pipeline is extensively a potential molecular mechanism that underlies their longevity- described in ref. 23. To enable a rapid genome-mapping process required for the associated transcriptomic signature. Using cutting-edge molecu- referenced-based assembly, the identical quality control reads, which consume lar techniques coupled with longitudinal field studies, we used large amounts of computational resources but make little contribution to the complexity of transcriptome assembly, were removed using FastUniq45, leaving only the diversity within nature to identify key targets and regions that one pair represented. These adaptor-free, high-quality and non-redundant paired- may regulate and control extraordinary ageing in mammals. These end reads were further used as inputs for assembly. We employed both de novo and results will drive a better understanding of the ageing process and reference-based methods to assemble individual transcriptomes, and the optimal can provide molecular targets for future ageing intervention stud- strategy was selected on the basis of the assembly quality. For the reference-based method, since the M. myotis genome is not yet available, the M. lucifugus genome ies, highlighting a new role for longitudinal field studies of non- was used as a reference as this is the phylogenetically closest, well-assembled model organisms. genome available. However, due to the genetic divergence (~30 million years) between these two species46, we carried out a series of tests to determine the Methods optimal number of mismatches per alignment for the genome-mapping step. Sampling. Bat capture and blood sampling were implemented in accordance In brief, the filtered RNA-Seq data were aligned to the M. lucifugus genome with the permits and ethical guidelines issued by ‘Arrêté’ by the Préfet du (MyoLuc 2.0) using Tophat2 (v.2.1.0)47, with different numbers of mismatches Morbihan and the University College Dublin ethics committee. Te sampling (~2–10). For all RNA-Seq samples, the mapping rates were categorized into procedures are exhaustively described in ref. 23. Briefy, female M. myotis bats, different groups based on the number of mismatches, and Student’s t-tests (all aged 0 to 7+ years (for example, frst marked with transponders as adult data followed normal distribution, Shapiro–Wilk test, P > 0.05) were conducted 6 years before subsequent recapture), were captured in Brittany, France. Tese between groups N2 and N3, groups N3 and N4 and so on, up to groups N9 and N10. individuals have been marked with unique transponders since 2010, and the Group N6 (six mismatches per alignment) was selected as the optimal mismatch initial ages of bats when frst captured were determined by examining the number, because there were no significant differences observed between groups N6 epiphyseal cartilage in their fnger bones. Individuals were recorded as juvenile and N7 based on the mapping rates (P > 0.05) (Supplementary Fig. 9). To exclude (0 years old) if the epiphyseal plates in their fnger bones were open; otherwise, any ambiguous mappings, we retained only the unique and concordant alignments 48 they were recorded as adult (1+ years old, true age unknown). Each year, using Samtools . The resulting BAM file was further used to reconstruct individuals recaptured were identifed by their transponder number and new transcripts by Cufflinks (v.2.2.1)49, with the parameters –GTF–guide and –F 0.1. captures (juvenile and adult) were transponded. For each individual, a volume of –GTF–guide allows the assembly of unannotated regions in the genome, while –F ~50–200 μl of blood, depending on the weight of the individual, was collected from 0.1 removes dubious isoforms with extremely low expression. The final assembly the uropatagial vein using a sterile needle (26 gauge). Blood samples were pipetted was determined by keeping those transcripts with fragment per kilobase of into cryotubes (2 ml, Nalgene labware) and immediately fash-frozen in liquid transcript per million (FPKM) mapped reads higher than zero. De novo assembly nitrogen. Haemostatic gel (Bloxang, Bausch & Lomb) was applied to the puncture was carried out using Trinity (v.2.1.1)50 in the strand-specific and paired-end mode to prevent further bleeding. Te bats were rapidly released afer being ofered water with the default parameters. For each sample, the quality of the reference-based 51 and food. Blood samples were maintained at −150 °C for long-term storage before and de novo assemblies was assessed using CEGMA (v.2.5) . On average, the RNA extraction. completeness of the reference-based assembly was 15% higher than that of the de novo assembly (Supplementary Fig. 10). Given their better quality, the reference- RNA extraction. Total RNA was extracted from whole blood using a RNAzol BD based assemblies were used to construct the ‘super’ transcriptome reference. kit (No. RB192, Molecular Research Center, Inc.), following the manufacturer’s instructions with minor modifications. RNA extraction is described in ref. 23. The ‘Super’ transcriptome assembly and annotation. The reference-based assemblies quantity and quality of RNA were assessed using a NanoDrop Spectrophotometer from all 100 samples were merged using Cuffmerge (v.2.2.1)49. The parameter (Thermo Scientific) and a Bioanalyzer 2100 (Aligent Technologies), respectively. —min-isoform-fraction 0.5 was used to discard the unreliable isoforms poorly All samples that met the criteria of having >2 μg of total RNA and an RNA supported by low sequencing coverage. The redundancy of the merged assembly 52 Integrity Number score >8.0 were chosen for mRNA-Seq Illumina sequencing was removed using CD-HIT with a sequence identity threshold of 95% (-c 0.95), 53 library preparation. For small RNA library preparation, 3 μg of total RNA per and FrameDP was employed to correct the misassembled transcripts with sample was required. unexpected ‘indels’ or stop codons. Functional characterization of the super transcriptome assembly was carried Library preparation and Illumina sequencing. For both RNA-Seq and small out using the following pipeline. Briefly, all transcripts were categorized into RNA library preparation, all qualified RNA samples were initially purified three groups: protein-coding RNA, lncRNA and miscRNA. To annotate the using a Turbo DNA-free kit (No. AM1907, Ambion) to remove residual DNA. protein-coding transcripts, the open reading frame (ORF) of each transcript was The Globin-Zero Gold rRNA Removal kit (Epicentre Illumina) was further predicted using FrameDP53. The transcripts with potential ORFs were queried employed to deplete unwanted rRNA and globin mRNA. Nevertheless, this against the Uniprot and Nr databases using BLASTX54, with an E-value <10−6. The kit was specifically designed for human and mouse, and performed less transcripts, with the effective hits which shared at least 60% similarity and 70% effectively on other species. A large amount of ribosomal RNA (mostly 7 s sequence coverage with the entries, were accordingly annotated as protein-coding rRNA) still remained after depletion based on the MiSeq result (Fasteris transcripts. The ORF potential transcripts with no effective hits in either database SA). Therefore, an additional step of 7 s rRNA depletion was carried out were regarded as miscRNA. lncRNA were predicted from the transcripts with no (Fasteris SA). The probes specifically designed for 7 s rRNA removal are obvious ORFs defined by FrameDP53. Firstly, these transcripts were compared to 55 54 5′-TCCTTAGGCAACCTGGTGG-3′, 5′-GGGAGGTCACCATATTGATG-3′ and the Pfam database to search for conserved domains using BLASTX with an 6 5′-GGCAACCTGGTGGCCCCCCGCTCCCGGGAGG-3′. To assess the efficiency E-value <10− . Those transcripts that contained or overlapped with any conserved of the 7 s rRNA probes, we constructed two libraries with and without 7 s rRNA motifs were excluded. In addition, those transcripts with coding potential scores depletion from the same sample, HWJ-8. The MiSeq titration run demonstrated >0.3, as evaluated by CPAT56, were also discarded. The threshold of the coding that the percentage of short reads that mapped to 7 s rRNA reduced from 31.65 to potential score is suggested in ref. 56. The final set of lncRNAs was determined 17.78 after depletion (Supplementary Fig. 8), implying that the 7 s rRNA depletion by removing those filtered transcripts with lengths shorter than 200 bp. The step performed effectively. remainder of the uncharacterized transcripts were considered as miscRNA. RNA-Seq libraries were prepared using the TruSeq stranded mRNA kit (Illumina), following the manufacturer’s protocols. Sequencing was performed Transcript expression analyses. For each RNA-Seq sample, adaptor-free, high- on Illumina HiSeq 2500 platforms, with the sequencing depth to a minimum quality reads (without redundancy removal) were used to quantify the reference of 50 million and 125-bp paired-end reads per sample. Small RNA libraries transcripts using Salmon (v.0.9)57 with the parameters –ISF and –F 31, indicating were constructed using the TruSeq small RNA kit (Illumina) and were further the library type (I, first-forward; S, strand-specific; F, paired-end) and k-mer sequenced on Illumina HiSeq 2500 platforms, to generate a minimum of 8 million length, respectively. Before expression analysis, RNA-Seq samples with a CEGMA 50-bp single-end reads per sample. index of the reference-based assembly <0.85, were removed. A transcript was

Nature Ecology & Evolution | VOL 3 | JULY 2019 | 1110–1120 | www.nature.com/natecolevol 1117 Articles Nature Ecology & Evolution considered expressed if it had a cumulative raw read count >100 across all samples. To gain a global view of transcriptomic shifts during ageing, for each species This led to a matrix containing expression data of 48,749 transcripts across PCA was performed based on expression data of all expressed genes using the 88 samples. R package prcomp (Supplementary Fig. 5). For each species, gene expression data To examine the complexity of the bat blood transcriptome, the contribution were normalized using TMM and further transformed to z-score. To compare of each transcript to total transcriptional output was measured in each sample. age-related gene expression patterns across species, for each gene we measured the To achieve this, all transcript per million (TPM) values from each sample were Spearman’s rank correlation coefficient between its expression and age across all sorted from most to least and their relative fractions were calculated by dividing four mammals. Since the age cohorts in human, mouse and wolf datasets started them by the sum of all TPM values respectively. A functional enrichment analysis from adult, samples from juveniles (0 years old) were excluded from correlation of the top 100 highly expressed transcripts was performed using Metascape58. We analyses in M. myotis. We plotted the distribution of Spearman’s correlation also explored the global transcriptomic similarity across 88 samples. Based on coefficient for each species and investigated the top 200 genes (100 positive and the expression data of 48,749 transcripts, pairwise Spearman’s rank correlation 100 negative) that exhibited the strongest correlation with age (Fig. 4a). Functional coefficient tests were carried out using the R package cor (v.3.0). Before the enrichment analyses of these top genes were performed using Metascape58. analysis, TPM values were log2-transformed (log2 (TPM + 1)). The average linkage To investigate the expression pattern at the pathway level, within each species method was applied to all correlation tests. we employed the median z-scores of all genes under each enriched gene ontology term to represent their overall pattern with age. The z-scores were transformed from Gene expression variance analyses. In this study, we focused largely on gene-level Spearman’s correlation coefficient within a species. These enriched gene ontology analysis and investigated only 31,460 protein-coding transcripts, as most lncRNAs terms were selected on the basis of their distinct expression patterns with age in and miscRNAs are currently functionally unknown. To obtain gene expression M. myotis (Fig. 2a–c) and were associated with ageing processes. The full list of gene estimates, raw expression counts of the transcripts which had the same BLAST ontology terms that were used for comparison is given in Supplementary Data 3. hits in the Uniprot or Nr databases were accordingly aggregated to the gene-level In addition, we also ascertained the expression levels of 307 genes curated in counts using the Bioconductor R package tximport59. Thus, 31,460 protein-coding the GenAge database (Build 19) that are highly associated with human ageing16, transcripts corresponded to 12,263 protein-coding genes. All downstream analyses across these four mammals. Only 207 genes were commonly expressed, so were conducted based on these 12,263 genes. their correlation with age was investigated (Supplementary Data 4). For the top For the age cohort analyses, samples with unambiguous ages (0, 1, 2, 3, 4, 5, 6) 20 most age-correlated genes (Fig. 4b), we further investigated their functional were employed, together with 7+ regarded as the oldest age cohort (Supplementary association with 207 human ageing-associated genes using the STRING database64 Fig. 1a). Gene expression counts (Supplementary Data 5) were trimmed mean (Supplementary Table 8). of M-values (TMM)-normalized, log2-transformed and further converted to z-scores. A linear mixed model (LMM) was used to evaluate the contribution of miRNA expression analyses. Additionally, we also sequenced 50 small RNA potential variables (see below) to gene expression variation. Normalized gene libraries from M. myotis whole-blood samples, 49 of which originated from the counts (n = 12,263) were considered as dependent variables, whereas age was same samples that had corresponding transcriptomes previously sequenced considered as an explanatory variable together with other variables including the (Supplementary Table 1). For each library, miRNA profiling and quantification individual, colony, year of capture and sequencing batch effect. With the exception were analysed using the miRDeep2 pipeline66. The detailed steps and parameters of age being modelled as a fixed effect, all other variables were modelled as used are described fully in ref. 19. By searching the miRBase (v.32)67, mature random effects. The LMM was implemented using the Bioconductor R package miRNA candidates were categorized into the ‘known’ and ‘novel’ groups. Here variancePartition60 with the following formula: we focused only on the ‘known’ miRNAs, as these are well documented and experimentally validated. The novel miRNAs were usually expressed at a lower Gene Expression ~+Age(1I∣+ndividual) (1∣Colony) level, thus having limited impact on gene regulation. For miRNA expression analysis, conserved miRNAs across all libraries were extracted as described +∣(1 Year − of − capture) +∣(1 Batcheffect) in ref. 19. Only miRNAs that were presented in at least 80% of all samples were used in the expression analysis. The samples from individuals 0, 1, 2, 3, 4, 5, 6 and 7+ years of age were used Gene co-expression network analyses. To identify gene expression for age cohort analysis. Quantile normalization was applied to the raw miRNA changes with age, pairwise differential gene expression analyses were expression counts (Supplementary Data 6). The normalized values were further performed using the R packages DESeq2 (ref. 61) and EdgeR62. For both log -transformed and converted to z-scores. PCA was performed using prcomp methods, FDR 0.05 and an absolute value of log (fold change) 0.5 were 2 < 2 > (v.3.0) in R68. Pairwise differential miRNA expression analysis was conducted using used to define differentially expressed genes. To reduce the rate of false-negatives DESeq2 (ref. 61), and differentially expressed miRNAs were defined according to and obtain a wide range of age-related candidates for pattern analysis, we the criteria stated above. maximized the number of differentially expressed genes from both methods. All network analyses were implemented in R. Gene co-expression analysis was miRNA–mRNA expression correlation analyses. miRNA–mRNA interaction performed using the R package WGCNA (v.1.63)63. Genes that exhibited no was investigated to elucidate the mechanisms of miRNA-directed regulation. First, differential expression between any pairs of age cohorts were excluded from this 3′-UTR sequences were initially extracted from the assembled protein-coding analysis because these non-varying genes usually represent noise for pattern transcripts (n = 31,460) using ExUTR44. miRNA targets were predicted using detection. The differentially expressed genes from pairwise differential expression miRanda69, with the parameters -strict and -en -20. For each miRNA–mRNA pair analyses were clustered into different modules, and the correlation between each predicted above, Spearman’s rank correlation coefficient was calculated. To achieve module and age was calculated (Supplementary information). After Benjamini– this, samples that had both transcriptome and miRNome sequenced were selected. Hochberg correction, modules with P 0.05 were considered to have significant < The raw expression counts were quantile-normalized and log -transformed. correlation with age. For each module, a functional enrichment analysis was 2 For each pair, the correlation coefficient was computed using cor (v.3.0) in R68. carried out using Metascape58. In particular, we established a gene product Negatively correlated miRNA–mRNA pairs were determined according to the network encompassing 107 genes that were enriched in DNA repair (Gene threshold of FDR < 0.1. Functional enrichment analysis of miRNA-mediated Ontology: 0006281) using STRING64. These genes were clustered in modules mRNA was performed using Metascape58. In particular, we investigated those that exhibited a positive correlation with age in M. myotis. The interaction was genes that were involved in cell cycle regulation and DNA repair pathways and that predicted based on different sources of evidence, including curated databases, are simultaneously regulated by miRNA in M. myotis. functional experiments and gene neighbourhood, co-occurrence and co- expression. Kyoto Encyclopedia of Genes and Genomes pathway enrichment Reporting Summary. Further information on research design is available in the analysis was performed using STRING64. Nature Research Reporting Summary linked to this article. Comparative transcriptomic analyses across mammals. To ascertain how the longevity-associated genes and pathways observed in M. myotis changed with Data availability age in other mammals, we took advantage of existing ageing blood transcriptome The raw data used in this study have been deposited in the National Center for datasets available from three independent cohort studies on other mammals Biotechnology Information’s BioProject under the accession PRJNA503704. The and compared their ageing transcriptomic signatures to those of bats. These additional data supporting the conclusions in this paper can be available in the comparative datasets included human (H. sapiens, n = 147, age ~25–75 years)24, Supplementary Data 1–6. mouse (Mus musculus, n = 25, ~0.2–2.5 years)15 and wolf (Canis lupus, n = 26, ~1–9 years)65. For human and mouse, raw gene expression counts were obtained Code availability 15 24 from refs. and , respectively, while for wolf we analysed the raw RNA-Seq data The custom scripts have been deposited in GitHub (https://github.com/ to generate gene expression estimates using the pipeline established for UCDBatLab/Longitudinal_myoMyo_transcriptome). M. myotis, as described above. For human and mouse, gene ontology term enrichment analyses of the top 100 highly expressed genes were performed using Metascape55, and the percentage of age-related gene expression variance was Received: 19 December 2018; Accepted: 1 May 2019; estimated using the same method as for M. myotis. Published online: 10 June 2019

1118 Nature Ecology & Evolution | VOL 3 | JULY 2019 | 1110–1120 | www.nature.com/natecolevol Nature Ecology & Evolution Articles

References 33. Ortega-Molina, A. et al. PTEN positively regulates brown adipose function, 1. Lopez-Otín, C., Blasco, M. A., Partridge, L., Serrano, M. & Kroemer, G. Te energy expenditure, and longevity. Cell Metab. 15, 382–394 (2012). hallmarks of aging. Cell 153, 1194–1217 (2013). 34. Baker, D. J. et al. Increased expression of BubR1 protects against aneuploidy 2. Gladyshev, V. N. Aging: progressive decline in ftness due to the rising and cancer and extends healthy lifespan. Nat. Cell Biol. 15, 96–102 (2013). deleteriome adjusted by genetic, environmental, and stochastic processes. 35. Orr, W. C. et al. Overexpression of glutamate-cysteine ligase extends life span Aging Cell 15, 594–602 (2016). in Drosophila melanogaster. J. Biol. Chem. 280, 37331–37338 (2005). 3. Kirkwood, T. B. Understanding the odd science of aging. Cell 120, 36. Satoh, A. et al. Sirt1 extends life span and delays aging in mice through 437–447 (2005). the regulation of Nk2 homeobox 1 in the DMH and LH. Cell Metab. 18, 4. Ageing and Health: Fact Sheet 404 (WHO, 2015). 416–430 (2013). 5. de Magalhaes, J. P. Te scientifc quest for lasting youth: prospects for curing 37. Hofmann, J. W. et al. Reduced expression of MYC increases longevity and aging. Rejuvenation Res. 17, 458–467 (2014). enhances healthspan. Cell 160, 477–488 (2015). 6. Austad, S. N. Methuselah’s Zoo: how nature provides us with clues 38. Zhang, G. et al. Hypothalamic programming of systemic ageing involving for extending human health span. J. Comp. Pathol. 142 (Suppl. 1), IKK-beta, NF-kappaB and GnRH. Nature 497, 211–216 (2013). S10–S21 (2010). 39. Henriksson, M. & Luscher, B. Proteins of the Myc network: essential 7. Munshi-South, J. & Wilkinson, G. S. Bats and birds: exceptional longevity regulators of cell growth and diferentiation. Adv. Cancer Res. 68, despite high metabolic rates. Ageing Res. Rev. 9, 12–19 (2010). 109–182 (1996). 8. Seluanov, A., Gladyshev, V. N., Vijg, J. & Gorbunova, V. Mechanisms 40. Jafri, M. A., Ansari, S. A., Alqahtani, M. H. & Shay, J. W. Roles of telomeres of cancer resistance in long-lived mammals. Nat. Rev. Cancer 18, and telomerase in cancer, and advances in telomerase-targeted therapies. 433–441 (2018). Genome Med. 8, 69 (2016). 9. Tian, X., Seluanov, A. & Gorbunova, V. Molecular mechanisms determining 41. Wu, K. J. et al. Direct activation of TERT transcription by c-MYC. Nat. lifespan in short- and long-lived species. Trends Endocrin. Met. 28, Genet. 21, 220–224 (1999). 722–734 (2017). 42. Sun, Q. et al. miR-146a functions as a tumor suppressor in prostate cancer by 10. Teeling, E. C. et al. Bat biology, genomes, and the bat1k project: to generate targeting Rac1. Prostate 74, 1613–1621 (2014). chromosome-level genomes for all living bat species. Annu. Rev. Anim. Biosci. 43. Komatsu, S. et al. Circulating miR-18a: a sensitive cancer screening biomarker 6, 23–46 (2018). in human cancer. Vivo 28, 293–297 (2014). 11. Ball, H. C., Levari-Shariati, S., Cooper, L. N. & Aliani, M. Comparative 44. Huang, Z. & Teeling, E. C. ExUTR: a novel pipeline for large-scale prediction metabolomics of aging in a long-lived bat: insights into the physiology of of 3′-UTR sequences from NGS data. BMC Genomics 18, 847 (2017). extreme longevity. PloS One 13, e0196154 (2018). 45. Xu, H. et al. FastUniq: a fast de novo duplicates removal tool for paired short 12. Foley, N. M. et al. Growing old, yet staying young: the role of telomeres in reads. PloS One 7, e52249 (2012). bats’ exceptional longevity. Sci. Adv. 4, eaao0926 (2018). 46. Ruedi, M. et al. Molecular phylogenetic reconstructions identify East Asia as 13. Jebb, D. et al. Population level mitogenomics of long-lived bats reveals the cradle for the evolution of the cosmopolitan genus Myotis (Mammalia, dynamic heteroplasmy and challenges the free radical theory of ageing. Chiroptera). Mol. Phylogenet. Evol. 69, 437–449 (2013). Sci. Rep. 8, 13634 (2018). 47. Kim, D. et al. TopHat2: accurate alignment of transcriptomes in the presence 14. Hughes, G. M., Leech, J., Puechmaille, S. J., Lopez, J. V. & Teeling, E. C. Is of insertions, deletions and gene fusions. Genome Biol. 14, R36 (2013). there a link between aging and microbiome diversity in exceptional 48. Li, H. et al. Te sequence alignment/map format and SAMtools. mammalian longevity? PeerJ 6, e4174 (2018). Bioinformatics 25, 2078–2079 (2009). 15. Aramillo Irizar, P. et al. Transcriptomic alterations during ageing refect the 49. Trapnell, C. et al. Diferential gene and transcript expression analysis shif from cancer to degenerative diseases in the elderly. Nat. Commun. 9, of RNA-seq experiments with TopHat and Cufinks. Nat. Protoc. 7, 327 (2018). 562–578 (2012). 16. de Magalhaes, J. P., Curado, J. & Church, G. M. Meta-analysis of age-related 50. Haas, B. J. et al. De novo transcript sequence reconstruction from RNA-seq gene expression profles identifes common signatures of aging. Bioinformatics using the Trinity platform for reference generation and analysis. Nat. Protoc. 25, 875–881 (2009). 8, 1494–1512 (2013). 17. Fushan, A. A. et al. Gene expression defnes natural changes in mammalian 51. Parra, G., Bradnam, K. & Korf, I. CEGMA: a pipeline to accurately annotate lifespan. Aging Cell 14, 352–365 (2015). core genes in eukaryotic genomes. Bioinformatics 23, 1061–1067 (2007). 18. Peters, M. J. et al. Te transcriptional landscape of age in human peripheral 52. Li, W. & Godzik, A. Cd-hit: a fast program for clustering and comparing blood. Nat. Commun. 6, 8570 (2015). large sets of protein or nucleotide sequences. Bioinformatics 22, 19. Huang, Z., Jebb, D. & Teeling, E. C. Blood miRNomes and transcriptomes 1658–1659 (2006). reveal novel longevity mechanisms in the long-lived bat, Myotis myotis. BMC 53. Gouzy, J., Carrere, S. & Schiex, T. FrameDP: sensitive peptide detection on Genomics 17, 906 (2016). noisy matured sequences. Bioinformatics 25, 670–671 (2009). 20. Kim, E. B. et al. Genome sequencing reveals insights into physiology and 54. Altschul, S. F., Gish, W., Miller, W., Myers, E. W. & Lipman, D. J. Basic local longevity of the naked mole rat. Nature 479, 223–227 (2011). alignment search tool. J. Mol. Biol. 215, 403–410 (1990). 21. Seim, I. et al. Genome analysis reveals insights into physiology and longevity 55. Finn, R. D. et al. Te Pfam protein families database: towards a more of the Brandt’s bat Myotis brandtii. Nat. Commun. 4, 2212 (2013). sustainable future. Nucleic Acids Res. 44, D279–D285 (2016). 22. Li, Y. & de Magalhaes, J. P. Accelerated protein evolution analysis reveals 56. Wang, L. et al. CPAT: Coding-Potential Assessment Tool using an alignment- genes and pathways associated with the evolution of mammalian longevity. free logistic regression model. Nucleic Acids Res. 41, e74 (2013). Age 35, 301–314 (2013). 57. Patro, R., Duggal, G., Love, M. I., Irizarry, R. A. & Kingsford, C. Salmon 23. Huang, Z. et al. A nonlethal sampling method to obtain, generate and provides fast and bias-aware quantifcation of transcript expression. Nat. assemble whole blood transcriptomes from small, wild mammals. Mol. Ecol. Methods 14, 417–419 (2017). Resour. 16, 150–162 (2016). 58. Tripathi, S. et al. Meta- and orthogonal integration of infuenza “OMICs” 24. Mele, M. et al. Human genomics. Te human transcriptome across tissues data defnes a role for UBR4 in virus budding. Cell Host Microbe 18, and individuals. Science 348, 660–665 (2015). 723–735 (2015). 25. Blagosklonny, M. V. Cell cycle arrest is not senescence. Aging 3, 59. Soneson, C., Love, M. I. & Robinson, M. D. Diferential analyses for 94 (2011). RNA-seq: transcript-level estimates improve gene-level inferences. F1000Res. 26. de Magalhaes, J. P. & Passos, J. F. Stress, cell senescence and organismal 4, 1521 (2015). ageing. Mech. Ageing Dev. 170, 2–9 (2018). 60. Hofman, G. E. & Schadt, E. E. variancePartition: interpreting drivers of 27. Franceschi, C., Garagnani, P., Vitale, G., Capri, M. & Salvioli, S. Infammaging variation in complex gene expression studies. BMC Bioinformatics 17, and ‘Garb-aging’. Trends Endocrin. Met. 28, 199–212 (2017). 483 (2016). 28. Kacprzyk, J. et al. A potent anti-infammatory response in bat macrophages 61. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and may be linked to extended longevity and viral tolerance. Acta Chiropt. 19, dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014). 219–228 (2017). 62. Robinson, M. D., McCarthy, D. J. & Smyth, G. K. edgeR: a Bioconductor 29. Wang, L. F., Walker, P. J. & Poon, L. L. Mass extinctions, biodiversity and package for diferential expression analysis of digital gene expression data. mitochondrial function: are bats ‘special’ as reservoirs for emerging viruses? Bioinformatics 26, 139–140 (2010). Curr. Opin. Virol. 1, 649–657 (2011). 63. Langfelder, P. & Horvath, S. WGCNA: an R package for weighted correlation 30. Gems, D. & Partridge, L. Genetics of longevity in model organisms: debates network analysis. BMC Bioinformatics 9, 559 (2008). and paradigm shifs. Annu. Rev. Physiol. 75, 621–644 (2013). 64. Szklarczyk, D. et al. Te STRING database in 2017: quality-controlled 31. Tacutu, R. et al. Human Ageing Genomic Resources: new and updated protein-protein association networks, made broadly accessible. Nucleic Acids databases. Nucleic Acids Res. 46, D1083–D1090 (2018). Res. 45, D362–D368 (2017). 32. Ortega-Molina, A. & Serrano, M. PTEN in cancer, metabolism, and aging. 65. Charruau, P. et al. Pervasive efects of aging on gene expression in wild Trends Endocrinol. Metab. 24, 184–189 (2013). wolves. Mol. Biol. Evol. 33, 1967–1978 (2016).

Nature Ecology & Evolution | VOL 3 | JULY 2019 | 1110–1120 | www.nature.com/natecolevol 1119 Articles Nature Ecology & Evolution

66. Friedlander, M. R., Mackowiak, S. D., Li, N., Chen, W. & Rajewsky, N. programme to Z.H.). The French fieldwork was supported by a Contrat Nature Grant miRDeep2 accurately identifes known and hundreds of novel microRNA awarded to Bretagne Vivante. genes in seven animal clades. Nucleic Acids Res. 40, 37–52 (2012). 67. Grifths-Jones, S., Grocock, R. J., van Dongen, S., Bateman, A. & Enright, A. Author contributions J. miRBase: microRNA sequences, targets and gene nomenclature. Nucleic E.C.T and Z.H. devised the study. M. myotis samples were collected by F.T., S.J.P., E.C.T., Acids Res. 34, D140–D144 (2006). E.J.P., N.M.F., D.J., C.V.W. and Z.H. RNA extraction was performed by C.V.W. and Z.H. 68. Team RC. R: A language and environment for statistical computing (2013). All data analyses were performed by Z.H. Z.H. is responsible for the Figures presented 69. Enright, A. J. et al. MicroRNA targets in Drosophila. Genome Biol. 5, throughout. The manuscript was written by E.C.T. and Z.H. with input from all authors. R1 (2003). Competing interests Acknowledgements The authors declare no competing interests. We acknowledge and thank the members of Bretagne Vivante and local volunteers and students from University College Dublin for their extensive help in sample collection, and the various owners/local authorities for allowing access to their sites. Additional information We would also like to thank M. Bekaert, M. Clarke, G. Hughes and J. Kacprzyk for Supplementary information is available for this paper at https://doi.org/10.1038/ helpful discussions of the analyses. We acknowledge the Irish Centre for High-End s41559-019-0913-3. Computing for the provision of computational facilities and support. This project was Reprints and permissions information is available at www.nature.com/reprints. funded by a European Research Council Research Grant (No. ERC-2012-StG311000 Correspondence and requests for materials should be addressed to E.C.T. to E.C.T.), a UCD Wellcome Institutional Strategic Support Fund, financed jointly by University College Dublin and SFI-HRB-Wellcome Biomedical Research Partnership Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in (No. 204844/Z/16/Z to E.C.T.), an Irish Research Council Consolidator Laureate Award published maps and institutional affiliations. (to E.C.T.) and a China Scholarship Council studentship (under the UCD-CSC funding © The Author(s), under exclusive licence to Springer Nature Limited 2019

Corresponding author(s): Emma C. TEELING

Last updated by author(s): Dec 18, 2018 Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Research policies, see Authors & Referees and the Editorial Policy Checklist.

Statistics For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section. n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one- or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section. A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals)

For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated

Our web collection on statistics for biologists contains articles on many of the points above. Software and code Policy information about availability of computer code Data collection No software was used for data collection

Data analysis Cutadapt (v1.5) NGS QC Toolkit (v2.3) FastUniq Tophat2 (v2.1.0) Samtools (v0.1.18) Cufflinks (v2.2.1) Trinity (v2.1.1) CEGMA (v2.5) CD-HIT (v4.5.7) FrameDP (v2.2.1) NCBI-BLAST (v2.2.25) CPAT (v1.2) Salmon (v0.9) Metascape R(v3.0) October 2018 tximport (v3.8) variancePartition (v1.12.0) DESeq2 EdgeR WGCNA (v1.63) GenAge (Build 19) STRING (v10.5) miRDeep (v2.0.0.8)

1 miRBase (v32)

EXUTR (v0.1.0) nature research | reporting summary miRanda (v3.3a) For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors/reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information. Data Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: - Accession codes, unique identifiers, or web links for publicly available datasets - A list of figures that have associated raw data - A description of any restrictions on data availability

The raw data used in this study have been deposited in the National Center for Biotechnology Information’s BioProject under the accession PRJNA503704. The additional data supporting the conclusions in this paper can be available in the Supplementary Data 1-6. Other intermediate datasets generated during this study are available from the corresponding author on reasonable request.

Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection. Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design All studies must disclose on these points even when the disclosure is negative. Sample size 150 samples collected from 71 individuals in five colonies were used for transcriptome sequencing (100 RNA-Seq) and small RNA sequencing (50 miRNA-Seq). 72 qualified RNA-Seq samples, with unambiguous age (0, 1, 2, 3, 4, 5, 6) and 7+ year-old cohort, were used in the longitudinal cohort analyses. 49 pairs of RNA-Seq and corresponding miRNA-Seq samples were used in the miRNA-mRNA network analyses.

Data exclusions 12 RNA-Seq samples were excluded from data analyses due to low CEGMA scores.

Replication The experimental findings were reliably reproduced. Adequate numbers of biological replicates were used for longitudinal transcriptome analyses.

Randomization The samples for RNA-Seq and miRNA-Seq were collected from five different colonies during the year 2013-2016. RNA extraction and library preparation were performed with the same protocol & procedure, and Illumina sequencing was done in the same company.

Blinding The investigators were not blinded to species, age, sex, year-of-capture and colony information during sample collection. However, clustering analyses were done with blinded information.

Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response. Materials & experimental systems Methods n/a Involved in the study n/a Involved in the study Antibodies ChIP-seq Eukaryotic cell lines Flow cytometry Palaeontology MRI-based neuroimaging Animals and other organisms

Human research participants October 2018 Clinical data

Animals and other organisms Policy information about studies involving animals; ARRIVE guidelines recommended for reporting animal research Laboratory animals This study did not involve laboratory animals

2 100 RNA-Seq samples and 50 miRNA-Seq samples were collected from 71 female Myotis myotis bats. These individuals were

Wild animals nature research | reporting summary caught in five colonies in Brittany during 2013-2016 (Supplementary Table 1-2). Of all individuals, 2 were caught four times in consecutive years, 2 were caught three times and 20 were caught twice, with the rest of individuals being caught only once.

Bats were caught in custom harp traps as they left the roost and were initially placed in individual cloth bags. For each bat, the uropatagium was extended to expose and choose a uropatagial vein. A sterile needle (26-gauge) was used to pierce the uropatagial vein on its dorsal side, creating a drop of blood on the uropatagium surface from which a volume of 80–200 microlitre (depending on the weight of the individual) was pipetted into a cryotube (2 ml, Nalgene labware). Blood samples were flash frozen in liquid nitrogen within a maximum of a few minutes after the vein was pierced. Haemostatic gel was applied on the vein to prevent further bleeding (Bloxang, Bausch & Lomb). Water and food were offered to the bats, and they were rapidly released. Blood samples were stored at -150 °C.

Field-collected samples Blood samples collected from the field were flash frozen in liquid nitrogen. The samples were delivered back to the lab using dry shippers within a few days. The samples were further transfered to -150°C freezers for longterm storage.

Ethics oversight Bat capture and blood sampling were implemented in accordance with the permits and ethical guidelines issued by ‘Arrêté’ by the Préfet du Morbihan and the University College Dublin ethics committee. Note that full information on the approval of the study protocol must also be provided in the manuscript. October 2018