Philippa Kaur

1.Bonan, G. B. Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science 320, 1444–1449 (2008).CAS 
Article 

Google Scholar 
2.Saatchi, S. S. et al. Benchmark map of forest carbon stocks in tropical regions across three continents. Proc. Natl Acad. Sci. USA 108, 9899–9904 (2011).CAS 
Article 

Google Scholar 
3.Veldkamp, E., Schmidt, M., Powers, J. S. & Corre, M. D. Deforestation and reforestation impacts on soils in the tropics. Nat. Rev. Earth Environ. 1, 590–605 (2020).Article 

Google Scholar 
4.Ceccherini, G. et al. Abrupt increase in harvested forest area over Europe after 2015. Nature 583, 72–77 (2020).CAS 
Article 

Google Scholar 
5.Mitchard, E. T. A. The tropical forest carbon cycle and climate change. Nature 559, 527–534 (2018).CAS 
Article 

Google Scholar 
6.Curran, L. M. et al. Lowland forest loss in protected areas of Indonesian Borneo. Science 303, 1000–1003 (2004).CAS 
Article 

Google Scholar 
7.Friedl, A. et al. MODIS Collection 5 Global Land Cover: Algorithm Refinements and Characterization of New Datasets, 2001–2012 Collection 5.1 (Boston University, 2010).8.Hansen, M. C. et al. High-resolution global maps of 21st-century forest cover change. Science 342, 850–853 (2013).CAS 
Article 

Google Scholar 
9.Margono, B. A., Potapov, P. V., Turubanova, S., Stolle, F. & Hansen, M. C. Primary forest cover loss in Indonesia over 2000–2012. Nat. Clim. Change 4, 730–735 (2014).Article 

Google Scholar 
10.Turubanova, S. et al. Ongoing primary forest loss in Brazil, Democratic Republic of the Congo, and Indonesia. Environ. Res. Lett. 13, 074028 (2018).Article 

Google Scholar 
11.Searchinger, T. et al. Creating a Sustainable Food Future: A Menu of Solutions to Feed Nearly 10 Billion People by 2050 (World Resources Institute, 2019).12.Gibbs, H. K. et al. Tropical forests were the primary sources of new agricultural land in the 1980s and 1990s. Proc. Natl Acad. Sci. USA 107, 16732–16737 (2010).CAS 
Article 

Google Scholar 
13.Tyukavina, A. et al. Congo Basin forest loss dominated by increasing smallholder clearing. Sci. Adv. 4, eaat2993 (2018).Article 

Google Scholar 
14.Achard, F. et al. Determination of tropical deforestation rates and related carbon losses from 1990 to 2010. Glob. Change Biol. 20, 2540–2554 (2014).Article 

Google Scholar 
15.Baccini, A. et al. Tropical forests are a net carbon source based on aboveground measurements of gain and loss. Science 358, 230–234 (2017).CAS 
Article 

Google Scholar 
16.Aide, T. M. et al. Woody vegetation dynamics in the tropical and subtropical Andes from 2001 to 2014: satellite image interpretation and expert validation. Glob. Change Biol. 25, 2112–2126 (2019).Article 

Google Scholar 
17.Song, X. P. et al. Global land change from 1982 to 2016. Nature 560, 639–643 (2018).CAS 
Article 

Google Scholar 
18.Körner, C. et al. A global inventory of mountains for bio-geographical applications. Alp. Bot. 127, 1–15 (2017).Article 

Google Scholar 
19.Zeng, Z. et al. Highland cropland expansion and forest loss in Southeast Asia in the twenty-first century. Nat. Geosci. 11, 556–562 (2018).CAS 
Article 

Google Scholar 
20.Zeng, Z., Gower, D. B. & Wood, E. F. Accelerating forest loss in Southeast Asian Massif in the 21st century: a case study in Nan Province, Thailand. Glob. Change Biol. 24, 4682–4695 (2018).Article 

Google Scholar 
21.Zarin, D. J. et al. Can carbon emissions from tropical deforestation drop by 50% in 5 years? Glob. Change Biol. 22, 1336–1347 (2016).Article 

Google Scholar 
22.Spracklen, D. & Righelato, R. Tropical montane forests are a larger than expected global carbon store. Biogeosciences 11, 2741–2754 (2014).CAS 
Article 

Google Scholar 
23.Miettinen, J., Shi, C. & Liew, S. C. Deforestation rates in insular Southeast Asia between 2000 and 2010. Glob. Change Biol. 17, 2261–2270 (2011).Article 

Google Scholar 
24.Austin, K. G. et al. What causes deforestation in Indonesia? Environ. Res. Lett. 14, 024007 (2019).Article 

Google Scholar 
25.Hansen, M. et al. Response to comment on ‘high-resolution global maps of 21st-century forest cover change’. Science 344, 981–981 (2014).CAS 
Article 

Google Scholar 
26.Chan, N., Xayvongsa, L. & Takeda, S. in Environmental Resources Use and Challenges in Contemporary Southeast Asia (eds Lopez, M. I. & Suryomenggolo, J.) 231–246 (Springer, 2018).27.Thompson, J. R., Carpenter, D. N., Cogbill, C. V. & Foster, D. R. Four centuries of change in northeastern United States forests. PLoS ONE 8, e72540 (2013).CAS 
Article 

Google Scholar 
28.Lawrence, D. & Vandecar, K. Effects of tropical deforestation on climate and agriculture. Nat. Clim. Change 5, 27–36 (2015).Article 

Google Scholar 
29.Zeng, Z. et al. Deforestation-induced warming over tropical mountain regions regulated by elevation. Nat. Geosci. 14, 23–29 (2021).CAS 
Article 

Google Scholar 
30.Senior, R. A., Hill, J. K., Benedick, S. & Edwards, D. P. Tropical forests are thermally buffered despite intensive selective logging. Glob. Change Biol. 24, 1267–1278 (2018).Article 

Google Scholar 
31.Senior, R. A., Hill, J. K., González del Pliego, P., Goode, L. K. & Edwards, D. P. A pantropical analysis of the impacts of forest degradation and conversion on local temperature. Ecol. Evol. 7, 7897–7908 (2017).Article 

Google Scholar 
32.Sodhi, N. S. et al. The state and conservation of Southeast Asian biodiversity. Biodivers. Conserv. 19, 317–328 (2010).Article 

Google Scholar 
33.Ahrends, A. et al. Current trends of rubber plantation expansion may threaten biodiversity and livelihoods. Glob. Environ. Change 34, 48–58 (2015).Article 

Google Scholar 
34.Edwards, D. P. et al. Degraded lands worth protecting: the biological importance of Southeast Asia’s repeatedly logged forests. Proc. R. Soc. B 278, 82–90 (2011).Article 

Google Scholar 
35.Srinivasan, U., Elsen, P. R. & Wilcove, D. S. Annual temperature variation influences the vulnerability of montane bird communities to land-use change. Ecography 42, 2084–2094 (2019).Article 

Google Scholar 
36.Rahbek, C. et al. Humboldt’s enigma: what causes global patterns of mountain biodiversity? Science 365, 1108–1113 (2019).CAS 
Article 

Google Scholar 
37.Guo, F., Lenoir, J. & Bonebrake, T. C. Land-use change interacts with climate to determine elevational species redistribution. Nat. Commun. 9, 1315 (2018).Article 
CAS 

Google Scholar 
38.Elsen, P. R., Monahan, W. B. & Merenlender, A. M. Topography and human pressure in mountain ranges alter expected species responses to climate change. Nat. Commun. 11, 1974 (2020).CAS 
Article 

Google Scholar 
39.Spracklen, D. V., Arnold, S. R. & Taylor, C. M. Observations of increased tropical rainfall preceded by air passage over forests. Nature 489, 282–285 (2012).CAS 
Article 

Google Scholar 
40.Spracklen, D. V. & Garcia-Carreras, L. The impact of Amazonian deforestation on Amazon Basin rainfall. Geophys. Res. Lett. 42, 9546–9552 (2015).Article 

Google Scholar 
41.Cheng, L. et al. Quantifying the impacts of vegetation changes on catchment storage–discharge dynamics using paired-catchment data. Water Resour. Res. 53, 5963–5979 (2017).Article 

Google Scholar 
42.Chappell, A., Baldock, J. & Sanderman, J. The global significance of omitting soil erosion from soil organic carbon cycling schemes. Nat. Clim. Change 6, 187–191 (2015).Article 
CAS 

Google Scholar 
43.Yue, Y. et al. Lateral transport of soil carbon and land–atmosphere CO2 flux induced by water erosion in China. Proc. Natl Acad. Sci. USA 113, 6617–6622 (2016).CAS 
Article 

Google Scholar 
44.Ziegler, A. D. et al. Carbon outcomes of major land-cover transitions in SE Asia: great uncertainties and REDD+ policy implications. Glob. Change Biol. 18, 3087–3099 (2012).Article 

Google Scholar 
45.Brinck, K. et al. High resolution analysis of tropical forest fragmentation and its impact on the global carbon cycle. Nat. Commun. 8, 14855 (2017).CAS 
Article 

Google Scholar 
46.Fox, J., Castella, J. C. & Ziegler, A. D. Swidden, rubber and carbon: can REDD+ work for people and the environment in montane mainland Southeast Asia? Glob. Environ. Change 29, 318–326 (2014).Article 

Google Scholar 
47.Harris, N. L. et al. Global maps of twenty-first century forest carbon fluxes. Nat. Clim. Change https://doi.org/10.1038/s41558-020-00976-6 (2021).48.Tachikawa, T., Hato, M., Kaku, M. & Iwasaki, A. Characteristics of ASTER GDEM version 2. In Proc. IEEE International Geoscience and Remote Sensing Symposium (IGARSS) 3657–3660 (IEEE, 2011).49.Burrough, P. A., McDonnell, R., McDonnell, R. A. & Lloyd, C. D. Principles of Geographical Information Systems (Oxford Univ. Press, 2015).50.Mokany, K., Raison, R. J. & Prokushkin, A. S. Critical analysis of root : shoot ratios in terrestrial biomes. Glob. Change Biol. 12, 84–96 (2006).Article 

Google Scholar 
51.Tyukavina, A. et al. Aboveground carbon loss in natural and managed tropical forests from 2000 to 2012. Environ. Res. Lett. 10, 074002 (2015).Article 
CAS 

Google Scholar 
52.Ryan, S. E. & Porth, L. S. A Tutorial on the Piecewise Regression Approach Applied to Bedload Transport Data (CreateSpace, 2015).53.Toms, J. D. & Lesperance, M. L. Piecewise regression: a tool for identifying ecological thresholds. Ecology 84, 2034–2041 (2003).Article 

Google Scholar 
54.Zeng, Z. et al. A reversal in global terrestrial stilling and its implications for wind energy production. Nat. Clim. Change 9, 979–985 (2019).Article 

Google Scholar 
55.Zaiontz, C. Real Statistics Using Excel (accessed 16 June 2021); http://www.real-statistics.com/ More

A combination of airborne and satellite remote sensing data were used to quantify changes in mangrove forest structure and function from Hurricane Irma (Supplementary Fig. 1). Findings based on multi-sensor airborne data were scaled to the entire study area using estimates of forest structure and vegetation phenology derived from satellite data.G-LiHT Airborne campaignDuring April 2017, NASA Goddard’s Lidar, Hyperspectral, and Thermal (G-LiHT) airborne imager conducted an extensive airborne campaign in South Florida covering >130,000 ha. The same flight lines were resurveyed with G-LiHT eight months later, during November and December of 2017, to quantify structural changes in coastal forests of South Florida and Everglades National Park (ENP) following Hurricane Irma (Fig. 1). Lidar data was collected with two VQ-280i (Riegl USA) and synced during flight using RiACQUIRE version 2.3.7. The plane flew at a nominal height of 335 m above ground level at a pulse repetition frequency of 300 kHz to collect ~12 laser pulses per square meter. The analysis of pre- and post-hurricane conditions used 1-m resolution lidar data products (Supplementary Fig. 2) and 3-cm resolution stereo aerial and ground photos to estimate changes in vegetation structure, fractional cover, and terrain heights across the study domain. G-LiHT lidar canopy height models, digital terrain models, and estimates of fractional vegetation canopy cover (FVC) were produced using standard processing methodology21. All Level 1 through 3 lidar data products and fine-resolution imagery are openly shared through the G-LiHT webpage (https://gliht.gsfc.nasa.gov/).High resolution stereo maps of canopy heightStereo imagery from high-resolution commercial satellites can be used to estimate canopy and terrain surfaces42,43. Here, we derived digital surface models (DSMs) from DigitalGlobe’s WorldView 2 Level 1B imagery. DigitalGlobe provides these data to U.S. Government agencies and non-profit organizations that support U.S. interests via the NextView license agreement44. The spatial resolution of these data depends on the degree of off-nadir pointing for each acquisition. In this study, image resolution ranged from 0.5 to 0.7 m. We selected along-track stereopairs within the study domain to identify stereo image strips (each ~17 km × 110 km) that were nominally cloud-free over the forested domain of interest for years 2012–2013, the most recent cloud-free stereo data available for the study region prior to Hurricane Irma. The DSMs were produced using the Ames Stereo Pipeline (ASP) v. 2.5.1 on the NASA Center for Climate Simulation’s Advanced Data Analytics Platform at Goddard Space Flight Center (ADAPT, https://www.nccs.nasa.gov/services/adapt). The Worldview DSMs have been shown to accurately estimate mangrove canopy height when compared to airborne lidar and radar interferometry42,43. The processing workflow was adapted from ref. 45, and was implemented semi-global matching algorithms with a 5 × 5 correlation kernel, and a 3 × 3 median-filter applied to the output point cloud prior to producing a 1 m DSM using a weighted average gridding rule46. The ASP processing yielded five DSMs at 1-m resolution that were used to capture pre-storm canopy surface elevations.Each of the five Worldview DSMs were individually calibrated using overlapping pre-storm G-LiHT lidar data to estimate mangrove canopy heights across the study region (Supplementary Fig. 1). We sampled 1000 points within the mangrove forest cover (see mangrove classification, below) to develop a bias-correction equation between G-LiHT lidar-derived canopy heights and stereo DSM elevations (Supplementary Fig. 6). The bias-corrected canopy height models from high-resolution stereo imagery were mosaicked together to generate a 1-m resolution CHM for the entire study region (Supplementary Fig. 7). A pre-storm canopy volume was calculated by summing the 1 m × 1 m WorldView CHM for the entire region of interest. Similarly, a post-storm canopy volume was derived using the canopy damage model (see the section below), the relationship between the pre-storm CHM and the max wind speed. This analysis was conducted in ArcMap 10.7.1.Landsat mangrove forest classificationLandsat 8 Operational Land Imager (OLI) imagery was used to map mangrove cover for the southern Florida study region. The imagery was preprocessed to surface reflectance47 and clouds were masked following methods outlined in ref. 48. The Surface Reflectance Tier 1 product in Google Earth Engine was used to create a cloud-free image mosaic for 2016 based on the median values of all cloud-free images for the year for all bands (Supplementary Fig. 1).Training points were hand-selected using contemporary Google Earth imagery, field photos, and expert knowledge of the region. Twenty-four polygons covering a mangrove area of 1243 ha and 17 polygons covering a non-mangrove area of 2759 ha were identified for training regions. Within each of the two classes (i.e., mangrove and non-mangrove), 100,00 points were sampled and used for the training data in a Random Forest Classification implemented in Google Earth Engine49. The Random Forest model used 20 trees and a bag fraction of 0.5. The Landsat-based mangrove map was validated using the Region 3 species land cover map developed by the National Park Service for Everglades National Park50. The National Park species map was reclassified into mangrove and non-mangrove land cover, and 500 randomly generated points were sampled within each of the two land cover classes. The resulting error matrix indicated an overall accuracy of 90.6%.Post-storm canopy coverTime series of Landsat data were used to estimate hurricane damages of mangrove forest cover through December 31, 2017. We combined data from Landsat 7 ETM+ and Landsat 8 OLI to create a dense time series of cloud-free observations. All images were pre-processed to surface reflectance and masked for clouds using the same methods as the mangrove classification. Landsat 7 and Landsat 8 data were then harmonized to account for differences in the sensor specifications following51. We calculated the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) for each image in the collection. We calculated two reference maps from the time series of Landsat imagery (Supplementary Fig. 1). A pre-storm reference was calculated as the median value for each reflectance and index band for all cloud-free imagery in the two years prior Hurricane Irma, August 31, 2015 through August 31, 2017. Similarly, a post-storm median mosaic image was made using Landsat data between October 1, 2017 and December 31, 2017.Pre- and post-storm wall-to-wall Fractional Vegetation Cover (FVC) maps were generated using a combination of lidar-based FVC metrics and Landsat imagery (Supplementary Fig. 1). First, lidar-based FVC was binned into five classes; 0–20%, 20–40%, 40–60%, 60–80%, and 80–100% (Supplementary Fig. 7). We then collected 1000 randomly generated points in each of the five FVC classes, a total of 5000 points, to be used as training data in the Landsat classification. Here, we implemented a Random Forest Classifier using 100 trees and a bag fraction of 0.5. These steps were applied to both the pre-storm and post-storm lidar-derived FVC and Landsat mosaic image metrics. Changes between the pre- and post-storm FVC were then calculated based on the five different FVC classes (Supplementary Fig. 7). For example, a pixel with pre-storm FVC of 80–100% and a post-storm FVC of 20–40%, a reduction of three FVC classes, was assigned a drop in FVC of 40–60% (Fig. 1).Recovery times and resilienceWe estimated the time to full recovery of pre-storm mangrove green canopy cover using the time series of Landsat NDVI during the first 15-months following Hurricane Irma. The pre-storm mean NDVI layer was used as a reference, as described in the previous section. Next we calculated the NDVI anomaly for each image during the post-storm period, September 17, 2017 through December 31, 2018 (Supplementary Fig. 1). We then summed the individual anomaly values from each Landsat image and normalized by the total number of valid pixels (i.e., pixels meeting quality control measures) to estimate the average change in NDVI within the 15 months after the storm. We used anomaly values to identify mangrove forests with large decreases in the 15 months after the storm using a threshold of 0.2 for the 15-month NDVI average anomaly19,52. These areas suffered large losses of canopy material and limited new growth during the post-storm period. We used the slope in NDVI values for each pixel during 2018 to estimate the time in years to full recovery to pre-storm NDVI values, excluding data from October to December 2017 to remove delayed browning of damaged vegetation and spurious NDVI values from surface water features following the storm. Areas with a negative NDVI slope were not assigned a recovery time.We used a combination of the NDVI slope, estimated time to full NDVI recovery, and the average change in NDVI between the pre- and post-storm periods to categorize mangrove forest resilience, the potential for mangroves to rebound to pre-disturbance conditions. The specific criteria for mangrove recovery rates and mangrove damage thresholds were adapted from field and remote sensing studies, respectively6,19,25. Regions of high resilience (a combination of high resistance and resilience) were identified based on rapid recovery and/or little to no immediate impact from the storm: (1) areas that were observed to recover to pre-disturbance conditions during 2018, (2) areas that were predicted to recover within 5 years regardless of the post-storm drop in NDVI6, and (3) regions with a post-storm change in NDVI 15 years or a negative NDVI slope that occurred in regions with the largest ( >0.2) post-storm drop in average NDVI25 (Supplementary Fig. 9). The resilience class map is available online for download53.Mangrove species and elevationWe used species level maps developed by the National Park Service for Everglades National Park50 to characterize the impact of Hurricane Irma on different mangrove species. For that study, dominant species were identified through photo-interpretation of stereoscopic, color-infrared aerial imagery. Grid cells of 50 m × 50 m covering an area (Region 3) of ~100,000 ha in southwest Florida were interpreted based on the majority cover type and validated using field observations. A total of 169 vegetation cover classes were identified in this region, however, only five mangrove cover classes were considered for these analyses: Avicennia germinans (Black Mangrove), Laguncularia racemosa (White Mangrove), Rhizophora mangle (Red Mangrove), Conocarpus erectus (Buttonwood), and a single mixed species mangrove class. Mangrove forest communities were defined as the dominant diagnostic species in the upper-most stratum50. The mangrove species data were reprojected to match the Landsat resolution and the resilience maps. We used the intersection of the resilience and species extent maps to estimate the proportion of each resilience class by dominant species.The USGS National Elevation Dataset (NED) was used to estimate the soil elevation across southwest Florida. The 1/9 arc second (~3 m × 3 m) products were acquired from NED, and reprojected to Landsat resolution to estimate the proportion of each resilience class by soil elevation.Additional data and analysisModeled maximum storm surge data for Hurricane Irma were acquired from Coastal Emergency Risks Assessment data portal. Storm surge is derived from the ADCIRC Prediction System that solves for time dependent, circulation, and transport in multiple dimensions54. Maximum sustained hurricane wind speed was modeled hourly at a 5 km × 5 km resolution for 2017 by NASA’s Global Modeling and Assimilation Office (GMAO)55. The storm maximum wind speed for each 5 km × 5 km grid cell was calculated and binned into six discrete classes of wind speeds at 5 m s−1 increments: 26–30, 31–35, 36–40, 41–45, 46–50, and >50.Statistical analysesCanopy height losses measured from NASA G-LiHT data were grouped by five pre-storm canopy height classes (0–5 m, 5–10 m, 10–15 m, 15–20 m, and >20 m). All valid pixels within the lidar footprint was used to calculate the mean, standard error, and area (sum of 1 m × 1 m pixels) for each class (Supplementary Table 1). These results were then tested for significant differences between canopy height losses and pre-storm canopy height classes between using a one-way ANOVA analysis with a post-hoc Tukey test in R (version 4.0.3). For testing the significance between environmental variables (i.e., pre-storm canopy height, canopy height loss, percent canopy height loss, surface elevation, and storm surge water level above ground) we employed a two-sided Kolmogorov–Smirnov test56 implemented in R (version 4.0.3). First, we created a multi-band stacked image which included each of the variable layers. Within each resilience class (i.e., Low, Intermediate, and High) with randomly selected 10,000–20,000 points using Google Earth Engine to sample from the environmental variables images. From that sample set we then randomly selected 500 samples within each of the resilience classes. Each class combination (1) Low-Intermediate, (2) Low-High, and (3) Intermediate-High were compared using the Kolmogorov–Smirnov test. We repeated this procedure using 5000 iterations in order to provide a robust estimate of the Kolmogorow–Smirnov statistic, including the mean and first and third quartiles, which were then compared to the critical value (Supplementary Table 2).Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More

1.Hoegh-Guldberg, O. et al. Coral reefs under rapid climate change and ocean acidification. Science 318, 1737–1742 (2007).CAS 
PubMed 
Article 
ADS 

Google Scholar 
2.Frölicher, T. L., Fischer, E. M. & Gruber, N. Marine heatwaves under global warming. Nature 560, 360–364 (2018).PubMed 
Article 
ADS 
CAS 

Google Scholar 
3.Oliver, E. C. J. et al. Projected marine heatwaves in the 21st century and the potential for ecological impact. Front. Mar. Sci. 6, 1–12 (2019).Article 

Google Scholar 
4.LaJeunesse, T. C. et al. Systematic revision of symbiodiniaceae highlights the antiquity and diversity of coral endosymbionts. Curr. Biol. 28, 2570-2580.e6 (2018).CAS 
PubMed 
Article 

Google Scholar 
5.Hoegh-Guldberg, O. Climate change, coral bleaching and the future of the world’ s coral reefs. Ove Hoegh-Guldberg (1998).6.Hughes, T. P. et al. Spatial and temporal patterns of mass bleaching of corals in the Anthropocene. Science (80-.) 359, 80–83 (2018).CAS 
Article 
ADS 

Google Scholar 
7.Van Hooidonk, R., Maynard, J. A., Manzello, D. & Planes, S. Opposite latitudinal gradients in projected ocean acidification and bleaching impacts on coral reefs. Glob. Chang. Biol. 20, 103–112 (2014).PubMed 
Article 
ADS 

Google Scholar 
8.Muller, E. M., Rogers, C. S., Spitzack, A. S. & Van Woesik, R. Bleaching increases likelihood of disease on Acropora palmata (Lamarck) in Hawksnest Bay, St John, US Virgin Islands. Coral Reefs 27, 191–195 (2008).Article 
ADS 

Google Scholar 
9.Cróquer, A. & Weil, E. Changes in Caribbean coral disease prevalence after the 2005 bleaching event. Dis. Aquat. Organ. 87, 33–43 (2009).PubMed 
Article 

Google Scholar 
10.Grottoli, A. G. et al. The cumulative impact of annual coral bleaching can turn some coral species winners into losers. Glob. Chang. Biol. 20, 3823–3833 (2014).PubMed 
Article 
ADS 

Google Scholar 
11.Schoepf, V. et al. Annual coral bleaching and the long-term recovery capacity of coral. Proc. R. Soc. B Biol. Sci. 282, 20151887 (2015).Article 
CAS 

Google Scholar 
12.Neal, B. P. et al. Caribbean massive corals not recovering from repeated thermal stress events during 2005–2013. Ecol. Evol. 7, 1339–1353 (2017).PubMed 
PubMed Central 
Article 

Google Scholar 
13.Alvarez-Filip, L., Dulvy, N. K., Gill, J. A., Côté, I. M. & Watkinson, A. R. Flattening of Caribbean coral reefs: Region-wide declines in architectural complexity. Proc. R. Soc. B Biol. Sci. 276, 3019–3025 (2009).Article 

Google Scholar 
14.Pratchett, M. S., Hoey, A. S., Wilson, S. K., Messmer, V. & Graham, N. A. J. Changes in biodiversity and functioning of reef fish assemblages following coral bleaching and coral loss. Diversity 3, 424–452 (2011).Article 

Google Scholar 
15.Poloczanska, E. S. et al. Responses of marine organisms to climate change across oceans. Front. Mar. Sci. 3, 1–21 (2016).Article 

Google Scholar 
16.Pinsky, M. L., Selden, R. L. & Kitchel, Z. J. Climate-driven shifts in marine species ranges: Scaling from organisms to communities. Ann. Rev. Mar. Sci. 12, 153–179 (2020).PubMed 
Article 

Google Scholar 
17.Kleypas, J. A., McManu, J. W. & Mene, L. A. B. Environmental limits to coral reef development: Where do we draw the line?. Am. Zool. 39, 146–159 (1999).Article 

Google Scholar 
18.Descombes, P. et al. Forecasted coral reef decline in marine biodiversity hotspots under climate change. Glob. Chang. Biol. 21, 2479–2487 (2015).PubMed 
Article 
ADS 

Google Scholar 
19.Vergés, A. et al. The tropicalization of temperate marine ecosystems: Climate-mediated changes in herbivory and community phase shifts. Proc. R. Soc. B Biol. Sci. 281, 20140846 (2014).Article 

Google Scholar 
20.Mies, M. et al. South Atlantic coral reefs are major global warming refugia and less susceptible to bleaching. Front. Mar. Sci. 7, 1–13 (2020).Article 
ADS 

Google Scholar 
21.Perry, C. T. & Larcombe, P. Marginal and non-reef-building coral environments. Coral Reefs 22, 427–432 (2003).Article 

Google Scholar 
22.Loiola, M. et al. Structure of marginal coral reef assemblages under different turbidity regime. Mar. Environ. Res. 147, 138–148 (2019).CAS 
PubMed 
Article 

Google Scholar 
23.Beger, M., Sommer, B., Harrison, P. L., Smith, S. D. A. & Pandolfi, J. M. Conserving potential coral reef refuges at high latitudes. Divers. Distrib. 20, 245–257 (2014).Article 

Google Scholar 
24.Glynn, P. W. Coral reef bleaching: Facts, hypotheses and implications. Glob. Chang. Biol. 2, 495–509 (1996).Article 
ADS 

Google Scholar 
25.Semmler, R. F., Hoot, W. C. & Reaka, M. L. Are mesophotic coral ecosystems distinct communities and can they serve as refugia for shallow reefs?. Coral Reefs 36, 433–444 (2017).Article 
ADS 

Google Scholar 
26.Rocha, L. A. et al. Mesophotic coral ecosystems are threatened and ecologically distinct from shallow water reefs. Science (8-.) 361, 281–284 (2018).CAS 
Article 
ADS 

Google Scholar 
27.Eckert, R. J., Studivan, M. S. & Voss, J. D. Populations of the coral species Montastraea cavernosa on the Belize Barrier Reef lack vertical connectivity. Sci. Rep. 9, 1–11 (2019).CAS 
Article 

Google Scholar 
28.Serrano, X. M. et al. Geographic differences in vertical connectivity in the Caribbean coral Montastraea cavernosa despite high levels of horizontal connectivity at shallow depths. Mol. Ecol. 23, 4226–4240 (2014).CAS 
PubMed 
Article 

Google Scholar 
29.Serrano, X. M. et al. Long distance dispersal and vertical gene flow in the Caribbean brooding coral Porites astreoides. Sci. Rep. 6, 1–12 (2016).Article 
CAS 

Google Scholar 
30.Morais, J. & Santos, B. A. Limited potential of deep reefs to serve as refuges for tropical Southwestern Atlantic corals. Ecosphere 9, e02281 (2018).Article 

Google Scholar 
31.Torda, G. et al. Rapid adaptive responses to climate change in corals. Nat. Clim. Chang. https://doi.org/10.1038/nclimate3374 (2017).Article 

Google Scholar 
32.Danielle, C. et al. Dynamic symbioses reveal pathways to coral survival through prolonged heatwaves. Nat Commun. 11(1), https://doi.org/10.1038/s41467-020-19169-y. (2020)33.D’Angelo, C. & Wiedenmann, J. Impacts of nutrient enrichment on coral reefs: New perspectives and implications for coastal management and reef survival. Curr. Opin. Environ. Sustain. 7, 82–93 (2014).Article 

Google Scholar 
34.Donovan, M.K.et al. Local conditions magnify coral loss after marine heatwaves. Science 372(6545), 977–980. https://doi.org/10.1126/science.abd9464. (2021)CAS 
PubMed 
Article 

Google Scholar 
35.Hughes, T. et al. Climate change, Human impacts, and the resilience of coral reefs. Laser Induced Damage Opt. Mater. 2009 7504, 75041H (2003).36.Carilli, J. E., Norris, R. D., Black, B. A., Walsh, S. M. & McField, M. Local stressors reduce coral resilience to bleaching. PLoS ONE 4, 1–5 (2009).Article 
CAS 

Google Scholar 
37.Donner, S. D., Heron, S. F. & Skirving, W. J. Future scenarios: A review of modelling efforts to predict the future of coral reefs in an era of climate change. 159–173. https://doi.org/10.1007/978-3-540-69775-6_10 (2018).38.McLeod, E. et al. Warming seas in the coral triangle: Coral reef vulnerability and management implications. Coast. Manag. 38, 518–539 (2010).Article 

Google Scholar 
39.Maynard, J. A. et al. Vulnerability to coral reefs. 1–8 (2019).40.Leão, Z. M. A. N., Kikuchi, R. K. P. & Testa, V. Corals and coral reefs of Brazil. Latin Am. Coral Reefs https://doi.org/10.1016/B978-044451388-5/50003-5 (2003).Article 

Google Scholar 
41.Leão, Z. M. A. N. & Kikuchi, R. K. P. A relic coral fauna threatened by global changes and human activities, Eastern Brazil. Mar. Pollut. Bull. 51, 599–611 (2005).PubMed 
Article 
CAS 

Google Scholar 
42.Francini-Filho, R. B. & De Moura, R. L. Dynamics of fish assemblages on coral reefs subjected to different management regimes in the Abrolhos Bank, eastern Brazil. Aquat. Conserv. Mar. Freshw. Ecosyst. 18, 1166–1179 (2008).Article 

Google Scholar 
43.Moura, R. L. et al. Spatial patterns of benthic megahabitats and conservation planning in the Abrolhos Bank. Cont. Shelf Res. 70, 109–117 (2013).Article 
ADS 

Google Scholar 
44.Vergés, A. et al. Tropicalisation of temperate reefs: Implications for ecosystem functions and management actions. Funct. Ecol. 33, 1000–1013 (2019).Article 

Google Scholar 
45.Yamano, H., Sugihara, K. & Nomura, K. Rapid poleward range expansion of tropical reef corals in response to rising sea surface temperatures. Geophys. Res. Lett. 38, (2011).46.Precht, W. F. & Aronson, R. B. Climate flickers and range shifts of reef corals. Front. Ecol. Environ. 2, 307 (2004).Article 

Google Scholar 
47.Aued, A. W. et al. Large-scale patterns of benthic marine communities in the brazilian province. PLoS ONE 13, 1–15 (2018).Article 
CAS 

Google Scholar 
48.Sully, S., Burkepile, D. E., Donovan, M. K., Hodgson, G. & van Woesik, R. A global analysis of coral bleaching over the past two decades. Nat. Commun. 10, 1–5 (2019).CAS 
Article 

Google Scholar 
49.Phillips, N. A Companion to the e-Book “YaRrr!: The Pirate’s Guide to R”. (2017).50.R Core Team. R: A Language and Environment for Statistical Computing. (2020).51.Banha, T. N. S. et al. Low coral mortality during the most intense bleaching event ever recorded in subtropical Southwestern Atlantic reefs. Coral Reefs https://doi.org/10.1007/s00338-019-01856-y (2019).Article 

Google Scholar 
52.Oliveira, U. D. R., Gomes, P. B., Cordeiro, R. T. S., De Lima, G. V. & Pérez, C. D. Modeling impacts of climate change on the potential habitat of an endangered Brazilian endemic coral: Discussion about deep sea refugia. PLoS ONE 14, 1–24 (2019).
Google Scholar 
53.Cowen, R. K. & Sponaugle, S. Larval dispersal and marine population connectivity. Ann. Rev. Mar. Sci. 1, 443–466 (2009).PubMed 
Article 

Google Scholar 
54.Price, N. N. et al. Global biogeography of coral recruitment: Tropical decline and subtropical increase. Mar. Ecol. Prog. Ser. 621, 1–17 (2019).Article 
ADS 

Google Scholar 
55.Cacciapaglia, C. & van Woesik, R. Reef-coral refugia in a rapidly changing ocean. Glob. Chang. Biol. 21, 2272–2282 (2015).PubMed 
Article 
ADS 

Google Scholar 
56.Baird, A. H. et al. A decline in bleaching suggests that depth can provide a refuge from global warming in most coral taxa. Mar. Ecol. Prog. Ser. 603, 257–264 (2018).Article 
ADS 

Google Scholar 
57.Bongaerts, P., Ridgway, T., Sampayo, E. M. & Hoegh-Guldberg, O. Assessing the ‘deep reef refugia’ hypothesis: Focus on Caribbean reefs. Coral Reefs 29, 1–19 (2010).Article 

Google Scholar 
58.Coles, S. L. et al. Evidence of acclimatization or adaptation in Hawaiian corals to higher ocean temperatures. PeerJ 2018, 1–24 (2018).
Google Scholar 
59.Bay, R. A., Rose, N. H., Logan, C. A. & Palumbi, S. R. Genomic models predict successful coral adaptation if future ocean warming rates are reduced. Sci. Adv. 3, 1–10 (2017).Article 

Google Scholar 
60.Wooldridge, S., Done, T., Berkelmans, R., Jones, R. & Marshall, P. Precursors for resilience in coral communities in a warming climate: A belief network approach. Mar. Ecol. Prog. Ser. 295, 157–169 (2005).Article 
ADS 

Google Scholar 
61.Mazzei, E. F. et al. Newly discovered reefs in the southern Abrolhos Bank, Brazil: Anthropogenic impacts and urgent conservation needs. Mar. Pollut. Bull. 114, 123–133 (2017).CAS 
PubMed 
Article 

Google Scholar 
62.Duarte, G. A. S. et al. Heat waves are a major threat to turbid coral reefs in Brazil. Front. Mar. Sci. 7, 179 (2020).Article 

Google Scholar 
63.Ferreira L.C. et al. Different responses of massive and branching corals to a major heatwave at the largest and richest reef complex in South Atlantic. Mar. Biol. 168(5), https://doi.org/10.1007/s00227-021-03863-6. (2021)64.Teixeira, C. D. et al. Sustained mass coral bleaching (2016–2017) in Brazilian turbid-zone reefs: taxonomic, cross-shelf and habitat-related trends. Coral Reefs https://doi.org/10.1007/s00338-019-01789-6 (2019).Article 

Google Scholar 
65.França, F. M. et al. Climatic and local stressor interactions threaten tropical forests and coral reefs. Philos. Trans. R. Soc. B Biol. Sci. 375, 20190116 (2020).Article 

Google Scholar 
66.Mumby, P. J. & Harborne, A. R. Marine reserves enhance the recovery of corals on Caribbean reefs. PLoS ONE 5, 1–7 (2010).Article 
CAS 

Google Scholar 
67.Kwiatkowski, L. et al. Twenty-first century ocean warming, acidification, deoxygenation, and upper ocean nutrient decline from CMIP6 model projections. Biogeosci. Discuss. https://doi.org/10.5194/bg-2020-16 (2020).68.Jokiel, P. L. Evaluating the assumptions involved. ICES J. Mar. Sci. 73, 550–557 (2015).Article 

Google Scholar 
69.Matz, M. V., Treml, E. A. & Haller, B. C. Estimating the potential for coral adaptation to global warming across the Indo-West Pacific. Glob. Chang. Biol. 26, 3473–3481 (2020).PubMed 
Article 
ADS 

Google Scholar 
70.Tyberghein, L. et al. Bio-ORACLE: A global environmental dataset for marine species distribution modelling. Glob. Ecol. Biogeogr. 21, 272–281 (2012).Article 

Google Scholar 
71.Assis, J. et al. Bio-ORACLE v2.0: Extending marine data layers for bioclimatic modelling. Glob. Ecol. Biogeogr. 27, 277–284 (2018).Article 

Google Scholar 
72.Sbrocco, E. J. & Barber, P. H. MARSPEC: Ocean climate layers for marine spatial ecology. Ecology 94, 979–979 (2013).Article 

Google Scholar 
73.Hijmans, J. R. et al. Package ‘ raster ’ R topics documented (2016).74.Sappington, J. M., Longshore, K. M. & Thompson, D. B. Quantifying landscape ruggedness for animal habitat analysis: A case study using bighorn sheep in the Mojave desert. J. Wildl. Manag. 71, 1419–1426 (2007).Article 

Google Scholar 
75.IPCC. Climate Change 2014 Part A: Global and Sectoral Aspects. Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (2014).76.Fox, J. & Weisberg, S. Multivariate Linear Models in R. An R Companion to Appl. Regres. 1–31 (2011).77.Zuur, A. F., Ieno, E. N., Walker, N. J., Saveliev, A. A. & Smith, G. M. Required Pre-knowledge : A Linear Regression. Mixed Effects Models and Extensions in Ecology with R 1, (2009).78.Martins, T. G., Simpson, D., Lindgren, F. & Rue, H. Bayesian computing with INLA: New features. (2012).79.Rue, H., Martino, S. & Chopin, N. Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. J. R. Stat. Soc. Ser. B Stat. Methodol. 71, 319–392 (2009).MathSciNet 
MATH 
Article 

Google Scholar 
80.Lindgren, F., Rue, H. & Lindström, J. An explicit link between Gaussian fields and Gaussian MARKOV random fields: The stochastic partial differential equation approach. J. R. Stat. Soc. Ser. B Stat. Methodol. 73, 423–498 (2011).MathSciNet 
MATH 
Article 

Google Scholar 
81.Muñoz, F., Pennino, M. G., Conesa, D., López-Quílez, A. & Bellido, J. M. Estimation and prediction of the spatial occurrence of fish species using Bayesian latent Gaussian models. Stoch. Environ. Res. Risk Assess. 27, 1171–1180 (2013).Article 

Google Scholar 
82.Held, L., Schrödle, B. & Rue, H. Posterior and cross-validatory predictive checks: A comparison of MCMC and INLA. Stat. Model. Regres. Struct. Festschrift Honour Ludwig Fahrmeir 1–20 (2010). https://doi.org/10.1007/978-3-7908-2413-183.Watanabe, S. Asymptotic equivalence of Bayes cross validation and widely applicable information criterion in singular learning theory. J. Mach. Learn. Res. 11, 3571–3594 (2010).MathSciNet 
MATH 

Google Scholar 
84.Roos, M. & Held, L. Sensitivity analysis in Bayesian generalized linear mixed models for binary data. Bayesian Anal. 6, 259–278 (2011).MathSciNet 
MATH 
Article 

Google Scholar 
85.Fonseca, V. P., Pennino, M. G., de Nóbrega, M. F., Oliveira, J. E. L. & de Figueiredo Mendes, L. Identifying fish diversity hot-spots in data-poor situations. Mar. Environ. Res. 129, 365–373 (2017).CAS 
PubMed 
Article 

Google Scholar 
86.Pennino, M. G., Vilela, R., Bellido, J. M. & Velasco, F. Balancing resource protection and fishing activity: The case of the European hake in the northern Iberian Peninsula. Fish. Oceanogr. 28, 54–65 (2019).Article 

Google Scholar 
87.Martínez-Minaya, J. et al. A hierarchical Bayesian Beta regression approach to study the effects of geographical genetic structure and spatial autocorrelation on species distribution range shifts. Mol. Ecol. Resour. 19, 929–943 (2019).PubMed 
Article 

Google Scholar  More

Our framework to predict unknown associations between known viruses and potential mammalian hosts or susceptible species comprised three distinct perspectives: viral, mammalian and network. Each perspective produced predictions from a unique vantage point (that of each virus, each mammal, and the network connecting them respectively). Subsequently, their results were consolidated via majority voting. This approach suggested that 20,832 (median, 90% CI = [2,736, 97,062], hereafter values in square brackets represent 90% CI) unknown associations potentially exist between our mammals and their known viruses, (18,920 [2,440, 91,517] in wild or semi-domesticated mammals). Number of unknown associations predicted by each perspective individually were as follows: mammalian only = 41,537 [4,275, 23,8971], viral only = 21,352 [2,536, 95,630], and network only = 76,081 [27,738, 20,5814]. Our results indicated a ~4.29-fold increase ([~1.43, ~16.33]) in virus-mammal associations (~4.89 [~1.5, ~19.81] in wild and semi-domesticated mammals).Additionally, we trained an independent pipeline including only the 3534 supported by evidence extracted from meta-data accompanying nucleotide sequences, as indexed in EID2 (55.82% of all associations – see Methods section and Supplementary Results 8). Our sequence-evidence pipeline indicated that 15,721 (median, 90% CI = [1,603, 88,553]) unknown associations could potentially exist (13,930 [1,298, 83,043] in wild or semi-domesticated mammals).In the following subsections we first illustrate the mechanism of our framework via an example, then further explore the predictive power of our approach for viruses and mammals.ExampleOur multi-perspective framework generates predictions for each known or unknown virus-mammal association (2,722,656 possible associations between 1,896 viruses and 1,436 terrestrial mammals). We highlight this functionality using two examples (Fig. 1). West Nile virus (WNV) a flavivirus with wide host range, and the bat Rousettus leschenaultia (order: Chiroptera). We first consider each of our perspectives separately, and then showcase how these perspectives are consolidated to produce final predictions.Fig. 1: Example showcasing final and intermediate predictions of West Nile Virus (WNV), and Rousettus leschenaultii.Panel A Top 60 predicted mammalian species susceptible to WNV. Mammals were ordered by mean probability of predictions derived from mammalian (all models), viral (WNV models) and network perspectives, and top 60 were selected. Circles represent the following information in order: 1) whether the association is known (documented in our sources) or not (potential or undocumented). Hosts are omitted for known associations. 2) Mean probability of the three perspectives (per association). 3) Median mammalian perspective probabilities of predicted associations. These probabilities are obtained from 3000 models (50 replicate models for each mammal), trained with viral features – SMOTE class balancing. 4) Median viral perspective probabilities of predicted associations (50 WNV replicate models trained with mammalian features – SMOTE class balancing). 5) Median network perspective probabilities of predicted associations (100 replicate models, balanced under-sampling). 6) Taxonomic order of predicted susceptible species. Orders are shortened as follows: Artiodactyla (Art), Carnivora (Crn), Chiroptera (Chp), primates (Prm), Rodentia (Rod), and Others (Oth). Panel B Top 50 predicted viruses of R. leschenaultii. Viruses were ordered by mean probability of predictions derived from mammalian (R. leschenaultii models), viral (all models) and network perspectives. Circles as per Panel A. Baltimore represents Baltimore classification. Panel C Median probability of predicted WNV-mammal associations in each of the three perspectives per mammalian order. Points represent susceptible species predicted by voting (at least two of the three perspectives – n = 137). Median ensemble probability is computed in each perspective (50 replicate models for each virus/mammal, 100 replicate network models). Predictions derived from each perspective at 0.5 probability cut-off. Supplementary Data 1 presents full WNV results. Panel D Median probability of virus-R. leschenaultii associations in the three perspectives per Baltimore group. Points represent susceptible species predicted by voting (at least two of the three perspectives – n = 64), predictions are derived as per panel C. Supplementary Data 2 lists full results for R. leschenaultii. Supplementary Fig. 7 illustrate the results when research effort into viruses and mammals is included in mammalian and viral perspectives, respectively.Full size image1) The mammalian perspective: our mammalian perspective models, trained with features expressing viral traits (Table 1), suggested a median of 90 [17, 410] unknown associations between WNV and terrestrial mammals could form when predicting virus-mammal associations based on viral features alone – a ~2.61-fold increase [~1.3, ~8.32]. Similarly, our results indicated that 64 [4, 331] new associations could form between our selected mammal (R. leschenaultia) and our viruses – a ~4.37-fold increase [~1.21, ~18.42] (Supplementary Results 4).Table 1 Viral traits & features used to build our mammalian models.Full size table(2) The viral perspective: our viral models, trained with features expressing mammalian traits (Table 2), indicated a median of 48 [0, 214] new hosts of WNV (~1.86- fold increase [~1, 4.82]). Results for our example mammal (R. leschenaultia) suggested 18 [3, 76], existing viruses could be found in this host (~1.95-fold increase [~1.16, ~5.00]) – Supplementary Results 5).Table 2 mammalian traits & features used to build our viral models.Full size table(3) The network perspective: Our network models indicated a median of 721 [448, 1,317] (~13.88 [9, 24.52] fold increase) unknown associations between WNV and terrestrial mammals, and that 246 [91, 336] existing viruses could be found in our selected host (R. leschenaultia), equivalent to a ~13.95 [~5.79, ~18.68] fold increase (Supplementary Results 6).Considering that each of the above perspectives approached the problem of predicting virus-mammal associations from a different angle, the agreement between these perspectives varied. In the case of WNV: mammalian and viral perspectives achieved 92.3% agreement [72.6%–98.5%]; mammals and network perspectives had 55.3% agreement [33.4%–69.5%]; and viruses and network had 52.9% agreement [19.8%–68.7%]. In the case of R. leschenaultia these numbers were as follows: 96.15% [82.44%, 99.58%], 87.24% [76.37%, 95.04%], and 87.61% [75.90%, 95.25%], respectively. The agreements between our perspectives across the 2,722,656 possible associations were as follows: 98.04% [90.36%, 99.73%] between mammalian and viral perspectives, 96.71% [88.62%, 98.92%] between mammalian and network perspectives, and 97.11% [91.57%, 98.95%] between viral and network perspectives.After voting, our framework suggested that a median of 117 [15, 509] new or undetected associations could be missing between WNV and terrestrial mammals (~3.45-fold increase [~1.3, ~12.2]). Similarly, our results indicated that R. leschenaultia could be susceptible to an additional 45 [5, 235] viruses that were not captured in our input (~1.37-fold increase [~1.26, ~13.37]). Figure 1 illustrates top predicted and detected associations for WNV (Supplementary Data 1) and R. leschenaultia (Supplementary Data 2). Supplementary Results 1 illustrate results with research effort into viruses, and mammals included as a predictor in our mammalian and viral perspective models, respectively. Predictions with and without research effort incorporated into models trained in these perspectives broadly agreed.Relative importance of viral featuresOur multi-perspective approach trained a suite of models for each mammalian species with two or more known viruses (n = 699, response variable = 1 if the virus is known to associate with the focal mammalian species, 0 otherwise). This enabled us to assess the relative importance (influence) of viral traits (Table 1) to each of our mammalian models. This in turn showcased variations of how these viral traits contribute to the models at the level of individual species (e.g. humans), and at an aggregated level (e.g. by order or domestication status). The results, highlighted in Fig. 2A, indicate that mean phylogenetic (median = 95.4% [75.6%, 100%]) and mean ecological (90.90% [43.50%, 100%]) distances between potential and known hosts of each virus were the top predictors of associations between the focal host and each of the input viruses. Maximum phylogenetic breadth was also important (74.7 0%, [16.60%, 100%]).Fig. 2: Results (viruses).Panel A Variable importance (relative contribution) of viral traits to mammalian perspective models. Variable importance is calculated for each constituent ensemble (n = 699) of our mammalian perspective (median of a suite of 50 replicate models, trained with viral features, with SMOTE sampling), and then aggregated (mean) per each reported group (columns). Panel B – Number of known and new mammalian species associated with each virus. Rabies lyssavirus was excluded from panel B to allow for better visualisation. Top 40 (by number of new hosts) are labelled. Species in bold have over 150 predicted hosts (Supplementary Data 3 lists details of these viruses including CI). Panel C Predicted number of viruses per species of wild and semi-domesticated mammals (group by mammalian order). Following orders (clockwise) are presented: Artiodactyla, Carnivora, Chiroptera, Perissodactyla, Primates, and Rodentia. Source of the silhouette graphics is PhyloPic.org. (Supplementary Data 4 lists aggregated results per mammalian order). Circles represent each mammalian species (with predicted viruses > 0), coloured by number of known viruses previously not associated with this species. Boxplots indicate median (centre), the 25th and 75th percentiles (bounds of box) and inter quantile range (whiskers) and are aggregated at the order level. Large red circles with error bars (90% CI) illustrate the median number of known viruses per species in each order. Number of species presented (n) is as follows: All = 1293 (Artiodactyla = 104, Carnivora = 177, Chiroptera = 548, Perissodactyla = 11, Primates = 171, and Rodentia = 282); Group I = 666 (94, 109, 156, 10, 160, 137); Group II = 371 (32, 120, 111, 1, 54, 53); Group III = 410 (87,62,123,9,51,78); Group IV = 739 (98, 102, 221, 9, 148, 161); Group V = 1129 (87, 173, 528, 8, 107, 226); Group VI = 358 (55, 64, 30, 6, 139, 64); and Group VII = 110 (3,2,53,1,43,8). Supplementary Fig. 8 presents results derived with research effort into mammalian hosts and viruses included in the constituent models trained in the viral and mammalian perspectives, respectively.Full size imageMammalian host rangeOur results suggested that the average mammalian host range of our viruses is 14.33 [4.78, 54.53] (average fold increase of ~3.18 [~1.23, ~9.86] in number of hosts detected per virus). Overall, RNA viruses had the average host range of 21.65 [7.01, 82.96] hosts (~4.00- fold increase [~1.34, ~14.15]). DNA viruses, on the other hand, had 7.85 [2.81, 29.47] hosts on average (~2.43 [~1.14, ~6.89] fold increase). Table 3 lists the results of our framework at Baltimore group level and selected family and transmission routes of our viruses. Figure 2 illustrates predicted mammalian host range of our viruses (Fig. 2B, Supplementary Data 3), and the increase in predicted number of viruses per species in species-rich mammalian orders of interest (Fig. 2C, Supplementary Data 4).Table 3 Predicted range of susceptible mammalian species of viruses per Baltimore group, family (top 15 families, ranked by fold increase) and transmission route.Full size tableRelative importance of mammalian featuresWe trained a suite of models for each virus species with two or more known mammalian hosts (n = 556, response variable = 1 if the mammal is known to associate with the focal virus species, 0 otherwise). This allowed us to calculate relative importance of mammalian traits (Table 2) to our viral models. We were also able to capture variations in how these features contribute to our viral models at various levels (e.g. Baltimore classification, or transmission route) as highlighted in Fig. 3A. Our results indicated that distances to known hosts of viruses were the top predictor of associations between the focal virus and our terrestrial mammals. The breakdown was: 1) mean phylogenetic distance – all viruses = 98.75% [93.01%, 100%], DNA = 99.48% [96.03%, 100%], RNA = [91.93%, 100%]; 2) mean ecological distance all viruses = 94.39% [71.86%, 100%], DNA = 96.36% [80.99%, 100%], RNA = [69.48%, 100%]. In addition, life-history traits significantly improved our models, in particular: longevity (all viruses = 60.9% [12.12%, 98.88%], DNA = 68.03% [11.22%, 99.69%], RNA = [13.55%, 96.37%]); body mass (all viruses = 62.92% [5.4%, 97.65%], DNA = 72.75% [18.49%, 100%], RNA = 57.45% [4.32%, 95.5%]); and reproductive traits (all viruses = 53.37% [5.67%, 95.99%]%, DNA = 59.46% [8.27%, 99.32%], RNA = 50.17% [4.85%, 92.17%]).Fig. 3: Results (Mammals).Panel A Variable importance (relative contribution) of mammalian traits to viral perspective models. Variable importance is calculated for each constituent model (n = 556) of our viral perspective (trained with mammalian features), and then aggregated (median) per each reported group (columns). Panel B Number of known and new viruses associated with each mammal. Labelled mammals are as follows: top 4 (by number of new viruses) for each of Artiodactyla, Carnivora, Chiroptera, Primates, Rodentia, and other orders. Species in bold have 100 or more predicted viruses (Supplementary Data 5). Panel C Top 18 genera (by number of predicted wild or semi-domesticated mammalian host species) in selected orders (Other indicated results for all orders not included in the first five circles). Each order figure comprises the following circles (from outside to inside): 1) Number of hosts predicted to have an association with viruses within the viral genus. 2) Number of hosts detected to have association. 3) Number of hosts predicted to harbour viral zoonoses (i.e. known or predicted to share at least one virus species with humans). 4) Number of hosts predicted to share viruses with domesticated mammals of economic significance (domesticated mammals in orders: Artiodactyla, Carnivora, Lagomorpha and Perissodactyla). 5) Baltimore classification of the selected genera (Supplementary Data 6). Supplementary Fig. 9 presents results derived with research effort into mammalian hosts and viruses included in the constituent models trained in the viral and mammalian perspectives, respectively.Full size imageWild and semi-domesticated susceptible mammalian hosts of virusesour framework indicated ~4.28 -fold increase [~1.2, ~14.64] of the number of virus species in wild and or semi-domesticated mammalian hosts (16.86 [4.95, 68.5] viruses on average per mammalian species). These results indicated an average of 13.45 [1.73, 65.04] unobserved virus species for each wild or semi-domesticated mammalian host (known viruses that are yet to be associated with these mammals). Our framework highlighted differences in the number of viruses predicted per order (Table 4). Figure 3 illustrates the predicted number of viruses in wild or semi-domesticated mammal by mammalian host range (Fig. 3B, Supplementary Data 5), and the top 18 virus genera (per number of host-virus associations) in selected orders (Fig. 3C, Supplementary Data 6). Supplementary Results 1 lists the results with the inclusion of research effort into mammalian species in our viral perspective models.Table 4 Predicted number of viruses per top 15 orders by fold increase in number of viruses predicted in wild or semi-domesticated mammalian hosts (per species).Full size tableNetwork perspective – Potential motifs
We quantified the topology of the network linking virus and mammal species by means of counts of potential motifs21. Figure 4 illustrates how potential motifs are captured in our network. Briefly, for each virus-mammal association for which we want to make predictions (n = 2,722,656, of which 6,331 are supported by our evidence, see methods section), we “force insert” this focal association into our network (Fig. 4A, B) and enumerate all instances of 3 (n = 2), 4 (n = 6), and 5-node (n = 20) potential motifs in which this association might feature if it actually existed21 (Fig. 4C visualises these different motifs). Following this process, a features-set is generated comprising the counts potential motifs for all included associations. Figure 4D illustrates the count of motifs (logged) grouped by mammalian order and virus Baltimore classification.Fig. 4: The network perspective – potential motifs (subgraphs) in our virus-host bipartite network.A The concept of potential motif. The association TBEV-P. leo is a forced insertion into the network prior to calculating motifs for the association. B Motifs space: networks represent 2 steps and 3 steps ego networks (union) of host (here P. leo) and virus (TBEV). 1, 2 and 3 step ego networks comprise the counting space for TBEV-P. leo potential motifs. Dark grey nodes represent viruses, light grey nodes represent hosts. Size of nodes is adjusted to represent overall number of hosts or viruses with known associations to the node. Red edges represent nodes reachable from the mammal (P. leo) in 1 or 2 steps (links). Blue edges represent nodes reachable from the virus (TBEV) with 1 or 2 steps (links). Humans and rabies virus were excluded from these networks. C 3, 4 and 5-node potential motifs in our virus-host bipartite network. Circles represent viruses and squares represent mammals. Red circles represent the focal virus (v), and blue squares represent the focal mammal (m) of the association v-m for which the motifs are being counted (dashed yellow line). This association has two states: either already known (documented in EID2), or unknown. Grey lines illustrate existing associations in our network. D Motifs counts. Heatmap illustrating distribution of motif-features (counts of potential motifs per each focal association) in our bipartite network, grouped by mammalian order and Baltimore classification. The counts are logged to allow for better visualisation. E Variable importance (relative contribution) of motif-features (variables) to our network perspective models (SVM-RW). Motifs (subgraphs) are coloured by the number of nodes (K = 3, 4, 5). Boxplots indicate median (centre), the 25th and 75th percentiles (bounds of box) and inter quantile range (whiskers). Points represent variable importance in individual runs (n = 100). Research effort into both viruses and mammals is included as independent variables in our network models (coloured in yellow).Full size imageRelative importance of network (motif) featuresFigure 4E illustrates that M4.1 was the most important feature in our network models: median = 100% [90.19%, 100%]. Followed by: M5.1 = 97.84% [89.19%, 99.93%], M5.7 = 98.8 97.22% [87.7%, 98.77%] and M4.6 = 96.75% [86.13%, 100%]. Research effort of viruses and mammals had relative importance = 90.26% [82.94%, 95.36%], 88.42% [78.38%, 94.87%] respectively. Overall, 5-node motif-features had median relative influence = 75.06% [1.21%, 98.14%]; whereas 3 and 4-node motif-features had relative influence = 71.69% [55.76%, 85.34%], and 61.06% [27.14%, 100%], respectively. Supplementary Fig. 29 illustrate the partial dependence of network perspective models on each of our network features.ValidationWe validated our framework in three ways: 1) against a held-out test set; 2) by systematically removing selected known viral-mammalian associations and attempting to predict them; and 3) against external data source, comprising viral-mammalian associations extracted using an exhaustive literature search targeting wild mammals and their viruses4,30.Our held-out test set comprised 15% of all data (randomly selected, n = 407,265; 954 known virus-mammal associations, see methods below). We removed this set from our network, computed network features (motifs), and trained constituent models in each perspective with the remainder data. We then estimated our framework performance metrics against the held-out test set. Our framework achieved overall AUC = 0.938 [0.862–0.959], F1-Score = 0.284 [0.464–0.124], and TSS = 0.876 [0.724–0.918], when trained without including research effort in its mammalian and viral perspectives. When research effort was included in these perspectives, performance metrics were as follows: AUC = 0.920 [0.823, 0.944], F1-Score = 0.272 [0.526, 0.093], and TSS = 0.840 [0.646, 0.888].The performance of our voting approach was better than any individual perspective, or combination of perspectives (Supplementary Tables 8–11). The most significant improvement was in F1-score, where individual perspectives scores were as follows: network = 0.104 [0.210–0.051], mammalian = 0.115 [0.009–0.064] (0.131 [0.284–0.035] with research effort), and viral = 0.181 [0.374–0.074] (0.196 [0.373–0.067]).Additionally, we conducted a systematic test to predict removed virus-mammal associations. In this test, we systematically removed one known virus-mammal association at a time from our framework, recalculated all inputs (including from network) and attempted to predict these removed associations. Our framework succeeded in predicting 90% of removed associations (90.70% for associations removed for viruses, 89.92% for associations removed from mammals, Supplementary Results 3).Finally, our framework predicted 84.02% [77.69%, 89.60%] of the externally obtained viral-mammalian associations (with detection quality  > 0) where both host and virus were included in our pipeline, and 77.82% [68.46%, 86.51%] (any detection quality). When including research effort in our mammalian and viral perspectives, these results were: 84.47% [78.15%, 89.60%], and 78.41% [68.83%, 86.37%], respectively. More

1.Cunningham, A. A., Daszak, P. & Wood, J. L. N. One health, emerging infectious diseases and wildlife: two decades of progress? Philos. Trans. R. Soc. B Biol. Sci. 372, 20160167 (2017).2.Suzan, G., Esponda, F., Carrasco-Hernández, R. & Aguirre, A. A. in New Directions in Conservation Medicine: Applied Cases of Ecological Health (eds. Aguirre, A. A., Ostfeld, R. & Daszak, P.). 135–150 (Oxford University Press USA, 2012).3.Hussain, S., Ram, M. S., Kumar, A., Shivaji, S. & Umapathy, G. Human presence increases parasitic load in endangered lion-tailed macaques (Macaca silenus) in its fragmented rainforest habitats in Southern India. PLoS ONE 8, 1–8 (2013).
Google Scholar 
4.Junge, R. E., Barrett, M. A. & Yoder, A. D. Effects of anthropogenic disturbance on indri (Indri indri) health in Madagascar. Am. J. Primatol. 73, 632–642 (2011).PubMed 
Article 

Google Scholar 
5.Friggens, M. M. & Beier, P. Anthropogenic disturbance and the risk of flea-borne disease transmission. Oecologia 164, 809–820 (2010).PubMed 
Article 

Google Scholar 
6.Woodroffe, R. et al. Contact with domestic dogs increases pathogen exposure in endangered African wild dogs (Lycaon pictus). PLoS ONE 7, e30099 (2012).7.Crowl, T. A., Crist, T. O., Parmenter, R. R., Belovsky, G. & Lugo, A. E. The spread of invasive species and infectious disease as drivers of ecosystem change. Front. Ecol. Environ. 6, 238–246 (2008).Article 

Google Scholar 
8.Keesing, F., Holt, R. D. & Ostfeld, R. S. Effects of species diversity on disease risk. Ecol. Lett. 9, 485–498 (2006).CAS 
PubMed 
Article 

Google Scholar 
9.Gámez-Virués, S. et al. Landscape simplification filters species traits and drives biotic homogenization. Nat. Commun. 6, 8568 (2015).10.Alberdi, A., Aizpurua, O., Bohmann, K., Zepeda-Mendoza, M. L. & Gilbert, M. T. P. Do vertebrate gut metagenomes confer rapid ecological adaptation? Trends Ecol. Evol. 31, 689–699 (2016).PubMed 
Article 
PubMed Central 

Google Scholar 
11.Hooper, L. V., Littman, D. R. & Macpherson, A. J. Interactions between the microbiota and the immune system. Science 336, 1268–1273 (2012).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
12.Shapira, M. Gut microbiotas and host evolution: scaling up symbiosis. Trends Ecol. Evol. 31, 539–549 (2016).PubMed 
Article 
PubMed Central 

Google Scholar 
13.Brugman, S. et al. A comparative review on microbiota manipulation: lessons from fish, plants, livestock, and human research. Front. Nutr. 5, 1–15 (2018).Article 
CAS 

Google Scholar 
14.Wasimuddin et al. Astrovirus infections induce age-dependent dysbiosis in gut microbiomes of bats. ISME J. 12, 2883–2893 (2018).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
15.Wasimuddin et al. Adenovirus infection is associated with altered gut microbial communities in a non-human primate. Sci. Rep. 9, 1–12 (2019).CAS 
Article 

Google Scholar 
16.Wilkins, L. J., Monga, M. & Miller, A. W. Defining dysbiosis for a cluster of chronic diseases. Sci. Rep. 9, 1–10 (2019).
Google Scholar 
17.Brüssow, H. Problems with the concept of gut microbiota dysbiosis. Microb. Biotechnol. 13, 423–434 (2020).PubMed 
Article 
PubMed Central 

Google Scholar 
18.Otto, S. P. Adaptation, speciation and extinction in the Anthropocene. Proc. R. Soc. B Biol. Sci. 285, 20182047 (2018).19.Amato, K. R. et al. Habitat degradation impacts black howler monkey (Alouatta pigra) gastrointestinal microbiomes. ISME J. 7, 1344–1353 (2013).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
20.Ingala, M. R., Becker, D. J., Bak Holm, J., Kristiansen, K. & Simmons, N. B. Habitat fragmentation is associated with dietary shifts and microbiota variability in common vampire bats. Ecol. Evol. https://doi.org/10.1002/ece3.5228 (2019)21.Juan, P. A. S., Hendershot, J. N., Daily, G. C. & Fukami, T. Land-use change has host-specificinfluenc on avian gut microbiomes. ISME J. https://doi.org/10.1038/s41396-019-0535-4 (2019)22.Barelli, C. et al. Habitat fragmentation is associated to gut microbiota diversity of an endangered primate: implications for conservation. Sci. Rep. 5, 14862 (2015).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
23.de Juan, S., Thrush, S. F. & Hewitt, J. E. Counting on β-diversity to safeguard the resilience of estuaries. PLoS ONE 8, 1–11 (2013).
Google Scholar 
24.Burnham, K. P. & Anderson, D. R. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach (Springer, 2002).25.Oksanen, J. et al. vegan: Community Ecology Package. R package version 2.5-5. https://github.com/vegandevs/vegan (2019).26.Nakagawa, S. & Cuthill, I. C. Effect size, confidence interval and statistical significance: a practical guide for biologists. Biol. Rev. 82, 591–605 (2007).PubMed 
Article 
PubMed Central 

Google Scholar 
27.Cohen, J. Statistical Power Analysis for the Behavioral Sciences (Lawrence Erlbaum Associates, 1988).28.Gillingham, M. A. F. et al. Offspring microbiomes differ across breeding sites in a panmictic species. Front. Microbiol. 10, 35 (2019).29.Anderson, M. J. Distance-based tests for homogeneity of multivariate dispersions. Biometrics 62, 245–253 (2006).PubMed 
Article 
PubMed Central 

Google Scholar 
30.Mandal, S. et al. Analysis of composition of microbiomes: a novel method for studying microbial composition. Microb. Ecol. Heal. Dis. 26, 1–7 (2015).
Google Scholar 
31.Douglas, G. M. et al. PICRUSt2 for prediction of metagenome functions. Nat. Biotechnol. 38, 669–673 (2020).Article 
CAS 

Google Scholar 
32.Louca, S. & Doebeli, M. Efficient comparative phylogenetics on large trees. Bioinformatics 34, 1053–1055 (2018).CAS 
PubMed 
Article 

Google Scholar 
33.Barbera, P. et al. EPA-ng: massively parallel evolutionary placement of genetic sequences. Syst. Biol. 68, 365–369 (2019).PubMed 
Article 

Google Scholar 
34.Czech, L., Barbera, P. & Stamatakis, A. Genesis and Gappa: processing, analyzing and visualizing phylogenetic (placement) data. Bioinformatics 36, 3263–3265 (2020).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
35.Nyhus, P. J. Human—wildlife conflict and coexistence. Annu. Rev. Environ. Resour. 41, 143–171 (2016).Article 

Google Scholar 
36.Foden, W. B. et al. Climate change vulnerability assessment of species. WIREs Clim. Chang. 10, 1–36 (2019).Article 

Google Scholar 
37.Beck, J. M. et al. Multicenter comparison of lung and oral microbiomes of HIV-infected and HIV-uninfected individuals. Am. J. Respir. Crit. Care Med. 192, 1335–1344 (2015).PubMed 
PubMed Central 
Article 

Google Scholar 
38.Pita, L., Rix, L., Slaby, B. M., Franke, A. & Hentschel, U. The sponge holobiont in a changing ocean: from microbes to ecosystems. Microbiome 6, 46 (2018).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
39.Rosado, P. M. et al. Marine probiotics: increasing coral resistance to bleaching through microbiome manipulation. ISME J. 13, 921–936 (2019).CAS 
PubMed 
Article 

Google Scholar 
40.Wang, L. et al. Corals and their microbiomes are differentially affected by exposure to elevated nutrients and a natural thermal anomaly. Front. Mar. Sci. 5, 1–16 (2018).Article 

Google Scholar 
41.Zaneveld, J. R., McMinds, R. & Thurber, R. V. Stress and stability: applying the Anna Karenina principle to animal microbiomes. Nat. Microbiol. 2, 17121 (2017).42.Rocca, J. D. et al. The Microbiome Stress Project: toward a global meta-analysis of environmental stressors and their effects on microbial communities. Front. Microbiol. 10, 3272 (2019).43.Gillingham, M. A. F. et al. Bioaccumulation of trace elements affects chick body condition and gut microbiome in greater flamingos. Sci. Total Environ. 761, 143250 (2020).44.Chase, J. M. Stochastic community assembly causes higher biodiversity in more productive environments. Science 328, 1388–1392 (2010).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
45.Jiménez, R. R., Alvarado, G., Estrella, J. & Sommer, S. Moving beyond the host: unraveling the skin microbiome of endangered Costa Rican amphibians. Front. Microbiol. 10, 1–18 (2019).Article 

Google Scholar 
46.Wang, J. et al. Phylogenetic beta diversity in bacterial assemblages across ecosystems: deterministic versus stochastic processes. ISME J. 7, 1310–1321 (2013).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
47.Chase, J. M. & Myers, J. A. Disentangling the importance of ecological niches from stochastic processes across scales. Philos. Trans. R. Soc. B Biol. Sci. 366, 2351–2363 (2011).Article 

Google Scholar 
48.Pound, K. L., Lawrence, G. B. & Passy, S. I. Beta diversity response to stress severity and heterogeneity in sensitive versus tolerant stream diatoms. Divers. Distrib. 25, 374–384 (2019).Article 

Google Scholar 
49.Zhou, J. & Ning, D. Stochastic Community Assembly: does it matter in microbial ecology? Microbiol. Mol. Biol. Rev. 81, 1–32 (2017).Article 

Google Scholar 
50.Nicholas, R. A. J. & Ayling, R. D. Mycoplasma bovis: disease, diagnosis, and control. Res. Vet. Sci. 74, 105–112 (2003).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
51.Ley, D. H. in Diseases of Poultry (eds. et al.) (Blackwell Publishing, 2008).52.Groebel, K., Hoelzle, K., Wittenbrink, M. M., Ziegler, U. & Hoelzle, L. E. Mycoplasma suis invades porcine erythrocytes. Infect. Immun. 77, 576–584 (2009).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
53.do Nascimento, N. C., Santos, A. P., Guimaraes, A. M. S., Sanmiguel, P. J. & Messick, J. B. Mycoplasma haemocanis—the canine hemoplasma and its feline counterpart in the genomic era. Vet. Res. 43, 66 (2012).54.Hardham, J. M. et al. Transfer of Bacteroides splanchnicus to Odoribacter gen. nov. as Odoribacter splanchnicus comb. nov., and description of Odoribacter denticanis sp. nov., isolated from the crevicular spaces of canine periodontitis patients. Int. J. Syst. Evol. Microbiol. 58, 103–109 (2008).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
55.Kaakoush, N. O. Insights into the role of Erysipelotrichaceae in the human host. Front. Cell. Infect. Microbiol. 5, 1–4 (2015).Article 
CAS 

Google Scholar 
56.Ormerod, K. L. et al. Genomic characterization of the uncultured Bacteroidales family S24-7 inhabiting the guts of homeothermic animals. Microbiome 4, 1–17 (2016).Article 

Google Scholar 
57.Herrmann, E. et al. RNA-based stable isotope probing suggests Allobaculum spp. as particularly active glucose assimilators in a complex murine microbiota cultured in vitro. Biomed Res. Int. 2017, 1829685 (2017).58.Greetham, H. L. et al. Allobaculum stercoricanis gen. nov., sp. nov., isolated from canine feces. Anaerobe 10, 301–307 (2004).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
59.Silva, Y. P., Bernardi, A. & Frozza, R. L. The role of short-chain fatty acids from gut microbiota in gut-brain communication. Front. Endocrinol. 11, 1–14 (2020).60.Wiegel, J., Tanner, R. & Rainey, F. A. in The Prokaryotes: Volume 4: Bacteria: Firmicutes, Cyanobacteria (eds. Dworkin, M., Falkow, S., Rosenberg, E., Schleifer, K.-H. & Stackebrandt, E.) 654–678 (Springer US, 2006).61.Tamanai-Shacoori, Z. et al. Roseburia spp.: a marker of health? Future Microbiol 12, 157–170 (2017).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
62.Freier, T. A., Beitz, D. C., Li, L. & Hartman, P. A. Characterization of Eubacterium coprostanoligenes sp. nov., a Cholesterol-Reducing Anaerobe. Int. J. Syst. Bacteriol. 44, 137–142 (1994).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
63.Venegas, D. P. et al. Short chain fatty acids (SCFAs)-mediated gut epithelial and immune regulation and its relevance for inflammatory bowel diseases. Front. Immunol. 10, 277 (2019).64.MetaCyc. MetaCyc Pathway: pyrimidine deoxyribonucleotides biosynthesis from CTP. https://biocyc.org/META/NEW-IMAGE?type=PATHWAY&object=PWY-7210&show-citations=T (2020).65.Caspi, R. et al. The MetaCyc database of metabolic pathways and enzymes. Nucleic Acids Res. 46, D633–D639 (2018).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
66.MetaCyc. MetaCyc Pathway: poly(glycerol phosphate) wall teichoic acid biosynthesis. https://biocyc.org/META/NEW-IMAGE?type=PATHWAY&object=TEICHOICACID-PWY (2020).67.Brown, S., Santa Maria, J. P. & Walker, S. Wall teichoic acids of gram-positive bacteria. Annu. Rev. Microbiol. 67, 313–336 (2013).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
68.MetaCyc. MetaCyc Pathway: L-lysine biosynthesis II. https://metacyc.org/META/NEW-IMAGE?type=PATHWAY&object=PWY-2941 (2020).69.Hutton, C. A., Perugini, M. A. & Gerrard, J. A. Inhibition of lysine biosynthesis: an evolving antibiotic strategy. Mol. Biosyst. 3, 458–465 (2007).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
70.Wanner, S. et al. Wall teichoic acids mediate increased virulence in Staphylococcus aureus. Nat. Microbiol. 2, 1–12 (2017).
Google Scholar 
71.MetaCyc. MetaCyc Pathway: formaldehyde assimilation II (assimilatory RuMP Cycle). https://biocyc.org/META/NEW-IMAGE?type=PATHWAY&object=PWY-1861 (2020).72.Chen, N. H., Djoko, K. Y., Veyrier, F. J. & McEwan, A. G. Formaldehyde stress responses in bacterial pathogens. Front. Microbiol. 7, 1–17 (2016).
Google Scholar 
73.Tauseef, S. M., Premalatha, M., Abbasi, T. & Abbasi, S. A. Methane capture from livestock manure. J. Environ. Manag. 117, 187–207 (2013).CAS 
Article 

Google Scholar 
74.Dale, V. H., Brown, S., Calderón, M. O., Montoya, A. S. & Martínez, R. E. Estimating baseline carbon emissions for the eastern Panama Canal watershed. Mitig. Adapt. Strateg. Glob. Chang 8, 323–348 (2003).Article 

Google Scholar 
75.Schmid, J. et al. Ecological drivers of Hepacivirus infection in a neotropical rodent inhabiting landscapes with various degrees of human environmental change. Oecologia https://doi.org/10.1007/s00442-018-4210-7 (2018)76.Adler, G. H. & Beatty, R. P. Changing reproductive rates in a neotropical forest rodent, Proechimys semispinosus. J. Anim. Ecol. 66, 472 (1997).Article 

Google Scholar 
77.Adler, G. H. Fruit and seed exploitation by Central American spiny rats, Proechimys semispinosus. Stud. Neotrop. Fauna Environ. 30, 237–244 (1995).78.Hoch, G. A. & Adler, G. H. Removal of black palm (Astrocaryum standleyanum) seeds by spiny rats (Proechimys semispinosus). J. Trop. Ecol. 13, 51–58 (1997).Article 

Google Scholar 
79.Endries, M. J. & Adler, G. H. Spacing patterns of a tropical forest rodent, the spiny rat (Proechimys semispinosus), in Panama. J. Zool. 265, 147–155 (2005).Article 

Google Scholar 
80.Adler, G. H. The island syndrome in isolated populations of a tropical forest rodent. Oecologia 108, 694–700 (1996).PubMed 
Article 
PubMed Central 

Google Scholar 
81.Caporaso, J. G. et al. Global patterns of 16S rRNA diversity at a depth of millions of sequences per sample. PNAS 108, 4516–4522 (2011).CAS 
PubMed 
Article 
PubMed Central 

Google Scholar 
82.Caporaso, J. G. et al. QIIME allows analysis of high-throughput community sequencing data. Nat. Methods 7, 335–336 (2010).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
83.Menke, S. et al. Oligotyping reveals differences between gut microbiomes of free-ranging sympatric Namibian carnivores (Acinonyx jubatus, Canis mesomelas) on a bacterial species-like level. Front. Microbiol. 5, 526 (2014).PubMed 
PubMed Central 
Article 

Google Scholar 
84.Bolyen, E. et al. Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2. Nat. Biotechnol. 37, 852–857 (2019).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
85.Callahan, B. J. et al. DADA2: high-resolution sample inference from Illumina amplicon data. Nat. Methods 13, 581–583 (2016).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
86.Callahan, B. J., Mcmurdie, P. J. & Holmes, S. P. Exact sequence variants should replace operational taxonomic units in marker-gene data analysis. ISME J. 11, 2639–2643 (2017).PubMed 
PubMed Central 
Article 

Google Scholar 
87.Quast, C. et al. The SILVA ribosomal RNA gene database project: Improved data processing and web-based tools. Nucleic Acids Res. 41, 590–596 (2013).Article 
CAS 

Google Scholar 
88.Yilmaz, P. et al. The SILVA and ‘All-species Living Tree Project (LTP)’ taxonomic frameworks. Nucleic Acids Res. 42, 643–648 (2014).Article 
CAS 

Google Scholar 
89.Glöckner, F. O. et al. 25 years of serving the community with ribosomal RNA gene reference databases and tools. J. Biotechnol. 261, 169–176 (2017).PubMed 
Article 
CAS 
PubMed Central 

Google Scholar 
90.Pedregosa, F. et al. Scikit-learn: machine learning in python. J. Mach. Learn. Res. 12, 2825–2830 (2011).91.Price, M. N., Dehal, P. S. & Arkin, A. P. FastTree 2—āpproximately maximum-likelihood trees for large alignments. PLoS ONE 5, e9490 (2010).92.Huson, D. H. & Scornavacca, C. Dendroscope 3: an interactive tool for rooted phylogenetic trees and networks. Syst. Biol. 61, 1061–1067 (2012).PubMed 
Article 
PubMed Central 

Google Scholar 
93.R. Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing. http://www.r-project.org/index.html (2017).94.McMurdie, P. J. & Holmes, S. phyloseq: An R package for reproducible interactive analysis and graphics of microbiome census data. PLoS ONE 8, e61217 (2013).95.Davis, N. M., Proctor, Di. M., Holmes, S. P., Relman, D. A. & Callahan, B. J. Simple statistical identification and removal of contaminant sequences in marker-gene and metagenomics data. Microbiome 6, 1–14 (2018).Article 

Google Scholar 
96.Shannon, C. E. A mathematical theory of communication. Bell Syst. Tech. J. 27, 379–423 (1948).Article 

Google Scholar 
97.Faith, D. P. Conservation evaluation and phylogenetic diversity. Biol. Conserv. 61, 1–10 (1992).Article 

Google Scholar 
98.Bates, D., Mächler, M., Bolker, B. M. & Walker, S. C. Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67, 1–48 (2015).99.Mcmurdie, P. J. & Holmes, S. Waste not, want not: why rarefying microbiome data is inadmissible. PLoS Comput. Biol. 10, e1003531 (2014).100.Kim, Y. S., Unno, T., Kim, B.-Y. & Park, M. Sex differences in gut microbiota. World J. Mens. Health 38, 48–60 (2020).PubMed 
Article 
PubMed Central 

Google Scholar 
101.Kolodny, O. et al. Coordinated change at the colony level in fruit bat fur microbiomes through time. Nat. Ecol. Evol. 3, 116–124 (2019).PubMed 
Article 
PubMed Central 

Google Scholar 
102.Kartzinel, T. R., Hsing, J. C., Musili, P. M., Brown, B. R. P. & Pringle, R. M. Covariation of diet and gut microbiome in African megafauna. Proc. Natl Acad. Sci. USA 116, 23588–23593 (2019).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
103.Zuur, A., Ieno, E. N., Walker, N., Saveliev, A. A. & Smith, G. M. Mixed Effects Models and Extensions in Ecology with R (Springer Science & Business Media, 2009).104.Nakagawa, S. & Schielzeth, H. A general and simple method for obtaining R2 from generalized linear mixed-effects models. Methods Ecol. Evol. 4, 133–142 (2013).Article 

Google Scholar 
105.Brooks, M. E. et al. glmmTMB balances speed and flexibility among packages for zero-inflated generalized linear mixed modeling. R. J. 9, 378–400 (2017).Article 

Google Scholar 
106.Lozupone, C. & Knight, R. UniFrac: a new phylogenetic method for comparing microbial communities. Appl. Environ. Microbiol. 71, 8228–8235 (2005).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
107.Lozupone, C. A., Hamady, M., Kelley, S. T. & Knight, R. Quantitative and qualitative β diversity measures lead to different insights into factors that structure microbial communities. Appl. Environ. Microbiol. 73, 1576–1585 (2007).CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
108.Anderson, M. J. Permutational Multivariate Analysis of Variance (PERMANOVA). https://doi.org/10.1002/9781118445112.stat07841. (2017)109.Anderson, M. J. & Walsh, D. C. I. PERMANOVA, ANOSIM, and the Mantel test in the face of heterogeneous dispersions: what null hypothesis are you testing? Ecol. Monogr. 83, 557–574 (2013).Article 

Google Scholar 
110.Li, H. et al. Pika population density is associated with the composition and diversity of gut microbiota. Front. Microbiol. 7, 1–9 (2016).
Google Scholar 
111.Weiss, S. et al. Normalization and microbial differential abundance strategies depend upon data characteristics. Microbiome 5, 1–18 (2017).Article 

Google Scholar 
112.Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. R. Stat. Soc. Ser. B 57, 289–300 (1995).
Google Scholar 
113.Fackelmann, G. gfackelmann/human-encroachment-into-wildlife-gut-microbiomes: Release 1.0.0. https://doi.org/10.5281/zenodo.4725220. (2021) More

Study siteThe experiment was conducted at Madingley Wood, Cambridgeshire, UK (0◦3.2´E, 52◦12.9´N) during summer 2018. Madingley Wood is an established field site with an ongoing long-term study of the blue tit and great tit populations. During the autumn and winter birds are caught from feeding stations using mist nets and they are fitted with British Trust of Ornithology (BTO) ID rings. Since 2012, blue tits and great tits have been fitted with RFID tags (BTO Special Methods permit to HMR), which enables collecting data remotely about their foraging behavior and social relationships. The study site has 90 nest boxes that are monitored annually during the breeding season. In 2018 chicks (n = 325) fledged successfully from 45 nest boxes (blue tits = 21, great tits = 24) and they were all ringed and fitted with RFID tags when they were approximately 10 days old. Because new juvenile flocks were arriving at our study site throughout the summer, we also conducted several mist-netting and ringing sessions in July and August to maintain a high proportion of blue tits and great tits ringed and RFID tagged for the experiments (on average 89%, see below). The study protocol was approved by the Animal Users Committee at the Department of Zoology, University of Cambridge.Food itemsWe investigated birds’ foraging choices by offering them colored almond flakes at bird feeders that were distributed throughout the wood. Before beginning the experiments, we allowed the birds to become familiar with the food items by providing plain ‘control’ almonds (plain and not colored) in paired feeders (1.5 m apart) at three locations (approximately 170 m from each other). The feeders were surrounded with metal cages to exclude larger birds, and we placed plastic buckets under the feeders to collect any spilled almonds and minimize birds’ opportunities to forage from the ground. We introduced the feeders at the beginning of June when the nestlings had fledged and were beginning to forage independently, and continued to provide these plain almonds in between our learning experiments (Fig. 3c).In the learning experiments, almond flakes were dyed with non-toxic food dye (Classikool Concentrated Droplet Food Colouring). We used three different color pairs: green (Leaf Green) and red (Bright Red), purple (Lavender Purple) and blue (Royal Blue), and orange (Satsuma Orange) and yellow (Dandelion Yellow). Almond flakes were dyed by soaking them for approximately 20 min in a solution of 900 ml of water and 30 ml of food dye and then left to air dry for 48 h. In the avoidance learning experiments, we made half of the almond flakes unpalatable by soaking them for one hour in 67% solution of chloroquine phosphate, following previously established methods from avoidance learning studies with birds in captivity14,23,24,25. The food dye was added to the solution during the last 20 min.Red and green are common colors used by aposematic, or cryptic prey, respectively9. Therefore, we investigated whether blue tits and great tits had initial color biases towards these colors before starting the main experiment. Because we did not want the birds in our study population to have any experience of the colors before the main experiment, this pilot study was conducted in Newbury, which is 130 km from our main study site. Birds were simultaneously presented with two feeders containing red and green almonds (both palatable) for 30 min and the number of almonds of each color taken by blue tits or great tits was recorded using binoculars. The position of the feeders was switched after 15 min to control for any preferences for feeder location, and the test was repeated on 9 different days. We did not find any evidence that birds had initial color preferences (t-test: t = 0, df = 15.69, p = 1). For the other two learning experiments, we chose color pairs that were available as a food dye and as different from red and green in the visible spectrum as possible to avoid generalization across experiments. These color pairs (blue/purple and yellow/orange) had similar contrast ratios as green and red, based on their RGB values (measured from photographs, see Supplementary Information). Although avian and human vision is different, the discriminability of colors is likely to be similar51, and rapid avoidance learning in each experiment shows that all colors were easily distinguishable. This was the main requirement for testing social information use, and subtle differences in color pair discriminability should only introduce noise to our data but not influence our conclusions.Learning experiments with colored almondsWe conducted three avoidance learning experiments with different color pairs throughout the summer: red/green, blue/purple, and yellow/orange (unpalatable/palatable). In addition, we conducted a reversal-learning experiment with the blue/purple color pair by making both colors palatable after birds had acquired avoidance to blue almonds. Each experiment followed a similar protocol, in which birds were presented with colored almonds at the same three feeding stations where they were previously offered plain almonds. Each feeding station had two feeders, where one contained the palatable color and the other contained the unpalatable color (except in the reversal learning test when both colors were palatable). We switched the side of the feeders every day to make sure that birds learned to associate palatability with an almond color and not a feeder position. The feeders were filled at least once a day (or more often if necessary) to make sure that birds always had access to both colors. We continued each avoidance learning experiment until >90% of all recorded visits were to the feeder with palatable almonds, indicating that most birds in the population had learned to discriminate the colors. This took 7 days in the red/green experiment and 8 days in the other two color pairs (blue/purple and yellow/orange). The reversal learning experiment was finished after 9 days when 50% of the visits were to the previously unpalatable color (blue), indicating that most birds had reversed their learned avoidance towards it.Recording visits to feedersWe monitored visits to all feeders using RFID antennas and data loggers (Francis Scientific Instruments, Ltd) that scanned birds’ unique RFID tag codes when they landed on a perch attached to the feeder. During the learning experiments, each day we also recorded videos from all three feeding stations (using Go Pro Hero Action Camera and Canon Legria HF R66 Camcorder). From the videos, we monitored the proportion of blue tits and great tits that did not have RFID tags and were therefore not recorded when visiting the feeders. We calculated the estimated RFID tag coverage for each day of the experiments by watching at least 100 visits to the feeders from the videos (divided equally among the three feeding stations) and recording whether blue tits and great tits had an RFID tag or not. We realized that the number of untagged individuals was very high (approximately 50% of all visiting birds) when we started the experiment with the first color pair (red/green; see Supplementary Fig. 3). We, therefore, stopped the experiment after two days and caught birds from the feeding stations with mist nets to fit RFID tags to new individuals. To maintain a high number of individuals RFID tagged for the other color pairs, we conducted a mist netting session a day before starting each experiment, as well as 4–5 days after it. We always switched the feeders back to containing plain almonds during mist-netting sessions to ensure that this would not interfere with the learning experiments. Apart from the first two days of the red/green experiment, the RFID tag coverage was on average 89% throughout the experiments (varying between 80 and 95%, Supplementary Fig. 3).Birds were recorded every time that they visited the feeders, i.e., landed on the RFID antenna. However, it is possible that birds did not take the almond during every visit. To get an estimate of how often birds landed on the antenna without taking the almond, and whether this differed between palatable and unpalatable colors, we analyzed the visits to the feeders from the video recordings. We watched videos from the five first days of each experiment (i.e., different color pairs) and analyzed 60 visits to each color (divided approximately equally among the three feeding stations). We recorded whether the feeding event happened (birds ate the almond at the feeder or flew away with it) or whether birds left the feeder without sampling the almond. Because the number of visits to the unpalatable feeder was low during the last days of the avoidance learning experiments, we decided not to analyze avoidance learning videos after day five (but recorded visits from all days of the reversal learning experiment). We found that in avoidance learning experiments birds started to ‘reject’ unpalatable almonds after two days, i.e., they sometimes landed on the feeder but flew away without taking the almond (see Supplementary Fig. 4a). This change was not observed at palatable feeders where birds continued to consume almonds at a similar rate as at the beginning of the experiment (Supplementary Fig. 4a). In reversal learning, the proportion of visits that did not include a feeding event did not differ between purple and blue almonds: birds showed similar hesitation towards both colors at the beginning of the experiment, but this wariness decreased when the experiment progressed, with birds taking the almond during most of their visits (Supplementary Fig. 4b).Statistical analyses and model validationForaging choices in learning experimentsWe first analyzed how birds’ foraging choices changed during the learning experiments using generalized linear mixed-effects models with a binomial error distribution. The number of times an individual visited each feeder on each day of the experiment was used as a bounded response variable, and this was explained by species (blue tit/great tit), individuals’ age (juvenile/adult), and day of the experiment (continuous variable), as well as bird identity as a random effect. When analyzing avoidance learning, initial exploration of data suggested that results were similar across all three experiments, so we combined the experiments in the same model. To investigate whether learning curves differed between the species or age groups, the day of the experiment was included as a second-order polynomial term, and we started model selections with models that included a three-way interaction between species, age, and day2. Best-fitting models were selected based on Akaike’s information criterion (see Supplementary Tables 1 and 2).Social networkTo investigate if birds used social information in their foraging choices, we first constructed a social network of the bird population based on their visits to feeders outside of the learning experiments, i.e., when birds were presented with plain almonds (in total 92 days, see Supplementary Information for the robustness of analysis to exclusion of network data before or after the experiment). We used only these data as individuals were likely to vary in their hesitation to visit novel colored almonds. We used a Gaussian mixture model to detect the clusters of visits (‘gathering events’) at the feeders52 and then calculated association strengths between individuals based on how often they were observed in the same group (gambit of the group approach). These associations (network edges) were calculated using the simple ratio index, SRI35.$$frac{x}{x+{y}_{{mathrm{A}}}+{y}_{{mathrm{B}}}+{y}_{{{mathrm{AB}},}}}$$
(2)
where x is the number of samples where individuals A and B co-occurred in the same group, yA is the number of samples where only individual A was seen, yB is the number of samples where only individual B was seen, and yAB is the number of samples where both A and B were observed in the same sample but not together. Network associations, therefore, estimated the probability that two individuals were in the same group at a given time, with the values scaled between 0 (never observed in the same group) and 1 (always observed in the same group).Social information use during avoidance learning: model descriptionIf social avoidance learning was occurring, then the more birds observed negative responses of others feeding on the unpalatable feeder, the less likely they would be to choose the unpalatable feeder themselves. Thus, we expected the probability of bird j choosing the unpalatable option at time t to decrease with ({R}_{-,j}left(tright)) (the real number of negative feeding events observed by j prior to time t). Likewise, if appetitive social learning was occurring, then the more birds observed positive responses of others feeding on the palatable feeder, the more likely they would be to choose the palatable feeder themselves (rather than the unpalatable feeder). So, we also expected the probability of j choosing the unpalatable option at time t to decrease as ({R}_{+,j}left(tright),)(the real number of positive events observed by j prior to time t) increased.However, we could not test for an effect of ({R}_{-,j}left(tright)) and ({R}_{+,j}left(tright)) directly, since birds often ate the almond away from the feeder, and therefore the real number of observed feeding events could not be measured. Instead, we aimed to test for a pattern following the social network that is consistent with these social learning processes. We reasoned that the probability that one individual i, observes a specific feeding event by another individual j, was proportional to the network connection between them, aij (probability they are in the same feeding group at a given time). Therefore, in each avoidance learning experiment (i.e., different color pair), we calculated the expected number of negative feeding events observed, prior to each choice (occurring at time t) as$${O}_{-,i}left(tright)={sum }_{j}{N}_{-,j}left(tright){a}_{{ij}},$$
(3)
where ({N}_{-,j}left(tright)) was the number of times j had visited unpalatable almonds prior to time t (i ≠ j), and summation is across all birds in the network, and likewise for the expected number of positive feeding events:$${O}_{+,i}left(tright)={sum }_{j}{N}_{+,j}left(tright){a}_{{ij}},$$
(4)
where ({N}_{+,j}left(tright)) was the number of times j had visited palatable almonds prior to time t (i ≠ j).We analyzed whether the expected observations of positive and/or negative feeding events of others influenced the foraging choices in the avoidance learning experiments using generalized linear mixed-effects models with a binomial error distribution. We used each choice (i.e., visit a feeder) as a binary response variable (1 = unpalatable chosen, 0 = palatable chosen), with the probability that unpalatable feeder is chosen on feeding event E given by ({p}_{E}={p}_{-,{i}(E)}left({t}_{E}right)), where i(E) is the individual that fed during event E and ({t}_{E}) is the time at which event E occurred. We then modeled the probability of i choosing the unpalatable option at time t as:$${p}_{-,i}left(tright)={rm{logit}}left(alpha +{beta }_{{rm{asoc}}+}{N}_{+,i}left(tright)+{beta }_{{rm{asoc}}-}{N}_{-,i}left(tright)+{beta }_{{rm{soc}}+}{O}_{+,i}left(tright)+{beta }_{s{rm{oc}}-}{O}_{-,i}left(tright)+{{{{rm{B}}}}}_{i}right),$$
(5)
where ({N}_{+,i}left(tright)) is the number of times a choosing individual had visited the palatable feeder (positive personal information), ({N}_{-,i}left(tright)) is the number of times a choosing individual had visited the unpalatable feeder (negative personal information), ({O}_{+,i}left(tright)) is the expected number of observed positive (positive social information) and ({O}_{-,i}left(tright)) observed negative feeding events (negative social information). Bird identity was included as a random effect, ({{rm{{B}}}}_{i}) (age and species were later added as variables, see below). Parameters ({beta }_{{rm{asoc}}+}) and ({beta }_{{rm{asoc}}-}) are the effects of asocial learning about the palatable and unpalatable foods, ({beta }_{{rm{soc}}+}) is the effect of social learning about the palatable food, and ({beta }_{{rm{soc}}-})is the effect of social avoidance learning about the unpalatable food. Estimation of these parameters, with associated Wald tests and confidence intervals, allowed us to make inferences about which effects were operating and the size of these effects. To aid model fitting we standardized all predictor variables and then back-transformed the effects to the original scale (see Supplementary Tables 3–5 for the model outputs). To assess the importance of asocial and social effects, we also ran separate models that excluded either asocial or social parameters and compared them to the initial model in Eq. (5) using Akaike’s information criterion (see Supplementary Table 6). However, in most cases, this reduced model fit significantly, and we, therefore, kept all parameters in the final models.Our approach took ({O}_{-,{i}}left(tright)) as a measure of ({R}_{-,j}left(tright)), and ({O}_{+,{i}}left(tright)) as a measure of ({R}_{+,j}left(tright))-, which we termed the ‘expected’ number of observations of each type. Strictly speaking, ({O}_{-,{i}}left(tright)) and ({O}_{+,{i}}left(tright)) were upper limits on the expected number of observations, assuming that birds observed all feeding events in the groups in which they were present, whereas only an unknown proportion of such events (({p}_{o})) was observed. Therefore, the real expected number of negative/positive observations would be (Eleft({R}_{-,j}left(tright)right)={p}_{o}{O}_{-,{i}}left(tright)) and (Eleft({R}_{+,j}left(tright)right)={p}_{o}{O}_{+,{i}}left(tright)) respectively. Thus, the coefficient, ({beta }_{{rm{soc}}-}), for the effect of ({O}_{-,{i}}left(tright)) could be interpreted as ({beta }_{s{rm{oc}}-}={p}_{o}acute{{beta }_{{rm{soc}}-}}) where (acute{{beta }_{s{rm{oc}}-}}) is the effect per observation. Note that since ({p}_{o}le 1), and(,{beta }_{{rm{soc}}-}=acute{{beta }_{s{rm{oc}}-}}{p}_{o}), ({beta }_{s{rm{oc}}-}) is more likely to underestimate than overestimate the effect per observation of a negative feeding event. An analogous argument applies to the coefficient, ({beta }_{{rm{soc}}+}), for the effect of ({O}_{+,{i}}left(tright)).Social information use during avoidance learning: extension to test for species effectsAfter fitting the initial model shown in Eq. (5), we further broke down the model to test whether individuals were more likely to learn socially by observing conspecifics than heterospecifics. This was done by splitting the expected number of observed positive and negative feeding events to observations of conspecifics (({O}_{+{rm{C}},i}left(tright)), ({O}_{-{rm{C}},i}left(tright))) and heterospecifics (({O}_{+{rm{H}},i}left(tright)), ({O}_{-{rm{H}},i}left(tright))), and including these in the model as separate explanatory variables thus:$${p}_{-,i}(t)={rm{logit}}left(begin{array}{c}alpha +{beta }_{{rm{asoc}}+}{N}_{+,i}(t)+{beta }_{{rm{asoc}}-}{N}_{-,i}(t)\ +{beta }_{{rm{soc}},+{rm{H}}}{O}_{+{rm{H}},i}(t)+{beta }_{{rm{soc}},-{rm{H}}}{O}_{-{rm{H}},i}(t)\ +{beta }_{{rm{soc}},+{rm{C}}}{O}_{+{rm{C}},i}(t)+{beta }_{{rm{soc}},-{rm{C}}}{O}_{-{rm{C}},i}(t) \ +{{{{rm{B}}}}}_{i}end{array}right)$$
(6a)
with ({beta }_{{rm{soc}},-{rm{H}}}) and ({beta }_{{rm{soc}},-{rm{C}}}) giving the effect of a negative observation of a heterospecific and conspecific, respectively, whereas ({beta }_{{rm{soc}},+{rm{H}}}) and ({beta }_{{rm{soc}},+{rm{C}}}) give the effect of positive observation of a heterospecific and conspecific, respectively. In general –/+ subscripts refer to negative/positive feeding events and C/H subscripts to feeding events by conspecifics/heterospecifics. By re-parameterizing the model thus:$${p}_{-,i}left(tright)={rm{logit}}left(begin{array}{c}alpha +{beta }_{{rm{asoc}}+}{N}_{+,i}left(tright)+{beta }_{{rm{asoc}}-}{N}_{-,i}left(tright) \ +{beta }_{{rm{soc}},{rm{H}}+}{O}_{+,i}left(tright)+{beta }_{{rm{soc}},{rm{H}}-}{O}_{-,i}left(tright)\ +left({beta }_{{rm{soc}},{rm{C}}+}-{beta }_{{rm{soc}},{rm{H}}+}right){O}_{+{rm{C}},i}left(tright)+left({beta }_{{rm{soc}},{rm{C}}-}-{beta }_{{rm{soc}},{rm{H}}-}right){O}_{-{rm{C}},i}left(tright)\ +{{{{rm{B}}}}}_{i}end{array}right)$$
(6b)
we were able to test for a difference between observations of negative feeds by conspecifics and heterospecifics (left({beta }_{{rm{soc}},{rm{C}}-}-{beta }_{{rm{soc}},{rm{H}}-}right)) and between observations of positive feeds by conspecifics and heterospecifics (left({beta }_{{rm{soc}},{rm{C}}+}-{beta }_{{rm{soc}},{rm{H}}+}right)).For all experiments there was no evidence for a difference between ({beta }_{{rm{soc}},-{rm{H}}}) and ({beta }_{{rm{soc}},-{rm{C}}}) (yellow/orange: Z = 0.803, p = 0.42; red/green: Z = 0.065, p = 0.95; blue/purple: Z = 1.113, p = 0.27). However, there was some evidence of a difference between ({beta }_{{rm{soc}},+{rm{H}}}) and ({beta }_{{rm{soc}},+{rm{C}}}) in two of the three experiments (yellow/orange: Z = 1.359, p = 0.17; red/green: Z = 1.417, p = 0.16; blue/purple: Z = 0.729, p = 0.47). Consequently, we reduced the model down to:$${p}_{-,i}left(tright)={rm{logit}}left(begin{array}{c}alpha +{beta }_{{rm{asoc}}+}{N}_{+,i}left(tright)+{beta }_{{rm{asoc}}-}{N}_{-,i}left(tright)\ +{beta }_{{rm{soc}},-}{O}_{-,i}left(tright)+{beta }_{{rm{soc}},+{rm{H}}}{O}_{+{rm{H}},i}left(tright)+{beta }_{{rm{soc}},+{rm{C}}}{O}_{+{rm{C}},i}left(tright)\ +{{{{rm{B}}}}}_{i}end{array}right)$$
(7)
for further analysis, i.e., with different effects for observations of conspecific/heterospecific positive feeds, but not of negative feeds. We did this for all color combinations (including blue/purple) to allow comparison across experiments (see Table 1). The R code used to run these models can be found in Supplementary data53 in ‘GLMM models Orange Yellow final.r’.Social information use during avoidance learning: simulations to test for a network effectNext, we tested whether the social effects we detected followed the social network. When using a network-based diffusion analysis (NBDA43), researchers can compare a network model with one in which the network has homogeneous connections among all individuals, but we found this to be unreliable for our model. Instead, we used a simulation approach to generate a null distribution for the null hypothesis of homogeneous social effects, taking the size of the social effects from the fitted models. We ran 1000 simulations (using the same procedure described above) for all social effects that were found to be significant in each avoidance learning model (each color pair; see Table 1). The total number of expected observations was kept equal, but we homogenized the observation effect across all birds by replacing the probability of bird i observing a feed by bird j, previously ({a}_{{ij}}), with ({sum }_{i}{a}_{{ij}}/n), where n is the number of birds in the experiment, (i.e., all birds had the same probability of observing each feeding event). The model was fitted to the simulated data each time to extract the Z value (Wald test statistic) of the social effect of interest. The distribution of these values was then used as a null distribution to test whether our observed social effect differed from the effects that did not follow the social network. To this end, we calculated the proportion of simulations that yielded a Z value as extreme or more extreme than that observed (judged by distance in either direction from the mean of the null distribution). The R code used to run these simulations can be found in Supplementary data53 in ‘Simulations to test if network effects follow network Orange Yellow.r’.Social information use during avoidance learning: extension to test for age effectsWe then aimed to test whether each of the three social effects detected differed based on the age class of the observed individual (adult versus juveniles). We, therefore, split the negative expected observations ({O}_{-,i}left(tright)) into the expected observations of adults ({O}_{-{rm{A}},i}left(tright)) and juveniles ({O}_{-{rm{J}},i}left(tright)), each with its associated coefficient in the model ({beta }_{{rm{soc}},-{rm{A}}}) and ({beta }_{{rm{soc}},-{rm{J}}}). Likewise, we split positive observations of conspecifics as ({O}_{+{rm{CA}},i}left(tright)) and ({O}_{+{rm{CJ}},i}left(tright)) and positive observations of heterospecifics as ({O}_{+{rm{HA}},i}left(tright)) and ({O}_{+{rm{HJ}},i}left(tright)) to give the model:$${p}_{-,i}left(tright)={rm{logit}}left(begin{array}{c}alpha +{beta }_{{rm{asoc}}+}{N}_{+,i}left(tright)+{beta }_{{rm{asoc}}-}{N}_{-,i}left(tright)\ +{beta }_{{rm{soc}},-{rm{A}}}{O}_{-{rm{A}},i}left(tright)+{beta }_{{rm{soc}},-{rm{J}}}{O}_{-{rm{J}},i}left(tright)\ +{beta }_{{rm{soc}},+{rm{HA}}}{O}_{+{rm{HA}},i}left(tright)+{beta }_{{rm{soc}},+{rm{HJ}}}{O}_{+{rm{HJ}},i}left(tright)\ +{beta }_{{rm{soc}},+{rm{CA}}}{O}_{+{rm{CA}},i}left(tright)+{beta }_{{rm{soc}},+{rm{CJ}}}{O}_{+{rm{CJ}},i}left(tright)\ +{{{{rm{B}}}}}_{i}end{array}right)$$
(8a)
As before, –/+ subscripts refer to negative/positive feeding events, C/H subscripts to feeding events by conspecifics/heterospecifics, and A/J subscripts to feeding events by adults/juveniles. We also fitted a re-parameterized version allowing us to test for a difference between expected observations of adults and observations of juveniles for each of the three social effects:$${p}_{-,i}left(tright)={rm{logit}}left(begin{array}{c}alpha +{beta }_{{rm{asoc}}+}{N}_{+,i}left(tright)+{beta }_{{rm{asoc}}-}{N}_{-,i}left(tright)\ +{beta }_{{rm{soc}},-{rm{J}}}{O}_{-,i}left(tright)+left({{beta }_{{rm{soc}},-{rm{A}}}-beta }_{{rm{soc}},-{rm{J}}}right){O}_{-{rm{A}},i}left(tright)\ +{beta }_{{rm{soc}},+{rm{H}}}{O}_{+{rm{H}},i}left(tright)+left({{beta }_{{rm{soc}},+{rm{HA}}}-beta }_{{rm{soc}},+{rm{HJ}}}right){O}_{+{rm{JA}},i}left(tright)\ +{beta }_{{rm{soc}},+{rm{C}}}{O}_{+{rm{C}},i}left(tright)+left({{beta }_{{rm{soc}},+{rm{CA}}}-beta }_{{rm{soc}},+{rm{CJ}}}right){O}_{+{rm{CA}},i}left(tright)\ +{{{{rm{B}}}}}_{i}end{array}right)$$
(8b)
The R code used to run these models can be found in Supplementary data53 in ‘GLMM models Orange Yellow final.r’. The main results of each model are presented in Table 2 and full model outputs in Supplementary Tables 3–5.Social information use during reversal learningTo investigate social information use during reversal learning, we used the order of acquisition diffusion analysis (OADA), a variant of NBDA43, which explores the order in which individuals acquire a behavioral trait44. The rate of social transmission between two individuals is assumed to be linearly proportional to their network connection, and the spread of trait acquisition is therefore predicted to follow the network patterns if individuals are using social information. We used NBDA to investigate whether the order of individuals’ first visit to the previously unpalatable blue almonds (mimics) followed the network. We fitted several different models that included (i) only asocial learning, (ii) social transmission of information following a homogeneous network (equal associations among all individuals), or (iii) social transmission of information following our observed network. Models that included social transmission were further divided into models with equal or different transmission rates from adults and juveniles, and from conspecifics and heterospecifics, by constructing separate networks for each adult/juvenile and conspecific/heterospecific combination. To investigate whether asocial or social learning rates differed between blue tits and great tits, we included species as an individual-level variable. We then compared different social transmission models that assumed that species differed in both asocial and social learning rates, only in asocial or only in social learning rates, or that they did not differ in either (see Table 3). The best-supported model was selected using a model-averaging approach with Akaike’s information criterion corrected for small sample sizes. All analyses were conducted with the software R.3.6.154, using lme455, asnipe56, and NBDA57 packages.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More

1.IPCC Climate Change 2014: Synthesis report. In Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (eds Core Writing Team et al.) (IPCC, 2014).
Google Scholar 
2.Keogan, K. et al. Global phenological insensitivity to shifting ocean temperatures among seabirds. Nat. Clim. Chang. 8, 313–318 (2018).ADS 
Article 

Google Scholar 
3.Chambers, L. E. et al. Phenological changes in the southern hemisphere. PLoS ONE 8, E75514 (2013).ADS 
CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
4.Thackeray, S. J. et al. Phenological sensitivity to climate across taxa and trophic levels. Nature 535, 241–245 (2016).ADS 
CAS 
PubMed 
Article 

Google Scholar 
5.Forrest, J. & Miller-Rushing, A. J. Toward a synthetic understanding of the role of phenology in ecology and evolution. Philos. Trans. R. Soc. Lond. Series B Biol. Sci. 365, 3101–3112 (2010).Article 

Google Scholar 
6.Møller, A. P., Rubolini, D. & Lehikoinen, E. Populations of migratory bird species that did not show a phenological response to climate change are declining. Proc. Natl. Acad. Sci. 105, 16195–16200 (2008).ADS 
PubMed 
PubMed Central 
Article 

Google Scholar 
7.Hipfner, J. M. Matches and mismatches: Ocean climate, prey phenology and breeding success in a zooplanktivorous seabird. Mar. Ecol. Prog. Ser. 368, 295–304 (2008).ADS 
Article 

Google Scholar 
8.Moyes, K. et al. Advancing breeding phenology in response to environmental change in a wild red deer population. Glob. Change Biol. 17, 2455–2469 (2011).ADS 
Article 

Google Scholar 
9.Poloczanska, E. S. et al. Global imprint of climate change on marine life. Nat. Clim. Chang. 3, 919–925 (2013).ADS 
Article 

Google Scholar 
10.Hauser, D. D. et al. Decadal shifts in autumn migration timing by Pacific Arctic beluga whales are related to delayed annual sea ice formation. Glob. Change Biol. 23, 2206–2217 (2017).ADS 
Article 

Google Scholar 
11.Beltran, R. S., Kirkham, A. L., Breed, G. A., Ward Testa, J. & Burns, J. M. Reproductive success delays moult phenology in a polar mammal. Sci. Rep. 9, 1–12 (2019).CAS 
Article 

Google Scholar 
12.Oosthuizen, W. C. et al. Individual heterogeneity in life-history trade-offs with age at first reproduction in capital breeding elephant seals. Popul. Ecol. 61, 421–435 (2019).Article 

Google Scholar 
13.Ling, J. K. The skin and hair of the southern elephant seal, Mirounga leonina (Linn). IV. Annual cycle of pelage follicle activity and moult. Aust. J. Zool. 60, 259–271 (2012).Article 

Google Scholar 
14.Johanos, T. C., Becker, B. L. & Ragen, T. J. Annual reproductive cycle of the female Hawaiian monk seal (Monachus schauinslandi). Mar. Mamm. Sci. 10, 13–30 (1994).Article 

Google Scholar 
15.Beltran, R. S., Burns, J. M. & Breed, G. A. Convergence of biannual moulting strategies across birds and mammals. Proc. R. Soc. B Biol. Sci. 285, 20180318 (2018).Article 

Google Scholar 
16.Boyd, I., Arnbom, T. & Fedak, M. Water flux, body composition, and metabolic rate during moult in female southern elephant seals (Mirounga leonina). Physiol. Zool. 66, 43–60 (1993).Article 

Google Scholar 
17.Postma, M., Bester, M. N. & De Bruyn, P. J. N. Spatial variation in female southern elephant seal mass change assessed by an accurate non-invasive photogrammetry method. Antarct. Sci. 25, 731–740 (2013).ADS 
Article 

Google Scholar 
18.Paterson, W. et al. Seals like it hot: Changes in surface temperature of harbour seals (Phoca vitulina) from late pregnancy to moult. J. Therm. Biol. 37, 454–461 (2012).Article 

Google Scholar 
19.Slip, D. J. & Burton, H. R. Population status and seasonal haulout patterns of the southern elephant seal (Mirounga leonina) at Heard Island. Antarct. Sci. 11, 38–47 (1999).ADS 
Article 

Google Scholar 
20.Kirkman, S. et al. Variation in the timing of moult in southern elephant seals at Marion Island. S. Afr. J. Wildlife Res. 33, 79–84 (2003).
Google Scholar 
21.Ling, J. & Bryden, M. Southern elephant seal Mirounga leonina (Linnaeus), 1758. In Handbook of Marine Mammals Vol. 2 (eds Ridgway, S. H. & Harrison, R. J.) 297–327 (Academic Press, 1981).
Google Scholar 
22.Constable, A. J. et al. Climate change and southern ocean ecosystems I: how changes in physical habitats directly affect marine biota. Glob. Change Biol. 20, 3004–3025 (2014).ADS 
Article 

Google Scholar 
23.Rogers, A. et al. Antarctic futures: An assessment of climate-driven changes in ecosystem structure, function, and service provisioning in the Southern Ocean. Ann. Rev. Mar. Sci. 12, 87–120 (2020).CAS 
PubMed 
Article 

Google Scholar 
24.Hindell, M. A. & Burton, H. R. Seasonal haul-out patterns of the southern elephant seal (Mirounga leonina L.), at Macquarie Island. J. Mammal. 69, 81–88 (1988).Article 

Google Scholar 
25.Postma, M., Bester, M. N. & De Bruyn, P. J. N. Age-related reproductive variation in a wild marine mammal population. Polar Biol. 36, 719–729 (2013).Article 

Google Scholar 
26.Le Boeuf, B. J. & Laws, R. M. Elephant seals: An introduction to the genus. In Elephant Seals: Population Ecology, Behavior, and Physiology (eds Le Beouf, B. J. & Laws, R. M.) 1–28 (University Of California Press, 1994).Chapter 

Google Scholar 
27.De Bruyn, P. J. N. et al. Sex at sea: Alternative mating system in an extremely polygynous mammal. Anim. Behav. 82, 445–451 (2011).Article 

Google Scholar 
28.Oosthuizen, W. C., Pradel, R., Bester, M. N. & De Bruyn, P. J. N. Making use of multiple surveys: Estimating breeding probability using a multievent-robust design capture-recapture model. Ecol. Evolut. 9, 836–848 (2019).Article 

Google Scholar 
29.Bester, M. N. et al. The marine mammal programme at the Prince Edward Islands: 38 years of research. Afr. J. Mar. Sci. 33, 511–521 (2011).Article 

Google Scholar 
30.Pistorius, P. A., De Bruyn, P. J. N. & Bester, M. N. Population dynamics of southern elephant seals: A synthesis of three decades of demographic research at Marion Island. Afr. J. Mar. Sci. 33, 523–534 (2011).Article 

Google Scholar 
31.Pistorius, P. A., Bester, M. N. & Kirkman, S. P. Survivorship of a declining population of southern elephant seals, Mirounga leonina, in relation to age, sex and cohort. Oecologia 121, 201–211 (1999).ADS 
CAS 
PubMed 
Article 

Google Scholar 
32.Oosthuizen, W. C., Bester, M. N., Altwegg, R., McIntyre, T. & De Bruyn, P. J. N. Decomposing the variance in southern elephant seal weaning mass: Partitioning environmental signals and maternal effects. Ecosphere 6, 139 (2015).Article 

Google Scholar 
33.Daniel, R. G., Jemison, L. A., Pendleton, G. W. & Crowley, S. M. Molting phenology of harbor seals on Tugidak Island, Alaska. Mar. Mamm. Sci. 19, 128–140 (2003).Article 

Google Scholar 
34.Chaise, L. L. et al. Local weather and body condition influence habitat use and movements on land of molting female southern elephant seals (Mirounga leonina). Ecol. Evol. 8, 6081–6090 (2018).PubMed 
PubMed Central 
Article 

Google Scholar 
35.Dingemanse, N. J. & Dochtermann, N. A. Quantifying individual variation in behaviour: Mixed-effect modelling approaches. J. Anim. Ecol. 82, 39–54 (2013).PubMed 
Article 

Google Scholar 
36.Burnham, K. P., Anderson, D. R. & Huyvaert, K. P. AIC model selection and multimodel inference in behavioral ecology: Some background, observations, and comparisons. Behav. Ecol. Sociobiol. 65, 23–35 (2011).Article 

Google Scholar 
37.Zuur, A., Ieno, E. N., Walker, N., Saveliev, A. A. & Smith, G. M. Mixed Effects Models and Extensions in Ecology with R (Springer Science & Business Media, 2009).MATH 
Book 

Google Scholar 
38.Bates, D., Mächler, M., Bolker, B. & Walker, S. Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67, 1–48 (2015).Article 

Google Scholar 
39.R Core Team. R: A language and environment for statistical computing. (R Foundation for Statistical Computing, Vienna, Austria, 2020). https://www.R-project.org/.40.Nakagawa, S. & Schielzeth, H. Repeatability for Gaussian and non-Gaussian data: A practical guide for biologists. Biol. Rev. 85, 935–956 (2010).PubMed 

Google Scholar 
41.Nakagawa, S. & Schielzeth, H. A general and simple method for obtaining R2 from generalized linear mixed-effects models. Methods Ecol. Evol. 4, 133–142 (2013).Article 

Google Scholar 
42.Grosbois, V. et al. Assessing the impact of climate variation on survival in vertebrate populations. Biol. Rev. 83, 357–399 (2008).CAS 
PubMed 
Article 

Google Scholar 
43.Forcada, J., Trathan, P. N. & Murphy, E. J. Life history buffering in Antarctic mammals and birds against changing patterns of climate and environmental variation. Glob. Change Biol. 14, 2473–2488 (2008).ADS 
Article 

Google Scholar 
44.Bradshaw, C. J., Hindell, M. A., Sumner, M. D. & Michael, K. J. Loyalty pays: Potential life-history consequences of fidelity to marine foraging regions by southern elephant seals. Anim. Behav. 68, 1349–1360 (2004).Article 

Google Scholar 
45.Cotté, C., Park, Y.-H., Guinet, C. & Bost, C.-A. Movements of foraging king penguins through marine mesoscale eddies. Proc. R. Soc. B Biol. Sci. 274, 2385–2391 (2007).Article 

Google Scholar 
46.Barbraud, C. & Weimerskirch, H. Antarctic birds breed later in response to climate change. Proc. Natl. Acad. Sci. 103, 6248–6251 (2006).ADS 
CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
47.Hindell, M. A. et al. Long-term breeding phenology shift in royal penguins. Ecol. Evol. 2, 1563–1571 (2012).PubMed 
PubMed Central 
Article 

Google Scholar 
48.Zimova, M. et al. Function and underlying mechanisms of seasonal colour moulting in mammals and birds: What keeps them changing in a warming world?. Biol. Rev. 93, 1478–1498 (2018).PubMed 
Article 

Google Scholar 
49.Thompson, P. & Rothery, P. Age and sex differences in the timing of moult in the common seal, Phoca vitulina. J. Zool. 212, 597–603 (1987).Article 

Google Scholar 
50.Cronin, M., Gregory, S. & Rogan, E. Moulting phenology of the harbour seal in south-west Ireland. J. Mar. Biol. Assoc. UK 94, 1079–1086 (2014).Article 

Google Scholar 
51.Kirkman, S. et al. Variation in the timing of the breeding haulout of female southern elephant seals at Marion Island. Aust. J. Zool. 52, 379–388 (2004).Article 

Google Scholar 
52.Hindell, M. A., Slip, D. J. & Burton, H. R. Body mass loss of moulting female southern elephant seals, Mirounga leonina, at Macquarie Island. Polar Biol. 14, 275–278 (1994).Article 

Google Scholar 
53.Carlini, A., Marquez, M., Daneri, G. & Poljak, S. Mass changes during their annual cycle in females of southern elephant seals at King George Island. Polar Biol. 21, 234–239 (1999).Article 

Google Scholar 
54.Worthy, G. A. J., Morris, P. A., Costa, D. P. & Le Boeuf, B. J. Moult energetics of the northern elephant seal (Mirounga angustirostris). J. Zool. 227, 257–265 (1992).Article 

Google Scholar 
55.Arnbom, T., Fedak, M. & Rothery, P. Offspring sex ratio in relation to female size in southern elephant seals, Mirounga leonina. Behav. Ecol. Sociobiol. 35, 373–378 (1994).Article 

Google Scholar 
56.Proffitt, K. M., Garrott, R. A., Rotella, J. J. & Wheatley, K. E. Environmental and senescent related variations in Weddell seal body mass: Implications for age-specific reproductive performance. Oikos 116, 1683–1690 (2007).Article 

Google Scholar 
57.Bérubé, C. H., Festa-Bianchet, M. & Jorgenson, J. T. Individual differences, longevity, and reproductive senescence in bighorn ewes. Ecology 80, 2555–2565 (1999).Article 

Google Scholar 
58.Oosthuizen, W. C., Péron, G., Pradel, R., Bester, M. N. & De Bruyn, P. J. N. Positive early-late life-history trait correlations in elephant seals. Ecology 102, e03288. https://doi.org/10.1002/ecy.3288 (2021).59.Kirby, V. L. & Ortiz, C. L. Hormones and fuel regulation in fasting elephant seals. In Elephant Seals: Population Ecology, Behavior, and Physiology (eds Le Beouf, B. J. & Laws, R. M.) 374–386 (University of California Press, 1994).Chapter 

Google Scholar 
60.Yochem, P. K. & Stewart, B. S. Hair and fur. In Encyclopedia of Marine Mammals 3rd edn (eds Würsig, B. et al.) 447–448 (Academic Press, 2018).Chapter 

Google Scholar 
61.Ashwell-Erickson, S., Fay, F., Elsner, R. & Wartzok, D. Metabolic correlates of molting and regeneration of pelage in Alaskan harbour and spotted seals (Phoca vitulina and Phoca largha). Can. J. Zool. 64, 1086–1092 (1986).CAS 
Article 

Google Scholar 
62.Laws, R. M. & Sinha, A. A. Reproduction. In Antarctic Seals: Research Methods and Techniques (ed. Laws, R. M.) 228–267 (Cambridge University Press, 1993).Chapter 

Google Scholar 
63.Smout, S., King, R. & Pomeroy, P. Environment-sensitive mass changes influence breeding frequency in a capital breeding marine top predator. J. Anim. Ecol. 89, 384–396 (2020).PubMed 
Article 

Google Scholar 
64.Clutton-Brock, T. & Sheldon, B. C. Individuals and populations: the role of long-term, individual-based studies of animals in ecology and evolutionary biology. Trends Ecol. Evol. 25, 562–573 (2010).PubMed 
Article 

Google Scholar 
65.Cordes, L. S. & Thompson, P. M. Variation in breeding phenology provides insights into drivers of long-term population change in harbour seals. Proc. R. Soc. B: Biol. Sci. 280, 20130847 (2013).Article 

Google Scholar 
66.Barbraud, C. et al. Contrasted demographic responses facing future climate change in Southern Ocean seabirds. J. Anim. Ecol. 80, 89–100 (2011).PubMed 
Article 

Google Scholar 
67.Atkinson, S., DeMaster, D. P. & Calkins, D. G. Anthropogenic causes of the western Steller sea lion Eumetopias jubatus population decline and their threat to recovery. Mamm. Rev. 38, 1–18 (2008).Article 

Google Scholar 
68.Desprez, M., McMahon, C. R., Hindell, M. A., Harcourt, R. & Gimenez, O. Known unknowns in an imperfect world: incorporating uncertainty in recruitment estimates using multi-event capture-recapture models. Ecol. Evol. 3, 4658–4668 (2013).PubMed 
PubMed Central 
Article 

Google Scholar 
69.Griffen, B. D. Reproductive skipping as an optimal life history strategy in the southern elephant seal, Mirounga leonina. Ecol. Evol. 8, 9158–9170 (2018).PubMed 
PubMed Central 
Article 

Google Scholar 
70.Rohwer, S., Viggiano, A. & Marzluff, J. M. Reciprocal tradeoffs between moult and breeding in albatrosses. Condor 113, 61–70 (2011).Article 

Google Scholar 
71.Grissot, A., Graham, I. M., Quinn, L., Bråthen, V. S. & Thompson, P. M. Breeding status influences timing but not duration of moult in the northern fulmar Fulmarus glacialis. Ibis 162, 446–459 (2019).Article 

Google Scholar 
72.Condit, R., Beltran, R.S., Robinson, P.W., Crocker, D.E. & Costa, D.P. Birth timing after the long feeding migration in elephant seals. https://doi.org/10.1101/2020.10.02.324210 (in review). More

AnimalsTissue samples for expression analysis were collected on August 24 and September 9 2016 from a population of spring-spawning herring kept in captivity at University of Bergen; the rearing of captive herring was approved by the Norwegian national animal ethics committee (Forsøksdyrutvalget FOTS ID-5072). The tissue samples used for ATAC-sequencing were collected at Hästskär on June 19 2019 from wild spring-spawning Atlantic herring in the Baltic Sea, which do not require ethical permission.Genome scan and genetic diversity at the TSHR locusWe used a 2 × 2 contingency X2 test to estimate the extent of SNP allele frequency differences between seven spring- and seven autumn-spawning herring populations from the Northeast Atlantic (Supplementary Table 2), and thus, identify the major genomic regions associated with seasonal reproduction. The SNP allele frequencies were generated in a previous study16 and were derived from Pool-seq data. For the X2 test, we formed two groups, spring and autumn spawners, and summed the reads supporting the reference and the alternative alleles for the pools included in each group.To characterize genetic diversity at the TSHR locus, we calculated nucleotide diversity (π) and Tajima’s D for the same seven spring- and seven autumn-spawning Atlantic herring populations used in the genome scan (n = ~41–100 per pool) (Supplementary Table 2). The whole-genome re-sequence data of these pools were previously reported by Han et al.16 (for details of the pools used here see Supplementary Table 2). Unbiased nucleotide diversity π and Tajima’s D were calculated for each pool using the program PoPoolation 1.2.236, which accounts for the truncated allele frequency spectrum of pooled data. In brief, a pileup file of chromosome 15 was generated from each pool BAM file using samtools v.1.1037,38. Indels and 5 bp around indels were removed to exclude spurious SNPs due to misalignments around indels. To minimize biases in the π and Tajima’s D calculations, which are sensitive to sequencing errors and coverage fluctuations39, the coverage of each pileup file was subsampled without replacement to a uniform value based on a per-pool coverage distribution (the target coverage corresponded to the 5th percentile of the coverage frequency distribution, which was used as the minimum coverage allowed for an SNP to be included in the analysis) (Supplementary Table 2). We also calculated the diversity parameters but skipping the coverage subsampling step and obtained very similar results with both approaches (Supplementary Fig. 5), thus, we decided to keep working with the subsampled datasets as coverage subsampling is recommended by the software developers36. To exclude spurious SNPs associated with repetitive sequences and copy number variants, we applied a maximum coverage filter equivalent to the per-pool 99th percentile of the coverage frequency distribution (Supplementary Table 2). A minimum base quality of 20, a minimum mapping quality of 20, and a minor allele count of 2 were required to retain high quality SNPs for further analysis. Both, π and Tajima’s D statistics were calculated using a sliding window approach with a window size of 10 kb and a step size of 2 kb (the selected window-step combination offered a good genomic resolution while reducing the noise from single SNPs, after testing windows of 5, 10, 20, 40, 50, and 100 kb for non-overlapping and overlapping windows with a step size equivalent to 20% of the window size). Only windows with a coverage fraction ≥ 0.5 were included in the computations. In addition, we estimated the effective allele frequency difference, or delta allele frequency (dAF), between spring and autumn spawning groups at the TSHR locus using the formula dAF = abs(mean(spring pools)−mean(autumn pools)). For each of the diversity parameters, we assessed whether the mean differences between sets of SNPs within chr 15: 8.85–8.95 Mb (215 SNPs) and outside (214 635 SNPs) the TSHR region were statistically significant among spring- and autumn-spawning groups using a Wilcoxon test. Data postprocessing, statistical tests, and plotting were performed in the R environment40 (for specific parameters used in PoPoolation, see the associated code to this publication).Identification of the 5.2 kb structural variantSequences spanning the entire TSHR gene plus 10 kb upstream and downstream from two reference assemblies, ASM96633v115 and Ch_v2.0.218, were aligned using BLAST41 and the output was subsequently processed with a custom R script40. Repeats were annotated by CENSOR21 for the region harboring the 5.2 kb structural variant. To validate the structural variant, long-range PCR was performed with genomic DNA from two spring- and autumn-spawning Atlantic herring in a 20 μL reaction containing 0.8 mM dNTPs, 0.3 μM each of the forward and reverse 5.2kb-confirm primers (Supplementary Table 1) and 0.75 U PrimeSTAR GXL DNA Polymerase (TaKaRa) following the program: 95 °C for 3 min, 35 cycles of 98 °C for 10 s, 58 °C for 20 s and 68 °C for 2 min 30 s, and a final extension of 10 min at 68 °C.ATAC-seq analysisBSH and brain without BSH were dissected from two spring-spawning herring caught in the Baltic Sea and transported to the lab on dry ice, then kept at –80 °C before nuclei isolation. ATAC-seq libraries were constructed according to the Omni-ATAC protocol with minor modifications42. Briefly, tissue was homogenized in 2 ml homogenization buffer (5 mM CaCl2, 3 mM Mg(Ac)2, 10 mM Tris-HCl (pH = 7.8), 0.017 mM PMSF, 0.17 mM ß-mercaptoethanol, 320 mM Sucrose, 0.1 mM EDTA and 0.1% NP-40) with a Dounce homogenizer on ice. 400 μl suspension was transferred to a 2 ml tube for the density gradient centrifugation with different concentrations of Iodixanol solution. After centrifugation, a 200 μl fraction containing the nuclei band was collected, stained with Trypan blue and counted with a Countess II Automated Cell Counter (Thermo Fisher Scientific). An aliquot of 100,000–200,000 nuclei was used as input in a 50 μl transposition reaction containing 2 X TD buffer and 100 nM assembled Tn5 transposase for a 30-min incubation at 37 °C. Tagmented DNA was purified with a Zymo clean kit (Zymo Research). Purified DNA was used for an initial pre-amplification for 5 cycles, and the additional amplification cycle was determined by qPCR based on the “R vs Cycle Number” plot43. Amplified libraries were purified with a Zymo clean kit again, and library concentrations and qualities were evaluated using the 2200 TapeStation System (Agilent Technologies).ATAC-seq was performed with a MiniSeq High Output Kit (150 cycles) on a MiniSeq instrument (Illumina) and 7–9 million reads were generated for each ATAC-seq library. Quality control, trimming, mapping, and peak calling of the sequenced reads were conducted following the ENCODE ATAC-seq pipeline (https://www.encodeproject.org/atac-seq/). The trimmed reads were aligned to the Atlantic herring reference genome (Ch_v2.0.2)18 with Bowtie244 and the mapping rate was 85–95%. Duplicate reads, reads with low mapping quality and those aligned to the mitochondria genome were removed. The remaining reads (4–5 million) were subjected to peak calling by MACS245, where 22–32 K peaks were called. Sequenced library qualities were further evaluated by calculating the TSS enrichment score and checking the library complexity with the Non-Redundant Fraction (NRF) and PCR Bottlenecking Coefficients (PBC1 and 2). Finally, conserved peaks between two biological replicates were identified by evaluating the irreproducible discovery rate (IDR).Genotyping of six differentiated variants and haplotype analysisAll six genetic variants, including the 5.2 kb structural variant, two non-coding SNPs, two missense SNPs and the copy number variant of C-terminal 22aa repeat, were genotyped in 45 spring-, 67 autumn-spawning Atlantic herring and 13 Pacific herring. TaqMan Custom SNP assays were performed to genotype the four SNPs in 5 μl reactions with a template of 20 ng genomic DNA (ThermoFisher Scientific). Copy number of the C-terminal 22aa repeat was determined by the PCR product size generated with geno22aa primers. Genotyping of the 5.2 kb structural variant was performed in a PCR reaction containing two forward primers (geno5.2kb-1F and geno5.2kb-2F) and one reverse primer (geno5.2kb-R), which generated PCR products with different sizes between spring and autumn spawners. All the primers used for genotyping are listed in Supplementary Table 1.Tissue expression profiles by quantitative PCRTotal RNA was prepared from gonad, heart, spleen, kidney, gills, intestine, hypothalamus and saccus vasculosus (BSH), and brain without BSH (brain) of six adult spring-spawning Atlantic herring using RNeasy Mini Kit (Qiagen). RNA was then reverse transcribed into cDNA with a High-Capacity cDNA Reverse Transcription Kit (ThermoFisher Scientific). TaqMan Gene Expression assay (ThermoFisher Scientific) containing 0.3 μM primers and 0.25 μM TaqMan probe (Integrated DNA Technologies) was performed to compare the relative expression levels of TSHR among different tissues. qPCR with SYBR Green chemistry was used for TSHB and DIO2 in a 10 μl reaction of SYBR Green PCR Master Mix (ThermoFisher Scientific) and 0.3 μM primers, with a program composed of an initial denaturation for 10 min at 95 °C followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min. Ct values were first normalized to the housekeeping gene ACTIN, then the average expression for each gene in the gonad was assumed to be 1 for the subsequent calculation of the relative expression in other tissues.Plasmid constructsThe coding sequence for the herring single-chain TSH (scTSH) was designed following a strategy previously used for mammalian gonadotropins46 that contained an in-frame fusion of the cDNA sequences (5′–3′) of herring TSH beta subunit (NCBI: XM_012836756.1) and alpha subunit (NCBI: XM_012822755.1) linked by six histidines and then the C-terminal peptide of the hCG beta subunit. The designed sequence should generate a protein with a size of 30.6 kDa. Both scTSH and spring herring TSHR cDNA sequences were synthesized in vitro and cloned in the expression vector pcDNA3.1 by Genscript (Leiden, Netherlands). pcDNA3.1 plasmid expressing human TSHR was kindly provided by Drs. Gilbert Vassart and Sabine Costagliola (Université libre de Bruxelles, Belgium). Then, the spring herring TSHR and human TSHR plasmids were used as templates for site-directed mutagenesis to generate constructs coding for different mutant herring or human TSHRs. Plasmids for the dual-luciferase assay, including pGL4.29[luc2P/CRE/Hygro] containing cAMP response elements (CREs) to drive the transcription of luciferase gene luc2P and pRL-TK monitoring the transfection efficiency, were purchased from Promega. Five ng of each plasmid was used to transform the XL1-Blue competent cells (Agilent), plasmid DNA was subsequently extracted from 200 ml overnight culture of a single transformant clone using an EndoFree plasmid Maxi Kit (Qiagen).Cell cultureChinese hamster ovary (CHO) (ATCC CCL-61) and human embryonic kidney 293 (HEK293) (ATCC CRL-1573) cells were maintained in DMEM supplemented with 5% (CHO) or 10% FBS (HEK293), 100 U/ml penicillin, 100 μg/ml streptomycin and 292 μg/ml l-Glutamine (ThermoFisher Scientific) at 37 °C with 5% CO2. Epithelioma Papulosum Cyprini (EPC) cells (ATCC CRL-2872) were cultured in EMEM (Sigma) supplemented with 10% FBS, 100 U/ml penicillin, 100 μg/ml streptomycin, 292 μg/ml l-Glutamine and 1 mM Sodium Pyruvate (ThermoFisher Scientific) at 26 °C with 5% CO2.Production of recombinant herring scTSHCHO cells were transfected with the scTSH expression plasmid using Lipofectamine 3000 (Invitrogen), stable clones were subsequently selected with 500 μg/ml G418 (Invitrogen) and screened for producing scTSH by western blot using a polyclonal antisera against the sea bass alpha subunit47. A positive clone was expanded in 225 cm2 cell culture flasks (Corning) in culture medium containing 5% FBS until confluence, then the cells were maintained in serum-free DMEM for hormone production for 7 days at 25 °C48. After 7 days, culture medium containing scTSH or without (negative control) was centrifuged at 15000 x g for 15 min and concentrated by ultrafiltration using Centricon Plus-70 / Ultracel PL-30 (Merck Millipore Ltd.). Then, western blotting was performed to confirm TSH production. Concentrated medium containing herring scTSH was denatured at 94 °C for 5 min in 0.1% SDS and 50 mM 2-mercaptoethanol, and then treated with 2.5 units of peptide-N-glycosidase F (Roche Diagnostics) at 37 °C for 2 h in 20 mM sodium phosphate with 0.5% Nonidet P-40, pH 7.5. All samples were run in 12% SDS-PAGE in the reducing condition and transferred to a PVDF membrane (Immobilon P; Millipore Corp.), then blocked overnight with 5% skimmed milk at 4 °C. After blocking, the membrane was incubated with polyclonal antisera against the sea bass alpha subunit (dilution 1:2000) for 90 min at room temperature, washed, and then further incubated with 1:25000 goat anti-rabbit immunoglobulin G (IgG) horseradish peroxidase conjugate (Bio-Rad Laboratories) for 60 min at room temperature. Immunodetection was performed by chemiluminescence with a Pierce ECL Plus Western Blotting Substrate kit (ThermoFisher Scientific).Cell surface expressionA Rhotag (MNGTEGPNFYVPFSNKTGVVYEE) was inserted at the N-terminus of herring TSHR for flow cytometry analysis of receptor cell surface expression. Anti-Rhotag polyclonal antibody was kindly provided by Drs. Gilbert Vassart and Sabine Costagliola (Université Libre de Bruxelles, Brussels, Belgium). PBS containing 1% BSA and 0.05% sodium azide was prepared as the flow cytometry (FCM) buffer for the washing and antibody incubation steps. 2.2 × 106 EPC cells were seeded in a 100 mm poly-d-Lysine-treated petri dish the day before transfection. Each dish was transfected with 10 μg TSHR or empty pcDNA3.1 expression plasmid using 20 μl jetPRIME transfection reagent in 500 μl jetPRIME transfection buffer (Polyplus transfection). Cells were harvested 24 h after transfection, then washed once in cold PBS and fixed in 2% PFA for 10 min at room temperature. After fixation, cells were washed three times with FCM buffer, then incubated with anti-Rhotag antibody or FCM buffer (negative control) for 1 h at room temperature. Cells were washed again with FCM buffer three times and stained with Alexa Fluor 488-labeled chicken anti-mouse IgG (H + L) antibody (1:200 dilution, ThermoFisher Scientific) or FCM buffer (negative control) for 45 min in the dark. After the fluorescent staining, cells were washed three times and resuspended in FCM buffer before analysis on a CytoFLEX instrument (Beckman Coulter). A minimum of 100,000 events was recorded for each sample, fluorescence intensities of negative control and cells transfected with empty pcDNA3.1 plasmid were used as the background for gating strategy. Cell surface expression was represented by the mean fluorescence intensity of the positively stained cell population.Dual-luciferase reporter assayEPC or HEK293 cells were plated in a 48-well plate at a density of 1 × 105 cells/well the day before transfection. A total of 250 ng plasmid mixture containing pGL4.29[luc2P/CRE/Hygro], TSHR expression plasmid (or empty pcDNA3.1) and pRL-TK with the ratio of 20:5:1 was prepared to transfect each well of cells using jetPRIME transfection reagent (Polyplus). Medium was replaced by fresh medium containing 10% FBS (TSH-induced condition) or serum-free medium (constitutive activity condition) 4 h after transfection. On day three, cells were treated with serum-free medium containing different dilutions of the concentrated scTSH medium for 4 h (TSH-induced condition) or directly subjected to the luminescence measurement without TSH induction (constitutive activity condition). Luminescence was measured using a Dual-Luciferase Reporter assay (Promega) on an Infinite M200 Microplate Reader (Tecan Group Ltd., Switzerland), and luciferase activity was represented as the ratio of firefly (pGL4.29[luc2P/CRE/Hygro]) to Renilla (pRL-TK) luminescence.5′-RACE to identify the herring DIO2 TSSTotal RNA was prepared from brain of a spring-spawning Atlantic herring using the RNeasy Mini Kit (Qiagen). Six μg of the isolated RNA was used for 5′-RACE with a FirstChoiceTM RLM-RACE Kit (ThermoFisher Scientific). One μl cDNA or Outer RACE PCR product was used as PCR template in a 20 μL reaction containing 0.8 mM dNTPs, 0.3 μM of each forward and reverse primer (Supplementary Table 1) and 0.75 U PrimeSTAR GXL DNA Polymerase (TaKaRa). Amplification was carried out with an initial denaturation of 3 min at 95 °C, followed by 35 cycles of 98 °C for 10 s, 58 °C for 20 s and 68 °C for 40 s, and a final extension of 10 min at 68 °C. The final 5′ RACE product was sequenced at Eurofins Genomics (Ebersberg, Germany).Sequence conservation analysisGenomic sequences covering the TSHR locus were extracted from Ensembl Genome Browser for Atlantic herring and 11 other fish species, including Amazon molly (Poecilia formosa), denticle herring (Denticeps clupeoides), goldfish (Carassius auratus), guppy (Poecilia reticulata), Neolamprologus brichardi, Japanese medaka (Oryzias latipes), northern pike (Esox lucius), orange clownfish (Amphiprion percula), spotted gar (Lepisosteus oculatus), three-spined stickleback (Gasterosteus aculeatus) and spotted green pufferfish (Tetraodon nigroviridis). The extracted sequences were firstly aligned using progressiveCactus49,50, and a subsequent alignment was generated using the hal2maf program from halTools51 with Atlantic herring assembly (Ch_v2.0.2)18 as the co-ordinate backbone. This alignment was used for the downstream phastCons score calculation by running phyloFit24 and phastCons25 from the PHAST package with default parameters. Peaks were called by grouping signals with a minimum phastCons score of 0.2 within 500 bp region.Structure modeling of human and herring TSHRsIn order to explore the possible interactions of the variant residues with other receptor interacting proteins and to study intramolecular interactions, we built a structural homology model for the herring TSHR (herrTSHR) complexed with herring TSH and Gs-protein. The TSHR hinge region that harbors the Q370H substitution and the C-terminus containing the 22aa repeat were excluded from the homology model due to the lack of structural templates for these regions. The homology model was constructed by using the following structural templates of evolutionarily related class A GPCRs: (i) the leucine-rich repeat domain (LRRD) complexed with hormone was modeled based on the solved FSHR LRRD – FSH complex structure (Protein Data Bank (PDB) ID: 4AY9)52,53, this part of model included herring TSHR Cys33 – Asn296 and fragments of the hinge region Gln297 – Thr312 and Ser393 – Ile421; (ii) the available structural complex of β2-adrenoreceptor with Gs-protein (PDB ID: 3SN6)54 was used as the template to model the seven-transmembrane helix domain (7TMD) of herring TSHR in the active conformation; (iii) the extracellular loop 2 (ECL2) was built by using the ECL2 of μ-opioid receptor (PDB ID: 6DDE)55. To prepare the template for herring TSHR modeling, the fused T4-lysozyme and bound ligand of β2-adrenoreceptor were deleted, the ECL1 and ECL3 loops were adjusted manually to the loop length of herring TSHR. Due to the lack of third intracellular loop (ICL3) in the β2-adrenoreceptor structure, amino acid residues of herring TSHR ICL3 were manually added to the template. Since herring TSHR does not have the TMH5 proline, which is highly conserved among all class A GPCRs and responsible for the helical kinks and bulges within this region56, we assumed a rather regular (stretched) helix conformation for the herring TSHR TMH5 and therefore replaced the kinked β2-adrenoreceptor TMH5 template with a regular α-helix. Moreover, the ECL2 template was substituted with μOR ECL2 structure because of its higher sequence similarity with herring TSHR in this region. Finally, amino acid residues of this chimeric 7TMD template and FSHR N-terminus were mutated to the corresponding spring herring TSHR residues and sequence of the heterodimeric FSH ligand was substituted by the herring TSH. All homology models were generated by using SYBYL-X 2.0 (Certara, NJ, US). The 7TMD structure was then fused with FSHR N-terminus at position 421. The assembled complex was subsequently optimized by the energy minimization under constrained backbone atoms (the AMBER F99 force field was used), followed by a 2 ns molecular dynamics simulation (MD) of the side chains. The entire TSHR complex was energetically minimized without any constraints until converging at a termination gradient of 0.05 kcal/mol*Å. Next, for autumn herrTSHR modeling, the spring TSHR sequence was substituted with autumn TSHR sequence. For humTSHR, the spring TSHR sequence was substituted with human TSHR, and the herring TSH ligand was replaced by the bovine TSH sequence. Both complex models were energetically minimized until converging at a termination gradient of 0.05 kcal/mol*Å.To investigate the microenvironment around the L471M mutation at TMH2 position 2.51, local short MD’s of 4 ns on Met4712.51 (spring herrTSHR), Leu471 (autumn herrTSHR) or Phe461 (humTSHR) and its surrounding amino acids were performed. During MD simulations, backbone atoms of the entire complexes as well as all side chains, except residues at positions 1.47, 1.51, 1.54, 2.48, 2.52, and 2.55 that form the hydrophobic patch around position 2.51, were constrained.Statistics and reproducibilityResults were presented as the mean + SD (standard deviation) calculated from at least four biological replicates for each experiment, and at least two independent experiments were conducted for each assay. Unpaired two-tailed Student’s t test was performed to calculate the P-values and means were judged as statistically significant when P ≤ 0.05.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More