Philippa Kaur

Birkeland, C. Coral Reefs in the Anthropocene (Springer, 2015).Book 

Google Scholar 
Kleypas, J. A., Mcmanus, J. W. & Meñez, L. A. B. Environmental limits to coral reef development: Where do we draw the line?. Am. Zool. 39(1), 146–159. https://doi.org/10.1093/icb/39.1.146 (1999).Article 

Google Scholar 
Perry, C. T. & Larcombe, P. Marginal and non-reef-building coral environments. Coral Reefs 22, 427–432. https://doi.org/10.1007/s00338-003-0330-5 (2003).Article 

Google Scholar 
Wilkinson, C. R. Global and local threats to coral reef functioning and existence: review and predictions. Mar. Freshw. Res. 50, 867–878. https://doi.org/10.1071/mf99121 (1999).Article 

Google Scholar 
Mies, M. et al. South atlantic coral reefs are major global warming refugia and less susceptible to bleaching. Front. Mar. Sci. 7, 514. https://doi.org/10.3389/fmars.2020.00514 (2020).Article 

Google Scholar 
Leão, Z. M. A. N. et al. Brazilian coral reefsin a period of global change: A synthesis. Braz. J. Oceanogr. 64, 97–116. https://doi.org/10.1590/S1679-875920160916064sp2 (2016).Article 

Google Scholar 
Coker, D. J., Wilson, S. K. & Pratchett, M. S. Importance of live coral habitat for reef fishes. Rev. Fish Biol. Fish. 24, 89–126. https://doi.org/10.1007/s11160-013-9319-5 (2014).Article 

Google Scholar 
Alvarez-Filip, L., Gill, J. A. & Dulvy, N. K. Complex reef architecture supports more small-bodied fishes and longer food chains on Caribbean reefs. Ecosphere 2, 118. https://doi.org/10.1890/ES11-00185.1 (2011).Article 

Google Scholar 
Wilson, S. K., Graham, N. A. J., Pratchett, M. S., Jones, G. P. & Polunin, N. V. C. Multiple disturbances and the global degradation of coral reefs: Are reef fishes at risk or resilient?. Glob. Change Biol. 12, 2220–2234. https://doi.org/10.1111/j.1365-2486.2006.01252.x (2006).ADS 
Article 

Google Scholar 
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, 1264. https://doi.org/10.1038/s41467-019-09238-2 (2019).ADS 
CAS 
Article 
PubMed 
PubMed Central 

Google Scholar 
Bellwood, D. R., Hughes, T. P., Folke, C. & Nystrom, M. Confronting the coral reef crisis. Nature 429, 827–833. https://doi.org/10.1038/nature02691 (2004).ADS 
CAS 
Article 
PubMed 

Google Scholar 
Hughes, T. P. et al. climate change, human impacts, and the resilience of coral reefs. Science 301, 929–933. https://doi.org/10.1126/science.1085046 (2003).ADS 
CAS 
Article 
PubMed 

Google Scholar 
Holbrook, N. J. et al. Keeping pace with marine heatwaves. Nat. Rev. Earth Environ. 1, 482–493. https://doi.org/10.1038/s43017-020-0068-4 (2020).ADS 
Article 

Google Scholar 
Bleuel, J., Pennino, M. G. & Longo, G. O. Coral distribution and bleaching vulnerability areas in Southwestern Atlantic under ocean warming. Sci. Rep. 11, 12833. https://doi.org/10.1038/s41598-021-92202-2 (2021).ADS 
CAS 
Article 
PubMed 
PubMed Central 

Google Scholar 
Fontoura, L. et al. The macroecology of reef fish agonistic behaviour. Ecography 43, 1278–1290. https://doi.org/10.1111/ecog.05079 (2020).Article 

Google Scholar 
Inagaki, K. Y., Pennino, M. G., Floeter, S. R., Hay, M. E. & Longo, G. O. Trophic interactions will expand geographically but be less intense as oceans warm. Glob. Change Biol. 26, 6805–6812. https://doi.org/10.1111/gcb.15346 (2020).ADS 
Article 

Google Scholar 
Longo, G. O., Hay, M. E., Ferreira, C. E. L. & Floeter, S. R. Trophic interactions across 61 degrees of latitude in the Western Atlantic. Glob. Ecol. Biogeogr. 28, 107–117. https://doi.org/10.1111/geb.12806 (2019).Article 

Google Scholar 
Pratchett, M. S. et al. Effects of climate-induced coral bleaching on coral-reef fishes: Ecological and economic consequences. Oceanogr. Mar. Biol. Annu. Rev. 46, 251–296. https://doi.org/10.1201/9781420065756.ch6 (2008).Article 

Google Scholar 
Graham, N. A. J. et al. Lag effects in the impacts of mass coral bleaching on coral reef fish, fisheries, and ecosystems. Conserv. Biol. 21, 1291–1300. https://doi.org/10.1111/j.1523-1739.2007.00754.x (2007).Article 
PubMed 

Google Scholar 
Strona, G. et al. Global tropical reef fish richness could decline by around half if corals are lost. Proc. R. Soc. B 288, 20210274. https://doi.org/10.1098/rspb.2021.0274 (2021).CAS 
Article 
PubMed 
PubMed Central 

Google Scholar 
McClenachan, L. Extinction risk in reef fishes 199–207 (Cambridge University Press, 2015).
Google Scholar 
Power, M. E. et al. Challenges in the quest for keystones. Bioscience 46, 609–620. https://doi.org/10.2307/1312990 (1996).Article 

Google Scholar 
Pereira, P. H. C. et al. The influence of multiple factors upon reef fish abundance and species richness in a tropical coral complex. Ichthyol. Res. 61, 375–384. https://doi.org/10.1007/s10228-014-0409-8 (2014).Article 

Google Scholar 
Coni, E. O. C. et al. An evaluation of the use of branching fire-corals (Millepora spp.) as refuge by reef fish in the Abrolhos Bank, eastern Brazil. Environ. Biol. Fish. 96, 45–55. https://doi.org/10.1007/s10641-012-0021-6 (2013).Article 

Google Scholar 
Graham, N. A. J. et al. Extinction vulnerability of coral reef fishes. Ecol. Lett. 14, 341–348. https://doi.org/10.1111/j.1461-0248.2011.01592.x (2011).Article 
PubMed 
PubMed Central 

Google Scholar 
Cornwell, W. K., Schwilk, D. W. & Ackerly, D. D. A trait-based test for habitat filtering: convex hull volume. Ecology 87, 1465–1471. https://doi.org/10.1890/0012-9658(2006)87[1465:ATTFHF]2.0.CO;2 (2006).Article 
PubMed 

Google Scholar 
Mouillot, D., Graham, N. A. J., Villéger, S., Mason, N. W. H. & Bellwood, D. R. A functional approach reveals community responses to disturbances. Trends Ecol. Evol. 28(3), 167–177. https://doi.org/10.1016/j.tree.2012.10.004 (2013).Article 
PubMed 

Google Scholar 
Pimiento, C. et al. Functional diversity of marine megafauna in the Anthropocene. Sci. Adv. 6, 7650. https://doi.org/10.1126/sciadv.aay7650 (2020).ADS 
Article 

Google Scholar 
Loiola, M. et al. Structure of marginal coral reef assemblages under different turbidity regime. Mar. Environ. Res. 147, 138–148. https://doi.org/10.1016/j.marenvres.2019.03.013 (2019).CAS 
Article 
PubMed 

Google Scholar 
Aued, A. W. et al. Large-scale patterns of benthic marine communities in the Brazilian Province. PLoS ONE 13, e0198452. https://doi.org/10.1371/journal.pone.0198452 (2018).CAS 
Article 
PubMed 
PubMed Central 

Google Scholar 
Leão, Z. M. A. N., Kikuchi, R. K. P. & Testa, V. Corals and Coral Reefs of Brazil 9–52 (Elsevier Publisher, 2003).
Google Scholar 
Pinheiro, H. T. et al. South-western Atlantic reef fishes: Zoogeographical patterns and ecological drivers reveal a secondary biodiversity centre in the Atlantic Ocean. Divers. Distrib. 24, 951–965. https://doi.org/10.1111/ddi.12729 (2018).Article 

Google Scholar 
Floeter, S. R. et al. Atlantic reef fish biogeography and evolution. J. Biogeogr. 35, 22–47. https://doi.org/10.1111/j.1365-2699.2007.01790.x (2008).Article 

Google Scholar 
Cord, I. et al. Brazilian marine biogeography: A multi-taxa approach for outlining sectorization. Mar. Biol. 169(5), 61. https://doi.org/10.1007/s00227-022-04045-8 (2022).Article 

Google Scholar 
Leal, I. C. S., Araújo, M. E. D., Cunha, S. R. D. & Pereira, P. H. C. The influence of fire-coral colony size and agonistic behaviour of territorial damselfish on associated coral reef fish communities. Mar. Environ. Res. 108, 45–54. https://doi.org/10.1016/j.marenvres.2015.04.009 (2015).CAS 
Article 
PubMed 

Google Scholar 
Kéry, M. & Royle, J. A. Applied hierarchical modeling in ecology: Analysis of distribution abundance and species richness in R and BUGS. In Prelude and Static Models Vol. 1 (eds Kéry, M. & Royle, J. A.) (Academic Press, 2016).MATH 

Google Scholar 
Hadj-Hammou, J., Mouillot, D. & Graham, N. A. J. Response and effect traits of coral reef fish. Front. Mar. Sci. https://doi.org/10.3389/fmars.2021.640619 (2021).Article 

Google Scholar 
McLean, M. et al. Trait similarity in reef fish faunas across the world’s oceans. PNAS 118(12), e2012318118. https://doi.org/10.1073/pnas.2012318118 (2021).CAS 
Article 
PubMed 
PubMed Central 

Google Scholar 
Brandl, S. J. et al. Coral reef ecosystem functioning: eight core processes and the role of biodiversity. Front. Ecol. Environ. 17, 445–454. https://doi.org/10.1002/fee.2088 (2019).Article 

Google Scholar 
Eggertsen, L. et al. Seaweed beds support more juvenile reef fish than seagrass beds in a south-western Atlantic tropical seascape. Estuar. Coast. Shelf S. 196, 97–108. https://doi.org/10.1016/j.ecss.2017.06.041 (2017).ADS 
Article 

Google Scholar 
Mouillot, D. et al. Functional over-redundancy and high functional vulnerability in global fish faunas on tropical reefs. PNAS 111, 13757–13762. https://doi.org/10.1073/pnas.1317625111 (2014).ADS 
CAS 
Article 
PubMed 
PubMed Central 

Google Scholar 
Briggs, J. C. Marine Zoogeography (McGraw-Hill, 1974).
Google Scholar 
Garcia, G. S., Dias, M. S. & Longo, G. O. Trade-off between number and length of remote videos for rapid assessments of reef fish assemblages. J. Fish Biol. 99(3), 896–904. https://doi.org/10.1111/jfb.14776 (2021).Article 
PubMed 

Google Scholar 
Quimbayo, J. P. et al. Life-history traits, geographical range, and conservation aspects ofreef fishes from the Atlantic and Eastern Pacific. Ecology 102, e03298. https://doi.org/10.1002/ecy.3298 (2021).Article 
PubMed 

Google Scholar 
Katsanevakis, S. et al. Monitoring marine populations and communities: methods dealing with imperfect detectability. Aquat. Biol. 16, 31–52. https://doi.org/10.3354/ab00426 (2012).Article 

Google Scholar 
Villéger, S., Mason, N. W. H. & Mouillot, D. New multidimensional functional diversity indices for a multifaceted framework in functional ecology. Ecology 89, 2290–2301. https://doi.org/10.1890/07-1206.1 (2008).Article 
PubMed 

Google Scholar 
Maire, E., Grenouillet, G., Brosse, S. & Villéger, S. How many dimensions are needed to accurately assess functional diversity? A pragmatic approach for assessing the quality of functional spaces. Glob. Ecol. Biogeogr. 24, 728–740. https://doi.org/10.1111/geb.12299 (2015).Article 

Google Scholar 
R Core Team. R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2021)Kellner, K. jagsUI: A Wrapper Around ‘rjags’ to Streamline ‘JAGS’ Analyses. R package version 1.5.2. https://CRAN.R-project.org/package=jagsUI (2021)Wickham, H. ggplot2: Elegant Graphics for Data Analysis (Springer-Verlag, 2016).Book 

Google Scholar 
Ferreira, C. E. L., Gonçalves, J. E. A. & Coutinho, R. Community structure of fishes and habitat complexity on a tropical rocky shore. Environ. Biol. Fish. 61, 353–369 (2001).Article 

Google Scholar 
Fulton, C. J. et al. Macroalgal meadow habitats support fish and fisheries in diverse tropical seascapes. Fish Fish. 21, 700–717. https://doi.org/10.1111/faf.12455 (2020).Article 

Google Scholar 
Ferreira, L. C. L. 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, 54. https://doi.org/10.1007/s00227-021-03863-6 (2021).CAS 
Article 

Google Scholar 
Lonzetti, B. C., Vieira, E. A. & Longo, G. O. Ocean warming can help zoanthids outcompete branching hydrocorals. Coral Reefs 41, 175–189. https://doi.org/10.1007/s00338-021-02212-9 (2022).Article 

Google Scholar 
Grillo, A. C., Candido, C. F., Giglio, V. J. & Longo, G. O. Unusual high coral cover in a Southwestern Atlantic subtropical reef. Mar. Biodivers. 51, 77. https://doi.org/10.1007/s12526-021-01221-9 (2021).Article 

Google Scholar 
Matheus, Z. et al. Benthic reef assemblages of the Fernando de Noronha Archipelago, tropical South-west Atlantic: Effects of depth, wave exposure and cross-shelf positioning. PLoS ONE 14(1), e0210664. https://doi.org/10.1371/journal.pone.0210664 (2019).CAS 
Article 
PubMed 
PubMed Central 

Google Scholar 
Meirelles, P. M. et al. Baseline assessment of mesophotic reefs of the vitória-trindade seamount chain based on water quality, microbial diversity, benthic cover and fish biomass data. PLoS ONE 10(6), e0130084. https://doi.org/10.1371/journal.pone.0130084 (2015).CAS 
Article 
PubMed 
PubMed Central 

Google Scholar 
Ferreira, C. E. L., Floeter, S. R., Gasparini, J. L., Ferreira, B. P. & Joyeux, J. C. Trophic structure patterns of Brazilian reef fishes: A latitudinal comparison. J. Biogeogr. 31, 1093–1106. https://doi.org/10.1111/j.1365-2699.2004.01044.x (2004).Article 

Google Scholar 
Fontoura, L. et al. Climate-driven shift in coral morphological structure predicts decline of juvenile reef fishes. Glob. Change Biol. 26, 557–567. https://doi.org/10.1111/gcb.14911 (2020).ADS 
Article 

Google Scholar 
MacNeil, M. A. et al. Accounting for detectability in reef-fish biodiversity estimates. Mar. Ecol.-Prog. Ser. 367, 249–260. https://doi.org/10.3354/meps07580 (2008).ADS 
Article 

Google Scholar 
Capitani, L., de Araujo, J. N., Vieira, E. A., Angelini, R. & Longo, G. O. Ocean warming will reduce standing biomass in a Tropical Western Atlantic reef ecosystem. Ecosystems 25, 843–857. https://doi.org/10.1007/s10021-021-00691-z (2022).Article 

Google Scholar 
Fogliarini, C. O., Longo, G. O., Francini-Filho, R. B., McClenachan, L. & Bender, M. G. Sailing into the past: Nautical charts reveal changes over 160 years in the largest reef complex in the South Atlantic Ocean. PECON 20(3), 231–239. https://doi.org/10.1007/10.1016/j.pecon.2022.05.003 (2022).Article 

Google Scholar 
Gasparini, J. L., Floeter, S. R., Ferreira, C. E. L. & Sazima, I. Marine ornamental trade in Brazil. Biodivers. Conserv. 14, 2883–2899. https://doi.org/10.1007/s10531-004-0222-1 (2005).Article 

Google Scholar 
Francini-Filho, R. B. et al. Brazil 163–198 (Springer, 2019).
Google Scholar 
Bellwood, D. R., Goatley, C. H. R. & Bellwood, O. The evolution of fishes and corals on reefs: Form, function and interdependence. Biol. Rev. 92, 878–901. https://doi.org/10.1111/brv.12259 (2017).Article 
PubMed 

Google Scholar 
Nunes, L. T. et al. Ecology of Prognathodes obliquus, a butterflyfish endemic to mesophotic ecosystems of St. Peter and St. Paul’s Archipelago. Coral Reefs 38, 955–960. https://doi.org/10.1007/s00338-019-01822-8 (2019).ADS 
Article 

Google Scholar 
Liedke, A. et al. Abundance, diet, foraging and nutritional condition of the banded butterflyfish (Chaetodon striatus) along the western Atlantic. Mar. Biol. 163, 6. https://doi.org/10.1007/s00227-015-2788-4 (2016).CAS 
Article 

Google Scholar  More

Animals, housing, and treatmentsAll procedures involving animals were approved by the University of Florida Institutional Animal Care and Use Committee (Protocol #201709925). All methods were performance in accordance with the relevant guidelines and regulations, and permission and informed consent was obtained from the University of Florida (owners) for the use of the steers in this experiment.The experiment was carried out during July and August of 2017 at the Feed Efficiency Facility of University of Florida, North Florida Research and Education Center, located in Marianna, Florida (30°52′N, 85°11″W, 35 m asl). Both ‘Argentine’ bahiagrass and ‘Florigraze’ rhizoma peanut hays were obtained from commercial producers. The hay bales were stored in enclosed barns throughout the duration of the experiment.Twenty-five Brahman × Angus crossbred steers (Bos sp.) were utilized (average BW = 341 ± 17 kg, approx. 16 months of age). The steers were grazing bermudagrass (Cynodon dactylon) pastures, a C4 grass, prior to the start of the study. The day prior to the start of the experiment (e.g. day-1), steers brought to working facilities, where they remained 16 h off feed and water, in order to obtain shrunk bodyweights. On day 0 of the experiment, steers were weighed, blocked by bodyweight, and allocated to five treatments (5 steers per treatment) and housed in grouped pens. Hay intake was recorded utilizing GrowSafe© systems (GrowSafe Systems Ltd., Calgary, AB, Canada), which utilize radio frequency identification to record feed intake by weight change measured to the nearest gram. Water was available ad libitum. Forage treatments were offered ad libitum by providing sufficient hay to maintain full feed troughs throughout each day of the experiment. Treatments were five proportions of ‘Florigraze’ rhizoma peanut hay in ‘Argentine’ bahiagrass hay: (1) 100% bahiagrass hay (0% RP); (2) 25% rhizoma peanut hay + 75% bahiagrass hay (25% RP); (3) 50% rhizoma peanut hay + 50% bahiagrass hay (50% RP); (4) 75% rhizoma peanut hay + 25% bahiagrass hay (75% RP); (5) 100% rhizoma peanut hay (100% RP). Diet chemical composition is presented in Table 1. All treatment proportions were weighed and mixed on as-fed basis. Mixing of diets was done manually; no hay mixers or choppers were used, to minimize leaf shatter.Sample collectionSteers were housed for 32 days and sampling occurred on 0, 8, 16, 24, and 32 days after initiation of treatment diets; exception was for feces, which were collected on d-1 given steers were fasted on d-0 of the experiment. The hay mixtures offered to the steers were collected (10 samples of each diet) and analyzed for nutritive value (Table 1), at the start of the experiment. All sampling occurred between 0700 and 1000 h on each of the sampling days.Fecal samples were collected directly from the rectum and placed in quart-sized plastic bags to avoid contamination. The feces were frozen at −20 °C. All fecal samples were thawed, dried at 55 °C for 72 h, and ground to pass a 2-mm stainless steel screen using a Wiley Mill (Model 4, Thomas-Wiley Laboratory Mill, Thomas Scientific, Swedesboro, NJ, USA). Samples were then ball milled using a Mixer Mill MM400 (Retsch GmbH, Haan, Germany) at 25 Hz for 9 min.Blood was obtained through jugular venipuncture using 10-mL K2 EDTA vials (Becton Dickinson and Company, Franklin Lakes, NJ, USA), and stored in ice until centrifugation. All tubes were centrifuged at 714 G for 20 min using an Allegra X-22R centrifuge (Beckman Coulter, Brea, CA, USA). A 10-mL sample of plasma was collected and placed in a separate glass vial, the remaining plasma, white blood cell, and platelet fractions were discarded. The remaining RBC pellet was re-suspended with 9 vol. 0.9% NaCl solution and mixed at room temperature for 15 min at 2 Hz orbital shaker. The tubes were then centrifuged at 714 G for 20 min. The saline solution from the centrifuged tubes was discarded after centrifugation. The rinse procedure was repeated two more times for a total of three rinses. After the third rinse procedure, a 500-µL sample was removed, frozen at −20 °C, and subsequently freeze-dried for isotopic analyses.Hair clippings were obtained from an area of 20 × 20 cm on the left hindquarter, utilizing veterinary hair clippers (Sunbeam-Oster Inc., Boca Raton, FL, USA). Hair clippings were collected, placed in nylon bags (Ankom Technology, Macedon, NY, USA), and frozen for subsequent analysis. Clippings were always collected in the same location from each animal in order to ensure new hair growth would be analyzed. All hair samples were cleaned using soapy water and defatted following protocols for other keratin-based tissues 31,34. Each sample was sonicated twice for 30 min in a methanol and chloroform solution (2:1, v/v), rinsed with distilled water, and oven dried overnight at 60 °C. Each hair sample was ball milled using a Mixer Mill MM400 (Retsch GmbH, Haan, Germany) at 25 Hz for 9 min.CalculationsAfter processing, all samples were analyzed for total C and N using a CHNS analyzer through the Dumas dry combustion method (Vario MicroCube, Elementar Americas Inc., Ronkonkoma, NY, USA) coupled to an isotope ratio mass spectrometer (IsoPrime 100, Elementar, Elementar Americas Inc., Ronkonkoma, NY, USA). Certified standards of L-glutamic acid (USGS40, USGS41; United States Geological Survey) were used for isotope ratio mass spectrometer calibration. Isotope ratios were as follows: δ13C of −26.39, + 37.63‰, and δ15N of −4.52, 47.57‰ for USGS40 and USGS41, respectively. Calibration of the IRMS was conducted according to Cook, et al. 35, with an accuracy of ≤ 0.06 ‰ for 15N and ≤ 0.08 ‰ for 13C.The isotope ratio for 13C/12C was calculated as:$$delta^{{{13}}} {text{C}} = , left( {^{{{13}}} {text{C}}/^{{{12}}} {text{C}}_{{{text{sample}}}} {-}^{{{13}}} {text{C}}/^{{{12}}} {text{C}}_{{{text{reference}}}} } right)/ , left( {^{{{13}}} {text{C}}/^{{{12}}} {text{C}}_{{{text{reference}}}} times { 1}000} right)$$
(1)

where δ13C is the C isotope ratio of the sample relative to Pee Dee Belemnite (PDB) standard (‰), 13C/12Csample is the C isotope ratio of the sample, and 13C/12Creference is the C isotope ratio of PDB standard 5. The isotope ratio for 15N/14N was calculated as:$$delta^{{{15}}} {text{N}} = , left( {^{{{15}}} {text{N}}/^{{{14}}} {text{N}}_{{{text{sample}}}} -^{{{15}}} {text{N}}/^{{{14}}} {text{N}}_{{{text{reference}}}} } right)/left( {^{{{15}}} {text{N}}/^{{{14}}} {text{N}}_{{{text{reference}}}} times { 1}000} right)$$
(2)
where δ15N is the N isotope ratio of the sample relative to atmospheric nitrogen (‰), 15N/14Nsample is the N isotope ratio of the sample, and 15N/14Nreference is the N isotope ratio of atmospheric N (standard) 5. The fraction factor (Δ) in this study was considered to be the difference between the diet isotopic composition (δ13C or δ15N) and that of the given sample 5.The dietary proportion of rhizoma peanut hay was back-calculated using δ13C and δ15N of each plant on a DM basis 3. This method is advantageous in that it does not require further tissue processing and facilitates implementation at the field scale. The proportion of rhizoma peanut was estimated using Eq. (3), described by Jones et al. 3:$$%RP=100-left{100 times frac{A-C}{B-C}right}$$
(3)
where %RP is the proportion of RP in the diet, A is the δ13C or δ15N of the sample obtained, B is the δ13C or δ15N of bahiagrass, and C is the δ13C or δ15N of RP.Statistical analysisAll response variables were analyzed using linear mixed model procedures as implemented in SAS PROC GLIMMIX (SAS/STAT 15.1, SAS Institute). Individual animals were considered the experimental unit. Treatment, collection day, and their interaction were considered fixed effects, and block was considered a random effect in the model. The data were analyzed as repeated measures, considering collection day as the repeated measure. The best covariance matrix was selected according to the lowest AICC fit statistic. Least squares treatment means were compared through pairwise t test using the PDIFF option of the LSMEAN statement in the aforementioned procedure. Based on the recommendations by Milliken and Johnson 36 and Saville 37, no adjustment for multiple comparisons was made. Orthogonal polynomial contrasts (linear and quadratic effects) were used to test effects of RP inclusion on isotopic responses. The slice option was used when the treatment × collection day interaction was significant (P ≤ 0.05), using collection day as the factor, to test treatment effects across collection days. Significance was declared at P ≤ 0.05. The contrast statement was used to test for linear or quadratic effects. Regression analyses for the dietary predictions were conducted using PROC REG from SAS.Predictions of dietary proportions based on Eq. (3) were generated for 16 subgroups reflecting combinations of isotope (13C or 15N), day (8 or 32), and sample type (feces, plasm, RBC, or hair). Analyses comparing predicted versus actual diet proportions included several components. First, we computed the concordance correlation coefficient (CCC) following the recommendations from Crawford, et al. 38. The CCC is a measure of agreement that encompasses both precision and accuracy, along with corresponding 95% bias accelerated and corrected (BCa) bootstrap confidence intervals. For comparative purposes we calculated the Pearson correlation coefficient which only reflects precision. Both correlation coefficients range from −1.0 to 1.0 and we interpreted values ≥ 0.80 as indicating strong positive agreement/correlation. Next, we regressed the actual percentages on the predicted percentages using linear regression. Perfect prediction corresponds to the estimated regression line having an intercept of zero and a slope of 1.0. We then partitioned the mean square error (MSE) of the predicted (from Eq. (3), not the above linear regression) versus actual percentages as described in Rice and Cochran 39. This partitioning entails calculating the proportion of MSE attributable to three sources of error: the difference in mean predicted and actual values (mean component, denoted “MC”), the error resulting from the slope of the above linear regression deviating from 1.0 (slope component, denoted “SC”), and random error (random component, denoted “RC”). The results from the above analyses were examined to identify subgroups whose predictions were sufficiently good to warrant hypothesis testing. In this context “good” means that the predicted percentages were strongly correlated with the actual percentages and the magnitudes of the predicted percentages were similar to the actual percentages. The objective of the hypothesis testing was to formally evaluate whether dietary proportions could be directly predicted from Eq. (3) (in contrast to generating predictions using the equation from regressing actual dietary percentages on the predicted percentages from Eq. (3)). Paired two one-sided test (TOST) equivalence tests were conducted for the selected subgroups with α = 0.0540. These tests are formulated such that the null hypothesis is “non-equivalence” and the alternative hypothesis is “equivalence”. An input parameter to the test is the equivalence region, a range for which we consider the mean actual minus predicted difference to be unimportant (“equivalent”) from a practical standpoint. For each equivalence test we also computed the 90% confidence interval for the mean actual minus predicted difference which we refer to as the “minimum equivalence region”. Based on the structure of TOST equivalence tests, to reject the null hypothesis at the 0.05 level, the equivalence region specified for the test must completely contain the minimum equivalence region. For example, if the pre-specified equivalence region is (−15%, 15%) and the computed minimum equivalence region is (−16%, −6%) the null hypothesis would not be rejected. Finally, we re-ran all of the analyses described above for the selected subgroups where, prior to analysis, predicted percentages outside of the valid range were assigned the appropriate boundary value (i.e., predicted percentages  100% were assigned a value of 100%). More

Global climate change combined with regional and local anthropic activities suggest an increase in recurrence and extent of wildfires on ecosystems worldwide31,47,48, affecting in particular regions with higher likelihood of fire occurrences31 and making natural systems more prone to fire occurrences21. Estimates of accumulated burned area in Brazil between 1985–2020 revealed that, among the Brazilian biomes, the Pantanal is the most affected by the fires (with accumulated burned area equivalent to 57.5% of the biome within Brazil)46. But 43% of 2020 burned area (≈13% of the Pantanal) had not burned since 200319. Therefore, it is impressive that nearly 1/3 of the Pantanal burned in a single year17,18,19 (Figs. 1, 2 and S1, S2). The high number of human-induced fires17,18,19,21 combined with the hottest and driest conditions since 198017,22,38,49 led 2020 to record the highest daily severity rating (DSR) index of fires for this time period17,49. With documented increase of 2 °C in the average temperature22 and a 40% shortage in rainfall26,38. But the fire risk got even higher with the simultaneous occurrence of dry and hot spells, between August and November, when the maximum temperature reached, on average, 6 °C above the normal, accounting for 55% of the burned area of 202049.Most fires started close to the agriculture frontiers21, but they predominantly affected the natural vegetation (reaching between 91–95% of it in occurrence of fire50,51 and 96% of it in estimated burned area)31,46, with tragic consequences for jaguars and the Pantanal biota17,19,26. Along with the fires, the severity of the 2020 drought22,52,53 dropped minimum river depths at around 86% below normal25,54 (Fig. 2 and S1, S3, S4). Consequently, resulting in several records of animal starvation, dehydration, and death17,19,26. And late mortality from indirect causes of fires certainly increased these numbers26. Besides, post-fire ecosystem and hydrology changes also had ecological effects with long-term impacts on ecosystem recovery and fire risk31, impacting resource quality, availability, and productivity26,31. Vegetation productivity declined below −1.5 σ over more than 30% of the natural areas and evaporation decreased (by ~ 9%)31. Burned vegetation made the soil more vulnerable to erosion, increasing the runoff (by ~ 5%) over the natural areas31, and the resulting charcoal and ash contaminated rivers17.Many reasons may have contributed to the intensity of the 2020 drought in the Pantanal, from climate8,22,24,49 to direct and indirect human impacts in the Upper Paraguay River Basin (UPRB)21,55,56. In fact, anthropic changes in land use also increased the biome sensitivity to fire-climate extremes)31. The shortage of rain throughout the UPRB, particularly in the summer season, is among the main factors, as the basin water balance controls the hydroclimatological dynamics in the Pantanal (Fig. 2 and S3–S9)22. The shortage of rain may also be a consequence of increased deforestation in the Amazon rainforest57,58, as summer rainfall in the Pantanal is strongly associated with the climate of the Amazon59. Furthermore, the reduction in wetland flooded areas is historically correlated with the spread of fires (Fig. 2 and S1)22,28,29. Low water levels led to the absence of flooding and reduced wetland areas, and the remaining dry vegetation provided flammable material and created favourable conditions for fires to occur22,23,24. In addition, the lack of governmental and human resources and delayed response at federal and local levels58,60,61 amplified the negative effects of water shortage17,19,58.Although historical hydrological series show that extreme drought events occurred in the past22,25,38,62 (e.g., from the late 1960s to early 1970s, Fig. S3), they also show that the recovery of the Pantanal was conditioned to the subsequent 15 years of regular to exceptional floods (1974 to early 1990s, Figs. S1, S3). Savanna-like vegetation, the predominant vegetation type in the Pantanal, usually recovers from the effects of fires in relatively short periods (months to a few years)23, depending on the severity and frequency of fires and climate conditions in the subsequent years23,28,29. But the resilience of many species may decrease with the annual repetition of extreme fire events28,29,30. Thus, human interventions to prevent (instead to promote) sequential fire events in the same area are paramount19,23,62,63.Estimating the effects that uncontrolled extensive fires can cause to the apex predator of the Neotropics in a region considered one of the strongholds for the species can contribute to the conservation of jaguar and other wildlife species, as well as to the debate regarding potential cumulative impact of recurrent wildfires on ecosystems26,31,51,62,63. Our results revealed the drastic impact of fire on estimated numbers of jaguars, home ranges, and priority areas for jaguar conservation in the Pantanal was exceptionally high in 2020 and proportionally more severe than the nominal 31% of burned area across the Pantanal (e.g., fires affected 45% of the jaguars and 79% of their HRs). Moreover, the annual comparison showed that 2019 was the second-worst year regarding fire impacts (only behind 2020) and equally extreme compared to historical means22. Although the Pantanal is well known for its annual and pluri-annual cycles of wet and dry seasons7,64, the unusual levels of droughts22,25,65,66 and fires17,20,21 in subsequent years are alarming. Furthermore, climate assessment and projections of warmer and dryer conditions for the region in the coming years are equally worrying22,24,37,38.We found that 45% of the jaguar population estimated for the Pantanal occupied areas affected by the 2020 fires (Fig. 1). This finding suggests that the fires heavily impacted the jaguars in the Pantanal, even if we assume that the major effects were only temporary displacement. This potential displacement may make it more difficult for jaguars to find new suitable areas, thus increasing territorial disputes and decreasing survival and reproductive success. Furthermore, 2019 ranked as the second-highest year of impact of fire on jaguar population estimates among the 16 years considered (Table 1, Fig. 1). Importantly, we did not consider cumulative impacts on sequential years or fire recurrence in these estimates. Moreover, the available estimates for jaguar abundance we used36 are very conservative and probably underestimated jaguar populations from the Pantanal by a maximum of 3 jaguars/100 km2. However, the reported density of jaguars may reach up to 12.4 jaguars/100 km2 in northern PAs5,67,68 and up to 6.5–7 jaguars/100 km2 in the southern Pantanal farms5,69,70. Considering that PAs in the northern Pantanal were severely damaged by the 2020 fires, our results show conservative figures for the actual number of jaguars affected by fires.We used densities estimated from an ecosystem-wide assessment of impacts as a proxy of the proportion of total population reached by fire each year on a regional scale. Fires affected a substantial proportion of estimated individuals in the Pantanal in 2019–2020. In 2020, for instance, 87% of all jaguars affected by fire were in the Brazilian Pantanal. In contrast, the smaller population in the Paraguayan and Bolivian Pantanal had a higher median percentage of individuals affected by fire between 2005–2019. While 45% of jaguars were affected by fire in a single year (2020) in the Pantanal, a study45 using the same conservative estimates36 for jaguar abundance in the Brazilian Amazon revealed that 1.8% of the population (1422 individuals) was killed or displaced by fire between 2016–2019. Another report estimated that more than 500 individuals were affected by the 2019 fires in the Brazilian and Bolivian Amazon71,72. Based on the same density estimates we found that in the Pantanal — a much smaller biome — more jaguars were affected by fire in single years (n = 513 in 2019 and n = 746 in 2020). This recent increase in the number of jaguars affected by fire raises a red flag to the supposed stability of the species in the Pantanal, which is currently globally and locally classified as Near Threatened1,5. Therefore, we recommend that future assessments by IUCN specialists carefully consider the frequency and intensity of fires as a potentially significant and growing threat to jaguars in the Pantanal, and their effects on long-term populational trends.Quantifying the occurrence of fire on HRs introduced a functional perspective to understanding the impact of fire on individual jaguars. Similarly, our estimates of the number of affected jaguars revealed a vast amount and extent of affected HRs in the last two years (Figs. 2 and 3). Jaguars are apex predators, often considered as a keystone73,74,75,76 and umbrella species45,77, highly dependent on large habitat areas78, dense native vegetation cover35,79,80, and abundance of prey67,81. Considering that jaguars often select areas with high environmental integrity35,68,78,79,80, the higher impact of recent fires on HRs corroborates reports showing the increase of natural areas burned in the Pantanal31,46,50,51. The proportion of burned forests, for instance, was 10 times higher in 2020 than the estimated median between 1985 and 201931. Sadly, it is likely that much of these burned forests in Northern Pantanal included areas pointed as suitable jaguar habitat and of great interest to the creation of additional PAs82.In the Pantanal, HRs are smaller35,83 and population densities are high5,67,68,69,70 because the biome is a highly productive system7,55,67, with an abundance of prey species and quality habitat, thus allowing jaguars to meet their spatial needs using smaller areas35,68,83. Consequently, floodplain jaguars are also usually larger44,84. However, a trend of increasing drought, rising temperatures, and repeated occurrences of exceptional fires would weaken the Pantanal’s resilience22,32. Importantly to note as well that the occurrence and intensity of fires are frequently higher in the dry season, peaking within jaguars HRs in the years with intense fire occurrence in the Pantanal. This apparent higher impact over jaguar habitat agrees with studies pointing out highest damage in PAs17,27 (Fig. S20), natural vegetation and particularly in forested areas in 202031,46,50,51. Recurrent impacts may particularly affect the most sensitive species28,29,30, resulting in a less productive environment32, which ultimately decreases the habitat quality of many species. These effects would likely push jaguars to expand their HRs, which would increase disputes for territories and favour a decrease in body size, consequently decreasing reproductive rates and population size.The extent of protected areas burned is another indicator of how fire can impact biodiversity. Like the HRs, the Pantanal PAs were affected differently in space and time, but the greatest fires occurred in recent years (2019 and 2020). In 2020, fires occurred in 62% of Brazilian PAs — particularly in northern Pantanal — where several portions of PAs overlapping with jaguar HRs were entirely or almost entirely affected by fires (Figs. 1–3). In 2019, however, fires affected the Pantanal PAs in Bolivia, Paraguay and southern Brazil more severely in areas that also overlapped with HRs (Figs. 1–3). Several causes can explain the spread of fires across PAs, including a combination of heat, drought, miscalculated human use of fires, lack of resources and personnel for surveillance and fire control improvement17,18,19,20,21,22,23.The displacement, injuries, and deaths caused by fire to animals within PAs are worrying because these areas are reportedly richer in diversity and biomass85,86 (including higher jaguars densities36,67,87 and are fundamental to safeguarding biodiversity and ensuring the long-term provision of ecosystem services88,89. Protected areas are important to jaguars because they provide larger continuous areas of natural dense vegetation cover (such as forests and shrublands), flooded habitats and limit contact with humans, attributes of great influence in jaguar habitat selection35,78,79,80,82, and particularly important to females90,91. However, although some PAs support up to 12.4 jaguars/100 km2 (e.g., Taiamã Ecological Station – TES)67, the currently availability of Pantanal PAs alone would not support viable jaguar populations for more than 50 years87. Therefore, sustainable management that allows coexistence in private lands is also fundamental for the conservation of jaguars in the Pantanal5,9,10,11. Protected areas of integral protection, such as TES, currently occupy only 5.7% of the Pantanal7 but were the most affected by fires in absolute area (Fig. S20, Table S5)27. The total number of PAs, including the sustainable use ones, corresponds to only 5% of the Brazilian Pantanal (Tables S1–S3)7,92,93,94,95,96 and around 10% of the entire Pantanal7, most of it in Bolivia97. These percentages are much lower than the minimum of 17% recommended in the Aichi goals for terrestrial ecosystems7,56. Furthermore, PAs are also scarce in the Pantanal headwaters (6% of the surrounding Cerrado uplands) (Tables S1–S3, Fig. S19)7,92,93,94,95,96. To make matters worse, PAs were reduced by almost 20% in the Brazilian Pantanal in 2007 and have not been expanded in the Cerrado uplands since 2006 (Tables S1–S3, Fig. S19)93. The relatively small coverage of protected areas in the Pantanal, which serve as refuges, increases the negative effects of fires, as jaguars are likely displaced into sub-optimal habitats. Consequently, jaguars and other species may struggle to find equally resource-rich sites after being displaced from PAs.For the long-term survival of the jaguar, it is essential to implement conservation plans that consider the dispersal and reproduction of the species along the Paraguay River98, increase the network and size of PAs82, and adequately allocate funding and personnel to maintain the PAs. Furthermore, careful implementation of strategies to mitigate the risk of fire18,19,62 and other human impacts outside PAs5,6,7,8,9,10,11,12,13,14,15,16,89,99 are urgent needs for conservation of the Pantanal. In any case, our results highlight that to sustain viable populations of jaguars and other species, conservation plans for the Pantanal must account for fire impact on PAs and other vital areas for biodiversity.Although jaguar HRs often overlap with PAs67,68,87, some individuals may settle in unprotected areas69,70. In our sample, we found that 38 HRs partially overlapped with PAs (Fig. 1) and 10 HRs did not. On the other hand, considering the sum of the HR extents and the total area overlapped with the PAs, we found that 20% of the HR extent matched the PAs. Notably, jaguars coexist with different levels of anthropic pressures outside the PAs4,5,9,10,11,12,13,14,15,16. Jaguar distribution range has been restricted to 63% of the Pantanal5 and even more restricted in the UPRB100. Agriculture expansion, particularly cattle ranching and soybean cultivation (Figs. S17, S18)65, has been identified as the main causes of jaguars’ disappearance or decline due to killing and habitat loss5,9,13.Sustainable use has been advocated as a conservation strategy in the Pantanal, mainly due to the characteristics of the region, where cattle ranching uses as pastures the natural areas restricted by the Pantanal flooding regime since the 17th century7,23. In recent years, ecotourism has also gained great importance55,101,102. However, there are risks in relying on sustainable use as a core strategy for 90% of the biome (95% of Brazilian Pantanal), and exposure to human-induced fires is one of them21,31.Fire is a fundamental factor acting on the dynamics of the Pantanal vegetation23,28,29. However, repeated uncontrolled fires can drastically impact forests and other habitats critical to the jaguars and increase the area for cattle ranching, therefore increasing the risk of livestock depredation and retaliatory hunting11. Thus, the conservation of the jaguar and other animal species in the Pantanal is critically linked to fire management and the use of private lands because the increased fire may extend and aggravate other anthropic impacts (Fig. 4). This work highlights the significant increase in the extent and severity of recent fires in the Pantanal and how these fires have affected jaguars. Further studies that estimate natural habitat recovery and fire recurrence and assess real-time and long-term effects of fire on jaguars and other species are critical to guide fire management and conservation.Fig. 4: Scheme summarizing the main impacts of fires in the Pantanal.The red arrows are intentionally larger and show a feedback loop linking increased negative human impacts, climate change, and drought to increased fires and burned areas, with a consequent negative impact on biodiversity. The blue arrows describe a feedback loop for fire control and impact mitigation. The dashed arrows denote other relevant effects in the biome (e.g., cumulative effects from infrastructure such as hydroelectric power plants, river waterways, water and soil pollution from legal and illegal mining and agriculture, poaching and illegal wildlife trade, opportunistic exploitation of burned areas, as well as natural climate constraints.Full size imageChanges in the climate8,22,24,37,38, landscape and water use in the UPRB over the last four decades7,18,56,65 are cumulative threats that may interfere with water recharge and vegetation resilience in the Pantanal. Global temperatures may increase up to 1.5 °C over the next five years37, in addition to the 2 °C already recorded since 1980. By the end of the 21st century, scientists estimate increases of 5 − 7 °C in the temperature and the frequency of climatic extremes and a 30% reduction in average rainfall8,37,38. Until 2019, pastures covered 15.5% of the Brazilian Pantanal and agriculture about 0.14%25. However, agriculture and pastures occupied 60–65% of the surrounding Cerrado uplands within the UPRB7,55,56, an occupation similar to the adjacent Paraguayan Chaco and Bolivian Chiquitano Forest7,103,104. And future projections estimate a loss of 14,005 km2 of native vegetation from 2018 through 2050105. Consequently, this land occupation impacted the main headwaters of the Pantanal rivers and ultimately the entire Pantanal6,56,106,107. Furthermore, by 2019, 47 hydroelectric power plants were installed or in operation, and another 133 were planned, totalling about 180 potential dam projects in the Brazilian UPRB108. Besides, most of these projected hydropower infrastructures will overlap with the distribution of jaguars, also in the adjacent biomes, impacting negatively the species particularly in Brazil15. These economic and infrastructure activities in the surrounding highlands frequently ignore their cumulative impacts109 and affect the Pantanal in different ways (Fig. 4, S17, S18), including its drainage dynamics and flood pulses, with consequent impacts on drought duration and fire spread17,19,22,23,24(Figs. 1–4, SI). This combination of factors probably intensifies the Pantanal droughts, particularly the periodic sequence of dry years.Therefore, a critical point is how human actions can exacerbate such extreme events7,21,31,55,106,110 and make fire control even more difficult19,23,62 or, on the opposite, contribute to minimize the overall impacts of drought and fires and promote biodiversity conservation19,63 (Fig. 4). Given that the rainfall remained below average in the last wet seasons53 (Figs. S1, S3–S8) and that a severe drought persisted in 2021111, a surveillance protocol for rapid response and programs for fire management, mitigation of human impacts and ecosystem recovery are needed19,23,62,63. If such measures keep lacking, a tragedy similar to the 2020 fires may be repeated in the coming years (Fig. 4). And Pantanal native vegetation may be reduced to only about 62% by 203021. To make matters worse, the government budget allocated for fire control and firefighting for 2021 was reduced to 65.5% of the 2019 budget61 and all funds allocated to the environment were reduced to the lowest level in 20 years61,112, with serious complaints of misuse113, embezzlement114 and wood-smuggling probe115.The extent of the recent wildfire in the Pantanal has signalled that fire is a potential threat to the long-term conservation of the jaguar. Furthermore, fires severely affected other species and human activities17,19,23, demanding an immediate mitigation plan18,19,62. In fact, permanent fire brigades have been established, and an animal rescue centre is under construction in response to the effects of the recent extensive fires in the Pantanal. Although actions are underway at local levels, the warming and drying trend22,24,37,38 is also a combination of global warming8,37 and rapid land-use changes7,18,65 (Figs. S17, S18), with cumulative impacts in the UPRB and Pantanal wetlands (Fig. 4). Therefore, the immediate reduction of deforestation in the Amazon and Pantanal and the establishment of a forest restoration plan in the UPRB are critical. The lack of sufficient mitigatory actions may throw the Pantanal into a perverse vortex (increasing feedback of cumulative negative impacts, (Fig. 4), thus affecting the survival of jaguars and the various species under their umbrella, as well as human welfare. More

Vega, F. E., Rosenquist, E. & Collins, W. Global project needed to tackle coffee crisis. Nature 425, 343 (2003).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Craparo, A. C. W., Van Asten, P. J. A., Läderach, P., Jassogne, L. T. P. & Grab, S. W. Coffea arabica yields decline in Tanzania due to climate change: global implications. Agric. For. Meteorol. 207, 1–10 (2015).ADS 
Article 

Google Scholar 
Davis, A. P. et al. High extinction risk for wild coffee species and implications for coffee sector sustainability. Sci. Adv. 5, eaav3473 (2019).ADS 
PubMed 
PubMed Central 
Article 

Google Scholar 
Davis, A. P., Gole, T. W., Baena, S. & Moat, J. The impact of climate change on indigenous arabica coffee (Coffea arabica): predicting future trends and identifying priorities. PLoS ONE 7, e47981 (2012).ADS 
CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
Davis, A. P., Mieulet, D., Moat, J., Sarmu, D. & Haggar, J. Arabica-like flavour in a heat-tolerant wild coffee species. Nat. Plants 7, 413–418 (2021).CAS 
PubMed 
Article 

Google Scholar 
Moat, J., Gole, T. W. & Davis, A. P. Least concern to endangered: applying climate change projections profoundly influences the extinction risk assessment for wild Arabica coffee. Global Change Biol. 25, 390–403 (2019).ADS 
Article 

Google Scholar 
Moat, J. et al. Resilience potential of the Ethiopian coffee sector under climate change. Nat. Plants 3, 17081 (2017).PubMed 
Article 

Google Scholar 
Kath, J. et al. Not so robust: Robusta coffee production is highly sensitive to temperature. Global Change Biol. 26, 3677–3688 (2020).ADS 
Article 

Google Scholar 
Liu, L. et al. Soil moisture dominates dryness stress on ecosystem production globally. Nat. Commun. 11, 1–9 (2020).ADS 
CAS 

Google Scholar 
Grossiord, C. et al. Plant responses to rising vapor pressure deficit. New Phytol. 226, 1550–1566 (2020).PubMed 
Article 

Google Scholar 
IPCC Climate Change 2022: Impacts, Adaptation, and Vulnerability (eds. Pörtner, H.-O. et al.) (Cambridge Univ. Press, 2022).Burke, M. et al. Higher temperatures increase suicide rates in the United States and Mexico. Nat. Clim. Change 8, 723–729 (2018).ADS 
Article 

Google Scholar 
Burke, M., Hsiang, S. M. & Miguel, E. Global non-linear effect of temperature on economic production. Nature 527, 235–239 (2015).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Duffy, K. A. et al. How close are we to the temperature tipping point of the terrestrial biosphere? Sci. Adv. 7, eaay1052 (2021).ADS 
CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
Scheffer, M. et al. Early-warning signals for critical transitions. Nature 461, 53–59 (2009).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Schneider, S. H. Abrupt non-linear climate change, irreversibility and surprise. Global Environ. Change 14, 245–258 (2004).Article 

Google Scholar 
Lenton, T. M. Early warning of climate tipping points. Nat. Clim. Change 1, 201–209 (2011).ADS 
Article 

Google Scholar 
Lenton, T. M. et al. Climate tipping points—too risky to bet against. Nature. 575, 592–595 (2019).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Lobell, D. B., Bänziger, M., Magorokosho, C. & Vivek, B. Nonlinear heat effects on African maize as evidenced by historical yield trials. Nat. Clim. Change 1, 42–45 (2011).ADS 
Article 

Google Scholar 
Lobell, D. B., Deines, J. M. & Tommaso, S. D. Changes in the drought sensitivity of US maize yields. Nat. Food 1, 729–735 (2020).Article 

Google Scholar 
Lobell, D. B. et al. Greater sensitivity to drought accompanies maize yield increase in the US Midwest. Science 344, 516–519 (2014).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Rigden, A., Mueller, N., Holbrook, N., Pillai, N. & Huybers, P. Combined influence of soil moisture and atmospheric evaporative demand is important for accurately predicting US maize yields. Nat. Food 1, 127–133 (2020).Article 

Google Scholar 
Schlenker, W. & Roberts, M. J. Nonlinear temperature effects indicate severe damages to US crop yields under climate change. Proc. Natl Acad. Sci. USA 106, 15594–15598 (2009).ADS 
CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
McDowell, N. G. et al. Mechanisms of woody-plant mortality under rising drought, CO2 and vapour pressure deficit. Nat. Rev. Earth Environ. 3, 294–308 (2022).ADS 
CAS 
Article 

Google Scholar 
Sinclair, T. R. et al. Limited-transpiration response to high vapor pressure deficit in crop species. Plant Sci. 260, 109–118 (2017).CAS 
PubMed 
Article 

Google Scholar 
López, J., Way, D. A. & Sadok, W. Systemic effects of rising atmospheric vapor pressure deficit on plant physiology and productivity. Global Change Biol. 27, 1704–1720 (2021).ADS 
Article 

Google Scholar 
McDowell, N. G. & Allen, C. D. Darcy’s law predicts widespread forest mortality under climate warming. Nat. Clim. Change 5, 669–672 (2015).ADS 
Article 

Google Scholar 
Abatzoglou, J. T., Dobrowski, S. Z., Parks, S. A. & Hegewisch, K. C. TerraClimate, a high-resolution global dataset of monthly climate and climatic water balance from 1958–2015. Sci. Data 5, 170191 (2018).PubMed 
PubMed Central 
Article 

Google Scholar 
You, L., Wood, S., Wood-Sichra, U. & Wu, W. Generating global crop distribution maps: from census to grid. Agric. Syst. 127, 53–60 (2014).Article 

Google Scholar 
Fong, Y., Huang, Y., Gilbert, P. B. & Permar, S. R. chngpt: threshold regression model estimation and inference. BMC Bioinformatics 18, 1–7 (2017).Article 

Google Scholar 
Qin, Y. et al. Agricultural risks from changing snowmelt. Nat. Clim. Change 10, 459–465 (2020).ADS 
Article 

Google Scholar 
Forster, P. M., Maycock, A. C., McKenna, C. M. & Smith, C. J. Latest climate models confirm need for urgent mitigation. Nat. Clim. Change 10, 7–10 (2020).ADS 
Article 

Google Scholar 
Forster, P. M. et al. Projections of when temperature change will exceed 2 °C above pre-industrial levels. Nat. Clim. Change 10, 407–412 (2011).
Google Scholar 
Joshi, M., Hawkins, E., Sutton, R., Lowe, J. & Frame, D. Projections of when temperature change will exceed 2 °C above pre-industrial levels. Nat. Clim. Change 1, 407–412 (2011).ADS 
Article 

Google Scholar 
IPCC, 2021: Summary for Policymakers. In Climate Change 2021: The Physical Science Basis (eds Masson-Delmotte, V. et al.) (Cambridge Univ. Press, in press).Lobell, D. B. et al. The critical role of extreme heat for maize production in the United States. Nat. Clim. Change 3, 497–501 (2013).ADS 

Google Scholar 
Sinclair, T. R., Hammer, G. L. & Van Oosterom, E. J. Potential yield and water-use efficiency benefits in sorghum from limited maximum transpiration rate. Funct. Plant Biol. 32, 945–952 (2005).PubMed 
Article 

Google Scholar 
Martins, M. Q. et al. Protective response mechanisms to heat stress in interaction with high [CO2] conditions in Coffea spp. Front. Plant Sci. 7, 947 (2016).PubMed 
PubMed Central 
Article 

Google Scholar 
Rodrigues, W. P. et al. Long‐term elevated air [CO2] strengthens photosynthetic functioning and mitigates the impact of supra‐optimal temperatures in tropical Coffea arabica and C. canephora species. Global Change Biol. 22, 415–431 (2016).ADS 
Article 

Google Scholar 
Ghini, R. et al. Coffee growth, pest and yield responses to free-air CO2 enrichment. Clim. Change 132, 307–320 (2015).ADS 
Article 

Google Scholar 
Rakocevic, M. et al. The vegetative growth assists to reproductive responses of Arabic coffee trees in a long-term FACE experiment. Plant Growth Regul. 91, 305–316 (2020).CAS 
Article 

Google Scholar 
Hammer, G. L. et al. Designing crops for adaptation to the drought and high‐temperature risks anticipated in future climates. Crop Sci. 60, 605–621 (2020).Article 

Google Scholar 
Gennari, P., Rosero-Moncayo, J. & Tubiello, F. N. The FAO contribution to monitoring SDGs for food and agriculture. Nat. Plants 5, 1196–1197 (2019).PubMed 
Article 

Google Scholar 
Lesk, C., Rowhani, P. & Ramankutty, N. Influence of extreme weather disasters on global crop production. Nature 529, 84–87 (2016).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Ortiz-Bobea, A., Ault, T. R., Carrillo, C. M., Chambers, R. G. & Lobell, D. B. Anthropogenic climate change has slowed global agricultural productivity growth. Nat. Clim. Change 11, 306–312 (2021).ADS 
Article 

Google Scholar 
Davis, A. P. et al. Hot coffee: the identity, climate profiles, agronomy, and beverage characteristics of Coffea racemosa and C. zanguebariae. Front. Sustain. Food Syst. 5, 740137 (2021).Article 

Google Scholar 
Sarmiento-Soler, A. et al. Disentangling effects of altitude and shade cover on coffee fruit dynamics and vegetative growth in smallholder coffee systems. Agric. Ecosyst. Environ. 326, 107786 (2022).CAS 
Article 

Google Scholar 
Wood, S. N. Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models. J. R. Stat. Soc. B 73, 3–36 (2011).MathSciNet 
MATH 
Article 

Google Scholar 
Barton, K. MuMIn: multi-model inference. R-Forge http://r-forge.r-project.org/projects/mumin/ (2009).R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing https://www.r-project.org/ (2021).Harrison, X. A. et al. A brief introduction to mixed effects modelling and multi-model inference in ecology. PeerJ 6, e4794 (2018).PubMed 
PubMed Central 
Article 

Google Scholar 
Najafi, E., Devineni, N., Khanbilvardi, R. M. & Kogan, F. Understanding the changes in global crop yields through changes in climate and technology. Earths Future 6, 410–427 (2018).ADS 
CAS 
Article 

Google Scholar 
Ovalle-Rivera, O. et al. Assessing the accuracy and robustness of a process-based model for coffee agroforestry systems in Central America. Agrofor. Syst. 94, 2033–2051 (2020).Article 

Google Scholar 
Varma, S. & Simon, R. Bias in error estimation when using cross-validation for model selection. BMC Bioinformatics 7, 1–8 (2006).Article 

Google Scholar 
Yuan, W. et al. Increased atmospheric vapor pressure deficit reduces global vegetation growth. Sci. Adv. 5, eaax1396 (2019).ADS 
CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
Son, H. & Fong, Y. Fast grid search and bootstrap-based inference for continuous two-phase polynomial regression models. Environmetrics 32, e2664 (2021).MathSciNet 
Article 

Google Scholar 
Wintgens, J. N. et al. Coffee: Growing, Processing, Sustainable Production. A Guidebook for Growers, Processors, Traders, and Researchers (Wiley, 2004). More

The pathosystemPlantago lanceolata L. is a perennial monoecious ribwort plantain that reproduces both clonally via the production side rosettes, and sexually via wind pollination. Seeds drop close to the mother plant and usually form a long-term seed bank47. Podospharea plantaginis (Castagne; U. Braun and S. Takamatsu) (Erysiphales, Ascomycota) is an obligate biotrophic powdery mildew that infects only P. lanceolata and requires living host tissue through its life cycle48. It completes its life cycle as localized lesions on host leaves, only the haustorial feeding roots penetrating the leaf tissue to feed nutrients from its host. Infection causes significant stress for host plant and may increase the host mortality31. The interaction between P. lanceolata and P. plantaginis is strain-specific, whereby the same host genotype may be susceptible to some pathogen genotypes while being resistant to others49. The putative resistance mechanism includes two steps. First, resistance occurs when the host plant first recognizes the attacking pathogen and blocks its growth. When the first step fails and infection takes place, the host may mitigate infection development. Both resistance traits vary among host genotypes49.Approximately 4000 P. lanceolata populations form a network covering an area of 50 × 70 km in the Åland Islands, SW of Finland. Disease incidence (0/1) in these populations has been recorded systematically every year in early September since 2001 by approximately 40 field assistants, who record the occurrence of the fungus P. plantaginis in the local P. lanceolata populations30. At this time, disease symptoms are conspicuous as infected plants are covered by white mycelia and conidia. The coverage (m2) of P. lanceolata in the meadows was recorded between 2001 and 2008 and is used as an estimate of host population size. In the field survey two technicians estimate Plantago population size by visually estimating how much ground/other vegetation P. lanceolata foliage covers (m2) in each meadow. The proportion of P. lanceolata plants in each population suffering from drought is also estimated annually in the survey. Data on average rainfall (mm) in July and August was estimated separately for each population using detailed radar-measured rainfall (obtained by Finnish Meteorological Institute) and it was available for years 2001–2008.Host population connectivity (SH)27 for each local population i was computed with the formula that takes into account the area of host coverage (m²) of all host populations surveyed, denoted with (Aj), and their spatial location compared to other host populations. We assume that the distribution of dispersal distances from a location are described by negative exponential distribution. Under this assumption, the following formula (1) quantifies for a focal population i, the effect of all other host populations, taking into account their population sizes and how strongly they are connected through immigration to it:$${S}_{i}^{H}=mathop{sum}limits_{jne i}{{{{{rm{e}}}}}}^{{-alpha d}_{{ij}}}sqrt{{A}_{j}}.$$
(1)
here, dij is the Euclidian distance between populations i and j and 1/α equals the mean dispersal distance, which was set to be two kilometres based on results from a previous study16.The annual survey data has demonstrated that P. plantaginis infects annually 2–16% of all host populations and persists as a highly dynamic metapopulation through extinctions and re-colonizations of local populations16. The number of host populations has remained relatively stable over the study period49. The first visible symptoms of P. plantaginis infection appear in late June as white-greyish lesions consisting of mycelium supporting the dispersal spores (conidia) that are carried by wind to the same or new host plants. Six to eight clonally produced generations follow one another in rapid succession, often leading to local epidemic with substantial proportion of the infected hosts by late summer within the host local population. Podosphaera plantaginis produces resting structures, chasmothecia, that appear towards the end of growing season in August–September31. Between 20% and 90% of the local pathogen populations go extinct during the winter, and thus the recolonization events play an important role in the persistence of the pathogen regionally16.Inoculation assay: Effect of connectivity and disease history on phenotypic disease resistanceHost and pathogen material for the experimentTo examine whether the diversity and level of resistance vary among host populations depending on their degree of connectivity (SH) and disease history, we selected 20 P. lanceolata populations for an inoculation assay. These populations occur in different locations in the host network, and were selected based on their connectivity values (S H of selected populations was 37–110 in isolated and 237–336 in highly connected category, Fig. 1). We did not include host populations in the intermediate connectivity category that was used in the population dynamic analyses in the inoculation assay due to logistic constraints. Podosphaera plantaginis is an obligate biotrophic pathogen that requires living host tissue throughout its life cycle, and obtaining sufficient inoculum for experiments is extremely time and space consuming. In both isolated and highly connected categories, half of the populations (IDs 193, 260, 311, 313, 337, 507, 1821, 1999, 2818 and 5206) were healthy during the study years 2001–2014, while half of the populations (IDs 271, 294, 309, 321, 490, 609, 1553, 1556, 1676 and 1847) were infected by P. plantaginis for several years during the same period. We collected P. lanceolata seeds from randomly selected ten individual plants around the patch area from each host population in August 2014.To acquire inoculum for the assay, we collected the pathogen strains as infected leaves, one leaf from ten plant individuals from four additional host populations (IDs 3301, 4684, 1784, and 3108) in August 2014. None of the pathogen populations were same as the sampled host populations and hence, the strains used in the assay all represent allopatric combinations. Both host and pathogen populations selected for the study were separated by at least two kilometres. The collected leaves supporting infection were placed in Petri dishes on moist filter paper and stored at room temperature until later use.Seeds from ten mother plants from each population were sown in 2:1 mixture of potting soil and sand, and grown in greenhouse conditions at 20 ± 2 °C (day) and 16 ± 2 °C (night) with 16:8 L:D photoperiod. Due to the low germination rate of collected seeds, population 260 (isolated and healthy population) was excluded from the study. Seedlings of ten different mother plants were randomly selected among the germinated plants for each population (n = 190), and grown in individual pots until the plants were eight weeks old.The pathogen strains were purified through three cycles of single colony inoculations and maintained on live, susceptible leaves on Petri dishes in a growth chamber 20 ± 2 °C with 16:8 L:D photoperiod. Every two weeks, the strains were transferred to fresh P. lanceolata leaves. Purified powdery mildew strains (M1–M4), one representing each allopatric population (3301, 4684, 1784 and 3108), were used for the inoculation assay. To produce enough sporulating fungal material, repeated cycles of inoculations were performed before the assay.Inoculation assay quantifying host resistance phenotypesIn order to study how the phenotypic resistance of hosts varies depending on population connectivity and infection history, we scored the resistance of 190 host genotypes, ten individuals from each study populations (n = 19), in an inoculation assay. Here, one detached leaf from each plant was exposed to a single pathogen strain (M1–M4) by brushing spores gently with a fine paintbrush onto the leaf. Leaves were placed on moist filter paper in Petri dishes and kept in a growth chamber at 20 ± 2 with a 16/8D photoperiod. All the inoculations were repeated on two individual Petri plates, leading to 760 host genotype—pathogen genotype combinations and a total of 1520 inoculations (19 populations * 10 plant genotypes * 4 pathogen strains * 2 replicates). We then observed and scored the pathogen infection on day 12 post inoculation, under dissecting microscope. The resulting plant phenotypic response was scored as 0 = susceptible (infection) when mycelium and conidia were observed on the leaf surface, and as 1 = resistance (no infection), when no developing lesions could be detected under a dissecting microscope. A genotype was defined resistant only if both inoculated replicates showed similar response (1), and susceptible if one or both replicates became infected (0).Statistical analysesBayesian spatio-temporal INLA model of the changes in host population sizeTo study how the pathogen infection influences on host population growth, we analyzed the relative change in host population size (m2) (defined as population size (t) − population size (t−1))/population size (t−1)) between consecutive years utilizing data from 2001 to 2008 in response to pathogen presence-absence status at t−1 (Supplementary Table 2). To assess whether this depends on host population connectivity, we estimated the separate effects of pathogen presence/absence in the previous year for connectivity categories—high-, low, and intermediate—that were based on the 0.2 and 0.8 quantiles of the host-connectivity values (Fig. 1A and Supplementary Figs. 1, 2). This allowed us to directly assess and compare the effect of the pathogen on host population growth in the extreme categories between isolated and highly connected host populations which were represented in the sampling for the inoculation study (Fig. 2).As covariates, we included the proportion (0–100%) of dry host plants measured each year within each local population as well as data on the amount of rainfall at the summer months (June, July, and August) obtained from the satellite images, as these were suggested be relevant for this pathosystem in an earlier analysis16. Observations where the change in host population size, or the host population coverage had absolute values larger than their 0.99 quantiles in the whole data, were regarded as outliers and omitted from the analysis. Before the analyses, all the continuous covariates were scaled and centred, and the categorical variables were transformed into binary variables.The relative changes in local host population size between consecutive years was analyzed by a Bayesian spatio-temporal statistical model that simultaneously considers the effects of a set of biologically meaningful predictors. The linear predictor thus consists of two parts (2,3):$$beta {X}_{t}+{z}_{t}{A}_{t}$$
(2)
where (beta) represents the correlation coefficients corresponding to the effects of environmental covariates, ({z}_{t}) corresponds to the spatiotemporal random effect, and ({X}_{t}) and ({A}_{t}) project these to the observation locations. For ({z}_{t}) we assume that the observations from a location in consecutive time points (t−1) and t are described by 1st order autoregressive process:$${z}_{t}=varphi {z}_{t-1}+{w}_{t}$$
(3)
where ({w}_{t}) corresponds to spatially structured zero-mean random noise, for which a Matern covariance function is assumed. Statistical inference then targets jointly the covariate effects (beta), the temporal autocorrelation (varphi), and the hyperparameters describing the spatial autocorrelation in wt. From these the overall variance, as well as spatial range—a distance after which spatial autocorrelation ceases to be significant—can be inferred (Supplementary Fig. 3). For more detailed description of the structure of the statistical model and how to do efficient inference with it using R-INLA, we refer to refs. 16,50.Identification of resistance phenotypesThe phenotype composition of each study population was defined by individual plant responses to the four pathogen strains, where each response could be “susceptible = 0” or “resistant = 1”. For example, a phenotype “1111” refers to a plant resistant to all four pathogen strains. The diversity of distinct resistance phenotypes within populations was estimated using the Shannon diversity index as implemented in the vegan software package51. The Shannon diversity index for all four study groups was then analyzed using a linear model with class predictors population type (well-connected or isolated), infection history (healthy or infected), and their interaction.Analysis of population connectivity and infection history effects on host resistanceTo test whether host population resistance varied depending on connectivity (SH) and infection history, we analyzed the inoculation responses (0 = susceptible, 1=resistant) of each host-pathogen combination by using a logit mixed-effect model in the lme4 package52. The model included the binomial dependent variable (resistance-susceptible; 1/0), and class predictors population type (well-connected or isolated), infection history (healthy or infected), mildew strain (M1, M2, M3, and M4) and their interactions. Plant individual and population were defined as random effects, with plant genotype (sample) hierarchically nested under population. Model fit was assessed using chi-square tests on the log-likelihood values to compare different models and significant interactions, and the best model was selected based on AIC-values. P-values for regression coefficients were obtained by using the car package53. We ran all the analyses in R software54.The metapopulation modelWe model the ecological and co-evolutionary dynamics of host and pathogen metapopulations to understand key features of the experimental system that impact on the qualitative patterns observed. The structure and parameters in our model are therefore not estimated using experimental data, but rather are chosen to cover a range of possibilities (e.g., low vs high transmission rates, variation in trade-off shapes for fitness costs). We construct the metapopulations in two stages to account for relatively well and poorly connected demes. All demes are identical in quality (i.e., no differences in intrinsic birth or death rates between demes) and only differ in their connectivity. Our metapopulation consists of an outer network of 20 demes, equally spaced around the unit square (0.2 units apart), and a 7×7 inner lattice of demes at a minimum distance of 0.2 units from the outer network (Fig. 3A), giving a total of 69 demes. Demes that are separated by a Euclidean distance of at most 0.2 are then connected to each other. This means that populations near the centre of the metapopulation are highly connected, while those on the boundary of the metapopulation are poorly connected. This also has the effect of making connections between well and poorly connected demes assortative (i.e., well/poorly connected demes tend to be connected to well/poorly connected demes). We relax the assumption of assortativity in a second type of network by randomly reassigning connections between demes, while maintaining the same degree distribution. (i.e., the probability of two demes being connected is proportionate to their degree). While well connected demes still have more connections to other well connected demes than to poorly connected demes, they are not more likely to be connected to a well connected deme than by chance based on the degree distribution. In both types of network structure, we classify a deme as well-connected if it is in the top 20% of the degree distribution and poorly connected if it is in the bottom 20%.We model the genetics using a multilocus gene-for-gene framework with haploid host and pathogen genotypes characterized by (L) biallelic loci, where 0 and 1 represent the presence and absence, respectively, of resistance and infectivity alleles. Host genotype (i) and pathogen genotype (j) are represented by binary strings: ({x}_{i}^{1}{x}_{i}^{2}ldots {x}_{i}^{L}) and ({y}_{j}^{1}{y}_{j}^{2}ldots {y}_{j}^{L}). Resistance acts multiplicatively such that the probability of host (i) being infected when challenged by pathogen (j) is ({Q}_{{ij}}={sigma }^{{d}_{{ij}}}), where (sigma) is the reduction in infectivity per effective resistance allele and ({d}_{{ij}}={sum }_{k=1}^{L}{x}_{i}^{k}big(1-{y}_{j}^{k}big)) is the number of effective resistance alleles (i.e., the number of loci where hosts have a resistance allele but pathogens do not have a corresponding infectivity allele). Hosts and pathogens with more resistance or infectivity alleles are assumed to pay higher fitness costs, ({c}_{H}left(iright)) eq. (4) and ({c}_{P}left(jright)) eq. (5) with:$${c}_{H}left(iright)={c}_{H}^{1}left(frac{1-{{{{{rm{e}}}}}}^{frac{{c}_{H}^{2}}{L}{sum }_{k=1}^{L}{x}_{i}^{k}}}{1-{{{{{rm{e}}}}}}^{{c}_{H}^{2}}}right)$$
(4)
and$${c}_{P}left(jright)={c}_{P}^{1}left(frac{1-{{{{{rm{e}}}}}}^{frac{{c}_{P}^{2}}{L}{sum }_{k=1}^{L}{y}_{j}^{k}}}{1-{{{{{rm{e}}}}}}^{{c}_{P}^{2}}}right)$$
(5)
where (0 , < , {c}_{H}^{1},; {c}_{P}^{1},le, 1) control the overall strength of the costs (i.e., the maximum proportional reduction in reproduction (hosts) or transmission rate (pathogens)) and ({c}_{H}^{2},; {c}_{P}^{2}in {{mathbb{R}}}_{ne 0}) control the shape of the trade-off. When ({c}_{H}^{2},; {c}_{P}^{2}, < , 0) the costs decelerate (increasing returns) and when ({c}_{H}^{2},; {c}_{P}^{2}, > , 0) the costs accelerate the costs accelerate (decreasing returns) (Supplementary Fig. 4). This formulation, therefore, allows for a wide-range of trade-off shapes that may occur in nature.The dynamics of the (finite) host and pathogen populations are modelled stochastically using the tau-leap method with a fixed step size of (tau=1). For population (p), the mean host birth rate at time (t) for host (i) (6) is$${B}_{i}^{p}left(tright)=left(aleft(1-{c}_{H}left(iright)right)-q{N}_{p}left(tright)right){S}_{i}^{p}left(tright)$$
(6)
where (a) is the maximum per-capita birth rate, (q) is the strength of density-dependent competition on births, ({N}_{p}left(tright)={S}_{i}^{p}left(tright)+{I}_{icirc }^{p}left(tright)) is the local host population size, ({S}_{i}^{p}left(tright)) and ({I}_{icirc }^{p}left(tright)={sum }_{j=1}^{n}{I}_{{ij}}^{p}left(tright)) are the local sizes of susceptible and infected individuals of genotype (i), and ({I}_{{ij}}^{p}left(tright)) is the local size of hosts of genotype (i) infected by pathogen (j). Host mutations occur at an average rate of ({mu }_{H}) per loci (limited to at most one mutation per time step), so that the mean number of mutations from host type (i) to ({i}^{{prime} }) is ({mu }_{H}{m}_{i{i}^{{prime} }}{B}_{i}^{p}left(tright)), where ({m}_{i{i}^{{prime} }}=1) if genotypes (i) and ({i}^{{prime} }) differ at exactly one locus, and is 0 otherwise.The mean local mortalities for susceptible and infected individuals are (b{S}_{i}^{p}left(tright)) and (left(b+alpha right){I}_{{ij}}^{p}left(tright)), respectively, where (b) is the natural mortality rate and (alpha) is the disease-associated mortality rate. The average number of infected hosts that recover is (gamma {I}_{{ij}}^{p}left(tright)), where (gamma) is the recovery rate.The mean number of new local infections of susceptible host type (i) by pathogen (j) eq. (7) is:$${INF}_{{ij}}^{p}left(tright)=beta left(1-{c}_{P}left(jright)right){Q}_{{ij}}{S}_{i}^{p}left(tright){Y}_{j}^{p}left(tright)$$
(7)
where (beta) is the baseline transmission rate and ({Y}_{j}^{p}left(tright)) is the local number of pathogen propagules following mutation and dispersal. Pathogen mutations occur in a similar manner to host mutations, with mutations from type (j) to ({j}^{{prime} }) occurring at rate ({mu }_{P}{m}_{j{j}^{{prime} }}{I}_{circ j}^{p}left(tright)) where ({mu }_{P}) is the mutation rate per loci (limited to at most one mutation per timestep) and ({I}_{circ j}^{p}left(tright)={sum }_{i=1}^{n}{I}_{{ij}}^{p}left(tright)) is the local number of pathogen (j.) Following mutation, the local number of pathogens of type (j) eq. (8) is:$${W}_{j}^{p}left(tright)={I}_{circ j}^{p}left(tright)left(1-{mu }_{P}Lright)+{mu }_{P}{m}_{j{j}^{{prime} }}{I}_{circ j}^{p}left(tright)$$
(8)
Pathogen dispersal occurs following mutation at a rate of (rho) between connected demes, given by the adjacency matrix ({G}_{{pr}}), with ({G}_{varSigma p}) the total number of connections for deme (p). The mean local number of pathogen propagules following mutation and dispersal eq. (9) is therefore:$${Y}_{j}^{p}left(tright)={W}_{j}^{p}left(tright)left(1-rho {G}_{varSigma p}right)+rho mathop {sum }limits_{r=1}^{{M}_{varSigma }}{G}_{{pr}}{W}_{j}^{r}left(tright)$$
(9)
We focus our parameter sweep on: (i) the structure of the network (assortative or random connections); (ii) the strength (left({c}_{H}^{1},; {c}_{P}^{1}right)) and shape (left({c}_{H}^{2},; {c}_{P}^{2}right)) of the trade-offs; (iii) the transmission rate (left(beta right)); and (iv) the dispersal rate (left(rho right)), fixing the remaining parameters as described in Supplementary Table 1 (preliminary investigations suggested they had less of an impact on the qualitative outcome) and conducting 100 simulations per parameter set. For each simulation we initially seed all populations with the most susceptible host type and place the least infective pathogen type in one of the well-connected populations to minimize the risk of early extinction. We then solve the dynamics for 10,000 time steps (preliminary investigations indicated this was a sufficient period for the metapopulations to reach a quasi-equilibrium in terms of overall resistance). We calculate the average level of resistance (proportion of loci with a resistance allele) between time steps 4001 and 5000 (transient dynamics) and over the final 1000 time steps (long-term dynamics) for well and poorly connected demes, categorized according to whether the disease is present in (infected) or absent from (uninfected) the local population at a given time point and discarding simulations where the pathogen is driven globally extinct.We compare the mean level of resistance in infected/uninfected poorly/well-connected populations across all simulations to the empirical results. We say that a simulation is a qualitative ‘match’ for the empirical findings if: (i) in poorly connected demes, the infected populations are on average at least 5% more resistant than uninfected populations; and (ii) in well-connected demes, the uninfected populations are on average at least 5% more resistant than infected populations. In other words, if ({R}_{{CS}}) is the mean resistance for a population with connectivity (C) ((C=W) and (C=P) for well and poorly connected demes, respectively) and infection status (S) ((S=U) and (S=I) for uninfected and infected populations, respectively), then a parameter set is a qualitative ‘match’ for the empirical findings if ({R}_{{WU}} > 1.05{R}_{{WI}}) and (1.05{R}_{{PI}}, > , 1.05{R}_{{PU}}). If these criteria are not met, then the parameter set is a qualitative ‘mismatch’ for the empirical findings. The model is not intended to be a replica of an empirical metapopulation, but rather is used to reveal the key factors which lead to qualitatively similar distributions of resistance and disease incidences observed in the study of the Åland islands. Hence, the purpose of the model is to determine which biological factors are likely to be crucial to the patterns observed herein.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More

Schmidt, J. Breeding places and migrations of the eel. Nature 111, 51–54 (1923).ADS 
Article 

Google Scholar 
Tucker, D. W. A new solution to the Atlantic eel problem. Nature 183, 495–501 (1959).ADS 
Article 

Google Scholar 
Voorhis, A. D. & Hersey, J. B. Oceanic thermal fronts in the Sargasso Sea. J. Geophys. Res. 69(18), 3809–3814 (1964).ADS 
Article 

Google Scholar 
Kleckner, R. C. & McCleave, J. D. The northern limit of spawning by Atlantic eels (Anguilla spp.) in the Sargasso Sea in relation to thermal fronts and surface water masses. J. Mar. Res. 46, 647–667 (1988).Article 

Google Scholar 
Ullman, D. S., Cornillon, P. C. & Shan, Z. On the characteristics of subtropical fronts in the North Atlantic. J. Geophys. Res: Oceans 112, C01010 (2007).ADS 

Google Scholar 
Miller, M. J. et al. Spawning by the European eel across 2000 km of the Sargasso Sea. Biol. Lett. 15, 20180835 (2019).Article 

Google Scholar 
Westerberg, H. et al. Larval abundance across the European eel spawning area: An analysis of recent and historic data. Fish. 19, 890–902 (2018).
Google Scholar 
Halliwell, G. R. Jr., Olson, D. B. & Peng, G. Stability of the Sargasso Sea subtropical frontal zone. J. Phys. Oceanogr. 24(6), 1166–1183 (1994).ADS 
Article 

Google Scholar 
van Ginneken, V. J. T. & Maes, G. E. The European eel (Anguilla anguilla, Linnaeus), its lifecycle, evolution and reproduction: A literature review. Rev. Fish Biol. Fish. 15, 367–398 (2005).Article 

Google Scholar 
Friedland, K. D., Miller, M. J. & Knights, B. Oceanic changes in the Sargasso Sea and declines in recruitment of the European eel. ICES J. Mar. Sci. 64, 519–530 (2007).Article 

Google Scholar 
Jacoby, D. M. P. et al. Synergistic patterns of threat and the challenges facing global anguillid eel conservation. Glob. Ecol. Conserv. 4, 321–333 (2015).Article 

Google Scholar 
Béguer-Pon, M. et al. Tracking anguillid eels: Five decades of telemetry-based research. Mar. Freshw. Res. 69, 199 (2018).Article 

Google Scholar 
Righton, D. et al. Important questions to progress science and sustainable management of anguillid eels. Fish 22, 762–788 (2021).
Google Scholar 
Aoyama, J. Life history and evolution of migration in catadromous eels (genus Anguilla). Aquat. Bio Sci. Monogr. 2, 1–42 (2009).
Google Scholar 
Tsukamoto, K., Aoyama, J. & Miller, M. J. Migration, speciation, and the evolution of diadromy in anguillid eels. Can. J. Fish. Aquat. Sci. 59, 1989–1998 (2002).Article 

Google Scholar 
Tesch, F.-W. Telemetric observations on the spawning migration of the eel (Anguilla anguilla) west of the European continental shelf. Env. Biol. Fish. 3, 203–209 (1978).Article 

Google Scholar 
Aarestrup, K. et al. Oceanic spawning migration of the European eel (Anguilla anguilla). Science 325, 1660 (2009).ADS 
CAS 
Article 

Google Scholar 
Westerberg, H. et al. Behaviour of stocked and naturally recruited European eels during migration. Mar. Ecol. Prog. Ser. 496, 145–157 (2014).ADS 
Article 

Google Scholar 
Amilhat, E. et al. First evidence of European eels exiting the Mediterranean Sea during their spawning migration. Sci. Rep. 6, 21817 (2016).ADS 
CAS 
Article 

Google Scholar 
Righton, D. et al. Empirical observations of the spawning migration of European eels: The long and dangerous road to the Sargasso Sea. Sci. Adv. 2, e1501694 (2016).ADS 
Article 

Google Scholar 
Verhelst, P. et al. Mapping silver eel migration routes in the North Sea. Sci Rep. 12, 318 (2022).ADS 
CAS 
Article 

Google Scholar 
Kuroki, M. et al. Hatching time and larval growth of Atlantic eels in the Sargasso Sea. Mar. Biol. 164, 118. https://doi.org/10.1007/s00227-017-3150-9 (2017).Article 

Google Scholar 
Acton, L. et al. What is the Sargasso Sea? The problem of fixing space in a fluid ocean. Polit. Geogr. 68, 86–100 (2019).Article 

Google Scholar 
GEBCO Compilation Group. GEBCO 2020 Grid. https://doi.org/10.5285/a29c5465-b138-234d-e053-6c86abc040b9 (2020).Miller, M. J. & Hanel, R. The Sargasso Sea Subtropical Gyre: The spawning and larval development area of both freshwater and marine eels. Sargasso Sea Alliance Science Report Series, 7, 20 pp (2011).Munk, P. et al. Oceanic fronts in the Sargasso Sea control the early life and drift of Atlantic eels. Proc. Biol. Sci. 277, 3593–3599 (2010).PubMed 
PubMed Central 

Google Scholar 
Béguer-Pon, M., Castonguay, M., Shan, S., Benchetrit, J. & Dodson, J. J. Direct observations of American eels migrating across the continental shelf to the Sargasso Sea. Nat. Commun. 6, 8705 (2015).ADS 
Article 

Google Scholar 
Westin, L. The spawning migration of European silver eel (Anguilla anguilla L.) with particular reference to stocked eel in the Baltic. Fish. Res. 38(3), 257–270 (1998).
Article 

Google Scholar 
Tesch, F.-W., Wendt, T. & Karlsson, L. Influence of geomagnetism on the activity and orientation of the eel, Anguilla anguilla (L.), as evident from laboratory experiments. Ecol. Freshw. Fish 1(1), 52–60 (1992).Article 

Google Scholar 
Tesch, F.-W. The Eel (Blackwell Science, Oxford, UK, 2003).Book 

Google Scholar 
Durif, C. M. F. et al. Magnetic compass orientation in the European eel. PLoS ONE 8(3), e59212 (2013).ADS 
CAS 
Article 

Google Scholar 
Schabetsberger, R. et al. Hydrographic features of anguillid spawning areas: Potential signposts for migrating eels. Mar. Ecol. Prog. Ser. 554, 141–155 (2016).ADS 
Article 

Google Scholar 
Naisbett-Jones, L. C., Putman, N. F., Stephenson, J. F., Ladak, S. & Young, K. A. A magnetic map leads juvenile European eels to the Gulf stream. Curr. Biol. 27, 1236–1240 (2017).CAS 
Article 

Google Scholar 
Dekker, W. Status of the European eel stock and fisheries. In Eel Biology (eds Aida, K. et al.) 237–254 (Springer, New York, 2003).Chapter 

Google Scholar 
Drouineau, H. et al. Freshwater eels: A symbol of the effects of global change. Fish Fish (Oxf) 19, 903–930 (2018).Article 

Google Scholar 
ICES. Joint EIFAAC/ICES/GFCM Working Group on Eels (WGEEL). ICES Scientific Reports. 2(85) (2020).Pike, C., Crook, V. & Gollock, M. Anguilla anguilla. The IUCN Red List of Threatened Species 2020: e.T60344A152845178 (2020).Durif, C., Dufour, S. & Elie, P. The silvering process of Anguilla anguilla: A new classification from the yellow resident to the silver migrating stage. J. Fish. Biol. 66, 1025–1043 (2005).Article 

Google Scholar 
Pankhurst, N. W. Relation of visual changes to the onset of sexual maturation in the European eel Anguilla Anguilla (L.). J. Fish Biol. 21, 127–140 (1982).Article 

Google Scholar 
Økland, F., Thorstad, E. B., Westerberg, H., Aarestrup, K. & Metcalfe, J. D. Development and testing of attachment methods for pop-up satellite archival transmitters in European eel. Anim. Biotelem. 1, 3 (2013).Article 

Google Scholar  More

Birks, H. J. B. Strengths and weaknesses of quantitative climate reconstructions based on late-quaternary biological proxies. Open Ecol. J. 3, 68–110 (2011).Article 

Google Scholar 
Chevalier, M. et al. Pollen-based climate reconstruction techniques for late Quaternary studies. Earth-Sci. Rev. 210, 103384 (2020).Article 

Google Scholar 
Bartlein, P. J. et al. Pollen-based continental climate reconstructions at 6 and 21 ka: a global synthesis. Clim. Dyn. 37, 775–802 (2011).Article 

Google Scholar 
Brierley, C. M. et al. Large-scale features and evaluation of the PMIP4-CMIP6 midHolocene simulations. Clim. Past 16, 1847–1872 (2020).Kageyama, M. et al. The PMIP4 Last Glacial Maximum experiments: preliminary results and comparison with the PMIP3 simulations. Clim. Past 17, 1065–1089 (2021).Article 

Google Scholar 
Harrison, S. BIOME 6000 DB classified plotfile version 1. https://doi.org/10.17864/1947.99. (2017).Loarie, S. R. et al. The velocity of climate change. Nature 462, 1052–1055 (2009).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Svenning, J. C. & Sandel, B. Disequilibrium vegetation dynamics under future climate change. Am. J. Bot. 100, 1266–1286 (2013).PubMed 
Article 

Google Scholar 
Neilson, R. P. et al. Forecasting regional to global plant migration in response to climate change. BioScience 55 https://academic.oup.com/bioscience/article/55/9/749/285963 (2005).Normand, S. et al. Postglacial migration supplements climate in determining plant species ranges in Europe. Proc. R. Soc. B: Biol. Sci. 278, 3644–3653 (2011).Article 

Google Scholar 
Seltzer, A. M. et al. Widespread six degrees Celsius cooling on land during the Last Glacial Maximum. Nature 593, 228–232 (2021).Tierney, J. E. et al. Glacial cooling and climate sensitivity revisited. Nature 584, 569 (2020).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Ray, N. & Adams, J. M. A GIS-based Vegetation Map of the World at the Last Glacial Maximum (25,000-15,000 BP). Internet Archaeol. 11, https://doi.org/10.11141/ia.11.2 (2001).Birks, H. J. B. & Willis, K. J. Alpines, trees, and refugia in Europe. Plant Ecol. Divers. 1, 147–160 (2008).Article 

Google Scholar 
Roberts, D. R. & Hamann, A. Glacial refugia and modern genetic diversity of 22 western North American tree species. Proc. R. Soc. B: Biol. Sci. 282, 20142903 (2015).Clark, J. S. Why trees migrate so fast: Confronting theory with dispersal biology and the paleorecord. Am. Nat. 152, 204–224 (1998).CAS 
PubMed 
Article 

Google Scholar 
Jackson, S. & Overpeck, J. Responses of plant populations and communities to environmental changes of the late Quaternary. Paleobiology 26, 194–220 (2000).Article 

Google Scholar 
Williams, J. W., Ordonez, A. & Svenning, J.-C. A unifying framework for studying and managing climate-driven rates of ecological change. Nat. Ecol. Evol. 5, 17–26 (2021).Harrison, S. P. & Goñi, M. F. S. Global patterns of vegetation response to millennial-scale variability and rapid climate change during the last glacial period. Quat. Sci. Rev. 29, 2957–2980 (2010).ADS 
Article 

Google Scholar 
Williams, J. W., Post, D. M., Cwynar, L. C., Lotter, A. F. & Levesque, A. J. Rapid and widespread vegetation responses to past climate change in the North Atlantic region. Geology 30, 971–974 (2002).ADS 
CAS 
Article 

Google Scholar 
Giesecke, T., Brewer, S., Finsinger, W., Leydet, M. & Bradshaw, R. H. W. Patterns and dynamics of European vegetation change over the last 15,000 years. J. Biogeogr. 44, 1441–1456 (2017).Article 

Google Scholar 
Ordonez, A. & Williams, J. W. Climatic and biotic velocities for woody taxa distributions over the last 16 000 years in eastern North America. Ecol. Lett. 16, 773–781 (2013).PubMed 
Article 

Google Scholar 
Svenning, J.-C. & Skov, F. Limited filling of the potential range in European tree species. Ecol. Lett. 7, 565–573 (2004).Article 

Google Scholar 
Talluto, M. V., Boulangeat, I., Vissault, S., Thuiller, W. & Gravel, D. Extinction debt and colonization credit delay range shifts of eastern North American trees. Nat. Ecol. Evol. 1, 1–6 (2017).Article 

Google Scholar 
Pearson, R. G. & Dawson, T. P. Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Glob. Ecol. Biogeogr. 12, 361–371 (2003).Article 

Google Scholar 
Webb, T. Is vegetation in equilibrium with climate? How to interpret late-Quaternary pollen data. Vegetatio 67, 75–91 (1986).Article 

Google Scholar 
Jackson, S. T. & Williams, J. W. Modern analogs in quaternary paleoecology: Here today, gone yesterday, gone tomorrow? Annu. Rev. Earth Planet. Sci. 32, 495–537 (2004).ADS 
CAS 
Article 

Google Scholar 
Cao, X., Tian, F., Dallmeyer, A. & Herzschuh, U. Northern Hemisphere biome changes ( >30°N) since 40 cal ka BP and their driving factors inferred from model-data comparisons. Quat. Sci. Rev. 220, 291–309 (2019).ADS 
Article 

Google Scholar 
He, F. Simulating transient climate evolution of the last deglaciation with CCSM3 Dissertation at the University of Wisconsin – Madison (2011).Osman, M. B. et al. Globally resolved surface temperatures since the Last Glacial Maximum. Nature 599, 239–244 (2021).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Shakun, J. D. et al. Global warming preceded by increasing carbon dioxide concentrations during the last deglaciation. Nature 484, 49–54 (2012).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Alley, R. B. The Younger Dryas cold interval as viewed from central Greenland. in Quaternary Science Reviews vol. 19 213–226 (Pergamon, 2000).He, C. et al. Hydroclimate footprint of pan-Asian monsoon water isotope during the last deglaciation. Sci. Adv. 7, eabe2611 (2021).ADS 
PubMed 
Article 

Google Scholar 
Reick, C. H., Raddatz, T., Brovkin, V. & Gayler, V. Representation of natural and anthropogenic land cover change in MPI-ESM. J. Adv. Modeling Earth Syst. 5, 459–482 (2013).Prentice, I. C., Guiot, J., Huntley, B., Jolly, D. & Cheddadi, R. Reconstructing biomes from palaeoecological data: a general method and its application to European pollen data at 0 and 6 ka. Clim. Dyn. 12, 185–194 (1996).Article 

Google Scholar 
Dallmeyer, A., Claussen, M. & Brovkin, V. Harmonising plant functional type distributions for evaluating Earth system models. Clim 15, 335–366 (2019).
Google Scholar 
Ni, J., Cao, X., Jeltsch, F. & Herzschuh, U. Biome distribution over the last 22,000 yr in China. Palaeogeogr. Palaeoclimatol. Palaeoecol. 409, 33–47 (2014).Article 

Google Scholar 
Williams, J. W. & Jackson, S. T. Novel climates, no-analog communities, and ecological surprises. Front. Ecol. Environ. 5, 475–482 (2007).Article 

Google Scholar 
Sobol, M. K., Scott, L. & Finkelstein, S. A. Reconstructing past biomes states using machine learning and modern pollen assemblages: a case study from Southern Africa. Quat. Sci. Rev. 212, 1–17 (2019).ADS 
Article 

Google Scholar 
Marinova, E. et al. Pollen‐derived biomes in the Eastern Mediterranean–Black Sea–Caspian‐Corridor. J. Biogeogr. 45, 484–499 (2018).Article 

Google Scholar 
Cao, X. et al. Pollen-based quantitative land-cover reconstruction for northern Asia covering the last 40 ka cal BP. Clim. Past 15, 1503–1536 (2019).Article 

Google Scholar 
Geng, R. et al. Modern pollen assemblages from lake sediments and soil in East Siberia and relative pollen productivity estimates for major taxa. Front. Ecol. Evol. 10, 508 (2022).Article 

Google Scholar 
Cao, X. et al. A taxonomically harmonized and temporally standardized fossil pollen dataset from Siberia covering the last 40 kyr. Earth Syst. Sci. Data 12, 119–135 (2020).ADS 
Article 

Google Scholar 
Sugita, S. Theory of quantitative reconstruction of vegetation I: pollen from large sites REVEALS regional vegetation composition. Holocene 17, 229–241 (2007).ADS 
Article 

Google Scholar 
Githumbi, E. et al. European pollen-based REVEALS land-cover reconstructions for the Holocene: Methodology, mapping and potentials. Earth Syst. Sci. Data 14, 1581–1619 (2022).ADS 
Article 

Google Scholar 
Snell, R. S. et al. Using dynamic vegetation models to simulate plant range shifts. Ecography 37, 1184–1197 (2014).Article 

Google Scholar 
Bullock, J. M. et al. Modelling spread of British wind-dispersed plants under future wind speeds in a changing climate. J. Ecol. 100, 104–115 (2012).Article 

Google Scholar 
Svenning, J. C., Normand, S. & Skov, F. Postglacial dispersal limitation of widespread forest plant species in nemoral Europe. Ecography 31, 316–326 (2008).Article 

Google Scholar 
Herzschuh, U. et al. Glacial legacies on interglacial vegetation at the Pliocene-Pleistocene transition in NE Asia. Nat. Commun. 7, 1–11 (2016).Article 

Google Scholar 
Herzschuh, U. Legacy of the Last Glacial on the present‐day distribution of deciduous versus evergreen boreal forests. Glob. Ecol. Biogeogr. 29, 198–206 (2020).Article 

Google Scholar 
Väliranta, M. et al. Plant macrofossil evidence for an early onset of the Holocene summer thermal maximum in northernmost Europe. Nat. Commun. 6, 1–8 (2015).Article 

Google Scholar 
Schulte, L., Li, C., Livsovski, S. & Herzschuh, U. Forest-permafrost feedbacks and glacial refugia help explain the unequal distribution of larch across continents. J. Biogeogr. 9, 0305–0270 (2022).
Google Scholar 
Davis, M. B., Shaw, R. G. & Etterson, J. R. Evolutionary responses to changing climate. Ecology 86, 1704–1714 (2005).Article 

Google Scholar 
Urban, M. C., Tewksbury, J. J. & Sheldon, K. S. On a collision course: competition and dispersal differences create no-analogue communities and cause extinctions during climate change. Proc. R. Soc. B Biol. Sci. 279, 2072–2080 (2012).Article 

Google Scholar 
Pennington, W. Lags in adjustment of vegetation to climate caused by the pace of soil development. Evidence from Britain. Vegetatio 67, 105–118 (1986).Article 

Google Scholar 
MacDonald, G. M., Kremenetski, K. V. & Beilman, D. W. Climate change and the northern Russian treeline zone. Philos. Trans. R. Soc. B: Biol. Sci. 363, 2285–2299 (2008).CAS 
Article 

Google Scholar 
Prentice, I. C., Bartlein, P. J. & Webb, T. Vegetation and climate change in eastern North America since the last glacial maximum. Ecology 72, 2038–2056 (1991).Article 

Google Scholar 
Cao, X. Y., Herzschuh, U., Telford, R. J. & Ni, J. A modern pollen-climate dataset from China and Mongolia: Assessing its potential for climate reconstruction. Rev. Palaeobot. Palynol. 211, 87–96 (2014).Article 

Google Scholar 
Leroy, S. A. G., Arpe, K., Mikolajewicz, U. & Wu, J. Climate simulations and pollen data reveal the distribution and connectivity of temperate tree populations in eastern Asia during the Last Glacial Maximum. Clim 16, 2039–2054 (2020).
Google Scholar 
Kaufman, D. et al. A global database of Holocene paleotemperature records. Sci. Data 7, 115 (2020).Mottl, O. et al. Global acceleration in rates of vegetation change over the past 18,000 years. Science 372, 860–864 (2021).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Nolan, C. et al. Past and future global transformation of terrestrial ecosystems under climate change. Science 361, 920–923 (2018).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Mauritsen, T. et al. Developments in the MPI-M Earth System Model version 1.2 (MPI-ESM1.2) and Its Response to Increasing CO2. J. Adv. Model. Earth Syst. 11, 998–1038 (2019).ADS 
PubMed 
PubMed Central 
Article 

Google Scholar 
Reick, C. et al. JSBACH 3—The land component of the MPI Earth System Model: documentation of version 3.2. Hamburg: MPI für Meteorologie. Berichte zur Erdsystemforsch. (2021).Brovkin, V., Raddatz, T., Reick, C. H., Claussen, M. & Gayler, V. Global biogeophysical interactions between forest and climate. Geophys. Res. Lett. 36, L07405 (2009).ADS 
Article 

Google Scholar 
Berger, A. L. Long-term variations of daily insolation and Quaternary climatic changes. J. Atmos. Sci. 35, 2361–2367 (1978).ADS 
Article 

Google Scholar 
Köhler, P., Nehrbass-Ahles, C., Schmitt, J., Stocker, T. F. & Fischer, H. A 156 kyr smoothed history of the atmospheric greenhouse gases CO2, CH4, and N2O and their radiative forcing. Earth Syst. Sci. Data 9, 363–387 (2017).ADS 
Article 

Google Scholar 
Tarasov, L., Dyke, A. S., Neal, R. M. & Peltier, W. R. A data-calibrated distribution of deglacial chronologies for the North American ice complex from glaciological modeling. Earth Planet. Sci. Lett. 315–316, 30–40 (2012).ADS 
Article 

Google Scholar 
Loana Meccia, V. & Mikolajewicz, U. Interactive ocean bathymetry and coastlines for simulating the last deglaciation with the Max Planck Institute Earth System Model (MPI-ESM-v1.2). Geosci. Model Dev. 11, 4677–4692 (2018).ADS 
Article 

Google Scholar 
Riddick, T., Brovkin, V., Hagemann, S. & Mikolajewicz, U. Dynamic hydrological discharge modelling for coupled climate model simulations of the last glacial cycle: the MPI-DynamicHD model version 3.0. Geosci. Model Dev. 11, 4291–4316 (2018).ADS 
Article 

Google Scholar 
Kapsch, M., Mikolajewicz, U., Ziemen, F. & Schannwell, C. Ocean response in transient simulations of the last deglaciation dominated by underlying ice‐sheet reconstruction and method of meltwater distribution. Geophys. Res. Lett. 49, e2021GL096767 (2022).ADS 
Article 

Google Scholar 
Murton, J. B., Bateman, M. D., Dallimore, S. R., Teller, J. T. & Yang, Z. Identification of Younger Dryas outburst flood path from Lake Agassiz to the Arctic Ocean. Nature 464, 740–743 (2010).ADS 
CAS 
PubMed 
Article 

Google Scholar 
Rehfeld, K., Marwan, N., Heitzig, J. & Kurths, J. Comparison of correlation analysis techniques for irregularly sampled time series. Nonlinear Process. Geophys. 18, 389–404 (2011).ADS 
Article 

Google Scholar 
Braconnot, P. et al. Evaluation of climate models using palaeoclimatic data. Nat. Clim. Change 2, 417–424 (2012).ADS 
Article 

Google Scholar 
Cao, X. Y., Ni, J., Herzschuh, U., Wang, Y. B. & Zhao, Y. A late Quaternary pollen dataset from eastern continental Asia for vegetation and climate reconstructions: set up and evaluation. Rev. Palaeobot. Palynol. 194, 21–37 (2013).Article 

Google Scholar 
Bigelow, N. H. et al. Climate change and Arctic ecosystems: 1. Vegetation changes north of 55°N between the last glacial maximum, mid-Holocene, and present. J. Geophys. Res. Atmos. 108, 8170 (2003).Ramankutty, N. & Foley, J. A. Estimating historical changes in global land cover: Croplands from 1700 to 1992. Glob. Biogeochem. Cycles 13, 997–1027 (1999).ADS 
CAS 
Article 

Google Scholar 
Berger, A. & Loutre, M. F. Insolation values for the climate of the last 10 million years. Quat. Sci. Rev. 10, 297–317 (1991).ADS 
Article 

Google Scholar 
Deplazes, G. et al. Links between tropical rainfall and North Atlantic climate during the last glacial period. Nat. Geosci. 6, 213–217 (2013).ADS 
CAS 
Article 

Google Scholar 
Wessel, P. et al. Generic mapping tools: improved version released. EOS Trans. AGU 94, 409–410 (2013).ADS 
Article 

Google Scholar  More

To compare soil viral community composition within and across terrestrial habitats on a regional scale, viromes were generated from 34 near-surface (top 15 cm) soil samples, with a total of 30 viromes included in downstream ecological analyses (see Supplementary Methods). The analyzed viromes were collected from four distinct habitats (wetlands, grasslands, chaparral shrublands, and woodlands, each with 7, 14, 4, and 5 viromes, respectively) across five field sites (Fig. S1 for sampling scheme, Table S1 for soil properties). Following quality filtering, the 30 viromes generated an average of 72,950,833 reads and 416 contigs ≥10 Kbp per virome (Table S2). Wetland viromes yielded more contigs ≥10 Kbp than viromes from other habitats, both in total and on average per virome (Table S2). We used VIBRANT to identify 3490 viral contigs in our assemblies, which were clustered into 3,432 viral operational taxonomic units (vOTUs), defined as ≥10 Kbp viral contigs sharing ≥ 95% average nucleotide identity over 85% contig length [17]. Constrained analysis of principal coordinates (CAP analysis) revealed strong clustering by habitat rather than by site, implying that, where environmental parameters are substantially different, environmental conditions are stronger drivers of viral community composition than geographic distance (Fig. S2).Multiple lines of evidence suggest that wetter soils harbored greater viral diversity than drier soils. We recovered the most vOTUs from wetlands, both in total (56% of all vOTUs were from wetlands) and per virome (on average, 307 vOTUs were recovered per wetland virome, compared to 116 from all habitats) (Fig. 1A). Unsurprisingly, wetlands had significantly greater moisture content than other habitats (Fig. 1B; ANOVA followed by Tukey multiple comparisons of means, p 100 Km distances here. Taken together, we propose that soil viral communities often display high heterogeneity within and among habitats, presumably due to a combination of host adaptations and microdiversity, dispersal limitation, and fluctuating environmental conditions over space and time. More