Philippa Kaur

Sample collection and RNA preparationWe analyzed 6 adult yellow-bellied toad individuals representative of distinct behavioral profiles, i.e. prolonged unken-reflex display vs no unken-reflex display (thereafter referred as “ + ” and “-“, respectively). Behavioral profiles were scored as in Chiocchio et al.12: 3 toads showed prolonged unken-reflex (+), whereas the other 3 did not show unken-reflex (−), as reported in Table 1. Sampling procedures were approved by the Italian Ministry of Ecological Transition and the Italian National Institute for Environmental Protection and Research (ISPRA; permit number: 20824, 18-03-2020). After dissection, brain tissue was immediately stored in RNAprotect Tissue Reagent (Quiagen) until RNA extraction. RNA extractions were performed using the RNeasy Plus Kit (Quiagen), according to the manufacturer’ instructions. RNA quality and concentration were assessed by means of both a spectrophotometer and a Bioanalyzer (Agilent Cary60 UV-vis and Agilent 2100, respectively – Agilent Technologies, Santa Clara, USA).Table 1 Summary of the 6 libraries deposited in the Sequence Read Archive (SRA) of NCBI, in terms of number of raw and trimmed reads per sample.Full size tableLibrary preparation and sequencingLibrary preparation and RNA sequencing were performed by NOVOGENE (UK) COMPANY LIMITED using Illumina NovaSeq platform. Library construction was carried out using the NEBNext® Ultra ™ RNA Library Prep Kit for Illumina®, following manufacturer instructions. Briefly, after the quality control check, the mRNA sample was isolated from the total RNA by using magnetic beads made of oligos d(T)25 (i.e. polyA-tail mRNA enrichment). Subsequently, mRNA was randomly fragmented, and a cDNA synthesis step proceeded using random hexamers and the reverse transcriptase enzyme. Once the synthesis of the first chain has finished, the second chain was synthesized with the addition of the Illumina buffer, dNTPs, RNase H and polymerase I of E.coli, by means of the Nick translation method. Then, the resulting products went through purification, repair, A-tailing and adapter ligation. Fragments of the appropriate size were enriched by PCR, the indexed P5 and P7 primers were introduced, and the final products were purified. Finally, the Illumina Novaseq 6000 sequencing system was used to sequence the libraries, through a paired-end 150 bp (PE150) strategy. We obtained on average 52.7 million reads for each library. The sequencing data are available at the NCBI Sequence Read Archive (project ID PRJNA76401320).Pre-assembly processing stageA total of 316,329,573 pairs of reads was generated by Illumina sequencing. All of them went to a cleaning analytic step. The quality of the raw reads was assessed with the FastQC 0.11.5 tool (http://www.bioinformatics.bbsrc.ac.uk/projects/fastqc), in order to estimate the RNAseq quality profiles. The quality estimators were generated for both the raw and trimmed data. The quality assessment metrics for trimmed data were aggregated across all samples into a single report for a summary visualization with MultiQC software tool21 v.1.9 (see Fig. 1). To remove low quality bases and adapter sequences, raw reads were also analyzed through a quality trimming step with Trimmomatic22, v.0.39 (setting the option SLIDINGWINDOW: 4: 15, MINLEN: 36, and HEADCROP: 13). All the unpaired reads were discarded. After the cleaning step and removal of low-quality reads, 297,354,405 clean reads (i.e. 94% of raw reads) were maintained for building the de novo transcriptome assembly (see Table 1).Fig. 1The cleaned reads from all samples were assessed with FastQC and visualized with MultiQC. (a) Read count distribution for mean sequence quality. (b) Mean quality scores distribution. (c) Read length distribution. (d) Per Sequence GC Content.Full size image
De novo transcriptome assembly and quality assessmentAs there is no reference genome for B. pachypus, we performed a de novo transcriptome assembly procedure. The workflow of the bioinformatic pipelines is shown in Fig. 2. All the described bioinformatics analyses were performed on the high-performance computing systems provided by ELIXIR-IT HPC@CINECA23.Fig. 2Workflow of the bioinformatic pipeline, from raw input data to annotated contigs, for the de novo transcriptome assembly of B. pachypus.Full size imageTo construct an optimized de novo transcriptome, avoiding chimeric transcripts, we employed rnaSPAdes24, a tool for de novo transcriptome assembly from RNA-Seq data implemented in the SPAdes v.3.14.1 package. rnaSPAdes automatically detected two k-mer sizes, approximately one third and half of the maximal read length (the two detected k-mer sizes were 45 and 67 nucleotides, respectively). At this stage, a total of 1,118,671 assembled transcripts were generated by rnaSPAdes runs, with an average length of 689.41 bp and an N50 of 1474 bp (Table 2).Table 2 Similarity rate of newly assembled transcripts versus the de novo transcriptome of B. pachypus.Full size tableResults from the assembly procedures were validated through three independent validator algorithms implemented in BUSCO25 v.4.1.4, DETONATE26 v.1.11 and TransRate27 v.1.0.3. These tools generate several metrics used as a guide to evaluate error sources in the assembly process and provide evidence about the quality of the assembled transcriptome. Busco provides a quantitative measure of transcriptome quality and completeness, based on evolutionarily-informed expectations of gene content from the near-universal, ultra-conserved eukaryotic proteins (eukaryota_odb9) database. Detonate (DE novo TranscriptOme rNa-seq Assembly with or without the Truth Evaluation) is a reference-free evaluation method based on a novel probabilistic model that depends only on the assembly and the RNA-Seq reads used to construct it. Transrate generates standard metrics and remapping statistics. No reference protein sequences were used for the assessment with Transrate. The main metrics resulted from the assembly validators are shown in Table 2 (“Before CD-HIT-est” column). After this triple assessment validation step, the result of the assembly procedure become the input for the CD-HIT-est v.4.8.128 program, a hierarchical clustering tool used to avoid redundant transcripts and fragmented assemblies common in the process of de novo assembly, providing unique genes. CD-HIT-est was run using the default parameters, corresponding to a similarity of 95%. Subsequently, a second validation step was launched on the CD-HIT-est output file. To refine the final transcriptome dataset, a further hierarchical clustering step was performed by running CORSET v1.0629. Then, the output of CORSET was validated by BUSCO, and quality assessment was performed with HISAT230,31 by mapping the trimmed reads to the reference transcriptome (unigenes). Results from all validation steps are shown in Table 2 and discussed in the “Technical Validation” paragraph.Finally, the CORSET output was run on TransDecoder32,33, the current standard tool that identifies long open read frames (ORFs) in assembled transcripts, using default parameters. TransDecoder by default performs ORF prediction on both strands of assembled transcripts regardless of the sequenced library. It also ranks ORFs based on their completeness, and determines if the 5 ‘end is incomplete by looking for any length of AA codons upstream of a start codon (M) without a stop codon. We adopted the “Longest ORF” rule and selected the highest 5 AUG (relative to the inframe stop codon) as the translation start site.Transcriptome annotationWe employed different kinds of annotations for the de novo assembly. We introduced DIAMOND34, an open-source algorithm based on double indexing that is 20,000 times faster than BLASTX on short reads and has a similar degree of sensitivity. Like BLASTX, DIAMOND attempts to determine exhaustively all significant alignments for a given query. Most sequence comparison programs, including BLASTX, follow the seed-and-extend paradigm. In this two-phase approach, users search first for matches of seeds (short stretches of the query sequence) in the reference database, and this is followed by an ‘extend’ phase that aims to compute a full alignment. The following parameter settings were applied: DIAMOND-fast DIAMOND BLASTX-t 48 -k 250 -min-score 40; DIAMOND-sensitive: DIAMOND BLASTX -t 48 -k 250 -sensitive -min-score 40.Contigs were aligned with DIAMOND on Nr, SwissProt and TrEMBL to retrieve the corresponding best annotations. An annotation matrix was then generated by selecting the best hit for each database. Following the analysis of BLASTX against Nr, SwissProt and TremBL, we obtained respectively: 123,086 (64.57%), 77,736 (40.78%), 122,907 (64.48%) contigs. The results obtained following the analysis with BLASTP against Nr, SwissProt and TrEMBL were 96,321 (50.53%), 57,877 (30.36%) and 97,256 (51.02%) contigs respectively. All the information on the resulting datasets is resumed in Table 3.Table 3 Summary of homology annotation hits on the different databases queried in this study.Full size tableThe output obtained by the BLASTX annotation consisted in a total of 77391 sequences simultaneously mapped on the three queried databases (i.e., Nr, SwissProt and TrEMBL). The output obtained following the BLASTP annotation consisted in a total of 57704 sequences simultaneously mapped on the three databases. Venn diagrams are presented in Fig. 3, showing the redundancy of the annotations in the different databases for both DIAMOND BLASTX (Fig. 3a) and DIAMOND BLASTP (Fig. 3b). Furthermore, the ten most represented species and the ten hits of the gene product obtained respectively with BLASTX and BLASTP by mapping the transcripts against the reference database Nr are shown in Figs. 4 and 5. Since BLASTX translated nucleotide sequence searches against protein sequences the BLASTX results are more exhaustive than BLASTP results. Contigs were also processed with InterProScan35 to detect InterProScan signatures. The InterPro database (http://www.ebi.ac.uk/interpro/) integrates together predictive models or ‘signatures’ representing protein domains, families and functional sites from multiple, diverse source databases: Gene3D, PANTHER, Pfam, PIRSF, PRINTS, ProDom, PROSITE, SMART, SUPERFAMILY and TIGRFAMs. The obtained InterProScan results for all the unigenes are available on Figshare in the form of Tab Separated Values (tsv) file format, which includes the GO and KEGG annotated contigs, respectively.Fig. 3Venn diagrams for the number of contigs annotated with DIAMOND (BLASTX (a) and BLASTP (b) functions) against the three databases: Nr, SwissProt, TREMBL.Full size imageFig. 4Most represented species and gene product hits. Top 10 best species (a) and protein (b) hits present in the reference database (Nr, BLASTX).Full size imageFig. 5Most represented species and gene product hits. Top 10 best species (a) and protein (b) hits present in the reference database (Nr, BLASTP).Full size imageComparison with Bombina orientalis brain transcriptomeWe compared the brain de novo transcriptome of B. pachypus with the brain de novo transcriptome of B. orientalis, recently produced in the frame of a prey-catching conditioning experiment17,18. The B. orientalis transcriptome resource was downloaded from GEO archive of NCBI (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE171766). To make the datasets comparable, we first performed ORF prediction on B. orientalis trascriptome using Transdecoder, using default settings. Results from the B. orientalis trascriptome ORF prediction are available in Figshare at the following link https://doi.org/10.6084/m9.figshare.20319633/). We also applied the makedb function implemented in DIAMOND to create the protein database index. Then, we aligned the B. pachypus predicted coding sequences and proteins (query files) against the B. orientalis protein database (reference) using DIAMOND BLASTX and BLASTP, respectively. We obtained 167041 matches from BLASTX and 156248 matches for BLASTP. Results from the BLASTX and BLASTP comparisons, and the most matched proteins, are available on Figshare36 (link available in next paragraph). More

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

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

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

Keith, D. A. et al. Nature https://doi.org/10.1038/s41586-022-05318-4 (2022).Article 

Google Scholar 
Domesday Book, or, The Great Survey of England of William the Conqueror A.D. MLXXXVI (Ordnance Survey Office, 1862).McMahon, G. et al. Environ. Manage. 28, 293–316 (2001).PubMed 
Article 

Google Scholar 
Spalding, M. D. et al. BioScience 57, 573–583 (2007).Article 

Google Scholar 
Holdridge, L. R. Science 105, 367–368 (1947).PubMed 
Article 

Google Scholar 
Köppen, W. in Handbuch der Klimatologie (eds Köppen, W. & Geiger, G. C.) 1–44 (Gebrüder Borntraeger, 1936).
Google Scholar 
Whittaker, R. H. Communities and Ecosystems (Macmillan, 1975).
Google Scholar 
Keddy, P. A. Trends Ecol. Evol. 9, 231–234 (1994).PubMed 
Article 

Google Scholar 
United Nations. Convention on Biological Diversity (UN, 1992).
Google Scholar 
MacArthur, R. H. Geographical Ecology: Patterns in the Distribution of Species (Princeton Univ. Press, 1972).
Google Scholar 
Schoener, T. W. in Community Ecology (eds Diamond, J. D. & Case, T. ) 467–479 (Harper & Row, 1986).
Google Scholar 
Winemiller, K. O., Fitzgerald, D. B., Bower, L. M. & Pianka, E. R. Ecol. Lett. 18, 737–751 (2015).PubMed 
Article 

Google Scholar  More

Johnson, N. et al. (eds) Global Soil Biodiversity Atlas (EU, 2016).
Google Scholar 
FAO et al. State of Knowledge of Soil Biodiversity — Status, Challenges and Potentialities (FAO, 2020).
Google Scholar 
Cameron, E. K. et al. Nature Ecol. Evol. 2, 1042–1043 (2018).PubMed 
Article 

Google Scholar 
van den Hoogen, J. et al. Nature 572, 194–198 (2019).PubMed 
Article 

Google Scholar 
Phillips, H. R. P. et al. Science 366, 480–485 (2019).PubMed 
Article 

Google Scholar 
Guerra, C. A. et al. Nature https://doi.org/10.1038/s41586-022-05292-x (2022).Article 

Google Scholar 
Moore, J. C. & de Ruiter, P. C. Energetic Food Webs: An Analysis of Real and Model Ecosystems (Oxford Univ. Press, 2012).
Google Scholar 
Wolters V. et al. Bioscience 50, 1089–1098 (2000).Article 

Google Scholar 
Schimel, J. P. & Schaeffer, S. M. Front. Microbiol. 3, 348 (2012).PubMed 
Article 

Google Scholar 
IPCC. In Climate Change 2022: Mitigation of Climate Change. Contribution of Working Group III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change Impacts, Adaptation, and Vulnerability: Summary for Policymakers (eds Shukla, P. R. et al.) 50 (Cambridge Univ. Press, 2022).
Google Scholar 
Chenu, C. et al. Soil Till. Res. 188, 41–52 (2019).Article 

Google Scholar 
Liang, C., Schimel, J. P. & Jastrow, J. D. Nature Microbiol. 2, 17105 (2017).PubMed 
Article 

Google Scholar 
Hannula, S. E. & Morriën, E. Geoderma 413, 115767 (2022).Article 

Google Scholar  More

We developed the IUCN Global Ecosystem Typology in the following sequence of steps: design criteria; hierarchical structure and definition of levels; generic ecosystem assembly model; top-down classification of the upper hierarchical levels; iterative circumscription of the units and ecosystem-specific adaptations of the assembly model; full description of the units; and map compilation. Some iteration proved necessary, as the description and review process sometimes revealed a need for circumscribing additional units.Design criteria and other typologiesUnder the auspices of the IUCN Commission on Ecosystem Management, we developed six design principles to guide the development of a typology that would meet the needs for global ecosystem reporting, risk assessment, natural capital accounting and ecosystem management: (1) representation of ecological processes and ecosystem functions; (2) representation of biota; (3) conceptual consistency throughout the biosphere; (4) scalable structure; (5) spatially explicit units; and (6) parsimony and utility (see Supplementary Table 1.1 and Supplementary Information, Appendix 1 for definitions and rationale).We assessed 23 existing ecological classifications with global coverage of terrestrial, freshwater, and/or marine environments against these principles to determine their fitness for IUCN’s purpose (Supplementary Information, Appendix 1). These include general classifications of land, water or bioclimate, as well as classifications of units that conform with the definition of ecosystems adopted in the United Nations Convention on Biological Diversity45 or an equivalent definition in the IUCN Red List of Ecosystems30. We reviewed documentation on methods of derivation, descriptions of classification units and maps to assess each classification against the six design principles (Supplementary Table 1.2 for details).Typology structure and ecosystem assemblyWe developed the structure of the Global Ecosystem Typology and the generic ecosystem assembly model at a workshop attended by 48 terrestrial, freshwater and marine ecosystem experts at Kings College London, UK, in May 2017. Participants agreed that a hierarchical structure would provide an effective framework for integrating ecological processes and functional properties (Supplementary Table 1.1, design principle 1), and biotic composition (principle 2) into the typology, while also meeting the requirement for scalability (principle 4). Although neither function nor composition were intended to take primacy within the typology, we reasoned that a hierarchy representing functional features in the upper levels is likely to support generalizations and predictions by leveraging evolutionary convergence13. By contrast, a typology reflecting compositional similarities in its upperlevels is less likely to be stable owing to dynamism of species assemblages and evolving knowledge on species taxonomy and distributions. Furthermore, representation of compositional relationships at a global scale would require many more units in upper levels, and possibly more hierarchical levels. Therefore, we concluded that a hierarchical structure recognizing compositional variants at lower levels within broad functionally based groupings at upper levels would be more parsimonious and robust (principle 6) than one representing composition at upper levels and functions at lower levels.Workshop participants initially agreed that three hierarchical levels for ecosystem function and three levels for biotic composition could be sufficient to represent global variation across the whole biosphere. Participants developed the concepts of these levels into formal definitions (Supplementary Table 3.1), which were reviewed and refined during the development process.To ensure conceptual consistency of the typology and its units throughout the biosphere (principle 3), we drew from community assembly theory to develop a generic model of ecosystem assembly. The traditional community assembly model incorporates three types of filters (dispersal, the abiotic environment and biotic interactions) that determine which biota from a larger pool of potential colonists can occupy and persist in an area13. We extended this model to ecosystems by: (1) defining three groups of abiotic filters (resources, ambient environment and disturbance regimes) and two groups of biotic filters (biotic interactions and human activity); (2) incorporating evolutionary processes that shape characteristic biotic properties of ecosystems over time; (3) defining the outcomes of filtering and evolution in terms of all ecosystem properties including both ecosystem-level functions and species-level traits, rather than only in terms of species traits and composition; and (4) incorporating interactions and feedbacks among filters and selection agents and ecosystem properties to elucidate hypotheses about processes that influence temporal and spatial variability in the properties of ecosystems and their component biota. In community assembly, only a small number of filters are likely to be important in any given habitat13. In keeping with this proposition, we used the generic model to identify biological and physical features that distinguish functionally different groups of ecosystems from one another by focusing on different ecological drivers that come to the fore in structuring their assembly and shaping their properties.Hierarchical levelsThe top level of classification (Fig. 2 and Extended Data Tables 1–4) defines five core realms of the biosphere based on contrasting media that reflect ecological processes and functional properties: terrestrial; freshwaters and inland saline waters (hereafter freshwater); marine; subterranean; and atmospheric. Biome gradient concepts25 highlight continuous variation in ecosystem properties, which is represented in the typology by transitional realms that mark the interfaces between the five core realms (for example, floodplains (terrestrial–freshwater), estuaries (freshwater–marine), and so on). In Supplementary Information, Appendix 3 (pages 3–16) and Supplementary Table 3.1, we describe the five core realms and review the hypothesized assembly filters and ecosystem properties that distinguish different groups within them. The atmospheric realm is included for comprehensive coverage, but we deferred resolution of its lower levels because its biota is poorly understood, sparse, itinerant and represented mainly by dispersive life stages46.Functional biomes (level 2) are components of the biosphere united by one or more major assembly processes that shape key ecosystem functions and ecological processes, irrespective of taxonomic identity (Supplementary Information, Appendix 3, page 17). Our interpretation aligns broadly with ‘functional biomes’ described elsewhere24,25,47, extended here to reflect dominant assembly filters and processes across all realms, rather than the more restricted basis of climate-vegetation relationships that traditionally underpin biome definition on land. Hence, the 25 functional biomes (Supplementary Information, Appendix 4, pages 52–186 and https://global-ecosystems.org/) include some ‘traditional’ terrestrial biomes47, as well as lentic and lotic freshwater systems, pelagic and benthic marine systems, and anthropogenic functional biomes assembled and usually maintained by human activity48.Level 3 of the typology defines 110 ecosystem functional groups described with illustrated profiles in Supplementary Information, Appendix 4 (pages 52–186) and at https://global-ecosystems.org/. These are key units for generalization and prediction, because they include ecosystem types with convergent ecosystem properties shaped by the dominance of a common set of drivers (Supplementary Information, Appendix 3, pages 17–19). Ecosystem functional groups are differentiated along environmental gradients that define spatial and temporal variation in ecological drivers (Figs. 2 and 3 and Supplementary Figs. 3.2 and 3.4). For example, depth gradients of light and nutrients differentiate functional groups in pelagic ocean waters (Fig. 3c and Extended Data Table 4), influencing assembly directly and indirectly through predation. Resource gradients defined by flow regimes (influenced by catchment precipitation and evapotranspiration) and water chemistry, modulated by environmental gradients in temperature and geomorphology, differentiate functional groups of freshwater ecosystems25 (Fig. 3b and Extended Data Table 3). Terrestrial functional groups are distinguished primarily by gradients in water and nutrient availability and by temperature and seasonality (Fig. 3a and Extended Data Table 1), which mediate uptake of those resources and regulate competitive dominance and productivity of autotrophs. Disturbance regimes, notably fire, are important global drivers in assembly of some terrestrial ecosystem functional groups49.Three lower levels of the typology distinguish functionally similar ecosystems based on biotic composition. Our focus in this paper is on global functional relationships of ecosystems represented in the upper three levels of the typology, but the lower levels (Supplementary Information, Appendix 3, pages 19 and 20) are crucial for representing the biota in the typology, and facilitate the scaling up of information from established local-scale typologies that support decisions where most conservation action takes place. These lower levels are being developed progressively through two contrasting approaches with different trade-offs, strengths and weaknesses. First, level 4 units (regional ecosystem subgroups) are ecoregional expressions of ecosystem functional groups developed from the top-down by subdivisions based on biogeographic boundaries (for example, in ref. 50) that serve as simple and accessible proxies for biodiversity patterns51. Second, level 5 units (global ecosystem types) are also regional expressions of ecosystem functional groups, but unlike level 4 units they are explicitly linked to local information sources by bottom-up aggregation52 and rationalization of level 6 units from established subglobal ecological classifications. Subglobal classifications, such as those for different countries (see examples for Chile and Myanmar in Supplementary Tables 3.3 and 3.4), are often developed independently of one another, and thus may involve inconsistencies in methods and thematic resolution of units (that is, broadly defined or finely split). Aggregation of level 6 units to broader units at level 5 based on compositional resemblance is necessary to address inconsistencies among different subglobal classifications and produce compositionally distinctive units suitable for global or regional synthesis.Integrating local classifications into the global typology, rather than replacing them, exploits considerable efforts and investments to produce existing classifications, already developed with local expertise, accuracy and precision. By placing national and regional ecosystems into a global context, this integration also promotes local ownership of information to support local action and decisions, which are critical to ecosystem conservation and management outcomes (Supplementary Information, Appendix 3, page 20). These benefits of bottom-up approaches come at the cost of inevitable inconsistencies among independently developed classifications from different regions, a limitation avoided in the top-down approach applied to level 4.Circumscribing upper-level unitsWe formed specialist working groups (terrestrial/subterranean, freshwater and marine) to develop descriptions of the units within the upper levels of the hierarchy, subdividing realms into functional biomes, and biomes into ecosystem functional groups. We used definitions of the hierarchical levels (Supplementary Table 3.1) and the conceptual model of ecosystem assembly (Fig. 1) to maintain consistency in defining the units at each level during iterative discussions within and between the working groups.Working groups agreed on preliminary lists of functional biomes and ecosystem functional groups by considering variation in major drivers along ecological gradients (Figs. 2 and 3 and Supplementary Figs. 3.2 and 3.4) based on published literature, direct experience and expertise of working group members, and consultation with colleagues in their respective research networks. After the workshop, working groups sought recent global reviews of the candidate units and recent case studies of exemplars to shape descriptions of the major groups of ecosystem drivers and properties for each unit. Circumscriptions and descriptions of the units were reviewed and revised iteratively to ensure clear distinctions among units, with a total of 206 reviews of descriptive profiles undertaken by 60 specialists, a mean of 2.4 reviews per profile (Supplementary Table 5.1). The working groups concurrently adapted the generic model of ecosystem assembly (Fig. 1) to represent working hypotheses on salient drivers and ecosystem properties for each ecosystem functional group.Incorporating human influenceVery few of the ecological typologies reviewed in Supplementary Information, Appendix 1 integrate anthropogenic ecosystems in their classificatory frameworks. Anthropogenic influences create challenges for ecosystem classification, as they may modify defining features of ecosystems to a degree that varies from negligible to major transformation across different locations and times. We addressed this problem by distinguishing transformative outcomes of human activity at levels 2 and 3 of the typology from lesser human influences that may be represented either at levels 5 and 6, or through measurements of ecosystem integrity or condition that reflect divergence from reference states arising from human activity.Anthropogenic ecosystems grouped within levels 2 and 3 were thus defined as those created and sustained by intensive human activities, or arising from extensive modification of natural ecosystems such that they function very differently. These activities are ultimately driven by socio-economic and cultural-spiritual processes that operate across local to global scales of human organization. In many agricultural and aquacultural systems and some others, cessation of those activities may lead to transformation into ecosystem types with qualitatively different properties and organizational processes (see refs. 53,54 for cropland and urban examples, respectively). Indices such as human appropriation of net primary productivity55, combined with land-use maps56, offer useful insights into the distribution of some anthropogenic ecosystems, but further development of indices is needed to adequately represent others, particularly in marine, and freshwater environments. Beyond land-use classification and mapping approaches (Supplementary Information, Appendix 1, page 6), a more comprehensive elaboration of the intensity of human influence underpinning the diverse range of anthropogenic ecosystems requires a multidimensional framework incorporating land-use inputs, outputs, their interactions, legacies of earlier activity and changes in system properties17.Where less intense human activities occur within non-anthropogenic ecosystem types, we focused descriptions on low-impact reference states. Therefore, human activities are not shown as drivers in the assembly models for non-anthropogenic ecosystem groups, even though they may have important influences on the contemporary ecosystem distribution. This approach enables the degree and nature of human influence to be described and measured against these reference states using assessment methods such as the Red List of Ecosystems protocol30, with appropriate data on ecosystem change.Indicative distribution mapsFinally, to produce spatially explicit representations of the units at level 3 of the typology (principle 5), we sought published global maps (sources in Supplementary Table 4.1) that were congruent with the concepts of respective ecosystem functional groups. Where several candidate maps were available, we selected maps with the closest conceptual alignment, finest spatial resolution, global coverage, most recent data and longest time series. The purpose of maps for our study was to visualize global distributions. Prior to applications of map data to spatial analysis, we recommend critical review of methods and validation outcomes reported in each data source to ensure fitness for purpose (Supplementary Information, Appendix 4).Extensive searches of published literature and data archives identified high-quality datasets for some ecosystem functional groups (for example, T1.3 Tropical–subtropical montane rainforests; MT1.4 Muddy shorelines; M1.5 Sea ice) and datasets that met some of these requirements for a number of other ecosystem functional groups (see Supplementary Table 4.1 for details). Where evaluations by authors or reviewers identified limitations in available maps, we used global environmental data layers and biogeographic regionalizations as masks to adjust source maps and improve their congruence to the concept of the relevant functional group (for example, F1.2 Permanent lowland rivers). For ecosystem functional groups with no specific global mapping, we used ecoregions50,57,58 as biogeographic templates to identify broad areas of occurrence. We consulted ecoregion descriptions, global and regional reviews, national and regional ecosystem maps, and applied in situ knowledge of participating experts to identify ecoregions that contain occurrences of the relevant ecosystem functional group (for example, T4.4 Temperate woodlands) (see Supplementary Table 4.1 for details). We mapped ecosystem functional groups as major occurrences where they dominated a landscape or seascape matrix and minor occurrences where they were present, but not dominant in landscape–seascape mosaics, or where dominance was uncertain. Although these two categories in combination communicate more information about ecosystem distribution than binary maps, simple spatial overlays using minor occurrences are likely to inflate spatial statistics. The maps are progressively upgraded in new versions of the typology as explicit spatial models are developed and new data sources become available (see ref. 27 for a current archive of spatial data).The classification and descriptive profiles, including maps, for each functional biome and ecosystem functional group underwent extensive consultation, and targeted peer review and revision through a series of four phases described in Supplementary Information, Appendix 5 (pages 2–4). The reviewer comments and revisions from targeted peer review are documented in Supplementary Table 5.1. In all, more than 100 ecosystem specialists have contributed to the development of v2.1 of the typology.LimitationsUneven knowledge of Earth’s biosphere has constrained the delimitation and description of units within the typology. There is a considerable research bias across the full range of Earth’s ecosystems, with few formal research studies evaluating the relative influence of different ecosystem drivers in many of the functional groups, and abiotic assembly filters generally receiving more attention than biotic and dispersal filters. This poses challenges for developing standardized models of assembly for each ecosystem functional group. The models therefore represent working hypotheses, for which available evidence varies from large bodies of published empirical evidence to informal knowledge of ecosystem experts and their extensive research networks. Large numbers of empirical studies exist for some forest functional groups, savannas, temperate heathlands in Mediterranean-type climates, coral reefs, rocky shores, kelp forests, trophic webs in pelagic waters, small permanent freshwater lakes, and others (see references in the respective profiles (Supplementary Information, Appendix 4)). For example, Bond49 reviewed empirical and modelling evidence on the assembly and function of tropical savannas that make up three ecosystem functional groups, showing that they have a large global biophysical envelope that overlaps with tropical dry forests, and that their distribution and dynamics within that envelope is strongly influenced by top-down regulation via biotic filters (large herbivores and their predators) and recurrent disturbance regimes (fires). Despite the development of this critical knowledge base, savannas suffer from an awareness disparity that hinders effective conservation and management59. In other ecosystems, our assembly models rely more heavily on inferences and generalizations of experts drawn from related ecosystems, are more sensitive to interpretations of participating experts, and await empirical testing and adjustment as understanding improves. Empirical tests could examine hypothesized variation in ecosystem properties along gradients within and between ecosystem functional groups and should return incremental improvements on group delineation and description of assembly processes.High-quality maps at suitable resolution are not yet available for the full set of ecosystem functional groups, which limits current readiness for global analysis. The maps most fit for global synthesis are based on remote sensing and environmental predictors that align closely to the concept of their ecosystem functional group, incorporate spatially explicit ground observations and have low rates of omission and commission errors, ‘high’ spatial resolution (that is, rasters of 1 km2 (30 arcsec) or better), and time series of changes. Sixty of the maps currently in our archive27 aligned directly or mostly with the concept of their corresponding ecosystem functional group, while the remainder were based on indirect spatial proxies, and most were derived from polygon data or rasters of 30 arcsec or finer (Supplementary Table 4.1). Maps for 81 functional groups were based either on known records, or on spatial data validated by quantitative assessments of accuracy or efficacy. Therefore, we suggest that maps currently available for 60–80 of the 110 functional groups are potentially suitable for global spatial analysis of ecosystem distributions. Although, a significant advance on broad proxies such as ecoregions, the maps currently available for ecosystem functional groups would benefit from expanded application of recent advances in remote sensing, environmental datasets, spatial modelling and cloud computing to redress inequalities in reliability and resolution. The most urgent priorities for this work are those identified in Supplementary Table 4.1 as relying on indirect proxies for alignment to concept, qualitative evaluation by experts and coarse resolution ( >1 km2) spatial data.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More