Philippa Kaur

Sakschewski, B. et al. Leaf and stem economics spectra drive diversity of functional plant traits in a dynamic global vegetation model. Glob. Change Biol. 21, 2711–2725 (2015).Article 

Google Scholar 
Berzaghi, F. et al. Towards a new generation of trait-flexible vegetation models. Trends Ecol. Evol. 35, 191–205 (2020).Article 
PubMed 

Google Scholar 
Bruelheide, H. et al. Global trait–environment relationships of plant communities. Nat. Ecol. Evol. 2, 1906–1917 (2018).Article 
PubMed 

Google Scholar 
Joswig, J. S. et al. Climatic and soil factors explain the two-dimensional spectrum of global plant trait variation. Nat. Ecol. Evol. 6, 36–50 (2021).Article 
PubMed 
PubMed Central 

Google Scholar 
van Bodegom, P. M., Douma, J. C. & Verheijen, L. M. A fully traits-based approach to modeling global vegetation distribution. Proc. Natl Acad. Sci. USA 111, 13733–13738 (2014).PubMed Central 

Google Scholar 
Moreno Martínez, A. et al. A methodology to derive global maps of leaf traits using remote sensing and climate data. Remote Sens. Environ. 218, 69–88 (2018).Article 

Google Scholar 
Pérez-Harguindeguy, N. et al. New handbook for standardized measurment of plant functional traits worldwide. Aust. J. Bot. 23, 167–234 (2013).Article 

Google Scholar 
Kattge, J. et al. TRY—a global database of plant traits. Glob. Change Biol. 17, 2905–2935 (2011).Article 

Google Scholar 
Kattge, J. et al. TRY plant trait database-enhanced coverage and open access. Glob. Change Biol. 26, 119–188 (2020).Article 

Google Scholar 
Jetz, W. et al. Monitoring plant functional diversity from space. Nat. Plants 2, 16024 (2016).Article 
PubMed 

Google Scholar 
Butler, E. E. et al. Mapping local and global variability in plant trait distributions. Proc. Natl Acad. Sci. USA 114, E10937–E10946 (2017).Article 
PubMed 
PubMed Central 

Google Scholar 
Boonman, C. C. et al. Assessing the reliability of predicted plant trait distributions at the global scale. Glob. Ecol. Biogeogr. 29, 1034–1051 (2020).Article 
PubMed 
PubMed Central 

Google Scholar 
Madani, N. et al. Future global productivity will be affected by plant trait response to climate. Sci. Rep. 8, 2870 (2018).Vallicrosa, H. et al. Global distribution and drivers of forest biome foliar nitrogen to phosphorus ratios (N:P). Glob. Ecol. Biogeogr. 31, 861–871 (2022).Article 

Google Scholar 
Meyer, H. & Pebesma, E. Predicting into unknown space? Estimating the area of applicability of spatial prediction models. Methods Ecol. Evol. 12, 1620–1633 (2021).Article 

Google Scholar 
Schiller, C. et al. Deep learning and citizen science enable automated plant trait predictions from photographs. Sci. Rep. 11, 16395 (2021).Aguirre-Gutiérrez, J. et al. Pantropical modelling of canopy functional traits using sentinel-2 remote sensing data. Remote Sens. Environ. 252, 112–122 (2021).Article 

Google Scholar 
Homolova, L. et al. Review of optical-based remote sensing for plant trait mapping. Ecol. Complex. 15, 1–16 (2013).Article 

Google Scholar 
Van Cleemput, E. et al. The functional characterization of grass-and-shrubland ecosystems using hyperspectral remote sensing: trends, accuracy and moderating variables. Remote Sens. Environ. 209, 747–763 (2018).Article 

Google Scholar 
Kattenborn, T., Fassnacht, F. E. & Schmidtlein, S. Differentiating plant functional types using reflectance: which traits make the difference? Remote Sens. Ecol. Conserv. 5, 5–19 (2019).Article 

Google Scholar 
Hauser, L. T. et al. Explaining discrepancies between spectral and in-situ plant diversity in multispectral satellite earth observation. Remote Sens. Environ. 265, 112684 (2021).Article 

Google Scholar 
Wäldchen, J. & Mäder, P. Plant species identification using computer vision techniques: a systematic literature review. Arch. Comput. Methods Eng. 25, 507–543 (2018).Article 
PubMed 

Google Scholar 
Jones, H. G. What plant is that? Tests of automated image recognition apps for plant identification on plants from the British flora. AoB Plants 12, plaa052 (2020).Article 
PubMed 
PubMed Central 

Google Scholar 
Hampton, S. E. et al. Big data and the future of ecology. Front. Ecol. Environ. 11, 156–162 (2013).Article 

Google Scholar 
WÜest, R. O. et al. Macroecology in the age of big data—where to go from here? J. Biogeogr. 47, 1–12 (2020).Article 

Google Scholar 
Mäder, P. et al. The Flora Incognita app—interactive plant species identification. Methods Ecol. Evol. 12, 1335–1342 (2021).Article 

Google Scholar 
Di Cecco, G. J. et al. Observing the observers: how participants contribute data to iNaturalist and implications for biodiversity science. BioScience 71, 1179–1188 (2021).Article 

Google Scholar 
Mahecha, M. D. et al. Crowd-sourced plant occurrence data provide a reliable description of macroecological gradients. Ecography 44, 1131–1142 (2021).Article 

Google Scholar 
Botella, C. et al. Jointly estimating spatial sampling effort and habitat suitability for multiple species from opportunistic presence-only data. Methods Ecol. Evol. 12, 933–945 (2021).Article 

Google Scholar 
iNaturalist Research-Grade Observations (GBIF, accessed 5 January 2022); https://www.gbif.org/dataset/50c9509d-22c7-4a22-a47d-8c48425ef4a7Callaghan, C. T. et al. Three frontiers for the future of biodiversity research using citizen science data. BioScience 71, 55–63 (2020).
Google Scholar 
Dickinson, J. L., Zuckerberg, B. & Bonter, D. N. Citizen science as an ecological research tool: challenges and benefits. Ann. Rev. Ecol. Evol. Syst. 41, 149–172 (2010).Article 

Google Scholar 
Kosmala, M. et al. Assessing data quality in citizen science. Front. Ecol. Environ. 14, 551–560 (2016).Article 

Google Scholar 
Boakes, E. H. et al. Patterns of contribution to citizen science biodiversity projects increase understanding of volunteers’ recording behaviour. Sci. Rep. 6, 33051 (2016).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Bowler, D.E. et al. Temporal trends in the spatial bias of species occurrence records. Ecography 2022, e06219 (2022). https://doi.org/10.1111/ecog.06219GBIF Occurrence Download (GBIF, 4 January 2022); https://doi.org/10.15468/dl.34tjreBruelheide, H. et al. sPlot—a new tool for global vegetation analyses. journal of vegetation science. J. Veg. Sci. 30, 161–186 (2019).Article 

Google Scholar 
Sabatini, F. et al. sPlotOpen—an environmentally balanced, open access, global dataset of vegetation plots. Glob. Ecol. Biogeogr. 30, 1740–1764 (2021).Article 

Google Scholar 
Whittaker, R.H. et al. Communities and Ecosystems (Macmillan/Collier Macmillan, 1970).Olson, D. M. et al. Terrestrial ecoregions of the world: a new map of life on Earth: a new global map of terrestrial ecoregions provides an innovative tool for conserving biodiversity. BioScience 51, 933–938 (2001).Article 

Google Scholar 
Joswig, J., Wirth, C. & Schuman, M. Climatic and soil factors explain the two-dimensional spectrum of global plant trait variation. Nat. Ecol. Evol. 6, 36–50 (2022).Article 
PubMed 

Google Scholar 
Díaz, S. et al. The global spectrum of plant form and function. Nature 529, 167–171 (2016).Article 
PubMed 

Google Scholar 
Ploton, P. et al. Spatial validation reveals poor predictive performance of large-scale ecological mapping models. Nat. Commun. 11, 4540 (2020).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Meyer, H. & Pebesma, E. Spatial validation reveals poor predictive performance of large-scale ecological mapping models. Methods Ecol. Evol. 12, 1620–1633 (2021).Article 

Google Scholar 
Schrodt, F. et al. Bhpmf—a hierarchical Bayesian approach to gap filling and trait prediction for macroecology and functional biogeography. Glob. Ecol. Biogeogr. 24, 1510–1521 (2015).Article 

Google Scholar 
Kuppler, J. et al. Global gradients in intraspecific variation in vegetative and floral traits are partially associated with climate and species richness. Glob. Ecol. Biogeogr. 29, 992–1007 (2020).Article 

Google Scholar 
Scheiter, S., Langan, L. & Higgins, S. I. Next-generation dynamic global vegetation models: learning from community ecology. New Phytol. 198, 957–969 (2013).Article 
PubMed 

Google Scholar 
Taubert, F. et al. Confronting an individual-based simulation model with empirical community patterns of grasslands. PLoS ONE 15, e0236546 (2020).Roger, E. & Klistorner, S. (2016) Bioblitzes help science communicators engage local communities in environmental research. J. Sci. Commun. https://doi.org/10.22323/2.15030206 (2016).Legendre, P. & Legendre, L. Numerical Ecology 3rd edn (Elsevier, 2012).Warton, D. I. et al. Smatr 3—an R package for estimation and inference about allometric lines. Methods Ecol Evol 3, 257–259 (2012).Article 

Google Scholar 
Wolf, S. et al. iNaturalist_traits: iNaturalist trait maps version 1 (January 5, 2022) Zenodo https://doi.org/10.5281/zenodo.6671891 (2022). More

Bacterial strains and mediaA collection of 185 genome-sequenced bacterial isolates, described previously14, was utilized to assemble the synthetic community used in this work. These isolates were obtained from surface-sterilized Brassicaceae roots, primarily Arabidopsis thaliana, grown in two soils from North Carolina, USA35. This isolate collection includes strains V. paradoxus CL14, Arthrobacter CL28, Acinetobacter CL69 and Acinetobacter CL71, which are also used in this work in individual strain contexts. V. paradoxus CL14 ΔHS33, which has a clean deletion of genes with gene ID 2643613677 through 2643613653 was constructed previously14 and used here. Additional strains were obtained from the American Type Culture Collection (ATCC): E. soli LF7 (ATCC BAA-2102), R. pomeroyi (ATCC 700808) and B. japonicum (ATCC 10324). P. phytofirmans PsJN (DSMZ 17436) was obtained from the DSMZ-German Collection of Microorganisms and Cell Cultures. P. putida strain 1290 was generously provided by Johan Leveau (University of California Davis). Pseudomonas strain Root 562 was generously provided by Paul Schulze-Lefert (Max-Planck-Gesellschaft). All bacteria, with exceptions noted below, were routinely grown on LB agar plates (10 g l−1 tryptone, 5 g l−1 yeast extract, 10 g l−1 NaCl, 1.5% (w/v) agar) and in 2xYT liquid medium (16 g l−1 tryptone, 10 g l−1 yeast extract, 5 g l−1 NaCl) at 28 °C. The 175-member (185-member minus 10 Variovorax strains) synthetic community (SC185-10V) was grown on KB medium as was done previously to culture this synthetic community14. B. japonicum (ATCC 10324) was routinely grown on liquid and solidified YM medium (1 g l−1 yeast extract, 10 g l−1 mannitol, 0.5 g l−1 dipotassium phosphate, 0.2 g l−1 magnesium sulfate, 0.1 g l−1 NaCl, 1 g l−1 CaCO4, pH 6.8, solidified with 1.5% agar as necessary) at 28 °C. R. pomeroyi (ATCC 700808) was routinely grown on liquid and solidified LB medium supplemented with 2% sea salt (Millipore Sigma S9883) and solidified with 1.5% (w/v) agar as necessary. M9 base medium was formulated using 1x M9 minimal salts medium (Sigma M6030) supplemented with 2 mM MgSO4, 0.1 mM CaCl2 and 10 µM FeSO4. A carbon source or sources were added to this M9 base medium to support bacterial growth. Unique strains constructed in this study are available upon request.Bacterial 16S rRNA sequencingBacterial colonization of Arabidopsis roots was assessed using a method similar to the previous study14. Roots from 8–10 plants were collected into sterilized 2 ml tubes containing three 4 mm glass beads and root fresh weight in each tube was obtained. Five such samples were collected for each bacterial treatment. The roots were washed three times with sterile distilled water and stored at −80 °C until further processing. The roots were then lyophilised for 48 h using a Labconco freeze-dry system and pulverized using an MPBio tissue homogenizer. DNA was extracted from the root samples and bacterial cell pellets saved from the bacteria for input into the experiment using the DNeasy PowerSoil HTP 96 kit (Qiagen) according to manufacturer instructions. The V3-V4 region of the bacterial 16S rRNA gene was amplified and sequenced as previously described14.16S amplicon sequence data processingThe 16S sequencing data from synthetic community experiments were processed as previously described14. Briefly, usable read output from MT-Toolbox36 (reads with 100% primer sequences that successfully merged with their pair) were filtered for quality with Sickle37 by not allowing any window with Q score under 20. The resulting sequences were globally aligned to a 16S rRNA gene sequence reference dataset from genome assemblies of the synthetic community members. For strains that do not have an intact 16S rRNA sequence in their assembly, Sanger sequencing was used to obtain the 16S rRNA gene sequence of the strains for inclusion in the reference dataset. The reference dataset also included sequences from Arabidopsis organellar sequences and known bacterial contaminants. Sequence alignment was performed with USEARCH v.7.109038 using the optional usearch_global at a 98% identity threshold. On average, 85% of read sequences matched an expected isolate. The 185 isolates of our 185-member synthetic community could not all be distinguished from one another on the basis of the V3-V4 sequence. They were thus classified into 97 unique sequences encompassing a set of identical (clustered at 100%) V3-V4 sequences coming from a single or multiple isolate strains. An isolate abundance table was created from the sequence mapping results.We estimated 16S rRNA absolute abundance using a plasmid spike-in method39. Synthetic DNA was spiked at known quantities into samples before DNA extraction and the ratio of added to recovered synthetic DNA served as a conversion factor by which the total number of 16S rRNA molecules in a given sample was estimated. We designed a plasmid which included 16S V3-V4 primer binding sequences flanking a randomly generated DNA sequence matching the most frequent length and Guanine + Cytosine (GC) content of amplicons generated using the same primer sequences from wild soil. These sequences were synthesized by Geneart (Invitrogen) and supplied cloned in plasmid pMA-T. The plasmid was transformed into E. coli and isolated using a miniprep spin kit (Qiagen). Specific volumes of this isolated plasmid were then added to individual samples before DNA extraction to spike-in approximately 20% of the predicted 16S copies occurring within the sample. We performed colony-forming units (c.f.u.) counting using similarly treated plant samples (that is, growth on SynCom-inoculated agar plates) to obtain an estimate of the 16S copy number per mg fresh weight of plant roots. We plated serial dilutions of plant root samples ground in MgCl2 on LB to perform c.f.u. counts. The c.f.u. count multiplied by a given sample’s fresh weight were used to calculate sample-specific predicted 16S copy numbers.Plant growth conditions and root growth inhibition assayA. thaliana ecotype Col-0 seeds were sterilized in 70% household bleach, 0.2% Tween-20 for 10 min with vigorous agitation and then rinsed 10 times with sterile distilled water. Seeds were stratified at 4 °C in sterile distilled water for 1–2 d. Plants were germinated for 7 d on 0.5x MS agar medium (2.22 g l−1 PhytoTech Labs M-404: Murashige & Skoog modified basal medium with Gamborg vitamins, 0.5 g l−1 MES hydrate, pH adjusted to 5.7, solidified with 1% (w/v) agar) supplemented with 0.5% (w/v) sucrose in vertical 12 ×12 cm square plates under long-day conditions (21 °C/18 °C, 16 h light/8 h dark, day/night cycle). Then 8 to 10 plants were aseptically transferred to 12 ×12 cm plates containing 0.5x MS agar medium without sucrose where the medium surface was spread with the bacterial inoculum. For assays with IAA addition, 100 nM IAA was added to the medium before pouring the plates. The plant root tip location was marked on plates after transfer to record the initial root tip position. The plates containing the plants and bacteria were incubated vertically under short-day conditions (22 °C/18 °C, 9 h light/15 h dark, day/night cycle) for an additional 11 d. Plates were imaged on a document scanner and primary root elongation was determined using imageJ to quantify the change in root tip position from the initial to the final position.Bacterial inoculation of plantsIndividual bacterial strains were grown on agar plates of the media types specified above at 28 °C. Before plant inoculation, a single colony was picked into the appropriate liquid medium and grown at 28 °C to late exponential or early stationary phase. To remove the medium from the bacteria before inoculation, strains were washed three times in sterile 10 mM MgCl2. The optical density at 600 nm (OD600) was measured for each washed strain and normalized to OD600 of 0.01 in 10 mM MgCl2. For plant experiments with mono-association of an individual strain, 100 µl of OD600 = 0.01 washed bacteria was spread on the 12 ×12 cm plate before plant transfer. For experiments in duo-association with Arthrobacter CL28, 100 µl of OD600 = 0.01 washed Arthrobacter CL28 was spread along with 100 µl of OD600 = 0.01 of the second strain.The 175-member synthetic community (SC185-10V) was prepared as described for the 185-member synthetic community used previously14 by leaving out the 10 isolates from the genus Variovorax. Briefly, 7 d before plant transfer, strains were inoculated individually into 600 µl KB medium in a 96-well plate and grown at 28 °C for 5 d. At 2 d before plant transfer, 20 ul from these 5-day-old cultures were transferred to 380 ul fresh KB medium in a new set of 96-well plates and both sets of plates were returned to the incubator for 2 d. This resulted in two cultures of each strain, one 7 d old and the other 2 d old, which were combined. The OD600 of the strains in each well was measured and the strains were combined while normalizing the OD600 of each strain in the pool. This pool was washed twice with 10 mM MgCl2 and diluted to OD600 = 0.2. For experiments with the SC185-10V SynCom, 100 µl of this OD600 = 0.2 washed pool was spread on 12 ×12 cm plates. For treatments where an additional strain was added to the SC185-10V SynCom, the individual strain was washed as described above, diluted to OD600 = 0.0034 in 10 mM MgCl2, and 100 µl of this dilution was spread on the plates with the SC185-10V SynCom. This addition of the individual strain corresponded to an OD600 three times that of a single strain in the SC185-10V SynCom (0.0034 = (0.2/175) × 3). For the addition of the 10 Variovorax strains to the SC185-10V SynCom experiment, the 10 Variovorax strains were grown individually in 2xYT medium from colonies grown on plates. The OD600 values of the 10 cultures were measured and the 10 strains were pooled while normalizing the OD600 of each strain to the same value. This mixture of the 10 Variovorax strains was then treated as the individual strains for washing and addition of 100 µl of OD600 = 0.0034 to the SC185-10V SynCom on plates.Construction of vectors with Variovorax CL14 iad gene insertsPortions of the V. paradoxus CL14 IAA degradation locus were subcloned into broad host range vector pBBR1MCS-232. Primers JMC579 through JMC604 (Supplementary Table 8) were used to amplify 3–5 kb segments of the locus by PCR using Q5 DNA polymerase (New England Biolabs). These primers were designed to amplify sections beginning and ending at gene start codons and with appropriate overlapping sequences for Gibson assembly either into the pBBR1MCS-2 backbone or to the adjacent section to make larger vector inserts, as appropriate. The pBBR1MCS-2 vector backbone was prepared for Gibson assembly by amplifying the vector by PCR using primers JMC577 and JMC578 (Supplementary Table 8) and subsequently treating with DpnI to remove circular vector template. PCR fragments were cleaned up as necessary using the QIAquick PCR purification kit (Qiagen). Appropriate fragments were mixed to construct the vectors by Gibson assembly using HiFi DNA Assembly Mastermix (New England Biolabs) according to manufacturer instructions. Gibson assembly products were transformed into NEB 10beta chemically competent E. coli (New England Biolabs) and selected on LB plates supplemented with 50 µg ml−1 kanamycin. Vectors were miniprepped using either the ZR plasmid miniprep classic kit or Zymo BAC DNA miniprep kit (Zymo Research) and confirmed via restriction mapping with PstI-HF (New England Biolabs) and Sanger sequencing (Genewiz).To construct vectors that are derivatives of pBBR1::70–66, the Q5 site-directed mutagenesis kit (New England Biolabs) was used for gene deletion. Briefly, vector pBBR1::70–66 was used as a PCR template and portions of this vector were amplified by PCR using primers JMC641 through JMC650 (Supplementary Table 8) and Q5 DNA polymerase (New England Biolabs). PCR products were cleaned up and circularized using KLD Mastermix (New England Biolabs). The product was transformed into NEB 10beta chemically competent E. coli (New England Biolabs) and selected on LB plates supplemented with 50 µg ml−1 kanamycin. Vectors were miniprepped and Sanger sequenced as described above to confirm the construction of the correct vectors.Conjugation of vectors to V. paradoxus CL14 ΔHS33Vectors were conjugated into V. paradoxus CL14 ΔHS33 using tri-parental mating. The helper E. coli strain carrying plasmid pRK201340 and donor NEB 10beta E. coli strains containing the pBBR1MCS-2-based vectors with Variovorax IAA degradation locus gene inserts were cultured in LB media containing 50 µg ml−1 kanamycin at 37 °C. V. paradoxus CL14 ΔHS33 was grown in 2xYT medium containing 100 µg ml−1 ampicillin at 28 °C. V. paradoxus CL14 wild type and derivative strains such as ΔHS33 are naturally resistant to ampicillin and this ampicillin selection allows for recovery of only Variovorax from the conjugation reaction. To prepare for conjugation, all bacteria were pelleted by centrifugation at 5,000 × g for 5 min and washed 3 times in 2xYT medium without antibiotics. For each conjugation reaction, equal volumes (100–300 µl) of each of the three washed bacteria: recipient V. paradoxus CL14 ΔHS33, donor NEB 10beta E. coli containing a pBBR1MCS-2-based vector, and helper E. coli pRK2013 were mixed. Control conjugation mixtures of each pair of strains and individual strains alone were performed in parallel to ensure successful selection of exconjugants only from mixtures of all three strains together. Conjugation mixtures were pelleted by centrifugation at 5,000 × g for 5 min, resuspended in 50 µl 2xYT media, transferred to LB media plates without antibiotics and allowed to dry in a laminar flow hood. These conjugation plates were incubated overnight at 28 °C. After 18–24 h, exconjugants were selected by streaking from the pooled conjugation mixtures on the LB plate without antibiotics to LB plates containing 50 µg ml−1 kanamycin and 100 µg ml−1 ampicillin. This selects for only V. paradoxus CL14 ΔHS33 (ampicillin resistant) containing the pBBR1MCS-2-based vector (kanamycin resistant). Individual colonies were picked into and subsequently cultured in 2xYT medium containing 50 µg ml−1 kanamycin and 100 µg ml−1 ampicillin at 28 °C.Construction of V. paradoxus CL14 gene deletionsUnmarked gene deletions in V. paradoxus CL14 were constructed as described previously14 using the suicide vector backbone pMo130 originally developed for gene knockouts in Burkholderia spp.41. Primers JMC203 and JMC204 (Supplementary Table 8) were used to amplify the pMO130 vector backbone by PCR. This product was subsequently treated with DpnI (New England Biolabs) to digest circular template DNA. Primers JMC605 through JMC612 and JMC671 through JMC677 (Supplementary Table 8) were used to amplify flanking regions for the gene deletion targets from V. paradoxus CL14 genomic DNA. All PCR was performed using Q5 DNA polymerase (New England Biolabs) and products were cleaned up, as appropriate, with the QIAquick PCR purification kit (Qiagen). These PCR products were assembled into suicide vectors using HiFi Gibson Assembly Mastermix (New England Biolabs), transformed into chemically competent NEB 5alpha E. coli (New England Biolabs), and selected on LB plates with 50 µg ml−1 kanamycin. Vectors were miniprepped using the ZR plasmid miniprep classic kit (Zymo Research) and confirmed by Sanger sequencing (Genewiz). Confirmed vectors were transformed into the chemically competent bi-parental mating strain E. coli WM3064. Transformants were selected at 37 °C on LB media supplemented with 50 µg ml−1 kanamycin and 0.3 mM diaminopimelic acid (DAP), and single colonies picked into LB medium also with 50 µg ml−1 kanamycin and 0.3 mM DAP.Bi-parental mating was performed by growing E. coli WM3064 containing the appropriate suicide vector as described above, and V. paradoxus CL14 was grown in 2xYT medium containing 100 µg ml−1 ampicillin at 28 °C. Both E. coli and Variovorax were washed separately three times using 2xYT medium, then mixed in a 1:1 ratio and pelleted. All centrifugation steps were performed at 5,000 × g for 5 min. The pelleted conjugation mixtures were resuspended in 1/10 the volume of 2xYT, plated on LB agar with 0.3 mM DAP and grown at 28 °C overnight. Exconjugants from these plates were streaked out and grown on LB agar with 100 µg ml−1 ampicillin, 50 µg ml−1 kanamycin, and no DAP at 28 °C. These strains were purified by streaking and growing on plates of the same medium once more. These strains with suicide vector integration were then grown once in liquid LB containing 100 µg ml−1 ampicillin and 1 mM isopropyl 1-thio-d-galactopyranoside (IPTG) at 28 °C and then streaked on plates containing media with 10 g l−1 tryptone, 5 g l−1 yeast extract, 100 g l−1 sucrose, 1.5% agar, 100 µg ml−1 ampicillin and 1 mM IPTG. Colonies from these plates were picked and grown in the same liquid media. These strains were then assessed for gene deletion by PCR using primers JMC657 through JMC660 and JMC697 through JMC699 (Supplementary Table 8). The Quick-DNA miniprep kit (Zymo Research) was used to isolate all genomic DNA for PCR screening. To purify the knockout strains, they were streaked and grown out three times on LB plates containing 100 µg ml−1 ampicillin before a final PCR confirmation. To check the purity of the final strains, PCR was performed with one primer outside the deletion region and one inside the deleted gene to ensure no product is produced for the knockout strain. The sequences for the primers used for this PCR reaction (JMC691, JMC717, JMC718, JMC693 and JMC694) can be found in Supplementary Table 8.Measurement of bacterial growth and IAA degradationIndividual strains were grown in 5 ml cultures in various media types supplemented with IAA at 28 °C and 250 r.p.m. To screen the V. paradoxus CL14 ΔHS33 pBBR1 vector complemented mutants, 2xYT medium supplemented with 0.1 mg ml−1 IAA was used. For comparison of other V. paradoxus CL14 mutants, M9 medium with 15 mM succinate and 0.1 mg ml−1 (0.57 mM) IAA was used. For comparison of IAA-degrading strains from diverse genera, M9 medium with 0.1% (w/v) casamino acids (Bacto) and 0.1 mg ml−1 IAA was used. For R. pomeroyi, 2% (w/v) sea salts were added to this M9 medium with casamino acids and IAA. The pBBR1 vector library in E. coli NEB 10beta was screened in LB medium supplemented with 0.04 mg ml−1 IAA and grown at 37 °C and 250 r.p.m. For all media types, IAA was solubilized in 100% ethanol at 20 mg ml−1 and diluted to 0.1 mg ml−1 in the media, resulting in 0.5% (v/v) ethanol in the media.To measure growth, a 200 µl sample was taken from the growing cultures and OD600 was determined on an Infinite M200 Pro plate reader (Tecan). Subsequently, cells were pelleted by centrifugation at 4,200 × g for 15 min and 50 µl of supernatant was transferred to a new 96-well plate and frozen at −80 °C until further analysis. IAA degradation was determined by thawing the plates containing 50 µl aliquots of culture supernatant and combining this with 100 µl of Salkowski reagent (10 mM ferric chloride and 35% perchloric acid)42. This was performed alongside mixing 50 µl of IAA standards with 100 µl of Salkowski reagent in the same 96-well plate format. Colour development was allowed to proceed for 1 h and absorbance was read at 530 nm on the Infinite M200 Pro plate reader (Tecan). The absorbances measured were converted to IAA concentration on the basis of the absorbances measured for the IAA standards.Liquid Chromatography Dual Mass Spectroscopy (LC–MS/MS) metabolomics on Variovorax IAA degradationV. paradoxus CL14 was grown in 5 ml cultures of M9 minimal medium supplemented with either 0.1 mg ml−1 IAA, 0.1 mg ml−1 13C6-IAA (with the 6 carbons of the benzene ring of the indole labelled, Cambridge Isotope Laboratories CLM-1896-PK), and/or 15 mM succinate. Cultures and parallel media controls were incubated at 28 °C with shaking at 250 r.p.m. Cultures and media controls were centrifuged (4,200 × g for 15 min at 4 °C) to pellet cells; supernatants were transferred to new tubes and both pellets (intracellular fraction) and supernatants (extracellular fraction) were stored frozen at −80 °C until extraction. All subsequent work was performed over dry ice or in chilled cold blocks. Frozen pellets from the intracellular fraction were thawed for 3 h at 4 °C, then 800 µl of cold LCMS-grade water was added to the pellets with repeated pipetting to break up the pellet until visually homogeneous. Samples were then quickly returned to −80 °C to freeze the suspension. Frozen pellet suspensions and extracellular solutions were lyophilised until dry. The cells from the dried pellet suspensions were lysed and homogenized with a bead mill (BioSpec Mini-Beadbeater-96) using one sterile 3.2 mm steel ball in each tube for 3 rounds of 5 s each with 10 s breaks in between to reduce heat production. Dried extracellular samples were concentrated by resuspension in 100 µl LCMS- grade methanol, vortexed 3 times for 10 s each, water bath sonicated for 20 min, incubated at 4 °C overnight, centrifuged (1,000 × g, 4 °C, 5 min), and the methanol supernatant was dried using a speed vacuum concentrator. On the day of LC–MS/MS analysis, homogenized dry material was suspended in LCMS-grade methanol with internal standard mix (100 µM U-13C/15N-labelled amino acids, SIGMA 767964). Intracellular samples were suspended at 11.1 µl mg−1 of original sample cell pellet wet weight; extracellular samples were suspended at 38.9 µl mg−1 of corresponding cell pellet wet weight from the culture. The solutions were vortexed 3 times for 10 s each, bath sonicated in ice water for 10 min, chilled at −20 °C for 10 min, then centrifuged (10,000 × g, 5 min, 10 °C) to pellet insoluble material. Supernatants containing the methanol extracts were filtered through 0.22 µm PVDF microcentrifuge filtration tubes (10,000 × g, 5 min, 10 °C); filtrates were transferred to glass vials and immediately capped. Filtrates were then analysed by LC–MS/MS using an Agilent 1290 UHPLC system connected to a Thermo Q Exactive Hybrid Quadrupole-Orbitrap mass spectrometer equipped with a heated electrospray ionization (HESI-II) source probe. Extracts were chromatographically separated on a ZORBAX RRHD Eclipse Plus C18, 95 Å, 2.1 × 50 mm, 1.8 µm column (Agilent) for non-polar metabolomics. Separation, ionization, fragmentation and data acquisition parameters are specified in Supplementary Table 7. Briefly, metabolites were separated by gradient elution followed by MS1 and data-dependent (top 2 most abundant MS1 ions not previously fragmented in last 7 s) MS2 collection; targeted data analysis was performed by comparison of sample peaks to a library of analytical standards analysed under the same conditions or by searching the raw data files for predicted m/z values based on structural information of compounds of interest. Three parameters were compared: matching m/z, retention time and fragmentation spectra using Metabolite Atlas (https://github.com/biorack/metatlas)43,44. Identification and standard reference comparison details are provided in Supplementary Table 6. Raw and processed data are available for download at the JGI Joint Genome Portal under ID 1340427. Statistical comparisons were performed using R version 3.6.2, using package agricolae 1.3–5 and stats 3.6.245; boxplots were generated with base R graphics using the boxplot function.Phylogenomic analysisTo guide the delineation of the IAA degradation operons across the bacterial tree of life, we constructed two Hidden Markov Model (HMM) profiles of the genes iacC and iacD by subsetting all homologous genes from the previously validated operons (Extended Data Fig. 4). In parallel, we downloaded the assembly files for all available complete genomes deposited in the NCBI RefSeq 202 repository46. For the 220,000 assembly files downloaded, we performed open reading frame (ORF) prediction using prodigal. We then used the two HMM profiles described above to query the predicted ORFs. Utilizing ad hoc scripts, we constructed a table of HMM hits along the genomes scanned and subset genomic loci where both iacC and iacD genes appeared adjacent to one another. The logic of using the iacC and iacD genes as anchor genes for our search is that the adjacent physical location of both iacC and iacD homologues is a conserved feature across all previously experimentally validated IAA-degrading operons (Extended Data Fig. 4). Next, for each region containing the adjacent iacC and iacD homologue genes, we extracted the gene neighbourhood adjacent to the anchor hit by extracting the amino acid sequence of ORFs +10 kb and −10kb with respect to the anchor hit. Using hmmscan from the Hmmer v3.1.b2 suite47, we performed HMM profiling in all ORFs extracted via our neighbourhood delineation against the COG database version 2003. Finally, we used the COG profiles across the neighbourhoods to create a matrix describing the prevalence of COGs across the regions (candidate regions) with the adjacent iacC and iacD homologue genes.For each genome containing at least one candidate region, we performed taxonomic classification using the GTDB database48. Due to the size of our estimated genomic matrix and to reduce potential biases due to over-representation of certain lineages within RefSeq, we performed principal coordinate analysis (PCoA) using a reduced matrix containing one representative candidate region per species. Species labelling was obtained from the GTDB taxonomic classification described above. PCoA was performed using the oh.pco function from the ohchibi package49, taking as input a binary version of the gene matrix described above. We classified candidate reads into the two types of IAA-degrading operon (iac-like and iad-like), utilizing a majority count-based approach using marker COGs conserved between the previously experimentally validated IAA-degrading operons (Extended Data Fig. 4). Specifically, for each potential operon, we determined the prevalence of COGs that a priori (Extended Data Fig. 4) showed differential prevalence across the two degrading operons (for example, iacA, iacB and iacI are markers of the iac operon, while iorB/iadB and iotA/iadA are exclusive markers of the iad-like operon). Hybrid gene clusters were defined as operons that exhibited the hallmark COGs of both operons.In parallel, we performed phylogenetic inference over all the genomes belonging to genera with at least one representative strain harbouring any of the two types of IAA-degrading operon. This phylogenetic tree was constructed using a super-matrix-based approach as previously described35. Finally, for each genus with at least one assembly harbouring a positive IAA-degrading operon, we estimated the prevalence of the trait across the genus by dividing the total number of isolates with detectable IAA degradation locus by the total number of isolates belonging to that genus in the dataset. In addition, to see the phylogenetic evenness of the distribution of the IAA degradation trait across each genus, we calculated the phylogenetic ratio by calculating the ratio between the average phylogenetic distance (computed via the cophenetic.phylo function from the ape R package50) of isolates with a detectable IAA degradation locus and the total average phylogenetic distance of all isolates within that genus. We constructed the MarR phylogeny using the MarR sequences from candidate regions with 100% markers of one of the two types of IAA-degrading operon. Amino acid sequences of the MarR homologues were aligned using MAFFT51 and phylogenetic inference was performed using FastTree 252.RNA-seq on Variovorax strainsV. paradoxus CL14 was grown in 5 ml cultures of M9 minimal medium supplemented with 15 mM succinate and 0.5% (v/v) ethanol alone or containing IAA. IAA was at a final concentration of 0.1 mg ml−1 in the medium to which it was added. Cultures were prepared at a starting OD600 of 0.02 and incubated at 28 °C, shaking at 250 r.p.m. Cells from all samples were collected for RNA-seq at 18 h to ensure IAA was still present in the cultures of strains that degraded IAA most rapidly. Cells were pelleted by centrifuging the culture at 4,200 × g for 15 min and removing the supernatant. Cell pellets were frozen at −80 °C before RNA extraction. To extract RNA, cells were lysed in TRIzol reagent (Invitrogen) according to manufacturer instructions for lysis and phase separation. After these steps, RNA was purified from the aqueous phase using the RNeasy mini kit (Qiagen) including the optional on-column DNase digestion with RNase-free DNase set (Qiagen). Total RNA was quantified using the Qubit 2.0 fluorometer (Invitrogen) and RNA-seq libraries were prepared using the Universal Prokaryotic RNA-Seq Prokaryotic AnyDeplete kit (Tecan) according to manufacturer instructions. The resulting libraries were pooled and sequenced on the Illumina HiSeq4000 to generate 50 bp single-end reads.RNA-seq data analysisThe V. paradoxus CL14 RNA-seq sequence data were analysed as described previously14. Briefly, the raw reads were mapped to the V. paradoxus CL14 genome (fasta file available at https://github.com/isaisg/variovoraxRGI/blob/master/rawdata/2643221508.fna) using bowtie253 with the ‘very sensitive’ flag. Hits to each individual coding sequence were counted and annotated using the function featureCounts from the R package Rsubread54, inputting the V. paradoxus CL14 gff file (available at https://github.com/isaisg/variovoraxRGI/blob/master/rawdata/2643221508.gff) and using the default parameters with the flag allowMultiOverlap = FALSE. Finally, DESeq255 was used to estimate Differentially Expressed Genes (DEGs) between treatments, with the corresponding fold-change estimates and False Discovery Rate (FDR) adjusted P values. For visualization purposes, we performed z-score standardization of each gene across samples and we visualized this standardized expression values utilizing a heat map constructed using ggplot256. These data can be found in Supplementary Table 9.MarR protein expression and purificationThe coding sequence for each gene can be found in Supplementary Table 10. MarR expression plasmids were synthesized as codon-optimized genes for E. coli expression by BioBasic in the pLIC-His N-term vector (pMCSG7) and transformed into E. coli BL21 (DE3) Gold cells for expression. Cells were grown in the presence of ampicillin in LB medium with shaking at 225 r.p.m. at 37 °C to an OD600 of 0.5, at which point the temperature was reduced to 18 °C. At an OD600 of 0.8, protein expression was induced by the addition of 0.1 mM IPTG and incubation continued overnight. Cells were collected by centrifugation at 4,500 × g for 20 min at 4 °C in a Sorvall (model RC-3B) swinging bucket centrifuge. Cell pellets were resuspended in buffer A (20 mM potassium phosphate, pH 7.4, 50 mM imidazole, 500 mM NaCl), DNase, lysozyme and a Roche Complete EDTA-free protease inhibitor tablet. Resuspended cells were sonicated and clarified via centrifugation at 17,000 × g for 60 min in a Sorvall (model RC-5B) swinging bucket centrifuge. The lysate was applied to a nickel-nitrilotriacetic acid HP column (GE Healthcare) on an Aktaxpress Fast Performance Liquid Chromatography (FPLC) system (Amersham Bioscience) and washed with buffer A. Protein was eluted with buffer B (20 mM potassium phosphate, pH 7.4, 500 mM imidazole, 500 mM NaCl). Fractions containing the protein of interest were combined and passed over a HiLoadTM 16/60 SuperdexTM 200 gel filtration column. Proteins were eluted in S200 buffer (20 mM HEPES, pH 7.4, 300 mM NaCl). Fractions were combined and concentrated for long-term storage at −80 °C.MarR mutant proteins were created by site-directed mutagenesis using primers from Integrated DNA Technologies. The mutant plasmids were sequenced to confirm the mutations. The mutants were produced and purified using E. coli BL21 (DE3) Gold as described above.Ligand binding studies by isothermal titration calorimetry (ITC)All ITC measurements were performed at 25 °C using an Auto-ITC200 microcalorimeter (MicroCal/GE Healthcare). The buffer employed was 20 mM HEPES, pH 7.4, 50 mM NaCl and 0.5% dimethly sulfoxide (DMSO) for protein/ligand binding and 20 mM HEPES, pH 7.4 and 300 mM NaCl for DNA/protein binding experiments. For ligand binding experiments, the calorimetry cell (volume 200 ml) was loaded with MarR wild-type, mutant or homologue protein at a concentration of 50 μM. The syringe was loaded with a ligand concentration of 0.5 or 2 mM. For DNA binding experiments, wild-type MarR_73 did not bind any of the DNA oligos examined; however, we hypothesized that this arose from the ability of this native receptor to remain bound to ligands retained from its recombinant expression in E. coli. Thus, we employed the MarR_73 S28A protein with reduced ligand binding capacity. Here, the calorimetry cell was loaded with duplex oligo at a concentration of 25 μM and the syringe was loaded with MarR S28A mutant protein, which was necessary to prevent ligand binding during expression and purification, at a concentration of 0.5 mM. A typical injection protocol included a single 0.2 μl first injection followed by 20 1.5 μl injections of the syringe sample into the calorimetry cell. The spacing between injections was kept at 180 s and the reference power at 8 μcal s−1. The data were analysed using Origin for ITC version 7.0 software supplied by the manufacturer and fit well to a one-site binding model. Two independent ITC measurements were performed for each condition. A non-integer N value (for example, 0.73 in Fig. 2a) indicates that some protein monomers may not be in an active conformation, and thus do not bind ligand. Additionally, small measurement errors in assessing the protein or ligand concentrations may also contribute to non-integer N values in ITC. To confirm that 300 mM NaCl did not negatively impact DNA binding, MarR_73 S28A was examined by ITC in 150 mM NaCl. In this condition, the KD for the 22 bp duplex was 0.428 ± 0.002 μM (N = 1.75 ± 0.014), while the KD for 24 bp duplex was 0.151 ± 0.025 μM (N = 2.51 ± 0.26).Protein crystallographyV. paradoxus MarR_73 was crystallized using the sitting drop vapour diffusion method at 20 °C in conditions outlined in Supplementary Table 4. Crystallization drops were set up using the Oryx4 protein crystallization robot (Douglas Instruments) and contained 0.15 μl protein and 0.15 μl well solution. For all V. paradoxus MarR_73 wild-type conditions, ligands were added at 10-fold molar excess before crystallization trials and crystals appeared within 2–5 d. V. paradoxus MarR_73 with the S28A and R46A mutations was crystallized in similar conditions as the wild-type protein. Similarly, P. putida MarR_iacR, B. japonicum MarR_Bj1, A. baumannii MarR_Ab and E. soli MarR_Es were crystallized using vapour diffusion methods in sitting drop trays at 20 °C and crystals appeared within 3–5 d. All crystallization conditions are outlined in Supplementary Table 4. Crystal specimens were cryoprotected with the well solution supplemented with glycerol to 20% (v/v) (Supplementary Table 4). X-ray diffraction data were collected at the Advanced Photon Source beamline 23-ID-D (Supplementary Table 3). Diffraction images were reduced using either XDS or Denzo and scaled with either Aimless or Scalepack57,58,59. The V. paradoxus MarR_73 structure in complex with IAA was determined by molecular replacement using the structure of 3CDH as a search model in Phaser60. All subsequent structures of V. paradoxus MarR_73 were determined using the V. paradoxus MarR_73 IAA complex structure (PDB: 7KFO) as a search model. The P. putida MarR_iacR and B. japonicum MarR_Bj1 structures were determined by molecular replacement using the structure of 3CJN as the search model. P. putida MarR_iacR (PDB: 7KUA) was subsequently used as the search model for molecular replacement to solve A. baumannii MarR_Ab and E. soli MarR_Es. A nickel ion was placed in the model of MarR_Ab. The following ions or molecules were examined and refined in this location in the MarR_Ab structure: water, Na, Mg, K, Ca, Mn, Fe, Co, Ni, Cu, Zn and Ba. Water, Na, Mg, K, Ca and Ba were deemed unacceptable in this site due to poor difference density. Of the remaining ions considered, there were no sources of Mn, Fe, Co, Cu or Zn in the protein expression media, protein purification buffers, protein storage buffer, crystallization condition or cryoprotectant solutions. Thus, we concluded that the ion present in this structure is Ni due to the use of a nickel-affinity column during the protein’s purification. It is unclear why this ion remained bound to MarR_Ab even after the subsequent size exclusion chromatography purification step, or why such an ion is only observed in this structure of the proteins examined. All structures were refined with either Phenix.refine or Refmac using iterative model building in Coot to the final parameters outlined in Supplementary Table 361,62. MarR_73 is a dimer with one protein monomer in the asymmetric unit and the dimer generated by crystallographic symmetry. PDB accession codes and associated crystallographic data are reported in Supplementary Table 3.Statistics and reproducibilityNo statistical method was used to predetermine sample size, but our sample sizes are similar to those reported in previous publications14,63,64. No data were excluded from the analyses. The experiments were randomized. The investigators were not blinded to allocation during experiments and outcome assessment. Where not stated, data distribution was assumed to be normal, but this was not formally tested.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More

Field samplingLake Masoko fish were chased into fixed gill nets and SCUBA by a team of professional divers at different target depths determined by diver depth gauge (12× male benthic, 12× male littoral). Riverine fish (11× Mbaka River and 1× Itupi river) were collected by local fishermen. On collection, all fish were euthanized using clove oil. Collection of wild fish was done in accordance with local regulations and permits in 2015, 2016, 2018 and 2019. On collection, fish were immediately photographed with color and metric scales, and tissues were dissected and stored in RNAlater (Sigma-Aldrich); some samples were first stored in ethanol. Only male specimens (showing bright nuptial coloration) were used in this study for the practical reason of avoiding any misassignment of individuals to the wrong population (only male individuals show clear differences in phenotypes and could therefore be reliably assigned to a population). Furthermore, we assumed that any epigenetic divergence relevant to speciation should be contributing to between-population differences in traits possessed by both sexes (habitat occupancy, diet). To investigate the role of epigenetics in phenotypic diversification and adaptation to different diets, homogenized liver tissue – a largely homogenous and key organ involved in dietary metabolism, hormone production and hematopoiesis – was used for all RNA-seq and WGBS experiments.Common-garden experimentCommon-garden fish were bred from wild-caught fish specimens, collected and imported at the same time by a team of professional aquarium fish collectors according to approved veterinary regulations of the University of Bangor, UK. Wild-caught fish were acclimatized to laboratory tanks and reared to produce first-generation (G1) common-garden fish, which were reared under the same controlled laboratory conditions in separate tanks (light–dark cycles, diet: algae flakes daily, 2–3 times weekly frozen diet) for approximately 6 months (post hatching). G1 adult males showing bright nuptial colors were culled at the same biological stages (6 months post hatching) using MS222 in accordance with the veterinary regulations of the University of Bangor, UK. Immediately on culling, fish were photographed and tissues collected and snap-frozen in tubes.Stable isotopesTo assess dietary/nutritional profiles in the three ecomorph populations, carbon (δ13C) and nitrogen (δ15N) isotope analysis of muscle samples (for the same individuals as RRBS; 12, 12 and 9 samples for benthic, littoral and riverine populations, respectively) was undertaken by elemental analyzer isotope ratio mass spectrometry by Iso-Analytical Limited. It is important to note that stable isotope analysis does not depend on the use of the same tissue as the ones used for the RRBS/WGBS samples45. Normality tests (Shapiro–Wilk, using the R package rstatix v.0.7.0), robust for small sample sizes, were performed to assess sample deviation from a Gaussian distribution. Levene’s test for homogeneity of variance was then performed (R package carData v.3.0-5) to test for homogeneity of variance across groups. Finally, Welch’s ANOVA was performed followed by Games–Howell all-pairs comparison tests with adjusted P value using Tukey’s method (rstatix v.0.7.0). Mean differences in isotope measurements and 95% CI mean differences were calculated using Dabestr v.0.3.0 with 5,000 bootstrapped resampling.Throughout this manuscript, all box plots are defined as follows: centre line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range; points, outliers.RNA-seqNext-generation sequencing library preparationTotal RNA from liver tissues stored in RNAlater was extracted using a phenol/chloroform approach (TRIzol reagent; Sigma-Aldrich). Of note, when tissues for bisulphite sequencing samples were not available, additional wild-caught samples were used (Supplementary Table 3). The quality and quantity of RNA extraction were assessed using TapeStation (Agilent Technologies), Qubit and NanoDrop (Thermo Fisher Scientific). Next-generation sequencing (NGS) libraries were prepared using poly(A) tail-isolated RNA fraction and sequenced on a NovaSeq system (S4; paired-end 100/150 bp; Supplementary Table 3), yielding on average 32.9 ± 3.9 Mio reads.Read alignment and differential gene expression analysisAdaptor sequence in reads, low-quality bases (Phred score  More

Data collectionIn phase 1 (2017), we first circulated a comprehensive multi-language questionnaire and associated guidelines on protected area personnel numbers to major national protected area agencies, focusing on the 50 countries listed in the WDPA as having the most protected areas. The questionnaire requested information on personnel numbers, type of employers and management levels (from executive to skilled practical workers). Protected area personnel were defined as those spending at least 50% of their work time on protected area-related tasks. The questionnaire also requested information about job titles used for personnel equivalent to rangers. This phase produced usable data for 28 countries/territories.In phase 2 (2018 onwards), we conducted online searches for published data on protected area personnel numbers in the countries/territories not included in the questionnaire survey or where questionnaire responses were incomplete or unclear. The resulting information came from official organizational reports (10 countries/territories), published external studies, project documents and journal papers (35 countries/territories) and websites of protected area organizations or individual sites (9 countries/territories).In phase 3 (2018–2021), we directly requested personal contacts to locate or supply information from official sources both for the remaining countries/territories and to improve or verify data from phases 1 and 2. The minimum data requested were the overall number of protected area personnel, the number of those personnel that could be categorized as rangers, the terrestrial area of protected areas managed by the listed personnel and the source of the information. This phase contributed usable data for 68 countries and territories. Data for a further 17 countries/territories were assembled from multiple sources.The final dataset covered 176 countries/territories: 167 surveyed countries/territories and a further 9 countries/territories that have no WDPA-listed protected areas (Supplementary Table 1), with contributions from more than 150 individuals.Initial data processingTo assess and, where necessary, improve the reliability of data obtained in a wide range of formats and levels of detail and from multiple sources, we scored the data for each country/territory from 0 to 5 for each of four criteria—detail, accuracy, source and age of the data—with a maximum score of 20 (Supplementary Table 1 and Supplementary Fig. 1). For all low-scoring records (a score of less than 15), we sought more-reliable sources in later phases of the study, rejecting any final scores of less than 10.On reviewing the data, we excluded from the analysis protected areas identified in the WDPA as predominantly or entirely marine, Antarctica and countries/territories categorized in the WDPA as polar (Greenland, French Southern Territories, Bouvet Island, Heard Island and McDonald Islands, South Georgia and the South Sandwich Islands). These large, remote and/or largely uninhabited areas are likely to have quite different management models and scales of staffing from terrestrial protected areas (although marine protected areas are also widely understaffed11). For example, in 2012 the 972,000 km2 of Northeast Greenland Protected Area (categorized by the WDPA as polar) was only periodically visited by six two-person teams of naval personnel47, and the 2008 management plan of the 1.51 million km2 Papahānaumokuākea Marine National Monument (Hawai’i, USA) specifies just nine personnel, working in conjunction with several other agencies48. Data for one country were supplied by officials on the agreement that the country was not specifically identified in publications (the country is given the three-letter code ZZZ in relevant tables and figures).Because the format, completeness and level of detail of the data varied widely, from comprehensive personnel lists to single figures, we restricted our raw dataset to six variables that could be consistently extracted from data obtained for each country/territory:

1.

Total number of non-ranger personnel (if known)

2.

Total number of rangers (if known)

3.

Total number of protected area personnel (either the sum of 1 and 2 or provided as an undifferentiated total)

4.

Terrestrial area of protected areas covered by surveyed personnel (km2)

5.

Total terrestrial area of protected areas of the country/territory (km2)

6.

Year of the data

We used the WDPA, official publications and websites to determine (or verify) the area of terrestrial protected areas covered by the personnel listed for each country/territory, using WDPA data if there were discrepancies. Total national terrestrial protected area coverage was taken from the WDPA, with the exception of Turkey, where the area officially reported to the WDPA is significantly less than the nationally published area.The raw data from the survey are shown in Supplementary Table 1.Candidate predictorsTo predict the number of rangers and non-rangers in countries and territories for which we had no data (Statistical analysis), we collected information on the following set of variables, hereafter referred to as candidate predictors:Location dataThe WGS84 latitude and longitude of the centroid of the largest land mass associated with each country/ territory (to obtain the polygons defining the land masses, we used the R package rnaturalearth version 0.1.0; https://github.com/ropensci/rnaturalearth)2020 data from the World Bank (https://data.worldbank.org/indicator)

Area of the country/territory

Population density: the mid-year population divided by land area

Gross domestic product (GDP) in US dollars

GDP per capita in US dollars (GDP divided by mid-year population)

Growth rate of GDP

The proportion of rural inhabitants

The proportion of unemployed inhabitants

The forested proportion of the country/territory

2020 data for each country/territory from the WDPA (https://www.protectedplanet.net/)

The total terrestrial area of WDPA-listed protected areas

The proportion of the terrestrial area of all IUCN-categorized protected areas (Categories I–VI) that falls within protected areas in Category I or II

The proportion of the terrestrial area of all IUCN-categorized protected areas (Categories I–VI) that falls within protected areas in Categories I–IV

2020 data from the Yale Center for Environmental Law and Policy Environmental Performance Index (https://epi.yale.edu/)

Environmental Performance Index (EPI): a composite index using 32 performance indicators across 11 categories

Ecosystem Vitality Index (EVI): an indicator of how well countries preserve, protect and enhance ecosystems and the services they provide

Species Protection Index (SPI): an indicator of the species-level ecological representativeness of each country’s/territory’s protected area network

Not all this information was available for all countries/territories. Most of the missing data were for small territories that account for only a very small proportion of the total area of protected areas worldwide (Supplementary Table 2c).Statistical analysisOur primary objective was to estimate the total number of all personnel engaged in managing all the world’s WDPA-listed terrestrial protected areas and the number categorized as rangers. Our raw data collection yielded full, partial or no information on total personnel and ranger numbers for each country/territory (Supplementary Table 1 shows the completeness of all the data collected). Our first task, therefore, was (1) to impute the information for unsurveyed protected areas on the basis of information from surveyed protected areas within the same countries/territories and (2) to predict those numbers for countries/territories where no information was available on overall personnel numbers and/or ranger numbers on the basis of relationships we could establish between available information and candidate predictors in other countries/territories (Supplementary Table 7). A brief description of these two approaches follows, and full details on the analysis are provided in Supplementary Information.Data imputationFor countries/territories where we had obtained information about numbers of personnel and/or rangers for only some protected areas, our strategy was to populate the unsurveyed protected areas in proportion to the densities of personnel or rangers from the surveyed protected areas of the same countries/territories. For example, for Spain we obtained evidence that there are 619 rangers responsible for protected areas covering 44,328 km2, out of a national total protected area system covering 142,573 km2. To impute the number of rangers for the remaining 98,245 km2, we used the density of rangers in the surveyed area (one ranger per 44,328/619 = 71.6 km2) and applied that to the unsurveyed area, giving a total of 1,991 rangers (619 + (98,245/71.6)). This imputation assumes that unsurveyed areas are staffed at the same density as surveyed areas, whereas in reality the relative densities are likely to vary in unknown ways within different countries/territories. To study the sensitivity of our results to the assumed proportion, we repeated our analysis using the following proportions of the observed densities: 0, 0.25, 0.50, 0.75 and 1.00. This provided a range of personnel numbers from a minimum (based on a proportion of 0) to a presumed maximum (based on a proportion of 1.00). From the data obtained, it was not possible to calculate the actual proportions, but based on the experience of the practitioners in the author team, the unsurveyed areas are highly unlikely to be staffed at higher densities than surveyed areas and, on average, are very likely to be staffed at lower densities. After all, most survey respondents were national or subnational agencies responsible for protected areas subject to stronger formal requirements for protection and management and therefore likely to have larger workforces. Unsurveyed protected areas are more likely to be managed by local entities, with fewer resources, less-stringent management obligations and therefore fewer personnel. The range of proportions we considered to populate unsurveyed areas should therefore yield predictions encompassing the actual (unknown) numbers of rangers and non-rangers with a conservative margin of error. In the main text, we have reported the results of imputation assuming a proportion of 1, which is probably the most optimistic assessment of the current workforce in protected areas within the proportions of the observed densities considered. Results using lower proportions are shown in Extended Data Fig. 2 and Supplementary Tables 4 and 5.Data predictionOur imputation approach was not possible for countries/territories where (1) zero ranger or personnel data had been obtained and (2) specific data had not been obtained that allowed imputation either for rangers or for total personnel (where only total personnel numbers or only ranger numbers had been obtained). To predict the missing information, we used two different statistical approaches: linear mixed models (LMMs)49 and a general implementation of random forests, which we term RF/ETs because it encompasses both random forests sensu stricto (RFs)50 and a variant called extremely randomized trees (ETs)51. LMMs and RFs have been extensively discussed and reviewed in the literature49,52,53. We adopted these approaches because both have proved successful in producing accurate predictions for a wide range of applications and because both are well suited to our data since they both produce predictions from a set of predictors and allow for the consideration of spatial effects54,55. Furthermore, comparing predictions generated through very different methods informs us about the robustness of our results with respect to key statistical assumptions. LMMs come from the ‘data modelling culture’56 and belong to parametric statistics; RF/ETs come from the ‘algorithmic modelling culture’ and belong to non-parametric statistics.We followed the same workflow for both statistical approaches, comprising eight steps: (1) general data preparation; (2) preparation of initial training datasets; (3) selection of predictor variables and of the method used for handling spatial autocorrelation; (4) preparation of final training datasets; (5) fine tuning; (6) final training; (7) preparation of datasets for predictions and simulations; and (8) predictions and simulations (see Supplementary Information for details).Both approaches yielded very similar results with our data. We chose to present the LMM results in the main text, but we provide and compare the results obtained by both approaches in Supplementary Information.SoftwareWe performed all the data analyses using the free open-source statistical software R version 4.157. We used the R package spaMM version 3.9.13 to implement LMMs58 and the R package ranger version 0.13.1 to implement RF/ETs59. To reformat and plot the data, we used the Tidyverse suite of packages60. Details are provided in an R package we specifically developed so that findings presented in this paper can readily be reproduced (see Code availability). Using a workstation with an AMD Ryzen Threadripper 3990 × 64-core processor and 256 GB of RAM, our complete workflow ran in ~3,000 CPU hours.Estimation of required numbers and densities of personnelTo estimate the numbers of personnel and rangers required for effective management of existing protected areas, we referred to ref. 25. This estimates that the minimum budget needed to adequately manage the existing protected area system is US$67.6 billion per year and that current annual expenditure is US$24.3 billion. From these figures, we can calculate that resources invested in the current global system of protected areas are approximately 36% of what is required. We consulted data from https://ourworldindata.org to determine that the proportion of global public expenditure on employee compensation has remained between 21.01% and 23.33% in the years from 2006 to 2019. We obtained these figures from the ‘Government Spending’ section of the site, consulting the chart ‘Share of employee compensation in public spending, 2002 to 2019’ and selecting data for ‘World’. On the basis of this broadly constant proportion and the assumption that total employee compensation is an indicator of total employee numbers, we inferred that current numbers of protected area employees are also around 36% of what is required. We therefore multiplied our estimations of personnel and ranger numbers by 1/0.36 and recalculated the densities on this basis (current requirement = 1/0.36 × current estimate).To estimate staffing requirements for 30% global coverage of protected areas—the global target intended to be reached by 2030—we used the mean personnel and ranger densities calculated as being required at present to ‘populate’ a global area of terrestrial protected areas if increased from the percentage at the time of our study (15.7%) to 30% (current requirement × (0.300/0.157)).Economic calculationsWe based our calculations on published data from 202025, which estimate that expanding the protected areas to 30% would generate higher overall output (revenues) than non-expansion (an extra US$64–454 billion per year by 2050). This figure is only an indicative, partial estimate, generated for the purposes of comparison and to illustrate the substantial return on investment that protected area staff investments imply. Using these figures and our estimates of personnel requirements to ensure effective management of 30% coverage, we calculated the range of sums that each additional protected area staff member has the potential to generate (Supplementary Table 8). For clarity, we rounded these figures to the nearest hundred US dollars in the main text.Our estimates of the gross value added per worker in forestry and agriculture (sectors responsible for similar proportions of the world as protected areas) are included to provide a point of comparison for the figures showing the economic benefit generated per protected area personnel member (see the preceding). The data for the gross annual value of world agricultural production (US$3,550,231,736,000) and the number of workers employed in agriculture (343,527,711) come from the Food and Agriculture Organization of the United Nations30, providing an average gross value of annual agricultural production per worker of US$10,335. We adjusted these 2018 data to 2020 price levels using a deflator based on the US consumer price index (CPI) from the World Economic Outlook database61 (Supplementary Table 9). This ensures that all the economic value data we present are directly comparable for protected area, agricultural and forestry workers. We calculated the gross value of forest production per worker on the basis of direct contribution of forestry of more than US$539 billion to world GDP in 201162 and total forest-sector employment of 11.881 million full-time-equivalent jobs in 201032. These were the most up-to-date global estimates we could locate from credible sources that presented comparable estimates of forest-sector employment and contribution to GDP. This gives an average gross value of forest production per worker of US$45,367 per year. We used the same method as for agriculture to bring these figures to 2020 price levels (Supplementary Table 9). These figures are rounded to the nearest hundred US dollars in the main text. More

De Meyer, M., Robertson, M., Peterson, A. & Mansell, M. Ecological niches and potential geographical distributions of Mediterranean fruit fly (Ceratitis capitata) and Natal fruit fly (Ceratitis rosa). J. Biogeogr. 35, 270–281 (2008).
Google Scholar 
Nguyen, A. D. et al. Trade-offs in cold resistance at the northern range edge of the common woodland ant Aphaenogaster picea (Formicidae). Am. Nat. 194, E151–E163 (2019).Article 
PubMed 

Google Scholar 
Gilioli, G. et al. Non-linear physiological responses to climate change: the case of Ceratitis capitata distribution and abundance in Europe. Biol. Invasions 24, 261–279 (2022).Article 

Google Scholar 
Lancaster, L. T., Dudaniec, R. Y., Hansson, B. & Svensson, E. I. Latitudinal shift in thermal niche breadth results from thermal release during a climate-mediated range expansion. J. Biogeogr. 42, 1953–1963 (2015).Article 

Google Scholar 
Hallas, R., Schiffer, M. & Hoffmann, A. A. Clinal variation in Drosophila serrata for stress resistance and body size. Genet. Res. 79, 141–148 (2002).Article 
PubMed 

Google Scholar 
Hoffmann, A. A., Anderson, A. & Hallas, R. Opposing clines for high and low temperature resistance in Drosophila melanogaster. Ecol. Lett. 5, 614–618 (2002).Article 

Google Scholar 
Ragland, G. & Kingsolver, J. Influence of seasonal timing on thermal ecology and thermal reaction norm evolution in Wyeomyia smithii. J. Evol. Biol. 20, 2144–2153 (2007).Article 
CAS 
PubMed 

Google Scholar 
MacMillan, H. A. & Sinclair, B. J. Mechanisms underlying insect chill-coma. J. Insect Physiol. 57, 12–20 (2011).Article 
CAS 
PubMed 

Google Scholar 
Neilson, E. W. et al. There’sa storm a-coming: Ecological resilience and resistance to extreme weather events. Ecol. Evol. 10, 12147–12156 (2020).Article 
PubMed 
PubMed Central 

Google Scholar 
Overgaard, J., Hoffmann, A. A. & Kristensen, T. N. Assessing population and environmental effects on thermal resistance in Drosophila melanogaster using ecologically relevant assays. J. Therm. Biol. 36, 409–416 (2011).Article 

Google Scholar 
Maysov, A. Chill coma temperatures appear similar along a latitudinal gradient, in contrast to divergent chill coma recovery times, in two widespread ant species. J. Exp. Biol. 217, 2650–2658 (2014).Article 
PubMed 

Google Scholar 
David, R. J. et al. Cold stress tolerance in Drosophila: analysis of chill coma recovery in D. melanogaster. J. therm. biol. 23, 291–299 (1998).Article 

Google Scholar 
Overgaard, J. & MacMillan, H. A. The integrative physiology of insect chill tolerance. Annu. Rev. Physiol. 79, 187–208 (2017).Article 
CAS 
PubMed 

Google Scholar 
Andersen, M. K. & Overgaard, J. The central nervous system and muscular system play different roles for chill coma onset and recovery in insects. Comp. Biochem. Physiol. A: Mol. Integr. Physiol. 233, 10–16 (2019).Article 
CAS 

Google Scholar 
Macdonald, S., Rako, L., Batterham, P. & Hoffmann, A. Dissecting chill coma recovery as a measure of cold resistance: evidence for a biphasic response in Drosophila melanogaster. J. Insect Physiol. 50, 695–700 (2004).Article 
CAS 
PubMed 

Google Scholar 
Gibert, P., Moreteau, B., Pétavy, G., Karan, D. & David, J. R. Chill-coma tolerance, a major climatic adaptation among Drosophila species. Evolution 55, 1063–1068 (2001).Article 
CAS 
PubMed 

Google Scholar 
Ayrinhac, A. et al. Cold adaptation in geographical populations of Drosophila melanogaster: phenotypic plasticity is more important than genetic variability. Funct. Ecol. 18, 700–706 (2004).Article 

Google Scholar 
Castañeda, L. E., Lardies, M. A. & Bozinovic, F. Interpopulational variation in recovery time from chill coma along a geographic gradient: a study in the common woodlouse, Porcellio laevis. J. Insect Physiol. 51, 1346–1351 (2005).Article 
PubMed 

Google Scholar 
Tonione, M. A., Cho, S. M., Richmond, G., Irian, C. & Tsutsui, N. D. Intraspecific variation in thermal acclimation and tolerance between populations of the winter ant Prenolepis imparis. Ecol. Evol. 10, 4749–4761 (2020).Article 
PubMed 
PubMed Central 

Google Scholar 
Karl, I., Janowitz, S. A. & Fischer, K. Altitudinal life-history variation and thermal adaptation in the copper butterfly Lycaena tityrus. Oikos 117, 778–788 (2008).Article 

Google Scholar 
Ghalambor, C. K., Huey, R. B., Martin, P. R., Tewksbury, J. J. & Wang, G. Are mountain passes higher in the tropics? Janzen’s hypothesis revisited. Integr. Comp. Biol. 46, 5–17 (2006).Article 
PubMed 

Google Scholar 
Addo-Bediako, A., Chown, S. L. & Gaston, K. J. Thermal tolerance, climatic variability and latitude. In Proceedings of the Royal Society of London. Series B: Biological Sciences 267, 739–745 (2000).Poikela, N., Tyukmaeva, V., Hoikkala, A. & Kankare, M. Multiple paths to cold tolerance: the role of environmental cues, morphological traits and the circadian clock gene vrille. BMC ecol. Evol. 21, 1–20 (2021).
Google Scholar 
Andersen, J. L. et al. How to assess Drosophila cold tolerance: chill coma temperature and lower lethal temperature are the best predictors of cold distribution limits. Funct. Ecol. 29, 55–65 (2015).Article 

Google Scholar 
Papadopoulos, N., Katsoyannos, B., Carey, J. & Kouloussis, N. Seasonal and annual occurrence of the Mediterranean fruit fly (Diptera: Tephritidae) in northern Greece. Ann. Entomol. Soc. Am. 94, 41–50 (2001).Article 

Google Scholar 
Malacrida, A. et al. Globalization and fruitfly invasion and expansion: the medfly paradigm. Genetica 131, 1–9 (2007).Article 
CAS 
PubMed 

Google Scholar 
Egartner, A., Lethmayer, C., Gottsberger, R. A. & Blümel, S. In Joint Meeting of the IOBC-WPRS Working Groups “Pheromones and other semiochemicals in integrated production” & “Integrated Protection of Fruit Crops” at. 143–152.Nyamukondiwa, C., Kleynhans, E. & Terblanche, J. S. Phenotypic plasticity of thermal tolerance contributes to the invasion potential of mediterranean fruit flies (Ceratitis capitata). Ecol. Entomol. 35, 565–575 (2010).Article 

Google Scholar 
Weldon, C. W., Terblanche, J. S. & Chown, S. L. Time-course for attainment and reversal of acclimation to constant temperature in two Ceratitis species. J. Therm. Biol. 36, 479–485 (2011).Article 

Google Scholar 
Pujol-Lereis, L. M., Rabossi, A. & Quesada-Allué, L. A. Analysis of survival, gene expression and behavior following chill-coma in the medfly Ceratitis capitata: effects of population heterogeneity and age. J. Insect Physiol. 71, 156–163 (2014).Article 
CAS 
PubMed 

Google Scholar 
Pujol-Lereis, L. M., Fagali, N. S., Rabossi, A., Catalá, Á. & Quesada-Allué, L. A. Chill-coma recovery time, age and sex determine lipid profiles in Ceratitis capitata tissues. J. Insect Physiol. 87, 53–62 (2016).Article 
CAS 
PubMed 

Google Scholar 
Weldon, C. W., Nyamukondiwa, C., Karsten, M., Chown, S. L. & Terblanche, J. S. Geographic variation and plasticity in climate stress resistance among southern African populations of Ceratitis capitata (Wiedemann)(Diptera: Tephritidae). Sci. Rep. 8, 1–13 (2018).Article 
CAS 

Google Scholar 
Nyamukondiwa, C., Weldon, C. W., Chown, S. L., le Roux, P. C. & Terblanche, J. S. Thermal biology, population fluctuations and implications of temperature extremes for the management of two globally significant insect pests. J. Insect Physiol. 59, 1199–1211 (2013).Article 
CAS 
PubMed 

Google Scholar 
Mitchell, K. A., Boardman, L., Clusella-Trullas, S. & Terblanche, J. S. Effects of nutrient and water restriction on thermal tolerance: A test of mechanisms and hypotheses. Comp. Biochem. Physiol. A: Mol. Integr. Physiol. 212, 15–23 (2017).Article 
CAS 

Google Scholar 
Hoffmann, A. A. & Ross, P. A. Rates and patterns of laboratory adaptation in (mostly) insects. J. Econ. Entomol. 111, 501–509 (2018).Article 
PubMed 

Google Scholar 
Popa-Báez, Á. -D. et al. Climate stress resistance in male Queensland fruit fly varies among populations of diverse geographic origins and changes during domestication. BMC Genet. 21, 1–19 (2020).Article 

Google Scholar 
Beck, H. E. et al. Present and future Köppen-Geiger climate classification maps at 1-km resolution. Sci. Data 5, 1–12 (2018).Article 

Google Scholar 
Kozak, K. H., Graham, C. H. & Wiens, J. J. Integrating GIS-based environmental data into evolutionary biology. Trends Ecol. Evol. 23, 141–148 (2008).Article 
PubMed 

Google Scholar 
Oyen, K. J. et al. Body mass and sex, not local climate, drive differences in chill coma recovery times in common garden reared bumble bees. J. Comp. Physiol. B. 191, 843–854 (2021).Article 
PubMed 

Google Scholar 
Angert, A. L., Bontrager, M. G. & Ågren, J. What do we really know about adaptation at range edges?. Annu. Rev. Ecol. Evol. Syst. 51, 341–361 (2020).Article 

Google Scholar 
Terblanche, J. S. & Hoffmann, A. A. Validating measurements of acclimation for climate change adaptation. Curr. Opin. insect sci. 41, 7–16 (2020).Article 
PubMed 

Google Scholar 
Kourti, A. Patterns of variation within and between Greek populations of Ceratitis capitata suggest extensive gene flow and latitudinal clines. J. Econ. Entomol. 97, 1186–1190 (2004).Article 
CAS 
PubMed 

Google Scholar 
Hangartner, S., Lasne, C., Sgrò, C. M., Connallon, T. & Monro, K. Genetic covariances promote climatic adaptation in Australian Drosophila. Evolution 74, 326–337 (2020).Article 
PubMed 

Google Scholar 
Bontrager, M. & Angert, A. L. Gene flow improves fitness at a range edge under climate change. Evol. Let. 3, 55–68 (2019).Article 

Google Scholar 
Liu, Q. et al. Extension of the growing season increases vegetation exposure to frost. Nat. Commun. 9, 1–8 (2018).
Google Scholar 
Schwartz, M. D., Ahas, R. & Aasa, A. Onset of spring starting earlier across the Northern Hemisphere. Glob. Change Biol. 12, 343–351 (2006).Article 

Google Scholar 
Ma, Q., Huang, J. G., Hänninen, H. & Berninger, F. Divergent trends in the risk of spring frost damage to trees in Europe with recent warming. Glob. Change Biol. 25, 351–360 (2019).Article 

Google Scholar 
Unterberger, C. et al. Spring frost risk for regional apple production under a warmer climate. PLoS ONE 13, e0200201 (2018).Article 
MathSciNet 
PubMed 
PubMed Central 

Google Scholar 
Manrakhan, A., Daneel, J.-H., Stephen, P. R. & Hattingh, V. Cold Tolerance of Immature Stages of Ceratitis capitata and Bactrocera dorsalis (Diptera: Tephritidae). J. Econ. Entomol. 115(2), 482–492 (2022).Article 
PubMed 

Google Scholar 
Papadopoulos, N. T., Carey, J. R., Katsoyannos, B. I. & Kouloussis, N. A. Overwintering of the mediterranean fruit fly (Diptera: Tephritidae) in Northern Greece. Ann. Entomol. Soc. Am. 89, 526–534 (1996).Article 

Google Scholar 
Papadopoulos, N. T., Katsoyannos, B. I. & Carey, J. R. Temporal changes in the composition of the overwintering larval population of the Mediterranean fruit fly (Diptera: Tephritidae) in Northern Greece. Ann. Entomol. Soc. Am. 91, 430–434 (1998).Article 

Google Scholar 
Katsoyannos, B. I., Kouloussis, N. A. & Carey, J. R. Seasonal and annual occurrence of Mediterranean fruit flies (Diptera: Tephritidae) on Chios Island, Greece: Differences between two neighboring citrus orchards. Ann. Entomol. Soc. Am. 91, 43–51 (1998).Article 

Google Scholar 
Mavrikakis, P. G., Economopoulos, A. P. & Carey, J. R. Continuous winter reproduction and growth of the mediterranean fruit fly (Diptera: Tephritidae) in Heraklion, crete Southern Greece. Environ. Entomol. 29, 1180–1187 (2000).Article 

Google Scholar 
Israely, N., Ziv, Y. & Oman, S. D. Spatiotemporal distribution patterns of Mediterranean fruit fly (Diptera: Tephritidae) in the central region of Israel. Ann. Entomol. Soc. Am. 98, 77–84 (2005).Article 

Google Scholar 
Bahrndorff, S., Lauritzen, J. M., Sørensen, M. H., Noer, N. K. & Kristensen, T. N. Responses of terrestrial polar arthropods to high and increasing temperatures. J. Exp. Biol. 224, jeb230797 (2021).Article 
PubMed 

Google Scholar 
Sinclair, B. J. & Roberts, S. P. Acclimation, shock and hardening in the cold. J. Therm. Biol. 30, 557–562 (2005).Article 

Google Scholar 
Bahrndorff, S., Gertsen, S., Pertoldi, C. & Kristensen, T. N. Investigating thermal acclimation effects before and after a cold shock in Drosophila melanogaster using behavioural assays. Biol. J. Lin. Soc. 117, 241–251 (2016).Article 

Google Scholar 
Sarmad, M., Ishfaq, A., Arif, H. & Zaka, S. M. Effect of short-term cold temperature stress on development, survival and reproduction of Dysdercus koenigii (Hemiptera: Pyrrhocoridae). Cryobiology 92, 47–52 (2020).Article 
PubMed 

Google Scholar 
Steyn, V. M., Mitchell, K. A., Nyamukondiwa, C. & Terblanche, J. S. Understanding costs and benefits of thermal plasticity for pest management: Insights from the integration of laboratory, semi-field and field assessments of Ceratitis capitata (Diptera: Tephritidae). Bull. Entomol. Res., 1–11 (2022).Davis, H. E., Cheslock, A. & MacMillan, H. A. Chill coma onset and recovery fail to reveal true variation in thermal performance among populations of Drosophila melanogaster. Sci. Rep. 11, 1–10 (2021).Article 

Google Scholar 
Noh, S., Everman, E. R., Berger, C. M. & Morgan, T. J. Seasonal variation in basal and plastic cold tolerance: Adaptation is influenced by both long-and short-term phenotypic plasticity. Ecol. Evol. 7, 5248–5257 (2017).Article 
PubMed 
PubMed Central 

Google Scholar 
Bruins, H. J. Ancient desert agriculture in the Negev and climate-zone boundary changes during average, wet and drought years. J. Arid Environ. 86, 28–42 (2012).Article 

Google Scholar 
Hoffmann, A. A., Sørensen, J. G. & Loeschcke, V. Adaptation of Drosophila to temperature extremes: Bringing together quantitative and molecular approaches. J. Therm. Biol. 28, 175–216 (2003).Article 

Google Scholar 
Kawecki, T. J. & Ebert, D. Conceptual issues in local adaptation. Ecol. Lett. 7, 1225–1241 (2004).Article 

Google Scholar 
Nyamukondiwa, C. & Terblanche, J. S. Thermal tolerance in adult mediterranean and Natal fruit flies (Ceratitis capitata and Ceratitis rosa): Effects of age, gender and feeding status. J. Therm. Biol. 34, 406–414 (2009).Article 

Google Scholar 
Team, R. C. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria (2021).Fick, S. E. & Hijmans, R. J. WorldClim 2: New 1-km spatial resolution climate surfaces for global land areas. Int. J. Climatol. 37, 4302–4315 (2017).Article 

Google Scholar 
Mazerolle, M. J. Model selection and multimodel inference using the AICcmodavg package (2020).Therneau, T. A Package for Survival Analysis in R. R Package Version 3.2-13.(2021. (2021).Kassambara, A., Kosinski, M., Biecek, P. & Fabian, S. Survminer: Drawing Survival Curves using’ggplot2′. R package version 0.4. 9. 2021. (2021).Lenth, R. V. Emmeans: Estimated Marginal Means, aka Least-Squares Means. R package version 1.7.2. (2022). More

Bates, H. W. XXXII. Contributions to an insect fauna of the Amazon Valley. Lepidoptera: Heliconidæ. Trans. Linn. Soc. Lond 23, 495–566 (1862).Article 

Google Scholar 
Müller, F. Ituna and thyridia: A remarkable case of mimicry in butterflies. Trans. Entomol. Soc. Lond. 1879, 20–29 (1879).
Google Scholar 
Baxter, S. W. et al. Convergent evolution in the genetic basis of Müllerian mimicry in Heliconius butterflies. Genetics 180, 1567–1577 (2008).Article 
CAS 

Google Scholar 
Sheppard, P. M., Turner, J. R. G., Brown, K., Benson, W. & Singer, M. Genetics and the evolution of Muellerian mimicry in Heliconius butterflies. Philos. Trans R. Soc. Lond. B, Biol. Sci. 308, 433–610 (1985).Article 
ADS 

Google Scholar 
Mallet, J. & Gilbert, L. E. Jr. Why are there so many mimicry rings? Correlations between habitat, behaviour and mimicry in Heliconius butterflies. Biol. J. Lin. Soc. 55, 159–180 (1995).
Google Scholar 
Brower, A. V. Parallel race formation and the evolution of mimicry in Heliconius butterflies: A phylogenetic hypothesis from mitochondrial DNA sequences. Evolution 50, 195–221 (1996).Article 
CAS 

Google Scholar 
Wilson, J. S. et al. North American velvet ants form one of the world’s largest known Müllerian mimicry complexes. Curr. Biol. 25, R704–R706. https://doi.org/10.1016/j.cub.2015.06.053 (2015).Article 
CAS 
PubMed 

Google Scholar 
Wilson, J. S., Williams, K. A., Forister, M. L., Von Dohlen, C. D. & Pitts, J. P. Repeated evolution in overlapping mimicry rings among North American velvet ants. Nat. Commun. 3, 1272 (2012).Article 
ADS 

Google Scholar 
Wilson, J. S., Pan, A. D., Limb, E. S. & Williams, K. A. Comparison of African and North American velvet ant mimicry complexes: Another example of Africa as the ‘odd man out’. PLoS ONE 13, e0189482. https://doi.org/10.1371/journal.pone.0189482 (2018).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Plowright, R. & Owen, R. E. The evolutionary significance of bumble bee color patterns: A mimetic interpretation. Evolution 34, 622–637 (1980).Article 
CAS 

Google Scholar 
Williams, P. The distribution of bumblebee colour patterns worldwide: Possible significance for thermoregulation, crypsis, and warning mimicry. Biol. J. Lin. Soc. 92, 97–118 (2007).Article 

Google Scholar 
Hines, H. M. & Williams, P. H. Mimetic colour pattern evolution in the highly polymorphic Bombus trifasciatus (Hymenoptera: Apidae) species complex and its comimics. Zool. J. Linn. Soc. 166, 805–826 (2012).Article 

Google Scholar 
Koch, J. B., Rodriguez, J., Pitts, J. P. & Strange, J. P. Phylogeny and population genetic analyses reveals cryptic speciation in the Bombus fervidus species complex (Hymenoptera: Apidae). PLoS ONE 13, e0207080 (2018).Article 

Google Scholar 
Ezray, B. D., Wham, D. C., Hill, C. E. & Hines, H. M. Unsupervised machine learning reveals mimicry complexes in bumblebees occur along a perceptual continuum. Proc. R. Soc. B 286, 20191501 (2019).Article 

Google Scholar 
Bateson, W. The alleged “Aggressive Mimicry” of volucellæ. Nature 46, 585 (1892).Article 
ADS 

Google Scholar 
Poulton, E. B. The volucellœ as alleged examples of variation “almost unique among animals”. Nature 47, 126 (1892).Article 
ADS 

Google Scholar 
Cockerell, T. D. New social bees. Psyche A J. Entomol. 24, 120–128 (1917).Article 

Google Scholar 
Koch, J., Strange, J. & Williams, P. In: Bumble bees of the western United States (US Forest Service, San Francisco California, 2012).
Google Scholar 
Williams, P. H., Thorp, R. W., Richardson, L. L. & Colla, S. R. In: Bumble bees of North America: An identification guide Vol. 87 (Princeton University Press, Princeton, 2014).
Google Scholar 
Ruxton, G. D., Franks, D. W., Balogh, A. C. & Leimar, O. Evolutionary implications of the form of predator generalization for aposematic signals and mimicry in prey. Evol Int. J. Org. Evol. 62, 2913–2921 (2008).Article 

Google Scholar 
Rowe, C., Lindström, L. & Lyytinen, A. The importance of pattern similarity between Müllerian mimics in predator avoidance learning. Proc. R. Soc. Lond. Ser. B Biol. Sci. 271, 407–413 (2004).Article 

Google Scholar 
Beatty, C. D., Beirinckx, K. & Sherratt, T. N. The evolution of Müllerian mimicry in multispecies communities. Nature 431, 63 (2004).Article 
ADS 
CAS 

Google Scholar 
Chittka, L. & Osorio, D. Cognitive dimensions of predator responses to imperfect mimicry. PLoS Biol. 5, e339 (2007).Article 

Google Scholar 
Dittrigh, W., Gilbert, F., Green, P., McGregor, P. & Grewcock, D. Imperfect mimicry: A pigeon’s perspective. Proc. R. Soc. Lond. Ser. B Biol. Sci. 251, 195–200 (1993).Article 
ADS 

Google Scholar 
Sherratt, T. N., Whissell, E., Webster, R. & Kikuchi, D. W. Hierarchical overshadowing of stimuli and its role in mimicry evolution. Anim. Behav. 108, 73–79 (2015).Article 

Google Scholar 
Beatty, C. D., Bain, R. S. & Sherratt, T. N. The evolution of aggregation in profitable and unprofitable prey. Anim. Behav. 70, 199–208 (2005).Article 

Google Scholar 
Kazemi, B., Gamberale-Stille, G., Tullberg, B. S. & Leimar, O. Stimulus salience as an explanation for imperfect mimicry. Curr. Biol. 24, 965–969 (2014).Article 
CAS 

Google Scholar 
Kikuchi, D. W., Dornhaus, A., Gopeechund, V. & Sherratt, T. N. Signal categorization by foraging animals depends on ecological diversity. Elife. 8, e43965 (2019).Article 

Google Scholar 
Rapti, Z., Duennes, M. A. & Cameron, S. A. Defining the colour pattern phenotype in bumble bees (Bombus): A new model for evo devo. Biol. J. Lin. Soc. 113, 384–404 (2014).Article 

Google Scholar 
Wilson, J. S., Sidwell, J. S., Forister, M. L., Williams, K. A. & Pitts, J. P. Thistledown velvet ants in the desert mimicry ring and the evolution of white coloration: Müllerian mimicry, camouflage and thermal ecology. Biol. Lett. 16, 20200242 (2020).Article 

Google Scholar 
Ascher, J. & Pickering, J. Discover Life bee species guide and world checklist (Hymenoptera: Apoidea: Anthophila) (2019).iNaturalist. Available from https://www.inaturalist.org. Accessed [2022].Bombus Latreille, 1802 in GBIF Secretariat (2021). GBIF Backbone Taxonomy. Checklist dataset https://doi.org/10.15468/39omei accessed via GBIF.org on 2021-12-03. More

Wolf, S. et al. Nat. Ecol. Evol. https://doi.org/10.1038/s41559-022-01904-x (2022).Article 

Google Scholar 
Kattge, J. et al. Glob. Change Biol. 26, 119–188 (2020).Article 

Google Scholar 
Sabatini, F. M. et al. Glob. Ecol. Biogeogr. 30, 1740–1764 (2021).Article 

Google Scholar 
Łopucki, R., Kiersztyn, A., Pitucha, G. & Kitowski, I. Ecol. Modell. 468, 109964 (2022).Article 

Google Scholar 
Kosmala, M., Wiggins, A., Swanson, A. & Simmons, B. Front. Ecol. Environ. 14, 551–560 (2016).Article 

Google Scholar 
White, C. R. et al. Funct. Ecol. 35, 1572–1578 (2021).Article 

Google Scholar 
Xirocostas, Z. A., Debono, S. A., Slavich, E. & Moles, A. T. Methods Ecol. Evol. 13, 596–602 (2022).Article 

Google Scholar 
Callaghan, C. T. et al. Bioscience 71, 55–63 (2020).
Google Scholar  More

Sequencing and assembly of the H. schachtii genomeWe measured (Supplemental Fig. 1), sequenced (BioProject PRJNA722882), and assembled the genome of H. schachtii (population Bonn) using a combination of flow cytometry, Pacific Biosciences sequencing, and Illumina sequencing. H. schachtii has the largest genome (160–170 Mb) of any cyst nematode measured/sequenced to date (Supplementary Table 1). It was sequenced to 192-fold coverage using Pacific Biosciences sequencing (fragment n50 of 16 kb), and 144-fold coverage using Illumina sequencing (150 bp Paired-end reads). The final, polished, contamination-free (Supplemental Fig. 2), assembly (v1.2) included ~179 Mbp contained within 395 scaffolds: 90% of the sequence is contained on scaffolds longer than 281,463 bp (n = 154). The assembly is a largely complete haploid representation of the diploid genome, as evidenced by core eukaryotic genes being largely present, complete and single copy (CEGMA 93.15% complete with an average of 1.12 copies each, and BUSCO (Eukaryota odb9) 79% complete with 8.2% duplicated—Supplementary Table 2). Over three million variants were phased into haplotypes (2029 blocks, N50 239.5 kb, covering 94.7% of the reference) which can be used to predict true protein variants (Supplementary data 1), and 601 larger structural variants were identified (Supplementary data 2).The trans-kingdom, lifestage-specific, transcriptomes of H. schachtii and A. thaliana provide a holistic view of parasitismWe devised a sampling procedure to cover all major life stages/transitions of the parasitic life cycle to generate a simultaneous, chronological, and comprehensive picture of nematode gene expression, and infection-site-specific plant gene expression patterns. We sampled cysts and pre-infective second-stage juveniles (J2s), as well as infected segments of A. thaliana root and uninfected adjacent control segments of root at 10 hours post infection (hpi – migratory J2s, pre-establishment of the feeding site), 48 hpi (post establishment of the feeding site), 12 days post infection females (dpi – virgin), 12 dpi males (differentiated, pre-emergence, most if not all stopped feeding), and 24 dpi females (post mating), each in biological triplicate (Fig. 1A). We generated approximately nine billion pairs of 150 bp strand-specific RNAseq reads (Supplementary data 3) covering each stage in biological triplicate (for the parasite and the host): in the early stages of infection we generated over 400 million reads per replicate, to provide sufficient coverage of each kingdom.Fig. 1: Trans-kingdom, lifestage-specific, transcriptome of H. schachtii and A. thaliana.A Schematic representation of the life cycle of H. schachtii infecting A. thaliana, highlighting the 7 stages sampled in this study. For each stage, the average number of trimmed RNAseq read pairs per replicate is shown, with the proportion of reads mapping to either parasite or host in parentheses. B Principle components 1 and 2 for H. schachtii and A. thaliana expression data are plotted. Arrows indicate progression through the life cycle/real-time. Hours post infection (hpi), days post infection (dpi).Full size imageStrand-specific RNAseq reads originating from host and parasite were deconvoluted by mapping to their respective genome assemblies (H. schachtii v.1.2 and TAIR10). For the parasite, ~500 million Illumina RNAseq read pairs uniquely mapping to the H. schachtii genome were used to generate a set of 26,739 gene annotations (32,624 transcripts – detailed further in the next section), ~77% of which have good evidence of transcription in at least one lifestage (≥10 reads in at least one rep). Similarly for the host, ~2.8 billion Illumina RNAseq read pairs uniquely mapping to the A. thaliana genome show that ~77% of the 32,548 gene models have good evidence of transcription in at least one stage (≥10 reads in at least one rep, even though we only sampled roots). A principal component analysis of the host and parasite gene expression data offers several insights into the parasitic process. Principle component 1 (60% of the variance) and 2 (19% of the variance) of the parasite recapitulate the life cycle in PCA space (Fig. 1B). The 12 dpi female transcriptome is more similar to the 24 dpi female transcriptome than to the 12 dpi male transcriptome. Principle components 1 (75% of the variance) and 2 (10% of the variance) of the host show that the greatest difference between infected and uninfected plant tissue is at the early time points (10 hpi), and that the transcriptomes of infected and uninfected plant material converge over time, possibly due to systemic effects of infection. A 12 dpi male syncytium transcriptome is roughly intermediate between a control root transcriptome and a 12 dpi female syncytium transcriptome. Given that at this stage most if not all of the males will have ceased feeding, this could be due to inadequate formation of the feeding site, or regression of the tissue. In any case, by comparing both principal component analyses, we can see that what is a relatively small difference in the transcriptomes of the feeding sites of males and females is amplified to a relatively large difference in the transcriptomes of the males and females themselves (Fig. 1B).The consequences, and possible causes, of large-scale segmental duplication in the Heterodera lineageTo understand the evolutionary origin(s) of the relatively large number of genes in H. schachtii in particular, and Heterodera spp. in general, we analysed the abundance and categories of gene duplication in the predicted exome. Compared to a related cyst nematode, Globodera pallida (derived using comparable methodology and of comparable contiguity) the exomes of H. schachtii and H. glycines are characterised by a relatively smaller proportion of single-copy genes (as classified by MCSanX toolkit17, and a relatively greater proportion of segmental duplications (at least five co-linear genes with no >25 genes between them), with relatively similar proportions of dispersed duplications (two similar genes with >20 other genes between them), proximal duplications (two similar genes with  +0.5 or  More