MainThe modern chemical industry relies on the consumption of finite fossil resources at a pace and scale that is inherently unsustainable. These processes are energy intensive and generate consumer products that are ultimately disposed by landfill or incineration, resulting in the irretrievable loss of this valuable carbon as environmental pollution or CO2 in the atmosphere. By contrast, nature has evolved elegant and efficient mechanisms for carbon resource utilization, by-product recycling and sustainable chemical synthesis. As these processes are genetically encoded, they offer a blueprint for modern engineering biology to remediate and upcycle carbon embedded in industrial and post-consumer waste. The resulting engineered bioprocesses can reintegrate this carbon into the circular chemical economy, while simultaneously reducing pollution, greenhouse gas emissions and the underlying drivers of global climate change.Research in this area has focussed on the bioavailable polymers cellulose, chitin and lignin as substrates, but more recently has explored the use of plastic waste as a microbial feedstock. Enzymatic depolymerization of lignocellulose and chitin has yielded carbohydrate monomers for bacterial growth and upcycling pathways to biofuels1, polyhydroxyalkanoates2, terpenes3 and amino acids4, whereas aromatic monomers isolated from lignin have been used as cellular substrates for adipic acid5 and coniferyl alcohol6 synthesis. The bio-upcycling of plastic waste has been enabled by the discovery of enzymes capable of depolymerizing poly(ethylene terephthalate) (PET)7, catalysed by the discovery of Ideonella sakaiensis, a bacterium capable of depolymerizing and assimilating PET8. This breakthrough has since driven the engineering of IsPETase variants with enhanced catalytic performance for applications in industrial biotechnology (Fig. 1a). These efforts have yielded PET hydrolases with enhanced activity and thermal stability9,10, as well as metabolic pathways for upcycling PET-derived monomers into value-added products such as vanillin11, adipic acid12, paracetamol13 and other platform chemicals14,15. In addition, there exists a growing list of chemoenzymatic approaches to upcycle other plastic wastes such as poly(hydroxybutyrate) into acetone16, polyethylene into functionalized carboxylic acids17 and mixed plastic wastes into ß-ketoadipate and poyhydroxyalkanoates18. Chemical catalysts have also been interfaced with plastic upcycling pathways to create new biocompatible chemistry approaches and green synthetic chemistry methods19 for the mild conversion of a range of plastic wastes into industrially applicable second-generation chemical products. Of note, ref. 20 recently reported the use of chemical catalysis to convert post-consumer polystyrene to benzoic acid (Co(NO3)2 + Mn(NO3)2, N-hydroxyphthalimide, O2 (4 bar), acetic acid, 150 °C followed by microbial conversion to the complex small molecules ergothioneine, mutilin and pleuromutilin in engineered Aspergillus nidulans and as a sole carbon source to generate the atoxigenic biocontrol agent Aspergillus flavus Af36. Similarly, another study18 coupled the chemocatalysed autooxidation of mixed domestic plastic wastes to oxygenated carboxylic acid intermediates with biotransformation by engineered strains of Pseudomonas putida KT2440 to create a metabolic funnel to valuable synthetic building blocks. Inspired by recent advances and the emerging potential of engineering biology to upcycle plastic waste into high-value molecules, we report the microbial conversion of industrial PET waste and a post-consumer PET plastic bottle into levodopa (l-DOPA), a frontline treatment for the symptoms of Parkinson’s disease (Fig. 1b). l-DOPA is currently produced at a global scale of ~250 t yr−1 (ref. 21) with demand projected to rise as a result of increased disease prevalence22,23. Beyond its clinical importance, l-DOPA serves as a biosynthetic precursor to melanin and a variety of complex plant-derived natural products24,25. Despite the existence of several biotechnological routes to l-DOPA—including fermentative pathways from l-tyrosine, chitin or D-glucose and expression in transgenic tomatoes—commercial production remains reliant on fossil fuel-derived chemical or chemoenzymatic synthesis26,27,28,29,30. Each alternative faces notable challenges, such as poor carbon economy, oxidative degradation, inefficient feedstock utilization or regulatory hurdles. This work therefore offers a sustainable alternative to existing chemical and biological methods for producing l-DOPA from virgin petrochemicals and highlights the first application of engineering biology to valorize plastic waste into a therapeutic for neurological disease.Fig. 1: Plastic waste recycling and bio-upcycling strategies.Approaches to the recycling, upcycling and environmental disposal of PET plastic waste, including the proposed bio-upcycling of PET waste to the Parkinson’s medication l-DOPA in engineered bacteria. a, Current: closed-loop recycling. b, This work: microbial upcycling. Credit: photographs in a, Rawpixel (https://www.rawpixel.com); bacterial icon in b, Bioicons (https://bioicons.com).Full size imageResultsDe novo pathway design and optimizationl-DOPA synthesis from PET monomer terephthalic acid (TPA) was envisioned via a four-step biosynthetic pathway encoded by seven genes in the laboratory bacterium Escherichia coli BL21(DE3). The pathway proceeds via conversion of TPA to protocatechuate (PCA) catalysed by a terephthalate 1,2-dioxygenase complex (TphA2 and TphA3), and cognate reductase (TphA1; together referred as the TPADO complex) and dihydroxy-3,5-cyclohexadiene-1,4-dicarboxylic acid dehydrogenase (known as DCDDH or TphB) from Comamonas sp. Decarboxylation of PCA to catechol by AroY and prFMN cofactor regeneration enzyme KpdB from Klebsiella pneumoniae is then followed by a final C–C bond formation via electrophilic aromatic substitution between catechol and pyruvate in the presence of ammonia by the pyridoxal 5′-phosphate (PLP)-dependent tyrosine-phenol lyase (TPL) from Fusobacterium nucleatum to form l-DOPA (Fig. 2a). The tpado and dcddh genes were assembled on to a pGro7 derived backbone to generate pPCA1 (module 1), aroY and kpdB genes were assembled on to a pQLinkN vector to generate pCAT1 (module 2) and tpl was assembled in a joint universal modular plasmids (JUMP) assembly level 1 vector to generate pFnTPL (module 3). The activity of each module was examined by transforming E. coli BL21(DE3) cells with pPCA1, pCAT1, pPCA1 + pCAT1 or pFnTPL, confirming soluble protein expression after growth in LB medium by SDS–PAGE (Supplementary Figs. 1 and 2) and then incubating with the requisite pathway substrate or intermediate in a whole-cell biotransformation (optical density OD600 = 30).Fig. 2: Pathway design, construction and bottlenecks.a, De novo biosynthetic pathway to l-DOPA from PET monomer TPA. b, Whole-cell activity when pPCA1, pCAT1 and pFnTPL are singly and multiply expressed in E. coli BL21(DE3) as well as whole pathway with pPCA1_pCAT-FnTPL. c, Enhanced TPA conversion to PCA in strains E. coli_pPCA1 and E. coli_pPCA3 expressing tpaK from R. jostii. d, PCA and catechol conversions upon increasing terephthalate concentration in strains E. coli_pPCA3 and E. coli_pPCA3_pCAT1, respectively. Metabolite concentrations were determined by reverse-phase HPLC relative to an internal standard of caffeine (0.01 g l−1). Data are presented as an average of three replicate experiments to one standard deviation. All data shown are from 3 ml reactions performed in 15 ml Falcon tubes. In b, the supercripts are as follows: [a]TPA substrate; [b]PCA substrate; and [c]catechol substrate. Credit: photograph in a, Rawpixel (https://www.rawpixel.com); bacterial icons in a, Bioicons (https://bioicons.com).Source dataFull size imageCells containing TPADO-dcddh (module 1), AroY/KpdB (module 2) or TPL (module 3) enzymes were highly active in isolation and observed to convert >90% substrate to product in 24 h (Fig. 2b). We sought to improve diffusion of terephthalate across the negatively charged bacterial cell membrane. This is known to be most efficient at pH 5 (ref. 11) (pKa1 = 3.5; pKa2 = 4.3) and therefore to mitigate this at pH 7 we assembled tpaK from Rhodococcus jostii downstream of the tpado and dcddh genes via homologous end assembly to generate the modified plasmid pPCA3 (Supplementary Fig. 3). TpaK is part of the major facilitator superfamily class of membrane transporters and has been shown to import aromatic acids including TPA when heterologously expressed in P. putida31,32. Pleasingly, TPA conversion to PCA at pH 7 displayed accelerated product formation (Fig. 2c). To our knowledge this constitutes the first use of a TPA transporter for biocatalysis in E. coli.Co-transformation with pCAT1 generated the modified strain E. coli_pPCA3_pCAT1 for conversion of TPA to catechol. Despite high-level production of PCA by E. coli_pPCA3, conversion to catechol was efficient at low concentration (1–10 mM; >85%) but decreased to 1 mM PCA (Fig. 3d and Supplementary Fig. 6). Similar results were observed in vitro using His-tag purified FnTPL, with no l-DOPA produced at >2 mM PCA (Supplementary Fig. 6). Overall, this provided evidence to suggest the PCA-dependent inhibition of TPL during l-DOPA biosynthesis. To support this hypothesis, molecular docking simulations of catechol and PCA within the active site of FnTPL were performed using GNINA (Fig. 3). Key hydrogen bonding and electrostatic interactions were identified between PCA and conserved active site residues T127, K260, R384 and Y74. Free energy calculations showed that both PCA and catechol bind with identical and thermodynamically favourable energies (−5.99 kcal mol−1), suggesting that competitive binding probably occurs within the active site pocket. This aligns with recent reports of TPL inhibition by gallic acid and 3,5-dihydroxybenzoic acid in bacteria associated with mouse gut microbiota35. Together, these findings provide strong evidence of PCA-dependent TPL inhibition during l-DOPA biosynthesis and highlight the need to circumvent this bottleneck to enable efficient l-DOPA production from TPA in vivo.Fig. 3: In silico modelling of TPL and in vitro inhibition by PCA.Predicted binding modes of PCA or catechol with FnTPL–dimer–PLP complex. a, Model of the functional dimer of FnTPL, with covalently linked PLP and docked catechol in monomer A (light grey) shown as stick representation in the active site. b, Predicted binding of catechol in the active site mediated primarily by phenylalanine and arginine residues. The catalytic Tyr-74 contributed by monomer B is highlighted in dark grey. c, Docked PCA bound to FnTPL with interactions mediated by arginine residues augmented by Met-382, Thr-127 and Phe-120. Both substrates are predicted to have a favourable binding affinity of −5.99 kcal mol−1, with Arg-220, Arg-384, Arg-407 and Thr-52 contributing to the binding of both substrates. Molecular docking simulations were performed using the molecular docking package GNINA. d, Assaying pathway module inhibition by pathway intermediates using E. coli_pFnTPL cells incubated in the presence of catechol and varying concentrations of PCA. Data are presented as an average of three replicate experiments to one standard deviation.Source dataFull size imageTo this end, the three pathway modules were separated into two strains (E. coli_pPCA3_pCAT1 and E. coli_pFnTPL) to enable catechol accumulation before conversion to l-DOPA. Reaction optimizations to maximize the performance of module 3 within E. coli_pFnTPL identified pH, time and pyruvate concentration as critical to high l-DOPA conversion from exogenous catechol in this strain (Fig. 4b). As reported in the literature36, FnTPL was most efficient at pH 8 (Fig. 4b). Interestingly, l-DOPA concentrations also decreased by 40% from 4.2 mM to 2.5 mM over 17 h under these optimized reaction conditions. Although l-DOPA is known to polymerize via dopaquinone to poly(dopaquinone) under aerobic conditions37, the addition of antioxidants did not reverse the observed product loss over time. Liquid chromatography–mass spectrometry (LC-MS) analysis of reaction extracts instead confirmed the degradation of l-DOPA via a non-enzymatic Pictet–Spengler reaction with pyruvate to form a heterocyclic adduct (Supplementary Fig. 7)27. Together, the optimum reaction conditions for l-DOPA synthesis from 5 mM catechol by E. coli_pFnTPL at OD600 = 30 were concluded to be pH 8.0 in the presence of 60 mM pyruvate for 3 h at 21 °C (Fig. 4b).Fig. 4: Two-strain process optimization, CO2 capture and industrial waste upcycling.a, A one-pot two-strain approach to decouple PCA mediated inhibition of TPL for l-DOPA biosynthesis from TPA and PET. b, Whole-cell reaction optimization for conversion of exogenous catechol to l-DOPA by E. coli_pFnTPL. c, Time course of two strain system for conversion of TPA to l-DOPA, with 24 h of reaction from TPA to CAT by strain E. coli_pPCA3_pCAT1 before addition of strain E. coli_pFnTPL for 24 h. A scaling factor of 0.83 was applied to account for dilution effects from the addition of two strains. d, Capture of released CO2 from TPA whole-cell reactions in E. coli by C. reinhardtii CC1690. Peak areas of residual CO2 for C. reinhardtii grown on TAP media (control) compared with incubation with or without headspace gas mixture from E. coli (HSGM). e, Preparative microbial biosynthesis of l-DOPA salt from industrial HSF PET waste collected from API Foilmakers. Data are presented as an average of three replicate experiments to one standard deviation, except e which was performed as a scaled-up single replicate. Catechol was generated by strain E. coli_pPCA3_pCAT1. ND, not detected. Credit: photographs in a, Rawpixel (https://www.rawpixel.com); Chlamydomonas icon in d, DBCLS/TogoTV under a Creative Commons license CC BY 4.0.Source dataFull size imageTwo-strain bioconversionHaving optimized conditions for the conversion of TPA to catechol by E. coli_pPCA3_pCAT1 and catechol to l-DOPA by E. coli_pFnTPL, we moved on to combine these strains for the one-pot biotransformation of TPA to l-DOPA. As PCA-dependent inhibition of TPL had been observed at concentrations >250 µM, a reaction involving the sequential addition of E. coli_pFnTPL cells after incubation of E. coli_pPCA3_pCAT1 with TPA was envisaged. Additionally, analysis of E. coli_pPCA3_pCAT1 (OD600 = 30) cells incubated with 5 mM TPA at 21 °C (220 rpm) for 24 h showed >90% catechol was present in the cell supernatant, indicating this would be suitable for uptake by a second microorganism (Supplementary Fig. 8). To this end, TPA was incubated with E. coli_pPCA3_pCAT1 (OD600 = 30) at 21 °C (220 rpm) for 24 h, before the addition of E. coli_pFnTPL (OD600 = 30) and further incubation for 3 h. Pleasingly, this resulted in the production of l-DOPA from TPA in 0.68 g l−1 and 69% overall conversion as the major product by high-performance liquid chromatography (HPLC) (Fig. 4c and Supplementary Fig. 9).CO2 capture using microalgaeAs a proof-of-concept and preliminary assessment of whether CO2 released during the enzymatic decarboxylation of TPA to catechol could be offset, we tested whether the microalga Chlamydomonas reinhardtii CC1690 could recapture this CO2 via photosynthesis. Engineered E. coli cultures supplied with 5 mM TPA-generated elevated CO2 levels (Supplementary Fig. 10) and headspace gas from these cultures was transferred to actively growing C. reinhardtii. Within 12 h, CO2 levels fell below the detection limit only in algal cultures and these cultures showed increased optical density and chlorophyll content relative to controls, confirming both CO2 fixation and stimulated algal growth (Fig. 4d and Supplementary Figs. 11 and 12). While these preliminary findings suggest that TPA-derived CO2 can be assimilated into algal biomass, further development, quantitative analysis and system-level validation will be required to establish the extent to which coupling biosynthesis with microalgal CO2 capture can contribute to overall process carbon neutrality.Industrial waste valorization and product isolationFollowing on from the one-pot bioconversion of PET monomer TPA to l-DOPA, we next moved on to generate l-DOPA from industrial and post-consumer PET waste. In addition to the use of a post-consumer PET bottle and packaging, we examined the upcycling of industrial PET waste. Hot stamping foils (HSF) are a prolific source of plastic waste worldwide generated by the chemical industry from the depositing of ultrathin lacquer and adhesive labels. This industry is rapidly growing (US$2.9 billion market in 2022, 5.6% CAGR through 2032) and is estimated to generate 40,000 t of PET waste per annum globally (https://www.maximizemarketresearch.com/market-report/global-hot-stamping-foils-market/25706). We depolymerized a PET bottle (from discarded waste at the University of Edinburgh, UK) and HSF samples (from API Foilmakers) under alkaline conditions, generating TPA-containing product streams that were quantified by nuclear magnetic resonance spectroscopy (51% and 83% purity for PET bottle and HSF waste, respectively). Crude TPA-rich samples were then added to optimized one-pot reactions using strains E. coli_pPCA3_pCAT1 (pH 7, 24 h, 21 °C, 220 rpm) and E. coli_pFnTPL (pH 8, 3 h, 21 °C, 220 rpm). Pleasingly, l-DOPA was generated from all reactions at 2.0 mM (49%) and 2.3 mM (55%) conversion from bottle and stamping foil PET waste, respectively (Supplementary Fig. 13). Furthermore, enzymatic depolymerization of PET packaging film by LCCICCM generated 4.64 mM l-DOPA from the released TPA (Supplementary Fig. 14). Reduced product conversions were attributed to the presence of lower grade PET from post-consumer waste due to residual plasticizers. The reaction was then scaled-up to 0.5 l and yielded 0.9 g l−1 of l-DOPA in one-pot process directly from chemically depolymerized PET waste (Supplementary Fig. 13).We next sought to isolate l-DOPA from biotransformation reactions using preparative reverse-phase HPLC. For efficient chromatographic separation, the reactions were miniaturized to generate low-volume product streams with high l-DOPA concentrations. This was achieved using a redesigned plasmid, pPCA4, which encodes tpado, dcddh and tpaK genes on a pBR322/rop origin backbone, compared with the p15a origin in pPCA3. This modification led to markedly improved catechol production at 30 mM TPA loading, increasing from 12.6 mM with pPCA3 to 27.6 mM with pPCA4 (Supplementary Fig. 15). In a two-step reaction involving E. coli_pPCA4_pCAT1 and E. coli_pFnTPL, we achieved 25.3 mM l-DOPA (5.0 g l−1, 84% conversion) from 0.26 g of stamping foil-derived TPA following sequential incubations of 24 h (strain 1) and 3 h (strain 1 + 2) at 21 °C (Fig. 4e). l-DOPA was successfully isolated as a solid TFA salt via preparative reverse-phase HPLC, equivalent to several clinical doses typically prescribed for early-onset Parkinson’s disease.DiscussionWe acknowledge that global plastic waste generation (~100 million t annually) far exceeds pharmaceutical production volumes. This pathway is therefore not proposed as a standalone solution, but rather as one component of a broader bio-upcycling portfolio. This study specifically focuses on stamping foil waste—a specialized and currently underaddressed industrial PET stream—as a viable feedstock for sustainable chemical manufacturing.Further optimization towards industrial implementation will require pathway intensification to enable l-DOPA recovery via direct precipitation from fermentation broth, given its aqueous solubility (Article CAS Google Scholar Kumagai, H., Katayama, T., Koyanagi, T. & Suzuki, H. Research overview of L-DOPA production using a bacterial enzyme, tyrosine phenol-lyase. Proc. Japan Acad. B 99, 75–101 (2023).Article CAS Google Scholar Krishnaveni, R., Rathod, V., Thakur, M. S. & Neelgund, Y. F. Transformation of L-tyrosine to L-dopa by a novel fungus, Acremonium rutilum, under submerged fermentation. Curr. Microbiol. 58, 122–128 (2009).Article CAS Google Scholar Wei, T., Cheng, B. Y. & Liu, J. Z. Genome engineering Escherichia coli for L-DOPA overproduction from glucose. Sci. Rep. 6, 30080 (2016).Breitel, D. et al. Metabolic engineering of tomato fruit enriched in L-DOPA. Metab. Eng. 65, 185–196 (2021).Article CAS Google Scholar Patrauchan, M. A. et al. Catabolism of benzoate and phthalate in Rhodococcus sp. strain RHA1: redundancies and convergence. J. Bacteriol. 187, 4050–4063 (2005).Article CAS Google Scholar Salvador, M. et al. Microbial genes for a circular and sustainable bio-PET economy. Genes 10, 373 (2019).Schweigert, N., Zehnder, A. J. B. & Eggen, R. I. L. Chemical properties of catechols and their molecular modes of toxic action in cells, from microorganisms to mammals. Environ. Microbiol. 3, 81–91 (2001).Article CAS Google Scholar Li, Q., Aubrey, M. T., Christian, T. & Freed, B. M. Differential inhibition of DNA synthesis in human T cells by the cigarette tar components hydroquinone and catechol. Fundam. Appl. Toxicol. 38, 158–165 (1997).Kobayashi, T. et al. 3,5-Dihydroxybenzoic acid as a potent inhibitor of tyrosine phenol-lyase decreases fecal phenol levels in mice. J. Med. Chem. 68, 8786–8795 (2025).Article CAS Google Scholar Tang, X. L. et al. Process development for efficient biosynthesis of l-DOPA with recombinant Escherichia coli harboring tyrosine phenol lyase from Fusobacterium nucleatum. Bioprocess. Biosyst. Eng. 41, 1347–1354 (2018).Article CAS Google Scholar Zhang, X. et al. Endogenous 3,4-dihydroxyphenylalanine and dopaquinone modifications on protein tyrosine: links to mitochondrially derived oxidative stress via hydroxyl radical. Mol. Cell. Prot. 9, 1199–1208 (2010).Article CAS Google Scholar Valenzuela-Ortega, M. & French, C. Joint universal modular plasmids (JUMP): a flexible vector platform for synthetic biology. Synth. Biol. 6, ysab003 (2021).Download referencesAcknowledgementsB.R. acknowledges a PhD studentship from the Industrial Biotechnology Innovation Centre (IBioIC). S.W. acknowledges a Future Leaders Fellowship from UKRI (MR/S033882/1), Sustainable Manufacturing grant from EPSRC (EP/W019000/1), Engineering Biology Mission Hub grant from BBSRC (BB/Y007972/1) and Sustainable Manufacturing Hub grant (UKRI1891). C.W.W. acknowledges a sLOLA grant from BBSRC (BB/X003027/1). C.U. acknowledges NSRF funding via the Research and Innovation Acceleration Agency for Competitiveness and Area Development (RCAD) (Program Management Unit for Technology and Innovation for Future Industries (PMU-B): Brainpower for Future Industries; grant number B38G690002). B.E. acknowledges research assistant and studentship funds from VISTEC. We thank G. Leung and T. Hinchcliffe (Impact Solutions) for insightful discussions, B. French (API Foilmakers) for providing HSF samples, C. P. Lilly and A. Molnar for providing a strain of C. reinhardtii CC1690 and experimental assistance and R. Cox (C-Source Renewables) for supplying bread waste glucose syrups.Author informationAuthors and AffiliationsInstitute of Quantitative Biology, Biochemistry and Biotechnology, School of Biological Sciences, University of Edinburgh, Edinburgh, UKBenjamin Royer, Yuta Era, Marcos Valenzuela-Ortega, Thomas W. Thorpe, Connor L. Trotter, Kitty Clouston, John F. C. Steele, Nicoll Zeballos, Eugene Shrimpton-Phoenix, Christopher W. Wood & Stephen WallaceVidyasirimedhi Institute of Science and Technology, Wangchan Valley, Rayong, ThailandBhumrapee Eiamthong & Chayasith UttamapinantAuthorsBenjamin RoyerView author publicationsSearch author on:PubMed Google ScholarYuta EraView author publicationsSearch author on:PubMed Google ScholarMarcos Valenzuela-OrtegaView author publicationsSearch author on:PubMed Google ScholarThomas W. ThorpeView author publicationsSearch author on:PubMed Google ScholarConnor L. TrotterView author publicationsSearch author on:PubMed Google ScholarKitty CloustonView author publicationsSearch author on:PubMed Google ScholarJohn F. C. SteeleView author publicationsSearch author on:PubMed Google ScholarNicoll ZeballosView author publicationsSearch author on:PubMed Google ScholarEugene Shrimpton-PhoenixView author publicationsSearch author on:PubMed Google ScholarBhumrapee EiamthongView author publicationsSearch author on:PubMed Google ScholarChayasith UttamapinantView author publicationsSearch author on:PubMed Google ScholarChristopher W. WoodView author publicationsSearch author on:PubMed Google ScholarStephen WallaceView author publicationsSearch author on:PubMed Google ScholarContributionsThe study was designed by S.W. and B.R. Experimental work was performed and analysed by B.R., Y.E., M.V.-O., T.W.T., C.L.T., K.C., J.F.C.S., N.Z., E.S.-P. and B.E. C.U., C.W.W. and S.W. provided project support and experimental guidance. The paper was written by S.W., B.R., Y.E., M.V.-O., T.W.T., C.L.T., K.C., J.F.C.S. and E.S.-P. All authors have given approval to the final version of the paper.Corresponding authorCorrespondence to Stephen Wallace.Ethics declarationsCompeting interestsThe authors declare no competing interests.Peer reviewPeer review informationNature Sustainability thanks the anonymous reviewers for their contribution to the peer review of this work.Additional informationPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Supplementary informationSupplementary InformationSupplementary Methods, Figs. 1–22 and Tables 1–7.Reporting SummarySource dataSource Data Fig. 2Raw data.Source Data Fig. 3Raw data.Source Data Fig. 4Raw data.Rights and permissionsOpen Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.Reprints and permissionsAbout this article