Interactions between humans and wildlife promote the transmission of parasites and pathogens(4, 5), which can lead to infectious disease outbreaks  (6, 7) , resulting in dramatic death tolls and longlasting socio-economic impacts  (5, 8, 9) . Understanding what shapes the spread of zoonotic (i.e., human-transmissible) pathogens and parasites (hereafter pathogens) across species is therefore not only one of the overarching goals of disease ecology, but also a public health priority  (5, 10) .

The wildlife trade creates opportunities for wildlife-to-human pathogen spillover at each stage of the supply chain, including harvesting, transport, breeding, retail, consumption and companionship  (2, 5, (11) (12) (13) (14) . For instance, a person buying three variable squirrels in Laotian wildlife markets has been estimated to have an 83% chance of getting at least one leptospirosis-infected individual  (13) . The wildlife trade is therefore often the source of disease outbreaks in humans  (2, 15) , perhaps most famously including the COVID-19 pandemic, which has been traced back to live mammals sold in the Huanan Seafood Wholesale Market in Wuhan  (16) . While the hunting and consumption of wild meat has been linked to some major epidemics, such as the HIV pandemic  (17)  and some Ebola outbreaks  (18) , other types of wildlife use and trade are also responsible for human infectious disease outbreaks  (2, 15) . For example, the trade in exotic pets has been implicated in a monkeypox outbreak in 2003 linked to prairie dogs in North America  (19) , and recent hospitalizations due to a rare Salmonella strain traced back to pet bearded dragons  (20) .

Despite growing awareness of the epidemic risk posed by the wildlife trade, we still know little about how it affects the exchange of pathogens between animals and humans  (13, (21) (22) (23) (24) (25) . Recent research on the determinants of host-pathogen interactions have mostly focused on environmental, life history, ecological and evolutionary drivers  (7, 11, (26) (27) (28) (29) (30) (31) (32) (33) . However, the role of human-wildlife interactions associated with the wildlife trade remains poorly studied  (5, 34, 35)  and previous empirical studies have only focused on quantifying the number of zoonotic pathogens occurring in traded animals  (22, 24, 36, 37) . As a result, although traded species are known to frequently host zoonotic pathogens  (13, 21, 24) , it remains unclear whether species occurring in the wildlife trade are more likely to share pathogens with humans compared to non-traded species. In theory, traded wildlife should be more likely to share pathogens with humans because frequent and close contact with humans increase opportunities for pathogen transmission  (4, 5) . This can occur either through spillover, where pathogens from wildlife infect humans  (38)  (thus becoming zoonotic), or through reverse zoonosis, where human pathogens (or wildlife pathogens that spilled over humans) infect wildlife (e.g., farmed minks contracting SARS-CoV-2)  (4, 39, 40) . Moreover, the more frequently species are traded, the more pathogens they should share with humans.

Here, we tested whether traded wildlife species are more likely to share pathogens with humans compared to non-traded wildlife, and whether the number of years that a species spends on the global wildlife market (in the last 40 years) predicts the number of pathogens it shares with humans. We used the CITES  (41)  and LEMIS  (42, 43)  trade databases to assess the occurrence of mammal species in the global wildlife trade (Supplementary Fig.  1 ). The CITES trade database is the most comprehensive database on the international wildlife trade in live animals and animal products. It is global in scale and span over five decades (1975 to present), although, it only contains records for CITES-listed species (~13% of all mammal species)  (41, 44) . The LEMIS trade database compiles records of live animals and animal products imported in the United States between 2000 and 2014  (42, 43) . Although this database is limited to U.S. importations and covers a shorter timespan than the CITES trade database, it includes all species and thus, provides valuable information on the trade in species not recorded by CITES. Here, we focused on the legal trade and excluded all trade records labelled as illegal from both databases, because, by definition, they are not systematically detected and recorded  (45) . In a second analysis, we analysed over 170,000 international trade records in CITES-listed mammals, covering 40 years of legal global trade in wild animals, alive and as products. We focused our analysis on mammals because they are the main reservoir of zoonotic pathogens  (21, 31)  and the best documented vertebrate clade  (34, 46, 47) . We assessed which wild mammal species were known to share pathogens with humans by querying the largest-to-date hand-curated mammal-pathogen associations database(48) (>190,000 documented associations between mammals and viral, bacterial, fungal, helminth, and protozoan pathogens and parasites). We accounted for potential biases and confounding factors in all our analyses. Because human-wildlife pathogen exchanges are facilitated by species phylogenetic relatedness with humans(7), we accounted for mammals' phylogenetic distance to humans. Moreover, because pathogens are unequally distributed geographically  (10, 49, 50)  and among taxa  (31, 51)  and because they are also more likely to be detected in well studied species(29, 52), we controlled for taxonomic relatedness, (order, family and genus as nested random effects on the intercept), biogeography (biogeographic realm of origin as random effect on the intercept), and research effort (research effort index as fixed effect; see Methods). Finally, we also accounted for the potential confounding effect of synanthropy on the link between zoonotic pathogens and the wildlife trade. This effect could exist because synanthropic species, that are species that live in or near human-modified environments, are more likely to share pathogens with humans (because of frequent contacts with humans or domesticates  (11, 29, 30, 53) ) and to occur in the wildlife trade (because they are easier to harvest than species occurring only in remote areas  (15, 54) ).

Among 1,505 traded mammal species, half shared at least one pathogen with humans, while it is the case for only eight percent of non-traded mammals (Fig.  1abc ). Using a binomial mixed-effects model, we show that the probability that a mammal species shares pathogens with humans (i.e., risk ratio) is 2.4 times higher in traded than in non-traded species (Odds Ratio [CI95%]trade = 2.72 [2.06-3.6], ptrade < 0.0001; Fig.  1c ) and 1.7 times higher in synanthropic than in non-synanthropic species (Odds Ratio [CI95%]synanthropy = 1.94 [1.43-2.62], psynanthropy < 0.0001; Fig.  1c ; see details in Supplementary Text 1). We then used a structural equation model (SEM) to test whether the probability of sharing pathogens with humans is directly linked to trade, or the result of a confounding effect of synanthropy, while accounting for the links between these variables and research effort (Fig.  1d ). Our SEM supports a direct positive effect of trade on the probability that wild mammals share pathogens with humans (Std. est. = 0.14, R 2 marginal = 0.49; Fig.  1d ; Supplementary Text 2). In addition, synanthropy was positively associated with both zoonotic host status (Std. est. = 0.08) and trade status (Std. est. = 0.15). However, our SEM shows limited explanatory power for the effect of synanthropy on trade (R 2 marginal = 0.02; Fig.  1d ), and a weak effect of synanthropy on zoonotic host status (compared to other effects in the SEM; Fig.  1d ). This indicates that synanthropy does not act as a confounding factor on the link between trade and human-wildlife pathogen sharing. Interestingly, our SEM supports a positive effect of trade and synanthropy on research effort, thus suggesting that part of the positive link between research effort on zoonotic host status is mediated by these two variables (i.e., traded and synanthropic species are more studied and thus their pathogens are better known; Fig.  1d ). Overall, these results support our main hypothesis that, even after accounting for sampling bias, species that are present in the wildlife trade are more likely to share pathogens with humans, because repeated and close cross-species contact creates opportunities for pathogen transmission. However, while some species have been continuously traded in the last four decades, others only sporadically occur in the global wildlife trade(55, 56) (Fig.  2a ). Therefore, the opportunities for cross-species contact and thus pathogen transmission may strongly differ from one species to the other.

Species that have been traded more frequently and for longer periods of time have had more opportunities to exchange pathogens with humans and should therefore be more likely to host a greater number of zoonotic pathogens. To test this, we performed a temporal analysis of the global trade in wild mammals over the last 40 years (1980-2019) using the CITES trade database, the only long-term longitudinal dataset of the global wildlife trade  (41) . We analysed 171,749 trade records in 546 mammal species that have been traded at least once between 1980 and 2019 (among species listed in CITES Appendices I or II; see Methods) and calculated the frequency at which species were traded in the last 40 years, as the number of years with recorded trade (hereafter "time in trade'; Fig.  2a ). Using a hurdle negative binomial mixed-effect model (Supplementary Text 3-4 for detailed results), we show that, as expected, time in trade increases both the probability of sharing at least one pathogen Preprint 6 with humans (Zero-inflation sub-model: Odds Ratio [CI95%] = 4.4 [2.8-6.7], p < 0.0001; Fig  2b , c , d ) and the number of zoonotic pathogens that a mammal species hosts (Conditional sub-model: Incidence Rate Ratio [CI95%] = 1.5 [1.3-1.8], p < 0.0001; Fig  2b , c , d ). Using this model, we estimate that, on average over the period considered, a wild mammal species shares one additional pathogen with humans every 9 years of presence in trade (mean [CI95%] = 8.9 [7.4 -11.2], see Supplementary Text 5 for calculation details). This implies that pathogens hosted by traded species that currently do not infect humans are likely to do so in the future. These results highlight the dynamic nature of human-wildlife pathogen interaction networks and parallel the positive correlation observed between time since domestication and number of pathogens shared with humans in domesticated mammals  (58) . Although these two trends are difficult to compare (in part because animal domestication is much older than the global wildlife trade), they both highlight the importance of sustained and close contact between animals and humans in promoting cross-species pathogen transmission and in shaping the hostpathogen associations that exist today. Preprint 7 (fixed effects) in the two sub-models of the hurdle negative binomial mixed effect model (zero-inflation sub-model on the left, conditional sub-model on the right). The x-axis has been deliberately modified to allow an intuitive and undistorted comparison of positive and negative effects. c, On the left, time in trade among species sharing pathogens with humans or not. On the right, number of pathogens shared with humans as function of time in trade. Each point is a mammal species. The size of points is proportional to research effort. The vertical axis is square rooted to improve readability. d, Average effects (slope ± CI95%) of time in trade on the probability of sharing at least one pathogen with humans (left) and on pathogen richness (right).

Animal-to-human transmission (spillover) is the most likely driver of trade-related pathogen exchange between wild animals and humans, because of the asymmetry of human-animal interactions (e.g., humans regularly consume wildlife and livestock, while the opposite is extremely rare)(4), and because wildlife frequently acts as a reservoir for human-transmissible pathogens, while the opposite is less common  (59) . Understanding the precise processes responsible for the patterns highlighted in this study will require more detailed data on the temporal sequence of pathogen transmission across animals and humans  (32) . Genomic tools will be central to retrace pathogen transmission across species  (17, (60) (61) (62) (63)  and interfaces (e.g., Mycobacterium bovis transmission across wildlife and livestock in South Africa(64)), and will accelerate the compilation of detailed records of pathogen prevalence in human and animal populations (wild, captive and domesticated(63, 65-67)). Moreover, museomics (the study of genomic data obtained from ancient and historic DNA from specimens in museum collections(  68 )) now allows to identify and quantify the prevalence of pathogens that infected wild and captive animals decades to centuries ago  (69) (70) (71) . Our analysis also uses on a simplified representation of host-pathogen interactions (i.e., presence/absence) and thus overlook both inter-and intra-specific variation in pathogen prevalence  (63, 65) . Higher resolution pathogen testing data will be crucial in assessing whether traded species are reservoirs with a high risk of pathogen transmission, common but dead-end hosts with low transmission risk, or anecdotal hosts  (72) . This information is essential to accurately identify high-risk species and focus regulations on them, rather than using total trade bans that could dilute control efforts or even backfire if trade is massively diverted to illegal channels  (73) .

Overall, our findings underscore the urgent need to enhance efforts to screen traded animals and animal products for pathogens and evaluate their potential for transmission to humans  (12, 13) . As this study focuses on species traded legally at international scale, it overlooks local but widespread wildlife markets (e.g., wild meat markets  (74, 75)  or local exotic pet markets(76)), as well as illegal markets  (22) .

Accounting for these components of the wildlife trade might improve our ability to evaluate risks related to cross-species transmission, and ultimately, to better predict and prevent future zoonotic Preprint 8 disease outbreaks. However, it will require important international efforts to continuously survey and compile information on these markets (e.g., identity and abundance of traded species, purpose of trade, presence and intraspecific prevalence of pathogens). Currently, this information is rare and scattered among numerous independent studies with an uneven geographical and temporal representation  (13, 75, 77, 78) .

The COVID-19 pandemic sparked a thorough re-evaluation of current wildlife trade regulations, exposing critical gaps in our ability to monitor and limit disease outbreaks linked to exploited and traded species  (79) . Currently, the primary international agreement regulating the wildlife trade, CITES, focuses exclusively on preventing species extinction caused by overexploitation of natural populations  (80) . Our findings are thus critical for ensuring the effectiveness of upcoming regulations aimed at pandemic prevention, including potential reforms to CITES(81), the WOAH/CITES collaborative agreement  (82)  and the proposed WHO pandemic agreement(83). By showing that traded wildlife species share one additional pathogen with humans for every decade of presence in the global wildlife market, our findings underscore that mitigating future zoonotic pathogen emergence will require reducing the volume of wildlife trade -including both species that pose a current risk to human health, and those that might someday soon, given the opportunity.

To account for knowledge bias, we assessed research effort by retrieving the number of publications per mammal species using the easyPubMed R package  (12) . We computed two indices of research effort per species by searching for keywords in the title and abstract of publications listed in PubMed  (13) . First, we calculated the total number of publications per mammal species, based on all possible valid specific Latin names and all possible common names (in English) from MDD. Second, we calculated the proportion of all publications that was focused on diseases or pathogens by refining our previous search with the following string: "pathogen* OR zoono* 'public health' OR 'emerging infectious disease' OR 'disease' OR parasit* OR virus OR viral OR helminth* OR bacteria* OR fung* OR protozoa OR infect*". We finally computed a unique index of research effort per species by multiplying the total number of publications (log-transformed and rescaled to [0-1]) to the proportion of these publications focusing on diseases and pathogens (logit-transformed and rescaled to [0-1]). The equation for calculating this index is as followed:

With 𝑅𝐸 𝑖 the index of research effort for species i, 𝐷 𝑖 the proportion of all publications that focused on diseases or pathogens for species i (to avoid infinite values, the logit of 0 and 1 were calculated as the logit of 0.025 and 0.975, respectively), and 𝑇 𝑖 the total number of publications for species i. This index of research effort combines the information on the total number of publications and proportion of these publications focused on pathogens and diseases (see Fig.  1a ).

Phylogenetic distance to humans. We used the VertLife database  (14)  to access phylogenetic relationships between mammals (downloaded on February 29 th , 2023). We calculated each species average phylogenetic distance to humans (in million years) from 20 randomly selected trees.

We assessed mammals' propensity to live inside or near humans and human settlements using the IUCN Red List database  (15) . A mammal species was listed as synanthropic if it was known to occur in urban areas (habitat code 14.4), rural gardens (14.5), water storage areas (15.1), aquaculture ponds (15.3), wastewater treatment areas (15.6) and canals and drainage channels (15.9). A total of 875 mammal species (out of 6,456) were considered as synanthropic (14%).

We used the CITES trade database  (16)  and the LEMIS trade database  (17)  to assess whether species were traded or not (following recommendations in Preprint 16 characterizing wildlife trade(  18 )). The CITES trade database is the most comprehensive database on the international wildlife trade in live animals and animal products and contains records for CITES-listed species from 1975 to present  (16, 19) . The LEMIS trade database compiles records of wildlife and wildlife products imported in the United States between 2000 and 2014  (17, 20) . Although this database is limited to U.S. importations and covers a shorter timespan than the CITES trade database, it covers all species and thus, provides valuable information on the trade in species not listed in CITES appendices and therefore absent from the CITES trade database. In this work, we considered species traded legally alive or as products, for commercial, personal, medical, scientific, and hunting trophy purposes. These trade purposes represent the majority of CITES and LEMIS trade records (84% and 99%, respectively). We excluded illegal trade records because they are not systematically reported  (21)  and cannot be easily compared with legal trade records.

The CITES trade database is the only long-term (1975present) and global compilation of the trade in wild animals  (19) . We used the aggregated version of the CITES trade database(16) (Comparative tabulation version 2023.1) to avoid duplicated trade records (e.g. when both importers and exporters report the same trade event). The CITES trade database totals 405,066 records of trade in mammals since 1975 (after taxonomic harmonisation).

However, we applied several filters to the database, thus reducing the number of analysed records to 290,241. First, we chose to consider only the period 1980-2019 (40 years) because the first five years  (1975) (1976) (1977) (1978) (1979)  and the last three years (2020-2022) of recording are incomplete, due to non-systematic recording during the first years (many species were not yet listed in CITES and many countries were not yet parties of CITES before 1980(  22 )), and of a time lag in reporting for the recent years  (16) . We also excluded species that were listed in CITES after 1980 because their trade was not reported systematically over the period 1980-2019, and species listed in CITES appendix III because their trade is not systematically recorded at global scale as CITES agreements protect (an report) them only when they are sourced from some specific countries  (23) . Using this refined dataset, we then calculated for each traded mammal species the number of years that the species appeared in trade over the last 40 years.

All data processing, statistical analyses and visualizations were performed in R and Rstudio  (24) (25) (26) (27) (28) . Generalised linear models were computed using the glmmTMB package  (29) . The statistical validity of each model was systematically assessed using the performance  (30)  and DHARMa(31) packages. For mixed effects models, whenever random effects with an estimation very close to zero created a singularity issue (that can hinder model's convergence), the problematic factors were removed from the model to allow a non-singular fit (following package recommendations  (32) ).

We used the sjPlot package  (33)  to calculate effect sizes of our GLMs. To interpret effect sizes expressed as odd ratios, we converted them in risk ratios following (eqn. 2). Sequential equation models were computed using the piecewiseSEM package  (34) .

With RR the risk ratio, OR the odd ratio, and 𝑃 𝑟𝑒𝑓 the baseline prevalence (i.e., the incidence in the baseline group).

To assess whether, among all mammal species, traded species are more likely to share pathogens with humans, we used a binomial mixed-effects model (logit link function) with zoonotic host status (i.e., the species shares at least a pathogen with humans or not) as response variable and trade status (i.e., traded or not) as predictor (fixed effect). To account potential confounding effects, we added the index of research effort (see eqn.

1), synanthropic status (i.e., synanthropic or not), and phylogenetic distance to humans as fixed-effect covariates. To account for the taxonomic relatedness between species, we added mammalian order, family and genus as nested random effects on the intercept. Species belonging to genera, families and orders with less than three representatives were excluded to avoid issues in the computation of random effects. To account for the biogeographical origin of mammal species, we added their biogeographical realm as a random effect on the intercept. Species occurring in two or more biogeographical realms were assigned to the category 'multiple'.

To further test whether the statistical relationship between trade and the probability of sharing pathogens with humans was direct or the result of the confounding effect of synanthropy, we used a structural equation model (SEM)  (34) . SEM is a multivariate analysis technique that integrates multiple predictor and response variables into a single directed network of hypothetical causal links, represented through linear equations. SEM enables the estimation of the strength and significance of each relationship in the model (using standardized estimates) while accounting for all specified relationships within the network  (34) , therefore allowing to account for the potential confounding effect of synanthropy on the relationship between trade and the probability of sharing pathogens with humans. Moreover, this modelling technique also allows to specify networks of interdependence between our variables of interest (zoonotic host status, trade and synanthropy) and research effort.

Our SEM model consists in the combination of three mixed effects sub-models, all accounting for the taxonomic relatedness between species (using mammalian order, family and genus as nested random effects on the intercept), and for the biogeographical origin of mammal species (using biogeographical realm as a random effect on the intercept). The first sub-model is a Gaussian model with research effort as the response variable and trade status and synanthropy status as predictors. This sub-model results Preprint 18 from the hypothesis that species that are traded and synanthropic are more likely to be studied. The second sub-model is a binomial model (logit link function) with trade status as response and synanthropy as predictor. This sub-model results from the hypothesis that synanthropy increases the probability of being traded. The third sub-model is a binomial model (logit link function) with zoonotic host status as response and trade status, synanthropy status, phylogenetic distance to humans, and research effort as predictors. This sub-model results from the hypotheses that the probability of being known to share zoonotic pathogens with humans increases with phylogenetic proximity to humans, research effort, synanthropy and trade.

We tested whether the number of years that a species has occurred in trade is positively linked to the probability of sharing pathogens with humans and to the number of pathogens shared. Here, we focused on mammal species listed in CITES appendices I and II, that were traded at least once between 1980 and 2019, based on records of the CITES trade database  (16) . We also excluded species that were listed in CITES after 1980 because their trade was not reported systematically over the period 1980-2019, and species listed in CITES appendix III because their trade is not systematically recorded at global scale as CITES agreements protect them only when they are sourced from some countries  (23) . Our analysis thus included 546 mammal species for which international trade events were recorded globally and continuously between 1980 and 2019. We then tested whether time in trade is positively linked to the probability of sharing pathogens with humans and to the number of pathogens shared with humans using a hurdle negative binomial mixed-effects model with pathogen richness as response variable (count, between 0 and 84) and time in trade (number of years, between 1 and 40) as predictor (i.e., fixed effects). We also added synanthropy (binary, synanthropic or not), phylogenetic distance to humans and research effort (see eqn. 1) as a fixed-effect covariates. To account for the taxonomic relatedness between species, we added mammalian order, family and genus as nested random effects on the intercept, to account for the biogeographical origin of mammal species, we added their biogeographical realm as a random effect on the intercept. After running the full model (i.e., all effects in both zero-inflation and conditional sub-models), random effects with low variance were removed to avoid singularity issues (Supplementary Text 3).

To assess the robustness of our results to variations in the host-pathogen association data used to test our hypotheses, we applied two independent filters to the CLOVER dataset, resulting in four distinct combinations (and thus datasets). We then repeated our analyses on each dataset and compared the outcomes. First, we excluded host-pathogen associations from the EID2 Preprint 19 database (one of the four databases compiled in CLOVER), as EID2 includes a substantial number of records from experimental infections(6). These records could introduce artificial host-pathogen associations and potentially inflate the number of species sharing pathogens with humans. Second, we filtered out all host-pathogen associations involving bacteria, helminths, fungi, and protozoa, focusing exclusively on viruses. This allowed us to determine whether our findings hold when considering only the pathogen group that poses the highest pandemic risk in the future  (35) . All our results remained consistent across both sources of variation in the host-pathogen association dataset. For a detailed description of the methods and results of this sensitivity analysis, see Supplementary Text 6. 746 databases (CITES and LEMIS). This Euler 747 plot does not include domesticated and 748 extinct species. 749 750 751 752 753 754 755 756 757 758

Table  S1 : List of databases used in this study.

Table S2: Matching rates with the reference database (MDD) for the five main sources of 761 information used in this study. 762 Initial number of species Number of species matched to MDD Matching rate (%) CLOVER 1364 1341 98 CITES 838 783 93 LEMIS 1449 1348 93 IUCN (synanthropic species) 1770 1746 99 Mammal phylogeny 5911 5658 96 Table S3: List of mammal species excluded from the study and reason for exclusion 763 Excluded species Reason Homo_sapiens human Bos_domesticus domestic Bos_frontalis domestic Bos_grunniens domestic Bos_indicus domestic Bos_javanicus domestic Bos_mutus domestic Bos_taurus domestic Bubalus_arnee wild relative to domesticate Bubalus_bubalis domestic Camelus_bactrianus domestic Camelus_dromedarius domestic Camelus_ferus wild relative to domesticate Canis_familiaris domestic Canis_lupus wild relative to domesticate Capra_aegagrus wild relative to domesticate Capra_hircus domestic Cavia_porcellus domestic Cavia_tschudii wild relative to domesticate Equus_africanus wild relative to domesticate Equus_asinus domestic Equus_caballus domestic Equus_ferus wild relative to domesticate Felis_catus domestic Felis_lybica wild relative to domesticate Lama_glama domestic Lama_guanicoe wild relative to domesticate Lama_pacos domestic Lama_vicugna wild relative to domesticate Mus_musculus domestic Mustela_furo domestic Mustela_putorius wild relative to domesticate Neogale_vison domestic Oryctolagus_cuniculus domestic Ovis_aries domestic Ovis_gmelini wild relative to domesticate Rattus_norvegicus domestic Sus_domesticus domestic Sus_scrofa wild relative to domesticate Antillomys_rayi extinct Archaeolemur_edwardsi extinct Bettongia_anhydra extinct Bettongia_pusilla extinct Boromys_offella extinct Boromys_torrei extinct Bos_primigenius extinct Brotomys_voratus extinct Caloprymnus_campestris extinct Chaeropus_ecaudatus extinct Chaeropus_yirratji extinct Conilurus_albipes extinct Conilurus_capricornensis extinct Coryphomys_buehleri extinct Coryphomys_musseri extinct Cryptoprocta_spelea extinct Dusicyon_australis extinct Dusicyon_avus extinct Geocapromys_caymanensis extinct Geocapromys_columbianus extinct Geocapromys_thoracatus extinct Heteropsomys_insulans extinct Preprint 23 Hexolobodon_phenax extinct Hippopotamus_lemerlei extinct Hippopotamus_madagascariensis extinct Hippotragus_leucophaeus extinct Hydrodamalis_gigas extinct Hyperplagiodontia_araeum extinct Isolobodon_montanus extinct Isolobodon_portoricensis extinct Juscelinomys_candango extinct Lagorchestes_asomatus extinct Lagorchestes_leporides extinct Lagostomus_crassus extinct Lenomys_grovesi extinct Leporillus_apicalis extinct Lutra_nippon extinct Macrotis_leucura extinct Megaladapis_madagascariensis extinct Megalomys_desmarestii extinct Megalomys_georginae extinct Megalomys_luciae extinct Megaoryzomys_curioi extinct Melomys_rubicola extinct Melomys_spechti extinct Neogale_macrodon extinct Neomonachus_tropicalis extinct Nesophontes_edithae extinct Nesophontes_hemicingulus extinct Nesophontes_hypomicrus extinct Nesophontes_major extinct Nesophontes_micrus extinct Nesophontes_paramicrus extinct Nesophontes_zamicrus extinct Nesoryzomys_darwini extinct Nesoryzomys_indefessus extinct Noronhomys_vespuccii extinct Notamacropus_greyi extinct Notomys_amplus extinct Notomys_longicaudatus extinct Notomys_macrotis extinct Notomys_mordax extinct Notomys_robustus extinct Oligoryzomys_victus extinct Onychogalea_lunata extinct Oryzomys_antillarum extinct Oryzomys_nelsoni extinct Palaeopropithecus_ingens extinct Pennatomys_nivalis extinct Perameles_eremiana extinct Perameles_fasciata extinct Perameles_myosuros extinct Perameles_notina extinct Perameles_papillon extinct Peromyscus_pembertoni extinct Pipistrellus_murrayi extinct Pipistrellus_sturdeei extinct Plagiodontia_ipnaeum extinct Plagiodontia_spelaeum extinct Potorous_platyops extinct Prolagus_sardus extinct Pseudomys_auritus extinct Pseudomys_glaucus extinct Pteropus_allenorum extinct Pteropus_brunneus extinct Pteropus_coxi extinct Pteropus_pilosus extinct Pteropus_subniger extinct Pteropus_tokudae extinct Rattus_macleari extinct Rattus_nativitatis extinct Rattus_sanila extinct Rhizoplagiodontia_lemkei extinct Rucervus_schomburgki extinct Solenodon_arredondoi extinct Solenodon_marcanoi extinct Solomys_spriggsarum extinct Thylacinus_cynocephalus extinct Tonatia_saurophila extinct Detailed results for the binomial mixed effects model (presented in Fig. 1c) used to test 765 whether traded species are more likely to share pathogens with humans, while accounting for 766 potential confounding effects. 767 768 Number of obs: 5446, groups: BioRealm, 8; genus:family:order, 540; family:order, 105; order, 23 769 AIC BIC logLik deviance df.resid 2566.4 2625.8 -1274.2 2548.4 5437 Random effects Groups Name Variance Std.Dev. BioRealm (Intercept) 0.1836 0.4285 genus:family:order (Intercept) 0.3481 0.5900 family:order (Intercept) 0.2956 0.5437 order (Intercept) 0.2336 0.4833 Fixed effects Estimate Std. Error z value p-value (Intercept) -3.352565 0.538084 -6.231 4.65E-10 *** Trade [yes] 1.00201 0.141595 7.077 1.48E-12 *** Synanthropy [yes] 0.660347 0.153943 4.29 1.79E-05 *** Research effort 24.917564 1.010106 24.668 2.00E-16 *** Phylo. dist. to humans -0.005152 0.002541 -2.027 0.0426 * R² Conditional 0.62 Marginal 0.49 Effects size estimate conf.low conf.high statistic p.value Effect size Trade [yes] 2.7237589 2.0636819 3.5949642 7.07661 1.48E-12 2.72*** Synanthropy [yes] 1.9354506 1.4313489 2.6170901 4.289509 1.79E-05 1.94*** Research effort 5.7037099 4.9668204 6.5499261 24.668221 2.35E-134 5.7*** Phylo. dist. to humans 0.7762342 0.6076343 0.9916153 -2.027355 0.04262607 0.78* 770 771 Fig. S2: Average effects of trade and synanthropy 772 on the probability of sharing at least one 773 pathogen with humans. 774 Code Z Shares at least one pathogen with humans T Traded S Synanthropic r Research effort p Phylogenetic distance to humans 778 SEM summary Response Predictor Estimate Std.Error DF Crit.Value P.Value Std.Estimate r S 0.0373 0.0024 5446 15.3658 <0.0001 0.1821 r T 0.0623 0.0023 5446 27.3731 <0.0001 0.3589 T S 1.0323 0.1245 5446 8.2919 <0.0001 0.1495 Z T 1.002 0.1416 5446 7.0766 <0.0001 0.1417 Z S 0.6603 0.1539 5446 4.2896 <0.0001 0.0791 Z r 24.9176 1.0101 5446 24.6683 <0.0001 0.6122 Z p -0.0052 0.0025 5446 -2.0274 0.0426 -0.0891 R² Response Marginal Conditional r 0.16 0.48 T 0.02 0.56 Z 0.49 0.62 Tests of direct separation Independ.Claim Test.Type DF Crit.Value P.Value r ~ p + … coef 5446 -1.0812 0.2796 T ~ p + … coef 5446 -1.3282 0.1841 AIC -9397.027 Global goodness-of-fit: Chi-Squared = 3.054 with P-value = 0.217 and on 2 degrees of freedom Fisher's C = 5.933 with P-value = 0.204 and on 4 degrees of freedom 779 780 781 Fixed effects, zero-inflation Estimate Std. Error z value p-value (Intercept) 0.22561 0.52778 0.427 0.669 n.s. Time in trade -0.110438 0.016393 -6.737 <0.0001 *** Synanthropy [yes] 0.15854 0.452176 0.351 0.726 n.s. Research effort -15.311148 2.345604 -6.528 <0.0001 *** Phylo. dist. to humans 0.025876 0.004338 5.965 <0.0001 *** Fixed effects, conditional Estimate Std. Error z value p-value (Intercept) -0.223679 0.270244 -0.828 0.408 n.s. Time in trade 0.031125 0.005669 5.49 <0.0001 *** Synanthropy [yes] -0.055035 0.186671 -0.295 0.768 n.s. Research effort 9.424277 1.036089 9.096 <0.0001 *** Phylo. dist. to humans -0.008383 0.001803 -4.65 <0.0001 *** R² for zero-inflated models: 0.66 787 Effects size, zero-inflation estimate conf.low conf.high statistic p.value Effect size TradeFrequency 0.2298062 0.1498161 0.3525047 -6.7367423 <0.0001 0.23*** Synanthropy [yes] 1.171799 0.4830148 2.8427969 0.3506163 0.726 1.17 Research effort 0.2479337 0.1631061 0.3768782 -6.5272465 <0.0001 0.25*** Phylo. dist. to humans 5.2984494 3.0630723 9.1651662 5.9636804 <0.0001 5.3*** Effects size, conditional estimate conf.low conf.high statistic p.value Effect size TradeFrequency 1.5135212 1.3053556 1.7548831 5.4897721 <0.0001 1.51*** Synanthropy [yes] 0.9464512 0.6564525 1.3645615 -0.2948279 0.768 0.95 Research effort 2.3593736 1.9609093 2.8388074 9.0947895 <0.0001 2.36*** Phylo. dist. to humans 0.5826235 0.4639791 0.7316064 -4.6499482 <0.0001 0.58** Interpretable effect sizes. 796 estimate conf.low conf.high TradeFrequency 4.3514926 2.8368416 6.6748483 Synanthropy [yes] 0.8533887 0.3517663 2.0703299 Research effort 4.0333366 2.6533774 6.1309802 Phylo. dist. to humans 0.1887345 0.1091088 0.3264696 797 798

Estimation of the rate at which species share new pathogens with humans as a function of time in trade.

First, we computed the predicted average response (i.e., number of pathogens shared with humans) across the range of the predictor variable (i.e., time in trade) using ggaverage from the ggeffects package (Fig.  S3a ). Average response corresponds to population-level predictions that are representative of covariate distribution.

We then calculated the numerical derivative of this average effect using finite differences, calculated as:   S3b ). We finally calculated the mean±CI95% rate of change over the whole range of the predictor variable (Fig.  S3b ). Number of species with 1+ zoonotic pathogens: A1: 871, A2: 799, A3: 699, A4: 691. Detailed summary of each model: A1 We found only minor differences between the 4 datasets. The main one being the nonsignificant effect of phylogenetic distance to humans in datasets A3 and A4 (p=0.0548 and 0.0598, respectively). Detail summaries of each SEM are displayed hereafter.