restoration ecology research paper

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here .

Loading metrics

Open Access

Peer-reviewed

Research Article

A systematic review of ecological attributes that confer resilience to climate change in environmental restoration

* E-mail: [email protected]

Affiliations School for Marine and Environmental Affairs, University of Washington, Seattle, Washington, United States of America, Ocean Associates Inc., under contract to Northwest Fisheries Science Center, National Marine Fisheries Services, National Oceanic and Atmospheric Association, Seattle, Washington, United States of America

Contributed equally to this work with: Tim Beechie, Terrie Klinger

Affiliation Fish Ecology Division, Northwest Fisheries Science Center, National Marine Fisheries Service, National Oceanic and Atmospheric Association, Seattle, Washington, United States of America

Affiliation School for Marine and Environmental Affairs, University of Washington, Seattle, Washington, United States of America

• Britta L. Timpane-Padgham, 
• Tim Beechie, 
• Terrie Klinger

• Published: March 16, 2017
• https://doi.org/10.1371/journal.pone.0173812
• Reader Comments

Ecological restoration is widely practiced as a means of rehabilitating ecosystems and habitats that have been degraded or impaired through human use or other causes. Restoration practices now are confronted by climate change, which has the potential to influence long-term restoration outcomes. Concepts and attributes from the resilience literature can help improve restoration and monitoring efforts under changing climate conditions. We systematically examined the published literature on ecological resilience to identify biological, chemical, and physical attributes that confer resilience to climate change. We identified 45 attributes explicitly related to climate change and classified them as individual- (9), population- (6), community- (7), ecosystem- (7), or process-level attributes (16). Individual studies defined resilience as resistance to change or recovery from disturbance, and only a few studies explicitly included both concepts in their definition of resilience. We found that individual and population attributes generally are suited to species- or habitat-specific restoration actions and applicable at the population scale. Community attributes are better suited to habitat-specific restoration at the site scale, or system-wide restoration at the ecosystem scale. Ecosystem and process attributes vary considerably in their type and applicability. We summarize these relationships in a decision support table and provide three example applications to illustrate how these classifications can be used to prioritize climate change resilience attributes for specific restoration actions. We suggest that (1) including resilience as an explicit planning objective could increase the success of restoration projects, (2) considering the ecological context and focal scale of a restoration action is essential in choosing appropriate resilience attributes, and (3) certain ecological attributes, such as diversity and connectivity, are more commonly considered to confer resilience because they apply to a wide variety of species and ecosystems. We propose that identifying sources of ecological resilience is a critical step in restoring ecosystems in a changing climate.

Citation: Timpane-Padgham BL, Beechie T, Klinger T (2017) A systematic review of ecological attributes that confer resilience to climate change in environmental restoration. PLoS ONE 12(3): e0173812. https://doi.org/10.1371/journal.pone.0173812

Editor: Maura (Gee) Geraldine Chapman, University of Sydney, AUSTRALIA

Received: September 7, 2016; Accepted: February 26, 2017; Published: March 16, 2017

This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.

Data Availability: All relevant data are within the paper and its Supporting Information files.

Funding: The authors received no specific funding for this work.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Substantial degradation of earth’s ecosystems—and powerful legal mandates such as the U.S. Endangered Species Act, U.S. Clean Water Act, E.U. Water Framework Directive, and E.U. Habitats Directive—have led many governmental agencies, non-profit organizations, and private interest groups to invest in restoration efforts. This ‘restoration economy’ was recently estimated to contribute $24.86 billion and 221,000 jobs annually to the U.S. economy [ 1 ]. Yet despite such monumental investments, ecological restoration has often been unsuccessful in reducing extinction rates and slowing declines in habitat quality [ 2 – 5 ]. On the other hand, evidence of increased biodiversity and improved ecosystem function following restoration demonstrates that restoration can be successful in rehabilitating the condition of ecosystems [ 5 , 6 ], and restoration now serves as an accepted and widely practiced management action.

Ecological restoration proceeds in the face of advancing climate change, which imposes additional stress on systems already under pressure from human use and this can undermine the long-term success of restoration efforts [ 7 ]. To address this concern, many have suggested a shift away from static restoration end points and towards dynamic and adaptive ecological process goals [ 3 , 8 , 9 ]. Evidence suggests that climate change impacts on populations and communities are increasingly considered in the development of management priorities and adaptation plans. For example, recent climate change studies have utilized trait-based vulnerability assessments to identify both potential impacts and inherent natural sources of climate-change resilience for individual species [ 10 – 14 ]. These assessments have in turn informed the development of decision-support frameworks to incorporate climate change into restoration planning [ 15 , 16 ].

Integrating resilience concepts and attributes could help improve restoration and monitoring efforts under conditions of climate change. Resilience approaches to restoration can foster adaptation to future climate impacts [ 15 , 17 – 19 ] by restoring dynamic processes that promote natural variability and biodiversity within ecological systems, and reducing the risk of dramatic ecosystem change, sharp declines in populations, or loss of ecosystem services [ 20 – 22 ]. Ecological resilience incorporates concepts of dynamic feedbacks, unpredictable change, and variation [ 23 , 24 ]. Here we use the resilience perspective of Walker et al. [ 25 ] that defines resilience as the capacity of a system to absorb disturbance and reorganize in ways that retain essentially the same functions, structures, identities, and feedbacks. This definition includes two important mechanisms of resilience, namely resistance to change and recovery from change.

To understand how resilience attributes can be integrated into restoration practices under climate change, we first distilled common attributes of ecological resilience from the published literature. We then applied a ‘climate filter’ to identify attributes likely to confer resilience under changing climate conditions. We further classified these attributes according to their ecological scale of application. We provide three examples to illustrate how practitioners can select resilience attributes that are appropriate for specific management applications. Finally, we outline general strategies for integrating resilience into restoration planning and monitoring in a changing climate.

Literature selection and examination

We examined the scientific literature to extract attributes of species or ecosystems that have been reported to confer ecological resilience. Using Web of Knowledge, one of us (BLTP) searched using the following terms: (river* OR stream OR (wetland NOT in title) OR ecosystem OR environment*) AND (restor* OR recov* OR re-creat* OR rehabilitat*) AND (resilienc* OR “ecological integrity”), restricting our search to papers published from 2009–2013. From a total of 915 search results, 232 articles were selected for further examination if the title described a scientific study investigating the resilience of some ecological characteristic(s). Of the 232 articles, 111 were selected for full review based on relevance to the study objectives as inferred from the abstract. Fifty-nine additional articles were gleaned from the selected literature based on best professional judgment of their fit with the goals of this study. These articles were added to the analysis for a total of 170 articles examined in this study ( S1 Fig ). For consistency, and to reduce inter-observer variation, all examination of the literature was performed by BLTP.

Resilience attribute identification

Attributes of ecological resilience were selected for further consideration if they were (1) typical of more than one ecosystem or species, (2) distinct from other attributes, and (3) measureable. From the assembled attributes, we created a database in which every attribute from each publication was recorded, along with the source of publication, ecosystem context, metric(s) used to measure or monitor the attribute, and whether the attribute was identified as conferring resistance to or recovery from disturbance. We then grouped the attributes into major categories and combined attributes that were similar to produce a list of 51 resilience attributes classified into five major categories. The resilience attributes that we identified come from a wide-range of ecosystems and range from more general (e.g. energy flows) to more specific (e.g. soil and air carbon balance). Given that our primary purpose in this study was to broadly inform restoration practices under climate change, we elected to retain as many attributes as possible and to broadly define terms to maximize utility to practitioners working across a range of scales and contexts. Practitioners can choose to further refine attributes and definitions based on specific applications.

Climate change filter

We next evaluated the attributes to identify those that were considered to confer resilience to climate change. An attribute passed through the climate change filter if the article specifically mentioned an attribute in relation to climate change or climate impacts. For example, if the article discussed how an attribute might confer resilience to climate change or an ecological feature directly affected by climate change such as stream flow or temperature, the attribute was retained in our list of resilience attributes. A total of 45 (out of 51) attributes remained after the climate filter was applied. Attributes eliminated by the climate filter (population (beta) diversity, gamma diversity, food-web complexity, large woody debris, salinity, and historical flow-disturbance regimes) may confer resilience to climate change impacts in some situations, but that was not apparent in the articles evaluated.

Attribute classification

We classified the 45 resilience attributes from our review into five categories that roughly equate to ecological scale: (1) individual attributes, (2) population attributes, (3) community attributes, (4) ecosystem attributes, and (5) process attributes. We used best professional judgment to classify each attribute by two criteria that we felt were integral for any restoration project: restoration focus (e.g., is the restoration effort species-specific, habitat-specific, or system-wide focused?) and scale of application (e.g., do restoration actions take place at a population, site, or ecosystem scale?). ‘Restoration focus’ refers to the type of project an attribute is best suited for. For example, a population attribute such as density is likely more suitable for a restoration effort that aims to restore a species, whereas a community attribute such as functional diversity is more applicable to a restoration effort aiming to restore an ecosystem. Some attributes were assigned to more than one category because they are suitable for more than one restoration focus. ‘Scale of Application’ denotes the scale an attribute can be used to describe (e.g., generally population scales for biological attributes, and site or ecosystem scales for environmental attributes). Several attributes were assigned to more than one scale because scale varies depending on environmental context or project type. Our classification does not account for every potential application; consequently, users may need to tune some classifications to meet the needs of particular systems or projects.

In a practical sense, the resilience attributes all serve as ecological metrics that can be used for monitoring efforts (e.g. population size, presence of propagules, recovery time after disturbance) and/or setting ecological goals for restoration projects (e.g. genetic diversity, increase or establish refugia or support areas, release from competition or predation).

Attribute selection and sample applications

The attribute classifications can be used to create a decision support table (DST) by using a filtering function ( S1 Table ) so that practitioners can identify resilience attributes that are best suited to the focus and spatial scale of a specific restoration plan or project. To create a sub-list of suitable resilience attributes, a practitioner can sort attributes by asking: (1) what is the focus of the restoration project? and (2) what is the spatial scale of the specific needs? The output comprises a sub-set of resilience attributes that are more likely to be relevant to the specific plan or project.

To illustrate use of the DST in restoration planning, we created three sample applications. We selected three different restoration efforts focused at different spatial scales to demonstrate (1) how relevant resilience attributes can be identified for a specific project and (2) how the attributes selected will differ according to the type of project. We use the Kissimmee watershed system as an example of restoration at the ecosystem scale, a Pacific salmon ( Oncorhynchus spp.) population as an example of restoration at the population scale, and vulnerable coral species as an example of restoration at the site scale.

Results and discussion

Summary of literature examined.

Most articles referred to riverine and coral ecosystems (32 and 28 citations, respectively), followed by terrestrial, marine, and forest ecosystems ( Fig 1 ). Rocky shore, wetland, and grassland ecosystems were less commonly cited (4 citations each). While our search terms did include river or stream habitats, as that was our intended focus originally, we also include broader terms of ‘ecosystem’ or ‘environment’ which resulted in a diverse representation of habitat types. The number of times an individual attribute was cited varied from 1–20. By attribute type, ecosystem attributes were most frequently cited ( Table 1 ), but there were more total citations of process attributes (63) because more than one third of all attributes (16, or 36%) were classified as process attributes.

• PPT PowerPoint slide
• PNG larger image
• TIFF original image

https://doi.org/10.1371/journal.pone.0173812.t001

https://doi.org/10.1371/journal.pone.0173812.g001

More than half (33 of 45) of the resilience attributes were defined as equivalent to resistance (to perturbation), and many others (42 of 45) used resilience synonymously with recovery, or as an outcome of recovery ( Table 1 ). Across all studies, 30 of the 45 attributes were used in both ways (i.e., some studies considered resilience to mean recovery, while others considered it to mean resistance). However, only a few sources explicitly considered resilience to include both concepts: that of resistance, or the ability of an ecosystem or community to persist through a disturbance, and that of recovery, or rate at which a system or community returns to its functional state.

Several studies in our review consisted of a census of resilience attributes within a specific ecosystem type [ 15 , 51 , 54 , 58 , 66 , 113 , 126 ]. Maynard et al. [ 66 ] used a literature review to distill a list of 19 ‘resilience indicators’ that ‘conferred resilience’ within coral reef systems. In a study by McClanahan et al. [ 113 ], a group of 50 scientists ranked and scored an existing list of ‘resilience factors’ also in coral reef systems. Bernhardt and Leslie [ 126 ] conducted a comprehensive study exploring sources of resilience to climate change within coastal marine ecosystems and found three important ecological themes for conferring resilience: connectivity, biological diversity, and adaptability. Our review, which included the aforementioned studies, also found these three ecological themes to be widely cited in the literature, but to these we add habitat variability and condition, presence of refugia or support areas, and natural disturbance history as commonly-cited themes. We discuss these themes and the influence of human pressures on resilience in the following sections.

Connectivity

Connectivity was found to enhance capacity for self-organization and recovery at multiple scales, including interactions between species at community and population scales and connectivity of habitat types and ecosystems throughout both space and time [ 126 ]. Connectivity supports resilience by allowing movement of propagules, larvae and adults to recolonize a disturbed area or replenish an area with new genetic material and enhance local populations. Fritz and Dodds [ 42 ] observed how flooding events increasing invertebrate populations by connect intermittent pools in rivers and provided for colonization and dispersed young individuals. In coral reef systems, Olds et al. [ 78 ] found that connectivity between mangrove ecosystems and protected reefs in eastern Australia enhanced herbivore biomass and richness.

Connectivity of various healthy habitat types helps maintain species that use a variety of habitats for feeding, reproduction, resting, rearing, refuge, and migrating [ 51 ]. In riverine systems, ecological connectivity is important for maintaining natural variability and supporting productivity [ 102 ]. Many species, such as Pacific salmon, rely on movement throughout the system, including the mainstem, tributaries, floodplain habitats, and deltas. Removal of anthropogenic barriers to migration can help increase resilience of aquatic biota to climate change impacts such as changing flow regimes [ 151 ]. Ecosystem connectivity is also critical to help regulate essential abiotic and biotic processes such as flow, temperature, water quality, aquatic and terrestrial interactions and food webs.

Biodiversity and the insurance hypothesis

Alpha diversity, genetic diversity, and functional diversity were the most frequently cited diversity attributes. Duffy [ 154 ] found that on average, greater species richness increased resource use within trophic levels and accumulation of biomass, and that the variance in these responses was reduced over time. Moreover, diverse communities have a higher chance of including either disturbance-resistant species or species that are able to recover quickly from a variety of perturbations [ 126 , 176 ]. Ecosystems or communities with greater functional and response diversity are able to maintain important ecosystem processes that sustain function and result in ‘no net loss’ in productivity, often referred to as the insurance hypothesis [ 94 , 116 , 147 , 164 ]. In an experimental study Naeem and Li [ 147 ] tested the hypothesis that a greater number of species should enhance the probability that a system will provide a more “consistent level of performance” using microbes. They found that the greater number of species per functional group led to more consistent biomass and density measures within the replicated microbial microcosms. Genetic diversity can provide this benefit by increasing the critical response diversity among populations and can help maintain ecosystem function [ 112 , 126 ]. Additionally, increased genetic diversity has been shown to promote population growth and improve fitness [ 177 ].

There is ongoing debate over the association between biodiversity and its influence on resilience. Not all findings support the insurance hypothesis. For example, in a greenhouse experiment, Lanta et al. [ 57 ] found that high species richness and functional diversity provided less resistance against drought stressed conditions than less diverse species assemblages. The same study found no effect of diversity on community resistance under outdoor experimental conditions. Similarly, in a study examining species richness in aquatic food webs, Downing and Leibold [ 63 ] found that while respiration rates showed “higher resilience” in species-rich communities, they did not exhibit increased “resistance” to disturbance. In contrast, however, a number of studies have found strong causal linkages between diversity measurements and productivity or stability in a number of terrestrial and aquatic systems [ 154 ], including seagrass [ 130 , 136 ] and forests [ 38 , 178 ].

Habitat variability and condition

Spatial and temporal variability in habitats have been observed to maintain higher levels of biodiversity [ 94 ], and thus contribute to ecosystem resilience. A study conducted by Oliver and others [ 64 ] found landscape structure, including increased heterogeneity within habitat patches, to influence resilience of populations to extreme climatic events. A landscape with a more heterogeneous habitat structure was more likely to contain refuge microclimates to support survival of the ringlet butterfly, and greater heterogeneity among habitat patches increased the likelihood of harboring species more resilient to extreme events [ 64 ]. Within river systems, spatiotemporal variability in flow and temperature regimes was found to regulate suitable habitat and maintain flexible species adaptations [ 58 , 67 , 79 ]. Milner et al. [ 151 ] showed that maintaining habitat heterogeneity can maximize resilience of aquatic species to altered flow regimes associated with climate change. While habitat variability generally increases diversity at various scales, it also serves as a useful “measure of resilience to impending climate change” [ 165 ].

Refugia and support areas

Within the ecosystem category, presence of refugia or support areas was particularly important to ecosystem resilience. In freshwater and salt marsh ecosystems, presence and type of riparian vegetation was found to create micro-habitats that promoted community resistance to dry conditions [ 35 , 62 , 80 ]. Various soil health metrics were identified as crucial for aiding in recovery of forest ecosystems [ 107 , 132 ] and improving functional resilience in other terrestrial ecosystems [ 179 – 181 ]. Studies in coral reef systems identified water quality to be an important control on macroalgal growth, which can cause serious negative impacts to coral recruitment and overall reef resilience [ 59 , 78 , 135 ]. Refugia can also serve as areas where species are able to survive or rest from disturbance [ 19 , 29 , 55 , 67 , 68 , 79 , 82 , 141 ], and have been documented to provide propagules or seed sources for recovery in other affected areas [ 26 , 28 , 37 , 151 , 182 ]. These particular habitat attributes may not influence resilience in every ecosystem, but these findings suggest that identifying principal habitat characteristics may be an important consideration in monitoring resilience within an ecosystem.

Natural disturbance history and adaptability

A history of natural environmental fluctuations and disturbance is one process that maintains habitat heterogeneity, and the variability induced by disturbances favors biodiversity [ 94 ]. Specifically, disturbance can regulate habitat structure at multiple scales, with the potential to affect species richness many years into the future [ 52 , 83 , 86 , 107 ]. A substantial proportion of the literature identified presence of natural disturbance as an important determinant for recovery rates, creation of alternate trajectories, and building biological capacity to adapt to or resist change. Systems that are naturally subjected to a variety of disturbances contain biota that have evolved life history traits favoring adaptability or flexibility [ 61 , 114 , 182 ]. Li et al. [ 61 ] determined that bacterioplankton communities in a lake ecosystem had developed a number of life history attributes (e.g., high growth rates, phenotypic flexibility) that favored adaptation and explained their high resilience to natural pulses of Microcystis blooms. Within marine ecosystems, Neubauer et al. [ 45 ] confirmed that a history of moderate exploitation within fisheries populations can increase their rate of recovery.

Natural disturbance can influence biophysical characteristics of ecosystems and populations. For example, the size of a disturbed area can influence recovery rates because it effects how close it is to undisturbed neighboring areas that can provide material for re-colonization [ 162 ]. Some authors characterized entire ecosystems that are subject to high levels of natural disturbance as resilient. The hypothesis is that systems with high levels of disturbance have adapted with species and or processes that support quick recovery or resist complete change altogether [ 7 , 19 , 75 , 83 , 90 , 182 , 183 ]. In addition to disturbance, the magnitude and duration of an event proved to be an important attribute conferring resilience within many different systems. A number of studies found disturbance intensity to affect the degree of recovery [ 136 , 149 ] with more severe disturbance being a predictor of more rapid recovery [ 48 , 169 ]. Despite many systems demonstrating a considerable resilience to disturbance, prolonged disturbance is more likely to result in persistent habitat changes and reduce the ability of a system or populations to recover [ 175 ]. There is also considerable concern about future impacts on disturbance duration, magnitude, frequency, and timing from human induced climate change [ 149 , 166 ].

The effects of increased disturbance due to climate change do pose serious unknowns for resilience. For example, holm oak woodlands are historically highly resilient to fire frequencies of about 50 year intervals, but if the frequency of fire increases in response to climate change the system may not exhibit the same degree of resilience [ 149 ]. In a study examining resilience of fishes and invertebrates in streams exposed to prolonged drought, Bêche and others [ 175 ] found that both severity and duration of drought disturbance influenced the abundance, richness, and general recovery of aquatic communities.

Human pressures, cumulative effects

We found contradictory evidence regarding the effects of human pressures on resilience. A number of studies reported that isolation from human pressures or reduced exposure to anthropogenic stressors increased resilience within their systems [ 33 , 121 , 133 ]. Alternatively, in a study of coral assemblages distributed over a wide geographic range, Côté and Darling [ 54 ] found that if there is a positive co-tolerance between non-climatic disturbance and climatic impacts among coral species, then some degree of human-caused degradation may “increase the abundance of disturbance-tolerant species within a community and thus the ability of an ecosystem to resist impacts of climatic disturbance”. However, reduced abundance of less tolerant species (and increased proportions of disturbance-tolerant species) can also be considered an indicator of ecosystem degradation, at least in some contexts [ 184 ].

A number of resilience attributes we identified, including exposure to human pressures, were often discussed in context of cumulative impacts. This is an important consideration when measuring resilience in locations subject to multiple human stressors. The ability of ecosystems and their components to maintain resilience in the face of climate change when those systems are already under stress from cumulative human-generated impacts is a topic of evident concern in the literature [ 3 , 54 , 159 ]. Multiple co-occurring modes of disturbance can confound efforts to identify, measure, and monitor resilience within a system.

Restoration examples using the DST

Attributes classified by restoration focus and scale of measurement roughly sorted according to attribute category ( Table 2 ). For example, individual and population attributes (e.g., dispersal potential or genetic diversity) tended to be associated with species-specific restoration actions and with resilience at the population scale. Community attributes generally described the structure and diversity of ecosystems (e.g., community structure, functional diversity, or species diversity), and therefore were most often associated with site-specific or system-wide restoration. Roughly half of the ecosystem attributes (e.g., habitat area and condition, or refuge areas) were associated with all three restoration foci and at all three spatial scales. Process attributes were most diverse with respect to both focus and scale.

https://doi.org/10.1371/journal.pone.0173812.t002

We illustrate how resilience metrics might be used in conservation or management of species or ecosystems with three applied examples. In each example, we focus on how practitioners might select a sub-set of resilience attributes for characterizing or monitoring resilience of species or ecosystems using the DST. The examples we selected—restoration of the Kissimmee River system in Florida, recovery of an endangered salmon population, and coral species restoration—demonstrate how a sub-set of resilience attributes and metrics differ depending on biological and management contexts and the scale at which attributes are measured.

Restoration at the ecosystem scale–the Kissimmee River example.

The Kissimmee River once meandered for more than 100 miles through central Florida; connecting diverse habitats and supporting a thriving wetland ecosystem [ 185 ]. Restoration of the Kissimmee River System in Florida began two decades ago, and aims to reverse channelization and draining of wetlands to restore floodplain connectivity and restore ecosystem processes important to both the Kissimmee River and the Everglades ecosystem to which it drains. Based on the restoration focus (system) and scale (ecosystem) of the restoration effort, we derived 23 resilience attributes from the DST ( Table 2 ) that are appropriate as restoration or monitoring variables. These attributes represent the community, ecosystem, and process categories ( Table 3 ). Key resilience attributes within the community category are assemblage, diversity, redundancy, and connectivity. Not surprisingly, resilience attributes related to connectivity appear in all three major categories, as connectivity is a cornerstone of efforts to restore the Kissimmee River and Everglades ecosystem. In this case, each connectivity attribute increases resilience by allowing organisms and materials to move freely as suitable habitats shift in location. Within the ecosystem category, habitat area, condition, and variability are attributes that can support diversity or redundancy. Restoration efforts have largely focused on increasing natural habitat area and condition, including water quality and flow which are key metrics used to evaluate restoration success [ 185 ]. The remaining attributes in the process category tend to be features that also influence habitat condition and therefore support the community attributes. For example, energy flows is a broad and somewhat non-descript metric, however in this ecological context managers or restoration practitioners could consider (and already are) measuring how much water, sediment, and/or nutrients move between refuge areas.

https://doi.org/10.1371/journal.pone.0173812.t003

Restoration at the population scale–the pacific salmon Example.

Recovery of salmon populations listed under the Endangered Species Act focuses on achieving several important targets, including adequate population size, population growth rate, spatial distribution, and diversity [ 186 ]. Each of these targets are listed in the resilience attributes’ population category and they are monitored and population performance is assessed using these criteria every 5 years. That is, these resilience attributes were selected to characterize recovery of salmon populations in part because they indicate both recovery of number of fish and recovery of population attributes that buffer populations against environmental change. This comports well with our DST, which suggests that relevant resilience attributes include genetic diversity and connectivity as well as growth, size, abundance, and life history flexibility in individuals and populations ( Table 4 ). In addition, delisting criteria consider whether habitat factors contributing to listing have been abated. Consequently, various habitat-related resilience attributes are also appropriate for consideration in restoration planning or monitoring recovery. Habitat characteristics such as area, condition, and presence of refugia play an important role in the restoration of endangered populations and are often key components for the recovery of any species listed under the ESA. In addition to the metrics that align with current actions regarding salmon restoration, the DST provides several novel metrics that could be used to increase the resilience of endangered salmon populations to climate change, or to monitor changes in resilience among salmon populations.

https://doi.org/10.1371/journal.pone.0173812.t004

Restoration at the site scale–the coral reef example.

One common management strategy for coral reef protection and restoration is the creation of Marine Protected Areas (MPA’s) [ 187 ]. For example, the Coral Triangle Initiative, a multi-lateral effort to address threats to reefs, fisheries, and food security in the South Pacific, is working towards establishing regional connectivity between MPA’s [ 188 ]. Restoration activities that focus on specific sites or habitats are more common for sessile species where the focus is either on restoring habitat for a species or ‘seeding’ a species to initiate recovery at a site and many of the following attributes resulting from the DST depend on having available habitat. Individual, ecosystem, and process categories are represented in the list of suitable resilience attributes for this type of restoration ( Table 5 ). Individual attributes speak to a species’ ability to persist in an area. Ecosystem attributes are focused on habitat characteristics that may affect a species such as its condition, structure, or whether there are support areas present. Key process attributes that may affect habitat or species include structural legacies, disturbance, or degree of exposure to human pressures. Evidence suggests that conservation of sessile organisms such as coral reefs is most effective when an Ecosystem-based Management approach is taken. To address the many threats to coral reefs the creation of an MPA is coupled with land-based management to help reduce pollution sources [ 189 ].

https://doi.org/10.1371/journal.pone.0173812.t005

Resilient restoration

Explicit consideration of climate change in restoration design is an increasingly common request among federal and state governmental agencies [ 15 , 16 , 190 , 191 ], and many restoration projects are now required to evaluate the ability of a restored system or site to withstand impacts from climate change. Evidence suggests that when resilience is made an explicit planning objective, it offers a way to improve restoration projects as a whole [ 51 , 102 ].

The purpose of our analysis is to assist restoration practitioners in identifying appropriate resilience attributes to measure and monitor within particular systems. The focus of the management or restoration action (species, habitat, or system) is the first basis for categorizing the resilience attributes, because the overarching goal or motivation of restoration will dictate objective setting and monitoring design. The scale at which the attributes should be measured is the second basis for selecting attributes. Together these two criteria can help distill a subset of potential resilience attributes that are suitable for a specific restoration action or monitoring efforts. The attributes and their associated metrics should be part of an adaptive management framework to be evaluated for their usefulness in conferring resilience to climate change.

From our examination of recent ecological literature, we have extracted three key points that may be helpful in integrating resilience metrics into restoration plans. First, if made an explicit planning objective, as opposed to a component of existing objectives, resilience may be a way to improve restoration projects as a whole [ 51 , 102 ]. By planning and monitoring for resilience, we are forced to identify sources of adaptive capacity within restored and natural ecosystems and to define actions that foster resilience. Second, considering the restoration focus and scale of a plan or project is essential in choosing appropriate resilience metrics to inform restoration efforts. In the face of climate change, restoration approaches that promote natural sources of resilience are more likely to be successful than those that focus on creating optimal steady states. Third, certain ecological attributes, such as diversity and connectivity, are more commonly considered to confer resilience because they apply to a wide variety of species and ecosystems. Even so, we identified numerous additional metrics that could potentially be useful for resilience planning.

The need to understand the dynamic nature of ecological systems, especially in the context of climate change, is crucial for successful restoration work. Improving our understanding of how certain ecological attributes confer resilience will help practitioners develop best practices for successful restoration in a changing climate. Past trends in climate and streamflow, for example, make it clear that stationarity of the physical environment is no longer a valid assumption in restoration planning. Moreover, we should not assume continuous directional change in ecosystems, as climate cycles and other sources of natural variability drive annual or decadal variation in habitats and species. Hence, assumptions made about response and recovery trajectories can greatly influence restoration planning decisions. By monitoring the response and recovery of a variety of species and ecosystems, we can better understand which attributes most contribute to ecological resilience to climate change.

Supporting information

S1 fig. prisma 2009 flow diagram..

https://doi.org/10.1371/journal.pone.0173812.s001

S1 Table. Interactive decision support table (DST).

https://doi.org/10.1371/journal.pone.0173812.s002

S2 Table. PRISMA 2009 checklist.

https://doi.org/10.1371/journal.pone.0173812.s003

Acknowledgments

We thank Sarah Morley, George Pess and Lara Hansen for helpful reviews of the manuscript.

Author Contributions

• Conceptualization: BLTP TB TK.
• Data curation: BLTP.
• Formal analysis: BLTP.
• Investigation: BLTP.
• Methodology: BLTP TB TK.
• Project administration: BLTP TK.
• Supervision: TB TK.
• Validation: BLTP.
• Visualization: BLTP.
• Writing – original draft: BLTP.
• Writing – review & editing: BLTP TB TK.
• View Article
• PubMed/NCBI
• Google Scholar
• 11. Glick P, Stein B, Edelson N. Scanning the conservation horizon: a guide to climate change vulnerability assessment. Washington (DC): National Wildlife Federation. 2011.
• 13. Morrison WE, Nelson MW, Howard JF, Teeters EJ, Hare JA, Griffis RB, et al. Methodology for Assessing the Vulnerability of Marine Fish and Shellfish Species to a Changing Climate. In: Commerce USDo, editor.: NOAA Technical Memorandum; 2015. p. 48 p.
• 17. Hansen LJ, Biringer JL, Hoffman J. Buying time: a user’s manual for building resistance and resilience to climate change in natural systems. WWF, Washington. 2003.
• 18. Beatley T. Planning for coastal resilience: best practices for calamitous times: Island Press; 2012.
• 21. Walker B, Salt D. Resilience thinking: sustaining ecosystems and people in a changing world: Island Press; 2012.
• 43. Pernetta J. Mangrove forests, climate change and sea level rise: hydrological influences on community structure and survival, with examples from the Indo-West Pacific: IUCN; 1993.
• 81. Grimsditch GD, Salm RV. Coral reef resilience and resistance to bleaching. Gland, Switzerland: IUCN; 2006. Report No.: 2831709504.
• 102. Bisson PA, Dunham JB, Reeves GH. Freshwater ecosystems and resilience of Pacific salmon: habitat management based on natural variability. 2009.
• 111. Grime J. Plant strategies, vegetation processes, and ecosystem properties–John Wiley and Sons Ltd. Chichester, England. 2002.
• 128. Shuman JK, Shugart HH. Resilience and stability associated with conversion of boreal forest: INTECH Open Access Publisher; 2012.
• 185. South Florida Environmental Report Highlights South Florida Water Management District [Internet] http://my.sfwmd.gov/portal/page/portal/xweb%20about%20us/agency%20reports .
• 189. Conservation Methods: NOAA Coral Reef Conservation Program [Internet]; [2016 May 3]. http://coralreef.noaa.gov/conservation/methods/ .
• 190. Force IOPT. Final recommendations of the interagency ocean policy task force. Washington, DC: The White House Council on Environmental Quality. 2010.
• 191. WDFW Climate News Digest http://wdfw.wa.gov/conservation/climate_change/digest/climate_change_digest_2013_04.pdf .

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Int J Environ Res Public Health

Progress of Ecological Restoration Research Based on Bibliometric Analysis

1 Key Laboratory of Land Surface Pattern and Simulation, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China

2 School of Earth Sciences, Guilin University of Technology, Guilin 541000, China

3 Hebei Collaborative Innovation Center for Urban-Rural Integration Development, Shijiazhuang 050061, China

Xiangwen Cai

Associated data.

All relevant datasets in this study are described in the manuscript.

With the deterioration of the global/regional ecological environment, ecological restoration plays an important role in sustainable development. However, due to the differences in research methods, objectives, and perspectives, the research results are highly diverse. This makes it necessary to sort the publications related to ecological restoration, clarify the research status, grasp the research hotspots, and predict the future research trends. Here, 23,755 articles from the core database of Web of Science were retrieved, and bibliometric analysis was carried out to understand the global ecological restoration research progress from 1990 to 2022 from a macro perspective, with the aim to determine the future development direction. The results are as follows. (1) From 1990 to 2022, the number of publications in the field of ecological restoration constantly increased, and the fluctuation of the average annual citations increased. The most important articles were published in high-ranking journals. (2) Ecological restoration covers a wide range of research areas, including biodiversity, ecosystem services, climate change, land use, and ecological restoration theories and technologies. The four main hotspots in this field are heavy metal removal, soil microbial biomass carbon and nitrogen concentrations, grassland ecological restoration, and evaluation framework and modeling of ecological restoration’s effects. Currently, studies focus on river basin remediation, heavy metal removal, and forest restoration. (3) Future ecological restoration research should strengthen the multi-object aspect and multi-scale ecological restoration research, improve the ecological restoration effect evaluation system, and incorporate social and economic issues. This study identified current research hotspots and predicted potential future research directions, providing a scientific reference for future studies in the field of ecological restoration.

1. Introduction

Since the industrial revolution, with the rapid development of the global economy, people’s material living standards have been improving constantly. However, this goes along with issues such as global warming, environmental pollution, and loss of biodiversity [ 1 , 2 ]. Natural ecosystems, such as forests, grasslands, and wetlands, are in massive decline globally [ 3 ]. Ecosystem structure and function are being destroyed on a large scale, the quality of the ecological environment is deteriorating [ 4 ], natural resource scarcity is becoming increasingly critical [ 5 ], and the food supply and service capacity of ecosystems are declining sharply, seriously threatening the living environment of human beings and the sustainable development of society and the economy [ 6 ]. In this context, the restoration of degraded ecosystems has attracted great attention around the world.

Numerous countries and regions have taken a range of measures to address issues such as ecosystem degradation. In the 1950s–1960s, Europe and North America pioneered a series of ecological projects to address the ecological crisis [ 7 ]. In 2010, the Conference of the Parties to the Convention on Biological Diversity (CBD), in its Strategic Plan for Biodiversity 2011–2020 , specified the restoration of at least 15% of degraded ecosystems [ 8 ]. In a resolution in 2019, the United Nations General Assembly, led by the United Nations Environment Programme (UNEP) and the Food and Agriculture Organization of the United Nations (FAO), declared the United Nations Decade for Ecosystem Restoration (2021–2030) to address biodiversity loss, climate disruption, and increased pollution [ 9 ].

The term “ecological restoration” was proposed by Leopold in 1935 [ 10 ]. Ecological restoration focuses on the ecosystem as an object [ 11 ], aiming to repair damaged ecosystems and rebuild species composition and structure, with emphasis on the improvement and overall enhancement of ecosystems [ 12 ]. The goal is to establish a balanced ecosystem that meets the requirements of social and economic development through the reconstruction and restoration of the ecosystem, relying both on the inherent restoration ability of the ecosystem and on the intervention of science and technology [ 13 ]. Science and technology means mainly include physical, chemical, bioremediation, microbial, animal, and plant remediation methods. For example, Sun et al. [ 14 ] adopted the design method of horizontal subsurface-flow constructed wetlands to restore the ecological environment of the Yanxi River Basin, an important water resource conservation area in Beijing. Grant et al. [ 15 ] used smoldering technology to restore coal-tar-contaminated sites. Wu et al. [ 16 ] constructed a remediation system for cadmium-polluted soil using biochar, pro-growth bacteria, and super-accumulative plants. Ecological restoration can be classified into large-scale (e.g., on the Loess Plateau [ 17 ]), medium-scale (e.g., river channels [ 18 ], water bodies [ 19 ], forests [ 20 ], wetlands [ 21 ]), and small-scale ecological restoration (e.g., mines [ 22 ], landfills [ 23 ]). Based on the restoration object, scientists distinguish mountain [ 24 ], river [ 25 ], soil [ 26 ], and plant ecological restoration [ 27 ], among other types. There is also the view that ecological restoration needs to be enhanced through social channels [ 28 ]. As problems such as ecosystem degradation are caused by unsustainable human activities, as in the case of land degradation caused by slope planting, people who are highly dependent on these activities are urged to make some changes [ 29 ]. This will largely determine the success of ecological restoration and the healthy and sustainable development of the ecosystem if ecological restoration balances conflicting ecological and social interests. In addition, scientists are largely aware that ecological restoration also involves various challenges such as poverty, food security, and urbanization [ 30 , 31 ]. Ecological restoration must involve not only the restoration of destroyed ecosystems but also the protection and management of intact ecosystems [ 32 , 33 ]. The status quo and restoration effect of the ecosystem can be evaluated using GIS (Geographic Information System), RS (Remote Sensing), and other technologies [ 34 ].

Although in recent years ecological restoration has become a hot topic, research on ecological restoration lacks systematic induction and sorting, and the overall progress of research needs to be further explored. At the same time, given the large number of studies in this field, relevant methods, such as bibliometrics, are required to quickly understand the research progress. Bibliometrics can help to understand trends in the published literature in a given research field, the journals and disciplines involved, the authors, countries, and institutions, as well as the collaboration between them [ 35 ]. In addition, the main topics and future research directions of a specific field can be understood from the keywords and topics of the literature [ 36 ]. Bibliometrics research promotes a field in a novel and meaningful way to lay a solid foundation. It enables scientists to master a wealth of information about their research field, to derive new research ideas, and to locate their expected future contributions to the field [ 7 , 37 , 38 ].

In this study, to determine the development of research in the field of ecological restoration and gain a deeper understanding of the future trend, an R runtime environment was set up, and the Bibliometrix series were used to systematically summarize and sort publications in the field of ecological restoration published in the Web of Science database between 1990 and 2022. The research objectives are as follows:

• To provide an overview of the research and major research forces in the field of ecological restoration from 1990 to 2022 (countries/regions, institutions, publications, journals, among others).
• To analyze the popular research topics in this field and their characteristics.
• To explore potential research directions based on emerging trend analysis.

2. Materials and Methods

2.1. method.

Bibliometric analysis can help determine publication patterns and objectively evaluates the research status quo and development process for different countries, regions, scientific research institutions, or authors in specific fields using quantitative research methods such as mathematical statistics. Bibliometrics methods excel at exploring the underlying knowledge structures contained in the academic literature and at integrating visualized results to further analyze the field. Bibliometrics software can be used to quantitatively analyze large amounts of literature data and generate visualization and content analysis results. The visualization approach allows for a clearer relationship between the various research scopes and a scientifically effective understanding of the development direction and trends of scientific research [ 39 , 40 ]. Bibliometrix, developed by Massimo Aria and Corrado Cuccurullo in 2017, is a currently widely used R language software package for bibliometric analysis and scientific visualization [ 41 ]. The scientific knowledge map generated by the software package reflects the development trend of a certain discipline or knowledge field in a certain period and facilitates an accurate understanding of the evolution of a given scientific frontier. Content analysis effectively combines qualitative and quantitative analysis. It merges the relevant contents of the research object and statistical data to finally draw a qualitative conclusion [ 42 ]. Literature metrology, visualization, and content analysis methods are combined to analyze the literature related to the research field, which can objectively evaluate and explore the research status and development trend of a certain field.

Here, we used the Bibliometrix software package to quantify parameters such as the number of publications, countries, institutions, journals, and citations. We employed Excel, the R language, and other software packages to draw charts. Content analysis and visualization include the analysis of highly cited literature and of cooperative networks of research institutions, as well as clustering analysis of high-frequency keywords. All this can be performed using the Bibliometrix package.

2.2. Data and Processing

Web of Science is the world’s largest and most comprehensive academic information resource, covering more than 12,000 academic journals in a wide range of disciplines, including natural sciences, engineering, biomedical sciences, social sciences, arts, and humanities [ 43 ]. In this paper, the Web of Science core database was taken as the data retrieval source, the search scope was selected to be “subject”, and the search term TS = (“Ecological rehabilitation” OR “Ecological remediation” OR “Ecological restoration”) was applied. The time span of the search was 1990 to August 2022, and the literature type was limited to “Article”, “Review”, and “Database Review”. The retrieval language was set to English. Overall, 23,755 documents in the field of ecological restoration were obtained after data deduplication, irrelevant data removal, and other preprocessing steps. The retrieved data were saved as plain texts.

After data collection, the articles were analyzed using the Bibliometrix software package. First, we analyzed the evolution of the volume and citations of the documents in the field of ecological restoration for the period from 1990 to 2022. Second, we analyzed the main research disciplines involved in ecological restoration as well as the countries and institutions with a larger number of published articles. Third, we identified the major journals in the field of ecological restoration, along with their JCR (Journal Citation Reports) divisions, If (Impact Factor) and the volume of published literature in the field of ecological restoration between 1990 and 2022. Finally, we conducted a visual analysis of keywords and themes in the field to identify hot research topics and trends.

Figure 1 shows a summary of the bibliometrics process in the field of ecological restoration.

Bibliometrics process in the field of ecological restoration. (* as of 31 August 2022).

3. Results and Analysis

3.1. analysis of the publication volume in the field of ecological restoration.

Based on the statistics of the number of published articles in each year, the research process of each field can be understood to a certain extent. From 1990 to 2021, the number of publications in the field of ecological restoration increased gradually ( Figure 2 ), indicating that ecological restoration has attracted increased attention. From 1990 to 1996, the number of published articles was less than 100 per year, and the total number of published literatures was 356, accounting for only 1.50% of the total articles. The annual increase in the number of articles was slow, and this period can be termed the “initial stage”; publications mainly focused on qualitative analysis. Subsequently, from 1997 to 2017, the number of published articles increased considerably, with the annual number of published articles increasing from 131 to 1715, indicating that during this period, ecological restoration research was gaining more importance. As more scientists published in this field, issues such as ecological damage, ecological restoration theories, measures, technologies, policies, and other aspects became more prominent, with considerable interdisciplinarity. After 2017, the number of ecological restoration publications continued to increase substantially, reaching 2936 in 2021.

Volume of published articles on ecological restoration from 1990 to 2022.

3.2. Analysis of Literature Citations in the Field of Ecological Restoration

3.2.1. annual citation trend analysis.

The average annual citation frequency of the literature in the ecological restoration field showed a fluctuating increase ( Figure 3 ). The citation frequency was low from 1990 to 1996; even in 1996, the year with the highest citation frequency, it was as low as 1.64, indicating that this research field was still in its infancy and did not receive widespread attention. The average citation frequency during 1997–2002 and 2003–2022 was 2.97 and 3.58, respectively, with a peak in 2020 (4.47). To some extent, the citation frequency is closely related to the development stage of the research. In general, the increasing number of articles over time indicates the growing influence of the field of ecological restoration.

Average annual citations in the field of restoration ecology from 1990 to 2021.

3.2.2. Analysis of Highly Cited Articles

Ecological restoration has received considerable interest from relevant institutions and researchers, with most of the important findings published in high-ranking journals ( Table 1 ). Among the studies on ecological restoration published from 1990 to 2022, the most frequently cited globally was “Historical overfishing and the recent collapse of coastal ecosystems” by Professor Jeremy B. Jackson [ 44 ] of the University of California, San Diego, which was published in Science in 2001; this publication received 4346 citations worldwide. In their article, the authors suggest that more specific palaeoecological, archaeological, and historical data should be obtained to provide a framework for the restoration of coastal ecosystems by considering the extensive human disturbances to coastal ecosystems from a historical perspective and to construct achievable goals for the restoration and management of coastal ecosystems. It provides a novel perspective, namely a historical perspective, and laid the foundation for coastal ecosystem restoration. This has triggered a wave of research on driving factors such as pollution, water degradation, and climate change [ 45 , 46 ], as well as measures to restore coastal ecosystems [ 47 , 48 ]. The second most frequently cited article was “The value of estuarine and coastal ecosystem services” by Professor Edward B. Barbier [ 49 ] of the University of Wyoming, which was published in Ecological Monographs in 2011, with 2626 citations worldwide. This study assesses the value of estuarine and coastal ecosystems such as swamps, mangroves, nearshore coral reefs, seagrass beds, sandy beaches, and dunes and factors driving their decline. The authors propose the protection and enhancement of the immediate and long-term values of estuarine and coastal ecosystems by improving regulatory, institutional, and legal frameworks and developing measures to rehabilitate them. This article laid the foundation for the restoration of estuarine and coastal ecosystems and triggered an upsurge in research on the factors affecting estuarine and coastal ecosystems [ 50 ], mangrove restoration [ 51 ], and swamp restoration [ 52 ]. The third most frequently cited article was “Mycorrhizal fungal diversity determines plant biodiversity, ecosystem variability and productivity” by Professor Marcel G. A. van der Heijden [ 53 ] of the University of Basel, published in Nature in 1998, with 2301 citations worldwide. This study shows that microbial interactions can drive ecosystem functions such as plant biodiversity, productivity, and variability and postulates the protection of microorganisms to ensure the successful management and restoration of various ecosystems. Microbial restoration is suggested to restore various degraded ecosystems. Other scientists picked up this issue and performed comprehensive studies in the fields of soil salinization [ 54 ], wetland restoration [ 55 ], and grassland restoration [ 56 ], among others.

Top 10 most cited publications in the field of ecological restoration from 1990 to 2022.

Note: TC = total citation number.

3.3. Main Countries/Regions Conducting Research in the Field of Ecological Restoration

We identified the top ten countries conducting research in this field based on the literature volume ( Table 2 , Figure 4 ). Although many countries around the world performed studies in restoration ecology, the ten countries with most of the publications were the USA, China, Australia, the UK, Brazil, Canada, France, Germany, Spain, and Italy. Of these, the USA, China, Australia, the UK, and Brazil each published over 843 studies, with the USA and China publishing 5738 and 5662 studies, respectively. The publications of the top five countries respectively accounted for 19.10%, 12.66%, 8.58%, 6.60%, and 5.47% of the total publications and collectively for over 45% of the total publications. This indicates that these countries contributed the most to research on ecological restoration. For example, in the US, studies on the dynamic changes and mechanisms of ecosystems after destruction and disturbance were conducted, investigating northern broadleaf and mixed forests, among others. Australian scientists mainly investigated the degradation of arid land and the restoration of coniferous forests in the cold temperate zone [ 57 ].

Top 10 countries in terms of literature volume in the field of ecological restoration. MCP = Multiple-country publications; SCP = Single-country publications.

Top 10 countries in terms of the literature volume in the field of ecological restoration.

In addition, the numbers of single-country populations (SCP) and multiple-country populations (MCP) can be used to further analyze the scientific research strength of a given country. Throughout the study period, China had the highest MCP at 1176, whereas the USA had the highest SCP at 4940. This indicates that US scientists largely performed independent research, whilst their Chinese counterparts were more engaged in international collaboration. As ecological restoration research received increasing attention from scientists, regionalization and globalization became the focus of ecological restoration research.

3.4. Main Institutions Conducting Research in the Field of Ecological Restoration

To obtain an understanding of the research strength of individual institutions, we determined the top ten research institutions based on their publication volume ( Table 3 ), which were as follows: University of the Chinese Academy of Sciences, Chinese Academy of Sciences, Beijing Normal University, American Forest Service, Arizona State University, Northwestern University, University of Western Australia, University of Queensland, University of Sao Paulo, and Institute of Geographic Sciences and Resources Research. The University of the Chinese Academy of Sciences published 602 articles in this specific field, followed by the Chinese Academy of Sciences with 566 articles and Beijing Normal University with 561 articles. As the top three institutions were Chinese ones, China attaches great importance to ecological restoration research. In recent years, with the introduction of the Master Plan of National Major Projects for the Protection and Restoration of Important Ecosystems (2021–2035), China has committed itself to the systematic protection, overall restoration, and comprehensive management of mountains, rivers, forests, fields, lakes, and grasslands. Institutions in the US, Australia, and Canada have followed this example.

Top 10 research institutions in terms of literature volume in the field of ecological restoration.

We also mapped a collaboration network among 50 institutions ( Figure 5 ), using institutions as units of analysis to showcase published papers and collaborations. The radius size of the circles in the figure is proportional to the number of studies published collaboratively by the institutions, and the thickness of the line between two circles indicates the collaboration intensity. The thicker the line, the more intense the collaboration between institutions and vice versa. Clearly, institutions from China, the USA, and Australia ranked highly in terms of the number of articles published in collaboration with other institutions. Collaborations were mostly found between institutions of the same country, whereas large-scale multinational institutional collaboration groups have not yet been established. Regarding global ecological restoration research, there is a need to strengthen the cooperation between institutions and deepen research from multiple perspectives in the future.

Collaboration network of institutions publishing articles in the field of ecological restoration from 1990 to 2022.

3.5. Main Research Disciplines in the Field of Ecological Restoration

Bibliometric analysis was conducted on the disciplines covered by the ecological restoration research in the Web of Science to explore the major disciplines; the results are shown as percentages. Generally, the field of ecological restoration presented a trend of interdisciplinary development ( Figure 6 ). Among the many research disciplines, environmental science was the dominant one, accounting for 25% of the total, followed by ecology (19%), biodiversity conservation (6%), water resources (6%), forestry (5%), environmental studies (5%), and engineering environment (7%). Other disciplines, such as marine and freshwater biology, multidisciplinary geosciences, plant science, and soil science, did not account for more than 5% of the total. The key issues of ecological restoration research in environmental science involved environmental monitoring as well as water, air, and soil pollution control. The discipline of ecology is closely related to that of biodiversity conservation. Ecological restoration research in the discipline of ecology focuses on the adaptation and feedback effects of ecosystem processes to global changes, ecosystem service functions, and biodiversity conservation. In the discipline of biodiversity conservation, ecological restoration studies aim to protect endangered plants and animals and the various biological resources on Earth.

Average annual citations in the field of ecological restoration from 1990 to 2021.

3.6. Main Journals Publishing Articles in the Field of Ecological Restoration

This section shows the most prolific journals in the field of ecological restoration and analyzes their leading indicators ( Table 4 ). The top ten journals were published in the Netherlands, Switzerland, Germany, the UK, and the USA, covering a variety of fields such as environmental science and ecology, engineering technology, and agriculture and forestry sciences. The most productive journals were Ecological Engineering , Ecological Indicators , and Science of the Total Environment . Ecological Engineering , which was divided into the Q2 area by JCR partition in 2021, with an impact factor of 4.379, published 502 articles on ecological restoration from 1990 to 2022. Ecological Indicators , which was divided into the Q2 area by JCR partition in 2021, with an impact factor of 6.363, published 392 articles on ecological restoration from 1990 to 2022. Science of the Total Environment , which was divided into the Q2 area by JCR partition in 2021, with an impact factor of 10.753, published 311 articles on ecological restoration from 1990 to 2022.

Top 10 journals in terms of literature volume in the field of ecological restoration from 1990 to 2022.

The subject areas of these three journals are closely related to the field of ecological restoration research. Specific topics covered in Ecological Engineering include habitat reconstruction, ecotechnology, synthetic ecology, bioengineering, restoration ecology, ecology conservation, ecosystem rehabilitation, stream and river restoration, reclamation ecology, and non-renewable resource conservation. Specific topics covered by Ecological Indicators include broader assessment objectives and methods, e.g., biodiversity, biological integrity, and sustainability, through the use of indicators, resource-specific indicators such as landscape, agroecosystems, forests ecosystems, aquatic ecosystems, and wetlands. Specific topics covered by Science of the Total Environment include environmental remediation of soil and groundwater, nanomaterials, microplastics, and other emerging contaminants, novel contaminant (bio)monitoring and risk assessment approaches, stress ecology in marine, freshwater, and terrestrial ecosystems, trace metals and organics in biogeochemical cycles, water quality, and security.

3.7. Hot Research Topics and Trends in the Field of Ecological Restoration

3.7.1. high-frequency keyword analysis.

Keywords highly generalize the content of a research article [ 58 ]. High-frequency keyword analysis well reflected the current hot issues in the field of ecological restoration research. In bibliometrics, keywords are considered to represent the foundational elements of the knowledge concepts and are widely used to reveal the knowledge structure of the field of study.

The top 50 keywords were plotted in a keyword cloud graph ( Figure 7 ), with larger fonts indicating a higher frequency of occurrence. High-frequency keywords included “Management”, “Conservation”, “Biodiversity”, “Vegetation”, “Climate-change”, “Ecosystem services”, “Patterns”, and “Land use”. We noted that there was a wide range of studies in the field of ecological restoration, which is considered a solution to ecosystems issues and is closely related to climate change, land use, and management strategies, among others. In global ecological restoration research, biodiversity, ecosystem services, climate change, land use, ecological restoration patterns, and ecosystem management and protection, among other factors, were focused on. Among them, enhancing biodiversity and ecosystem services enables ecosystem restoration and regeneration [ 59 ]. Vulnerable ecosystems that have been damaged by climatic changes and human activities are closely linked to ecological restoration. Different restoration patterns are applicable to ecosystems of different types (such as soils, forests, or lakes) and the science of the restoration pattern determines the effectiveness of ecological restoration. Ecological patterns are established based on the theory of bioremediation, focusing on applying ecological principles to degraded ecosystems, combined with one or more of the corresponding ecological techniques for restoration [ 60 , 61 ]. In addition, the management and protection of the ecosystem by the government is crucial and will largely determine the success of ecological restoration programs.

Cloud graph of high-frequency keywords in the field of ecological restoration from 1990 to 2022.

Specifically, the field of ecological restoration research can be divided into numerous areas, such as forest, river, soil, and plant restoration, according to the restoration objectives. With respect to the restoration scale, large-scale ecological restoration (such as on the Loess Plateau region of China), meso-scale ecological restoration (such as that of rivers, forests, or wetlands), and small-scale (such as that of mines and landfill sites) ecological restoration projects can be distinguished. Among the specific issues, soil erosion, desertification, grassland degradation, forest resource shortages, water shortages, biodiversity reduction, and climate change can be considered [ 62 ]. At the same time, ecological restoration research needs to be combined with socio-economic aspects, such as ecological migration [ 63 ] or fishing bans [ 64 ]. In addition, the evaluation of ecological restoration effects is an important issue [ 65 ].

3.7.2. Cluster Analysis and Multiple Correspondence Analysis of High-Frequency Keywords

Cluster analysis is a common method in bibliometrics, and in statistics, it is a multivariate statistical analysis method for studying the “clustering of things” [ 66 , 67 ]. In this study, hierarchical clustering was applied to first take the keyword of each cluster as a category; subsequently, the keywords were merged into a higher-level cluster based on similarity, and finally, all individuals were grouped into categories. Here, we obtained four categories ( Figure 8 ).

Dendrogram of the system cluster analysis of keywords in the field of ecological restoration from 1990 to 2022.

The first category is mainly related to “Heavy metals”, “Pollution”, “Removal”, and “Restoration”. It focuses on exploring and removing heavy metals from contaminated soil, river ecosystems, and other ecosystems. The heavy metal industry is an important economic pillar for the development of a country’s regional economy, providing raw materials for industrial production. However, years of continuous mineral extraction have led to large areas of wasteland and severe damage to ecosystems. Heavy metals are characterized by high toxicity and low solubility. Wastewater from heavy metal extraction, smelting, and casting in industrial production is often discharged directly into the soil and surface water and has gradually become a key factor threatening ecosystem health [ 68 , 69 ]. The second category is mainly related to “Biomass”, “Soil”, “Nitrogen”, and “Carbon”. Studies in this category focus on soil microbial biomass carbon and nitrogen concentrations. Soil microorganisms play an important role in terrestrial ecosystem restoration as major decomposers. Carbon and nitrogen are often considered the most important elements in terrestrial ecosystems as their interactions play a key role in the global biogeochemical cycle and ecosystem functions [ 70 , 71 ]. The third category is mainly related to “Species richness”, “Diversity”, and “Grassland”, with a focus on grassland ecological restoration. Due to overgrazing, large-scale development, and construction, grassland resources have been seriously damaged. Grassland landscape fragmentation has been intensified, the ecological functions have declined, and the normal operation rules of the food chain, energy flow, and material circulation of grassland ecosystems have been seriously disturbed [ 72 ]. As generally, a higher species richness means a faster ecosystem restoration, there is an urgent need to increase grassland species richness to restore grassland resources [ 73 ]. The fourth category is mainly related to “Climate change”, “Water quality”, “Land use”, “Ecosystem services”, “Model”, and “Framework”, with an emphasis on frameworks and the evaluation of the effects of ecological restoration models. Ecological restoration effect evaluation mainly involves water quality, climate change, land use change, and ecosystem services [ 74 , 75 , 76 ].

3.7.3. Thematic Evolution Analysis

Research topic and anticipated research direction evolve over time. The research direction of ecological restoration has undergone tremendous changes in the past 32 years. In Figure 9 , the rectangles and squares from left to right depict the chronological order of the topic evolution. The topic development from 1990 to 2010 is shown on the left, that from 2011 to 2019 is shown in the middle, and that from 2020 to 2022 is shown on the right. The keywords are connected by grey lines of different shapes that connect rectangles of various colors. Based on the temporal distribution of the key topics, we summarized the development trend of ecological restoration research.

Evolution of research topics in the field of ecological restoration from 1990 to 2022.

Before 2011, ecological restoration research was in its infancy, and scientists started to explore subjects such as “Risk assessment” and “Plant species”. From 2011 to 2019, studies on ecological restoration were gradually enriched, with topics such as “Forest restoration”, “Risk assessment”, “Climate change”, “River basin”, and “Water quality”, presenting a trend of multidisciplinary and multi-perspective comprehensive analysis. In 2019, the United Nations announced the implementation of the Decade for Ecosystem Restoration (2021–2030) initiative to address biodiversity loss, climate damage, and increased pollution. From 2020 to 2022, ecological restoration research mainly focused on “Forest restoration”, “Plant species”, “River basin”, and “Heavy metals”. Forests represent the largest carbon pools, which makes them important in climate change mitigation. Generally, for every 1 m 3 stock volume growth of forests, an average of 1.83 t of carbon dioxide is absorbed, and 1.62 t of oxygen is released. Forest restoration will therefore contribute to the worldwide achievement of the ‘carbon neutrality’ target by 2060 [ 77 , 78 ]. Some governments and international organizations have launched forest restoration programs. For example, the German government and the International Society for the Conservation of Nature launched the “Bonn Challenge”, which aims to restore 150 million hectares of forest by 2020 [ 79 ]. The New York Declaration On Forests aims to restore 350 million hectares of degraded or logged land to forests by 2030 [ 80 ]. With the rapid development of the social economy, science and technology, industry and construction, and heavy metal pollution seriously threatens the safety of soil and river ecosystems. The number of studies on the technologies and methods for treating heavy-metal-polluted sites is constantly increasing, with theoretical and practical significance [ 81 ]. River basins are the basic units of hydrological response and the spatial scale for studying ecological restoration. The mechanisms of interaction between landscape patterns and hydrological processes at this scale are complex [ 82 ]. River basin restoration includes, but is not limited to, dam removal/renovation, fishway construction, flow modification, floodplain reconstruction, stormwater management, natural shoreline protection, habitat protection and restoration, riverbank stabilization, vegetation restoration, river channel reconstruction, water quality management, and ecological regulation [ 83 , 84 ].

4. Discussion

Ecosystems not only provide a variety of raw materials or products that are directly used by humans but also have various functions such as carbon storage, climate regulation, pollution purification, water conservation, soil and water conservation, wind prevention and sand fixation, and biodiversity maintenance [ 85 ]. In this sense, ecosystems guarantee the sustainable development of human society and the economy. With the rapid development of society and the economy, ecosystems are increasingly disturbed and destroyed by human activities [ 86 ]. In response to ecosystem degradation, biodiversity loss, and climate change caused by human activities, the concept and practice of ecological conservation and restoration are rapidly developing across the globe. Many countries/regions, such as China, India, and New Zealand, have set short-term and long-term development goals for ecological restoration [ 87 , 88 , 89 ]. Such detailed goals and tasks oscillate around one or several aspects of ecological restoration to address the many ecological issues than need to be solved. At the same time, academic research on ecological restoration has increased significantly over the past three decades, with significant positive trends.

The results show that the number of articles in the field of ecological restoration has been increasing continuously since 1990. Before 1996, few articles were published, and the research progress was slow. However, after 1996, ecological restoration received increased attention, and the development of this research field was accelerated. Various theories have been proposed, such as the community succession, the ecosystem stability, the community construction, and the ecological niche theory. Of these, the community succession theory has been widely used in the restoration of degraded forest, grassland, and wetland vegetation [ 90 ]. In addition, studies also focused on restoration techniques and measures for different objects. For example, grassland restoration techniques include no-till reseeding, livestock reduction, grassland cultivation, grassland tillage, conversion of farmland to grassland, and fertilization [ 91 ]. River restoration technologies mainly include physical restoration, such as aeration and aeration; sediment dredging; chemical remediation, such as the introduction of chemicals that react with pollutants; and bioremediation, namely the use of aquatic animals, plants, and microorganisms to absorb, degrade, and transform pollutants [ 92 ].

However, the research perspectives are not limited to the direct study of ecological restoration, and climate change, land use, policy making, and other related aspects are also considered. For example, man-made grassland degradation caused by overgrazing, climate change, and other driving factors is one of the most serious global disturbances, affecting about 49% of the global grassland area, with serious impacts on livelihoods, biodiversity, and ecosystem functions [ 93 ]. The rate of forest extinction has been accelerated by the rapid decline of forest resources due to the expansion of arable land and excessive commercial logging for fuel and infrastructure materials [ 94 ]. The study of ecological restoration policies is also crucial to the management and protection of ecosystems and largely determines the success of ecological restoration. The co-occurrence of keywords and the evolution of this theme also show that the four hotspots in this field are heavy metal pollution remediation, soil microbial biomass carbon and nitrogen concentrations, grassland ecological restoration, and ecological restoration effect evaluation frameworks and models. The main research directions include watershed restoration, heavy metal removal, and forest restoration.

In the future, cooperation between countries/regions should be strengthened to carry out ecological restoration research from multiple perspectives. European and American countries are leading the way in ecological restoration, not only because they have taken the lead in relevant research but also because of the extensive and close collaboration between countries. At present, there is still a large gap between the ecological restoration statuses of developing countries and those of developed countries. Therefore, it is of great significance to strengthen the cooperation with other countries to promote the healthy development of the ecological system [ 95 ]. Most ecological restoration studies only focus on the protection and restoration of specific ecosystems and targets such as river basins, heavy metal pollution, and forests. In the future, the research scope of ecological restoration should also be transformed from local and single-ecosystem ecological restoration to large-scale and global ecological restoration involving multiple ecosystems, with the aim to initiate multi-object systematic and multi-scale ecological restoration research.

From the study of the protection and restoration of specific ecosystems and targets, research has gradually shifted to the mechanism and path of the successful realization of ecological restoration. At the same time, ecological restoration is not only a natural and ecological process but also a social and economic process, involving economic loss, economic input, value restoration, and resource development and management, with a series of human value orientation motivations and goals. The integration of the conceptual framework and research methods of natural science and humanities in ecological restoration projects is gaining increasing attention [ 96 ]. Therefore, ecological restoration is not only a scientific but also a social and economic issue. In the future, an important direction of ecological restoration research is to cross-study ecological restoration with social disciplines and economic disciplines, such as supporting mechanisms of ecological restoration, including policy, legal, and management mechanisms [ 97 , 98 ]. In addition, the evaluation of ecological restoration effects is crucial and should involve water quality, climate change, land use change, and ecosystem services in the fields of ecology, environmental science, engineering, and geography, among others.

Multidisciplinary research in the field of ecological restoration is still in its infancy [ 99 ]. Integrating multidisciplinary theories to formulate ecological restoration policies and research frameworks for adaptation and deepening the system of multidisciplinary theories and methods will help optimize the evaluation system of ecological restoration effects and make ecological restoration more scientific and comprehensive [ 100 ]. In the future, it might be important to establish a scientific, reasonable, authoritative, and effective evaluation system according to local conditions.

The results obtained so far are encouraging. However, studies on ecological restoration are scattered and ignore the systematic characteristics of the research objects, which makes it difficult to form a unified logical main line and connect different research perspectives. In this paper, we look at the development of ecological restoration research from a macro perspective, determining the major research journals, institutions, countries, and disciplines and analyzing the findings of the major articles. With this approach, we clarify the cognitive context of the academic circle, help scientists to understand the hotspots and directions of ecological restoration, and provide references for in-depth research on ecological restoration. Although this paper has a certain contribution to the research field, it also has some shortcomings. For example, only the keywords “Ecological rehabilitation”, “Ecological remediation”, and “Ecological restoration” were selected. Other keywords, such as “Land reclamation” and “Comprehensive land management”, are also related to the field of ecological restoration, but they were not considered. The retrieval period was 1990 to 2022, although certainly, some important results were already published before 1990. Our study is based on a single database, the Web of Science core database, and ignores articles from Scopus, CNKI, and other databases. Therefore, in the following research, we will explore how to merge other databases and expand our research scope based on the current discussion. Finally, ecological restoration research involves many aspects and issues. In the future, the development trend of important branch issues will be further analyzed.

5. Conclusions

In this paper, we looked at the development of ecological restoration research from a macro perspective, identifying the trends in publications, major research journals, institutions, countries, and articles. The evolution of the research hotspots and themes was analyzed and the future research directions were discussed. From 1990 to 2022, the number of articles published in the field of ecological restoration research increased gradually, and the fluctuation of the annual average citation number increased. The most important articles were published in high-ranking journals. The US was the leading research country in this field, followed by China and Australia. American scientists seemed to prefer independent research, whereas Chinese scholars engaged more in international collaboration. The top three institutions in this field were located in China, with UNIV CHINESE ACAD SCI being the most prolific one. The field shows the interdisciplinary development trend of environmental science, ecology, and biodiversity conservation. The top three prolific journals were Ecological Engineering, Ecological Indicators , and Science of the Total Environment.

The results also indicated that the research field of ecological restoration covers a wide range, with biodiversity, ecosystem services, climate change, land use, and ecological restoration theories and technologies being the main topics. The four hotspots were heavy metal pollution, soil microbial biomass carbon and nitrogen concentrations, grassland ecological restoration, and evaluation frameworks and modeling of ecological restoration’s effects. At present, river basin restoration, heavy metal removal, and forest restoration are the main research directions in the field of ecological restoration. In the future, multi-object systematic research and multi-scale ecological restoration research should be strengthened, along with improvement of the ecological restoration effect evaluation system, and ecological restoration research must incorporate social and economic issues. We try to clarify the cognitive context of the academic circle, help scientists to understand the hotspots and directions of ecological restoration, and provide references for in-depth research on ecological restoration.

Funding Statement

This research was funded by the Second Tibetan Plateau Scientific Expedition and Research (grant number 2019QZKK0603), and the Project of National Natural Science Foundation of China (grant number 42071233).

Author Contributions

X.W.: Methodology, software, data curation, writing—original draft; W.S.: Conceptualization, funding acquisition; X.W. and W.S. Writing—review & editing; Y.S. and X.C.: Investigation, validation; X.W., W.S., Y.S. and X.C.: Proofreading. All authors have read and agreed to the published version of the manuscript.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Data availability statement, conflicts of interest.

The authors declare no conflict of interest.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 25 March 2024

The evolutionary drivers and correlates of viral host jumps

• Cedric C. S. Tan   ORCID: orcid.org/0000-0003-3536-8465 1 , 2 ,
• Lucy van Dorp   ORCID: orcid.org/0000-0002-6211-2310 1   na1 &
• Francois Balloux   ORCID: orcid.org/0000-0003-1978-7715 1   na1  

Nature Ecology & Evolution ( 2024 ) Cite this article

11k Accesses

2423 Altmetric

Metrics details

• Molecular evolution
• Viral evolution

Most emerging and re-emerging infectious diseases stem from viruses that naturally circulate in non-human vertebrates. When these viruses cross over into humans, they can cause disease outbreaks, epidemics and pandemics. While zoonotic host jumps have been extensively studied from an ecological perspective, little attention has gone into characterizing the evolutionary drivers and correlates underlying these events. To address this gap, we harnessed the entirety of publicly available viral genomic data, employing a comprehensive suite of network and phylogenetic analyses to investigate the evolutionary mechanisms underpinning recent viral host jumps. Surprisingly, we find that humans are as much a source as a sink for viral spillover events, insofar as we infer more viral host jumps from humans to other animals than from animals to humans. Moreover, we demonstrate heightened evolution in viral lineages that involve putative host jumps. We further observe that the extent of adaptation associated with a host jump is lower for viruses with broader host ranges. Finally, we show that the genomic targets of natural selection associated with host jumps vary across different viral families, with either structural or auxiliary genes being the prime targets of selection. Collectively, our results illuminate some of the evolutionary drivers underlying viral host jumps that may contribute to mitigating viral threats across species boundaries.

Similar content being viewed by others

Diversity and evolution of the animal virome

Erin Harvey & Edward C. Holmes

Epidemiological hypothesis testing using a phylogeographic and phylodynamic framework

Simon Dellicour, Sebastian Lequime, … Philippe Lemey

Predicting the global mammalian viral sharing network using phylogeography

Gregory F. Albery, Evan A. Eskew, … Kevin J. Olival

The majority of emerging and re-emerging infectious diseases in humans are caused by viruses that have jumped from wild and domestic animal populations into humans (that is, zoonoses) 1 . Zoonotic viruses have caused countless disease outbreaks ranging from isolated cases to pandemics and have taken a major toll on human health throughout history. There is a pressing need to develop better approaches to pre-empt the emergence of viral infectious diseases and mitigate their effects. As such, there is an immense interest in understanding the correlates and mechanisms of zoonotic host jumps 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 .

Most studies thus far have primarily investigated the ecological and phenotypic risk factors contributing to viral host range through the use of host–virus association databases constructed mainly on the basis of systematic literature reviews and online compendiums, including VIRION 11 and CLOVER 12 . For example, ‘generalist’ viruses that can infect a broader range of hosts have typically been shown to be associated with greater zoonotic potential 2 , 3 , 5 . In addition, factors such as increasing human population density 1 , alterations in human-related land use 4 , ability to replicate in the cytoplasm or being vector-borne 3 are positively associated with zoonotic risk. However, despite global efforts to understand how viral infectious diseases emerge as a result of host jumps, our current understanding remains insufficient to effectively predict, prevent and manage imminent and future infectious disease threats. This may partly stem from the lack of integration of genomics into these ecological and phenotypic analyses.

One challenge for predicting viral disease emergence is that only a small fraction of the viral diversity circulating in wild and domestic vertebrates has been characterized so far. Due to resource and logistical constraints, surveillance studies of novel pathogens in animals often have sparse geographical and/or temporal coverage 13 , 14 and focus on selected host and pathogen taxa. Further, many of these studies do not perform downstream characterization of the novel viruses recovered and may lack sensitivity due to the use of PCR pre-screening to prioritize samples for sequencing 15 . As such, our knowledge of which viruses can, or are likely to emerge and in which settings, is poor. In addition, while genomic analyses are important for investigating the drivers of viral host jumps 16 , most studies do not incorporate genomic data into their analyses. Those that did have mostly focused on measures of host 2 or viral 3 diversity as predictors of zoonotic risk. As such, despite the limited characterization of global viral diversity thus far, existing genomic databases remain a rich, largely untapped resource to better understand the evolutionary processes surrounding viral host jumps.

Further, humans are just one node in a large and complex network of hosts in which viruses are endlessly exchanged, with viral zoonoses representing probably only rare outcomes of this wider ecological network. While research efforts have rightfully focused on zoonoses, viral host jumps between non-human animals remain relatively understudied. Another important process that has received less attention is human-to-animal (that is, anthroponotic) spillover, which may impede biodiversity conservation efforts and could also negatively impact food security. For example, human-sourced metapneumovirus has caused fatal respiratory outbreaks in captive chimpanzees 17 . Anthroponotic events may also lead to the establishment of wild animal reservoirs that may reseed infections in the human population, potentially following the acquisition of animal-specific adaptations that could increase the transmissibility or pathogenicity of a virus in humans 13 . Uncovering the broader evolutionary processes surrounding host jumps across vertebrate species may therefore enhance our ability to pre-empt and mitigate the effects of infectious diseases on both human and animal health.

A major challenge for understanding macroevolutionary processes through large-scale genomic analyses is the traditional reliance on physical and biological properties of viruses to define viral taxa, which is largely a vestige of the pre-genomic era 18 . As a result, taxon names may not always accurately reflect the evolutionary relatedness of viruses, precluding robust comparative analyses involving diverse viral taxa. Notably, the International Committee on Taxonomy of Viruses (ICTV) has been strongly advocating for taxon names to also reflect the evolutionary history of viruses 18 , 19 . However, the increasing use of metagenomic sequencing technologies has resulted in a large influx of newly discovered viruses that have not yet been incorporated into the ICTV taxonomy. Furthermore, it remains challenging to formally assess genetic relatedness through multiple sequence alignments of thousands of sequences comprising diverse viral taxa, particularly for those that experience a high frequency of recombination or reassortment.

In this study, we leverage the ~12 million viral sequences and associated host metadata hosted on NCBI to assess the current state of global viral genomic surveillance. We additionally analyse ~59,000 viral sequences isolated from various vertebrate hosts using a bespoke approach that is agnostic to viral taxonomy to understand the evolutionary processes surrounding host jumps. We ascertain overall trends in the directionality of viral host jumps between human and non-human vertebrates and quantify the amount of detectable adaptation associated with putative host jumps. Finally, we examine, for a subset of viruses, signatures of adaptive evolution detected in specific categories of viral proteins associated with facilitating or sustaining host jumps. Together, we provide a comprehensive assessment of potential genomic correlates underpinning host jumps in viruses across humans and other non-human vertebrates.

An incomplete picture of global vertebrate viral diversity

Global genomic surveillance of viruses from different hosts is key to preparing for emerging and re-emerging infectious diseases in humans and animals 13 , 16 . To identify the scope of viral genomic data collected thus far, we downloaded the metadata of all viral sequences hosted on NCBI Virus ( n  = 11,645,803; accessed 22 July 2023; Supplementary Data 1 ). Most (68%) of these sequences were associated with SARS-CoV-2, reflecting the intense sequencing efforts during the COVID-19 pandemic. In addition, of these sequences, 93.6%, 3.3%, 1.5%, 1.1% and 0.6% were of viruses with single-stranded (ss)RNA, double-stranded (ds)DNA, dsRNA, ssDNA and unspecified genome compositions, respectively. The dominance of ssRNA viruses is not entirely explained by the high number of SARS-CoV-2 genomes, as ssRNA viruses still represent 80% of all viral genomes if SARS-CoV-2 is discounted.

Vertebrate-associated viral sequences represent 93% of this dataset, of which 93% were human associated. The next four most-sequenced viruses are associated with domestic animals ( Sus , Gallus , Bos and Anas ) and, after excluding SARS-CoV-2, represent 15% of vertebrate viral sequences, while viruses isolated from the remaining vertebrate genera occupy a mere 9% (Fig. 1a and Extended Data Fig. 1a ), highlighting the human-centric nature of viral genomic surveillance. Further, only a limited number of non-human vertebrate families have at least ten associated viral genome sequences deposited (Fig. 1b ), reinforcing the fact that a substantial proportion of viral diversity in vertebrates remains uncharacterized. Viral sequences obtained from non-human vertebrates thus far also display a strong geographic bias, with most samples collected from the United States of America and China, whereas countries in Africa, Central Asia, South America and Eastern Europe are highly underrepresented (Fig. 1c ). This geographical bias varies among the four most-sequenced non-human host genera Sus , Gallus , Anas and Bos (Extended Data Fig. 1b ). Finally, the user-submitted host metadata associated with viral sequences, which is key to understanding global trends in the evolution and spread of viruses in wildlife, remains poor, with 45% and 37% of non-human viral sequences having no associated host information provided at the genus level, or sample collection year, respectively. The proportion of missing metadata also varies extensively between viral families and between countries (Extended Data Fig. 2 ). Overall, these results highlight the massive gaps in the genomic surveillance of viruses in wildlife globally and the need for more conscientious reporting of sample metadata.

a , Proportion of non-SARS-CoV-2, vertebrate-associated viral sequences deposited in public sequence databases ( n  = 2,874,732), stratified by host. Viral sequences associated with humans and the next four most-sampled vertebrate hosts are shown. Sequences with no host metadata resolved at the genus level are denoted as ‘missing’. b , Proportion of host families represented by at least 10 associated viral sequences for the five major vertebrate host groups. c , Global heat map of sequencing effort, generated from all viral sequences deposited in public sequence databases that are not associated with human hosts ( n  = 1,599,672). d , Number of vertebrate viral species on NCBI Virus used for the genomic analyses in this study, stratified by viral family. The 32 vertebrate-associated viral families considered in this study are shown and the remaining 21 families that were not considered are denoted as ‘others’.

Humans give more viruses to animals than they do to us

To investigate the relative frequency of anthroponotic and zoonotic host jumps, we retrieved 58,657 quality-controlled viral genomes spanning 32 viral families, associated with 62 vertebrate host orders and representing 24% of all vertebrate viral species on NCBI Virus ( https://www.ncbi.nlm.nih.gov/labs/virus/vssi/#/ ) (Fig. 1d ). We found that the user-submitted species identifiers of these viral genomes are poorly ascribed, with only 37% of species names consistent with those in the ICTV viral taxonomy 20 . In addition, the genetic diversity represented by different viral species is highly variable since they are conventionally defined on the basis of the genetic, phenotypic and ecological attributes of viruses 18 . Thus, we implemented a species-agnostic approach based on network theory to define ‘viral cliques’ that represent discrete taxonomic units with similar degrees of genetic diversity, similar to the concept of operational taxonomic units 21 (Fig. 2a and Methods ). A similar approach was previously shown to effectively partition the genomic diversity of plasmids in a biologically relevant manner 22 . Using this approach, we identified 5,128 viral cliques across the 32 viral families that were highly concordant with ICTV-defined species (median adjusted Rand index, ARI = 83%; adjusted mutual information, AMI = 75%) and of which 95% were monophyletic (Fig. 2a ). Some clique assignments aggregated multiple viral species identifiers, while others disaggregated species into multiple cliques (Fig. 2b ; clique assignments for Coronaviridae illustrated in Extended Data Fig. 3 ). Despite the human-centric nature of genomic surveillance, viral cliques involving only animals represent 62% of all cliques, highlighting the extensive diversity of animal viruses in the global viral-sharing network (Extended Data Fig. 4a ).

a , Workflow for taxonomy-agnostic clique assignments. Briefly, the alignment-free Mash 53 distances between complete viral genomes in each viral family are computed and dense networks where nodes and edges representing viral genomes and the pairwise Mash distances, respectively, are constructed. From these networks, edges representing Mash distances >0.15 are removed to produce sparse networks, on which the community-detection algorithm, Infomap 54 , is applied to identify viral cliques. Concordance with the ICTV taxonomy was assessed using ARI and AMI. b , Sparse networks of representative viral cliques identified within the Coronaviridae (ssRNA), Picobirnaviridae (dsRNA), Genomoviridae (ssDNA) and Adenoviridae (dsDNA). Some viral clique assignments aggregated multiple viral species, while others disaggregated species into multiple cliques. Nodes, node shapes and edges represent individual genomes, their associated host and their pairwise Mash distances, respectively. The list of viral families considered in our analysis are shown on the bottom-left corner of each panel. Silhouettes were sourced from Flaticon.com and Adobe Stock Images ( https://stock.adobe.com ) with a standard licence.

We then identified putative host jumps within these viral cliques by producing curated whole-genome alignments to which we applied maximum-likelihood phylogenetic reconstruction. For segmented viruses, we instead used single-gene alignments as the high frequency of reassortment 23 precludes robust phylogenetic reconstruction using whole genomes. Phylogenetic trees were rooted with suitable outgroups identified using metrics of alignment-free distances (see Methods ). We subsequently reconstructed the host states of all ancestral nodes in each tree, allowing us to determine the most probable direction of a host jump for each viral sequence (approach illustrated in Fig. 3a ). To minimize the uncertainty in the ancestral reconstructions, we considered only host jumps where the likelihood of the ancestral host state was twofold higher than alternative host states (Fig. 3a and Supplementary Methods ). Varying the stringency of this likelihood threshold yielded highly consistent results (Extended Data Fig. 5a ), indicating that the inferred host jumps are robust to our choice of threshold. In total, we identified 12,676 viral lineages comprising 2,904 putative vertebrate host jumps across 174 of these viral cliques.

a , Illustration of ancestral host reconstruction approach used to infer the directionality of putative host jumps. Putative host jumps are identified if the ancestral host state has a twofold higher likelihood than alternative host states. The mutational distance (substitutions per site) represents the sum of the branch lengths between the tip sequence and the ancestral node for which the first host state transition occurred in a tip-to-root traverse. b , Number of distinct putative host jumps involving humans across all viral families considered ( n  = 32). Black dots represent the observed point estimates for each type of host jump. The violin plots show the bootstrap distributions of these estimates, where the host jumps within each viral clique were resampled with replacement for 1,000 iterations. Black lines show the 95% confidence intervals associated with these bootstrap distributions. Silhouettes were sourced from Flaticon.com and Adobe Stock Images ( https://stock.adobe.com ) with a standard licence. A two-tailed paired t -test was performed to test for a difference in the zoonotic and anthroponotic bootstrap distributions.

Among the putative host jumps inferred to involve human hosts (599/2,904; 21%), we found a much higher frequency of anthroponotic compared with zoonotic host jumps (64% vs 36%, respectively; Fig. 3b ). This finding was statistically significant as assessed via a bootstrap paired t -test ( t  = 227, d.f. = 999, P  < 0.0001) and a permutation test ( P  = 0.035; see Methods ). In addition, this result was robust to our choice of likelihood thresholds used during ancestral reconstruction (Extended Data Fig. 5b ), the tree depth at which the host jump was identified (Extended Data Fig. 5c ), and to sampling bias ( Supplementary Notes and Fig. 1 ). The highest number of anthroponotic jumps was contributed by the cliques representing SARS-CoV-2 (132/383; 34%), MERS-CoV (39/383; 10%) and influenza A (37/383; 10%). This is concordant with the repeated independent anthroponotic spillovers into farmed, captive and wild animals described for SARS-CoV-2 (refs. 13 , 24 , 25 , 26 , 27 ) and influenza A 28 , 29 . Meanwhile, there has only been circumstantial evidence for human-to-camel transmission of MERS-CoV 30 , 31 , 32 . Noting the disproportionate number of anthroponotic jumps contributed by these viral cliques, we reperformed the analysis without them and found a significantly higher frequency of anthroponotic than zoonotic jumps (53.5% vs 46.5%; bootstrap paired t -test, t  = 40, d.f. = 999, P  < 0.0001), suggesting that our results are not driven solely by these cliques. Further, 16/21 of the viral families were involved in more anthroponotic than zoonotic jumps (Extended Data Fig. 5d ), indicating that this finding is generalizable across most viruses. Overall, our results highlight the high but largely underappreciated frequency of anthroponotic jumps among vertebrate viruses.

Host jumps of multihost viruses require fewer adaptations

Before jumping to a new host, a virus in its natural reservoir may fortuitously acquire pre-adaptive mutations that facilitate its transition to a new host. This may be followed by the further acquisition of adaptive mutations as the virus adapts to its new host environment 16 .

For each host jump inferred, we estimated the extent of both pre-jump and post-jump adaptations through the sum of branch lengths from the observed tip to the ancestral node where the host transition occurred (Fig. 3a ). However, in practice, the degree of adaptation inferred may vary on the basis of different factors, including sampling intensity and the time interval between when the host jump occurred and when the virus was isolated from its new host. As such, for each viral clique, we considered only the minimum mutational distance associated with a host jump.

We first examined whether the minimum mutational distance associated with a host jump for each viral clique was higher than the minimum for a random selection of viral lineages not involved in host jumps (Fig. 3a and Methods ). Indeed, the minimum mutational distance for a putative host jump within each clique was significantly higher than that for non-host jumps (Fig. 4a ; two-tailed Mann–Whitney U -test, U  = 6,767, P  < 0.0001). Noting that both sampling intensity and the different mutation rates of viral families may confound these results, we corrected for these confounders using a logistic regression model but found a similar effect (odds ratio, OR host jump  = 1.31; two-tailed Z -test for slope = 0, Z  = 6.58, d.f. = 289, P  < 0.0001).

a , b , Distributions (Gaussian kernel densities and boxplots) of ( a ) minimum mutational distance and ( b ) minimum dN/dS for inferred host jump events and non-host jump controls on the logarithmic scale. Differences in distributions were assessed using two-sided Mann–Whitney U -tests. c , d , Scatterplots of the ( c ) minimum mutational distance and ( d ) minimum dN/dS for host jump and non-host jumps. Lines represent univariate linear regression smooths fitted on the data. We corrected for the effects of sequencing effort and viral family membership using Poisson regression models. The parameter estimates in these Poisson models and their statistical significance, as assessed using two-tailed Z -tests, after performing these corrections are annotated. For all panels, each data point represents the minimum distance or minimum dN/dS across all host jump or randomly selected non-host jump lineages in a single clique. Boxplot elements are defined as follows: centre line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range.

We then considered the commonly used measure of directional selection acting on genomes, the ratio of non-synonymous mutations per non-synonymous site (dN) to the number of synonymous mutations per synonymous site (dS). Comparing the minimum dN/dS for host jumps within each clique, we observed that minimum dN/dS was also significantly higher for host jumps compared with non-host jumps (Fig. 4b ; OR host jump  = 2.39; Z  = 4.84, d.f. = 263, P  < 0.0001). Finally, after correcting for viral clique membership, there were no significant differences in log-transformed mutational distance ( F (1,528)  = 2.23, P  = 0.136) or dN/dS estimates ( F (1,338)  = 1.66, P  = 0.198) between zoonotic and anthroponotic jumps, or between forward and reverse cross-species jumps (mutational distance: F (1,1588)  = 0.538, P  = 0.463; dN/dS: F (1,1168)  = 0.0311, P  = 0.860), indicating that there are no direction-specific biases in these measures of adaptation. Overall, these results are consistent with the hypothesized heightened selection following a change in host environment and additionally provide confidence in our ancestral-state reconstruction method for assigning host jump status.

However, the extent of adaptive change required for a viral host jump may vary. For instance, some zoonotic viruses may require minimal adaptation to infect new hosts while in other cases, more substantial genetic changes might be necessary for the virus to overcome barriers that prevent efficient infection or transmission in the new host. We therefore tested the hypothesis that the strength of selection associated with a host jump decreases for viruses that tend to have broader host ranges. To do so, we compared the minimum mutational distance between ancestral and observed host states to the number of host genera sampled for each viral clique. We found that the observed host range for each viral clique is positively associated with greater sequencing intensity (that is, the number of viral genomes in each clique; Pearson’s r  = 0.486; two-tailed t -test for r  = 0, t  = 34.9, d.f. = 3,932, P  < 0.0001), in line with the strong positive correlation between per-host viral diversity and surveillance effort reported in previous studies 2 , 3 , 8 . After correcting for both sequencing effort and viral family membership, we found that the mutational distance for host jumps tends to decrease with broader host ranges (Poisson regression, slope = −0.113; two-tailed Z -test for slope = 0, Z  = −9.40, d.f. = 129, P  < 0.0001). In contrast, the relationship between mutational distance and host range for viral lineages that have not experienced host jumps is only weakly positive (slope = 0.0843; Z  = 7.16, d.f. = 127, P  < 0.0001) (Fig. 4c ). Similarly, the minimum dN/dS for a host jump decreases more substantially for viral cliques with broader host ranges (slope = −0.427; Z  = −9.18, d.f. = 116, P  < 0.0001) than for non-host jump controls (slope = 0.143; Z  = 3.08, d.f. = 116, P  < 0.01) (Fig. 4d ). These trends in mutational distance and dN/dS were consistent when the same analysis was performed for ssDNA, dsDNA, +ssRNA and −ssRNA viruses separately (Extended Data Fig. 6 ). These results indicate that, on average, ‘generalist’ multihost viruses experience lower degrees of adaptation when jumping into new vertebrate hosts.

Host jump adaptations are gene and family specific

We next examined whether genes with different established functions displayed distinctive patterns of adaptive evolution linked to host jump events. Since gene function remains poorly characterized in the large and complex genomes of dsDNA viruses, we focused on the shorter ssRNA and ssDNA viral families. We selected for analysis the four non-segmented viral families with the greatest number of host jump lineages in our dataset: Coronaviridae (+ssRNA; n  = 2,537), Rhabdoviridae (−ssRNA; n  = 1,097), Paramyxoviridae (−ssRNA; n  = 787) and Circoviridae (ssDNA; n  = 695). For these viral families, we extracted all annotated protein-coding regions from their genomes and categorized them as either being associated with cell entry (termed ‘entry’), viral replication (‘replication-associated’) or virion formation (‘structural’), and classifying the remaining genes as ‘auxiliary’ genes.

For the Coronaviridae , Paramyxoviridae and Rhabdoviridae , the entry genes encode surface glycoproteins that could also be considered structural but were not categorized as such given their important role in mediating cell entry. The capsid gene of circoviruses, however, encodes the sole structural protein that is also the key mediator of cell entry and was therefore categorized as structural. To estimate putative signatures of adaptation in relation to lineages that have experienced host jumps for the different gene categories, we modelled the change in log 10 (dN/dS) in host jumps versus non-host jumps using a linear model, while correcting for the effects of clique membership (see Methods ). Contrary to our expectation that entry genes would generally be under the strongest adaptive pressures during a host jump, we found that the strength of adaptation signals for each gene category varied by family. Indeed, the strongest signals were observed for structural proteins in coronaviruses (effect = 0.375, two-tailed t -test for difference in parameter estimates, t  = 4.35, d.f. = 10,121, P  < 0.0001) and auxiliary proteins in paramyxoviruses (effect = 0.439, t  = 2.15, d.f. = 4,225, P  = 0.02) (Fig. 5 ). Meanwhile, no significant adaptive signals were observed in the entry genes of all families (minimum P  = 0.3), except for the capsid gene in circoviruses (effect = 0.325, t  = 2.68, d.f. = 1,367, P  = 0.004) (Fig. 5 ). These findings suggest that selective pressures acting on viral genomes in relation to host jumps are likely to differ by gene function and viral family.

The strength of adaptation signals in genes associated with host jump and non-host jump lineages were estimated using linear models for Coronaviridae ( n  = 10,129), Paramyxoviridae ( n  = 4,233), Rhabdoviridae ( n  = 3,321), and Circoviridae ( n  = 1,373). We modelled the effects of gene type and host jump status on log(dN/dS) while correcting for viral clique membership and, for each gene type, inferred the strength of adaptive signal (denoted ‘effect’) as the difference in parameter estimates for host jumps versus non-host jumps. Points and lines represent the parameter estimates and their standard errors, respectively. Differences in parameter estimates were tested against zero using a one-tailed t -test. Subpanels for each gene type were ordered from left to right with increasing effect estimates.

Given the lack of adaptive signals in the entry proteins, we further hypothesized that within each gene, adaptative changes are likely to be localized to regions of functional importance and/or that are under relatively stronger selective pressures exerted by host immunity. To test this, we focused on the spike gene (entry) of viral cliques within the Coronaviridae since the key region involved in viral entry is well characterized (that is, the receptor-binding domain (RBD)) 33 . We found that dN/dS estimates consistent with adaptive evolution were indeed localized to the RBDs, but also to the N-terminal domains (NTD), of SARS-CoV-2 (genus Betacoronavirus ), avian infectious bronchitis virus (IBV; Gammacoronavirus ) and MERS (genus Alphacoronavirus ) (Extended Data Fig. 7 ). This is consistent with the strong immune pressures exerted on these regions of the spike protein 34 , 35 and the central role of the RBD in host-cell recognition and entry 36 , 37 , 38 . Overall, our results indicate that the extent of adaptation associated with a host jump likely varies by gene function, gene region and viral family.

The post-genomic era has opened opportunities to advance our understanding of the diversity of viruses in circulation and the macroevolutionary principles of viral host range. Leveraging ~59,000 publicly available viral sequences isolated from vertebrate hosts, we inferred that humans give more viruses to other vertebrates than they give to us across the 32 viral families we considered. We further demonstrated that host jumps are associated with heightened signals of adaptive evolution that tend to decrease in viruses with broader host ranges. This indicates that there may be a minimum mutational threshold necessary for viruses to expand their host range. Finally, we showed that adaptive evolution linked to host jumps may vary by gene function and may be localized to specific gene regions of functional importance.

To bypass the limitations of existing viral taxonomies, we used a taxonomy-agnostic approach to define roughly equivalent units of viral diversity, which formed the basis for most of the analyses presented in this study. The use of operational taxonomic units rather than traditional taxonomic species names further allowed us to perform like-for-like analyses across the entire diversity of viruses. Our approach identified cliques that were largely concordant with traditional viral species nomenclature but also highlighted inconsistencies, where in some cases, single viral species appear to form distinct taxonomic groups while other groups of species seem to form a single group solely based on their genetic relatedness (Fig. 2 and Extended Data Fig. 3 ). However, we do not claim that our approach should supersede existing taxonomic classification systems, especially since a robust and meaningful species definition requires the integration of viral properties with finer-scale evolutionary analyses that was not necessary for our purposes. Nevertheless, we anticipate that the development and use of similar network-based approaches may pave the way for the development of efficient classification frameworks that can rapidly incorporate novel, metagenomically derived viruses into existing taxonomies.

Harnessing cliques as a mechanism of identifying clusters of related viruses for phylogenetic inspection allowed us to quantify the number and sources of recent host jump events. One important caveat to this approach is that the viral cliques involved in putative host jumps represent only a fraction of the viral diversity sequenced thus far (Extended Data Fig. 4b ) and the patterns we observed may change as more viruses are discovered. However, we consistently found higher frequencies of anthroponotic than zoonotic jumps across 16 of the 21 viral families (Extended Data Fig. 5d ). Since each of these families are associated with varying viral discovery effort, the consistency of this pattern makes it highly unlikely that surveillance biases are driving the excess of anthroponotic jumps we inferred. Another caveat is that our clique assignment approach clusters viruses within ~15% sequence divergence, which limits our analyses to relatively recent host jump events. However, the limited divergence of the sequences within each clique also allowed us to produce more robust alignments and hence evolutionary inferences.

Of the 599 recent host jumps identified, 64% were inferred as anthroponotic (Fig. 3b ). While the relative importance of anthroponotic versus zoonotic events has been speculated 13 , 29 , 39 , 40 , we provide a formal evaluation of the zoonotic-to-anthroponotic ratio in vertebrates, showing that anthroponoses are equally, if not more, critical to consider than zoonoses when assessing viral spillover dynamics. It stands to reason that the substantial global human population size and ubiquitous spatial distribution position us as a major source for viral exchange. However, it is also likely that behavioural factors might amplify the risk of anthroponotic transmission, for example, through changes in land use, agricultural methods or heightened interactions between humans and wildlife 4 . Overall, our results highlight the importance of surveying and monitoring human-to-animal transmission of viruses, and its impacts on human and animal health.

We observed heightened evolution and adaptive signals in association with host jumps (Fig. 4 ). This result is largely intuitive, since a virus jumping into a new host is likely to be under different selective pressures exerted directly by the novel host environment and indirectly by changes in host-to-host transmission dynamics. The evolutionary signals we captured may include pre-requisite adaptations that enable a virus to infect the new host. In addition, they probably also represent the burst of adaptive mutations which may be acquired following a host jump, which has been demonstrated for multiple viral systems 24 , 41 , 42 , 43 . Further, these signals could potentially reflect a relaxation of previous selective pressures no longer present in the novel host. We note that these signals of heightened evolution could also, in principle, be inflated by sampling bias, where two viruses circulating in the same host are more often drawn from the same population. However, this was largely controlled for in our analysis through comparisons to representative non-host jump lineages that are expected to be affected by the same sampling bias.

We observed lower mutational and adaptive signals associated with host jumps for viruses that infect a broader range of hosts (Fig. 4c,d ). The most likely explanation for this pattern is that some viruses are intrinsically more capable of infecting a diverse range of hosts, possibly by exploiting host-cell machinery that are conserved across different hosts. For example, sarbecoviruses (the subgenus comprising SARS-CoV-2) target the ACE2 host-cell receptor, which is conserved across vertebrates 44 , 45 , and the high structural conservation of the sarbecovirus spike protein 15 may explain the observation that single mutations can enable sarbecoviruses to expand their host tropism 46 . In other words, multihost viruses may have evolved to target more conserved host machinery that reduces the mutational barrier for them to productively infect new hosts. This may provide a mechanistic explanation for previous observations that viruses with broad host range have a higher risk of emerging as zoonotic diseases 2 , 3 , 5 .

Our approach to identifying putative host jumps hinges on ancestral-state reconstruction (Fig. 3a ), which has been shown to be affected by sampling biases 47 , 48 . However, we accounted for this, at least in part, by including sequencing effort as a measure of sampling bias in our statistical models, allowing us to draw inferences that were robust to disproportionate sampling of viruses in different hosts. Our approach also does not consider the epidemiology or ecology of viral transmission, as this is largely dependent on host features such as population size, social structure and behaviour for which comprehensive datasets at this scale are not currently available. We anticipate that future datasets that integrate ecology, epidemiology and genomics may allow more granular investigations of these patterns in specific host and viral systems. In addition, the patterns we described are broad and do not capture the idiosyncrasies of individual host–pathogen associations. These include a variety of biological features— intrinsic ones, such as the molecular adaptations required for receptor binding, as well as more complex ones including cross-immunity and interference with other viral pathogens circulating in a host population.

Overall, our work highlights the large scope of genomic data in the public domain and its utility in exploring the evolutionary mechanisms of viral host jumps. However, the large gaps in the genomic surveillance of viruses thus far suggest that we have only just scratched the surface of the true viral diversity in nature. In addition, despite the strong anthropocentric bias in viral surveillance, 81% of the putative host jumps identified in this study do not involve humans, emphasizing the large underappreciated scale of the global viral-sharing network (Extended Data Fig. 8 ). Widening our field of view beyond zoonoses and investigating the flow of viruses within this larger network could yield valuable insights that may help us better prepare for and manage infectious disease emergence at the human–animal interface.

Data acquisition, curation and quality control

The metadata of all partial and complete viral genomes were downloaded from NCBI Virus ( https://www.ncbi.nlm.nih.gov/labs/virus/vssi/#/ ) on 22 July 2023, with filters excluding sequences isolated from environmental sources, lab hosts, or associated with vaccine strains or proviruses ( n  = 11,645,803). Where possible, host taxa names in the metadata were resolved in accordance with the NCBI taxonomy 49 using the ‘taxizedb’ v.0.3.1 package in R. User-submitted viral species names were compared to the ICTV master species list version ‘MSL38.V2’ dated 6 July 2023.

To generate a candidate list of viral sequences for further genomic analysis, the metadata were filtered to include 53 viral families known to infect vertebrate hosts on the basis of information provided in the 2022 release of the ICTV taxonomy ( https://ictv.global/taxonomy ) 50 and with reference to that provided by ViralZone ( https://viralzone.expasy.org/ ) 51 . We then retained only sequences from viral families comprising at least 100 sequences of greater than 1,000 nt in length. Since the sequences of segmented viral families are rarely deposited as whole genomes and since the high frequency of reassortment 23 precludes robust phylogenetic reconstruction, we identified sequences for single genes conserved within each of these families for further analysis ( Arenaviridae : L segment; Birnaviridae : ORF1/RdRP/VP1/Segment B; Peribunyaviridae : L segment; Orthomyxoviridae : PB1; Picobirnaviridae : RdRP; Sedoreoviridae : VP1/Segment 1/RdRP; Spinareoviridae : Segment 1/RdRP/Lambda 3). These sequences were retrieved by applying text-based pattern matching (that is, ‘grepl’ in R) to query the GenBank sequence titles. For non-segmented genomes, we retained all non-human-associated sequences and subsampled the human-associated sequences as follows: we selected a random subsample of 1,000 SARS-CoV-2 genomes of greater than 28,000 nt from distinct countries, isolation sources and with distinct collection dates. For influenza B, we retained only human sequences with distinct country of origins, sample types and collection dates, and hosts of isolation. For other human-associated sequences, we retained viruses with distinct species, country, isolation source and collection date information. We then downloaded the final candidate list of viral sequences ( n  = 92,973) using ‘ncbi-acc-download’ v.0.2.8 ( https://github.com/kblin/ncbi-acc-download ). Further quality control of the genomes downloaded was performed using ‘CheckV’ (v.1.0.1) 52 , retaining sequences with more than 95% completeness (for non-segmented viruses) and less than 5% contamination (for all sequences). This resulted in a final genomic dataset comprising 58,657 observations (Supplementary Table 1 ) composed of gene sequences for segmented viruses and complete genomes for non-segmented viruses. For simplicity, we will henceforth refer to the gene sequences and complete genomes as ‘genomes’.

Taxonomy-agnostic identification of viral cliques

To identify viral cliques, we calculated the pairwise alignment-free Mash distances of genomes within each viral family via ‘Mash’ (v.1.1) 53 with a k -mer size of 13. This k -mer size ensures that the probability of observing a k -mer by chance, given the median genome length for each clique, is less than 0.01. Given a genome length, l , alphabet, Σ  = {A, T, G, C}, and the desired probability of observing a k -mer by chance, q  = 0.01, this was computed using the formula described previously 53 :

We then constructed undirected graphs for each viral family with nodes and edges representing genomes and Mash distances, respectively. From these networks, we removed edges with Mash distance values greater than a certain threshold, t , before we applied the community-detection algorithm, Infomap 54 . This community-detection algorithm performs well in both large (>1,000 nodes) and small (≤1,000 nodes) undirected graphs 55 and seeks to identify subgraphs within these undirected graphs that minimize the information required to constrain the movement of a random walker 54 . We refer to the subgraphs identified through this algorithm as ‘viral cliques’. Here we forced the community-detection algorithm to identify taxonomically relevant cliques by removing edges with Mash distance values greater than t , which resulted in sparser graphs with closely related genomes (for example, from the same species) being more densely connected than more distantly related genomes (for example, different species). The value of t was selected by maximizing the proportion of monophyletic cliques identified and the concordance of the viral cliques identified with the viral species names from the NCBI taxonomy, based on the commonly used clustering performance metrics, AMI and ARI (Supplementary Fig. 2 ). These metrics were computed using the ‘AMI’ and ‘ARI’ functions in ‘Aricode’ v.1.0.2. To assess whether the viral cliques identified fulfil the species definition criterion of being monophyletic 18 , we reconstructed the phylogenies of each viral family by applying the neighbour-joining algorithm 56 implemented in the ‘Ape’ v.5.7.1 R package on their pairwise Mash distance matrices. We then computed the proportion of monophyletic viral cliques using the ‘is.monophyletic’ function in Ape v.5.7.1 across the various values of t . Given the discordance between the NCBI and ICTV taxonomies, we applied the above optimization protocol to t using the viral species names in the ICTV taxonomy. Using the NCBI viral species names, t  = 0.15 maximized both the median AMI and ARI across all families (Supplementary Fig. 2a ), with 94.3% of the cliques identified being monophyletic (Supplementary Fig. 2b ). Using the ICTV viral species names, t  = 0.2 and t  = 0.25 maximized the median AMI and median ARI across families (Supplementary Fig. 2c ), with 93.7% and 87.8% of the cliques being monophyletic (Supplementary Fig. 2b ), respectively. Since t  = 0.15 produced the highest proportion of monophyletic clades that were approximately concordant with existing viral taxonomies, we used this threshold to generate the final viral clique assignments for downstream analyses (Supplementary Table 1 ).

Identification of putative host jumps

We retrieved all viral cliques that were associated with at least two distinct host genera and comprised at least 10 genomes ( n  = 215). We then generated clique-level genome alignments using the ‘FFT-NS-2’ algorithm in ‘MAFFT’ (v.7.490) 57 , 58 . We masked regions of the alignments that were poorly aligned or prone to sequencing error by replacing alignment sites that had more than 10% of gaps or ambiguous nucleotides with Ns. Clique-level genome alignments that had more than 20% of the median genome length masked were considered to be poorly aligned and thus removed from further analysis ( n  = 6; Supplementary Fig. 3 ). Following this procedure, we reconstructed maximum-likelihood phylogenies for each viral clique with ‘IQ-Tree’ (v.2.1.4-beta) 59 , using 1,000 ultrafast bootstrap (UFBoot) 60 replicates. The optimal substitution model for each tree was automatically determined using the ‘ModelFinder’ 61 utility native to IQ-Tree. To estimate the root position for each clique tree, we reconstructed neighbour-joining Mash trees for each viral clique, including 10 additional genomes whose minimum pairwise Mash distance to the genomes in each tree was 0.3–0.5, as potential outgroups. The most basal tips in these neighbour-joining Mash trees were identified and used to root the maximum-likelihood clique trees. This approach, as opposed to using maximum-likelihood phylogenetic reconstruction involving the outgroups, was used as it is difficult to reliably align clique sequences with highly divergent outgroups.

To identify putative host jumps, we performed ancestral-state reconstruction on the resultant rooted maximum-likelihood phylogenies with host as a discrete trait using the ‘ace’ function in Ape v.5.7.1. Traversing from a tip to the root node, a putative host jump is identified if the reconstructed host state of an ancestral node is different from the observed tip state, has a twofold greater likelihood compared with alternative states and is different from the host state of the sampled tip. Where the tip and ancestral host states were of different taxonomic ranks, we excluded putative host jumps where the ancestral host state is nested within the tip host state, or vice versa (for example, ‘ Homo ’ and ‘Hominidae’). Missing host metadata were encoded as ‘unknown’ and included in the ancestral-state reconstruction analysis. Host jumps involving unknown or non-vertebrate host states were excluded from further analysis. Separately, we extracted non-host jump lineages to control for any biases in our analysis approach. To do so, we randomly selected an ancestral node where the reconstructed host state is the same as the observed tip state and has a twofold greater likelihood than alternative host states, for each viral genome that is not involved in any putative host jumps. For the mutational distance and dN/dS analyses, we retained only viral cliques where non-host jump lineages could be identified. An analysis exploring the robustness of this host jump inference approach to sampling biases (Supplementary Fig. 1 ) and a more detailed description of the inference algorithm (Supplementary Fig. 4 ) are provided in Supplementary Information .

Implementation of this algorithm yielded a list of all viral lineages involving a host jump (Supplementary Table 2 ). Since multiple lineages may involve a host transition at the same ancestral node, we calculated the number of unique host jump events as the number of distinct nodes for each unique host pair. For example, the three lineages Node1 (host A)→Tip1 (host B), Node1 (host A)→Tip2 (host B) and Node1 (host A)→Tip3 (host C) would be considered as two distinct host jump events, one between hosts A and B and the other between hosts A and C. This counting approach was used for Fig. 3a and Extended Data Fig. 5 . The list of all 2,904 distinct host jumps is provided in Supplementary Table 3 .

Calculating mutational distances and dN/dS

Mutational distance and dN/dS estimates may be lineage specific and may depend on sampling intensity. In addition, there is a nonlinear relationship between dN/dS and branch length, that is, the estimated dN/dS decreases with increasing evolutionary distance 62 . Therefore, we opted to compare the minimum adaptive signal (that is, minimum dN/dS) associated with a host jump for each clique. For host jump lineages, mutational distances were calculated as the sum of the branch lengths between the tip sequence and the ancestral node for which the first host state transition occurred (in substitutions per site) using the ‘get_pairwise_distances’ function in the ‘Castor’ (v.1.7.10) 63 R package; this was then multiplied by the alignment length to obtain the estimated number of substitutions (Fig. 3a ). To calculate the dN/dS estimates, we reconstructed the ancestral sequences of ancestral nodes using the ‘-asr’ flag in IQ-Tree, which is based on an empirical Bayesian algorithm ( http://www.iqtree.org/doc/Command-Reference ). We then extracted coding regions from the clique-level masked alignments based on the user-submitted gene annotations on NCBI GenBank (in ‘gff’ format) of each viral genome. We then computed the dN/dS estimates using the method of ref. 64 implemented in the ‘dnastring2kaks’ function of the ‘MSA2dist’ v.1.4.0 R package ( https://github.com/kullrich/MSA2dist ). We calculated the minimum mutational distance and dN/dS across all host jump events in each clique for our downstream statistical analyses, which, in principle, represents the minimum evolutionary signal associated with a host jump in each viral clique. For non-host jump lineages, we similarly computed the minimum mutational distance and dN/dS across the randomly selected lineages. Estimates where dN = 0 or dS = 0 were removed. The list of all minimum mutational distance and minimum dN/dS estimates is provided in Supplementary Tables 4 and 5 , respectively. The dN/dS estimates for the analysis shown in Fig. 5 are provided in Supplementary Table 6 .

For the coronavirus spike gene analysis (Extended Data Fig. 7 ), spike sequences were extracted from the clique-level multiple sequence alignments, with gaps trimmed to the reference sequences (avian infectious bronchitis virus, EU714028.1; SARS-CoV-2, MN908947.3; MERS, JX869059.2). The genomic coordinates for the functional domains of the spike proteins were derived from previous studies 33 , 37 , 65 . Estimates where dN = 0 or dS = 0 were removed. The dN/dS estimates are provided in Supplementary Table 7 .

Statistical analyses

All statistical analyses were performed using the ‘stats’ package native to R v.4.3.1. To generate the bootstrapped distributions shown in Fig. 3b , we randomly resampled the host jumps within each clique with replacement (1,000 iterations) and performed two-tailed paired t -tests using the ‘t.test’ function. Mann–Whitney U -tests, analysis of variance (ANOVA), linear regressions, and Poisson and logistic regressions were implemented using ‘wilcox.test’, ‘anova’, ‘lm’ and ‘glm’ functions, respectively.

A permutation test was performed to assess whether the higher proportion of anthroponotic versus zoonotic jumps was statistically significant. We randomly permuted the host states in each clique for 500 iterations while preserving the number of host-jump and non-host-jump lineages (illustrated in Supplementary Fig. 5 ). The P value was calculated as the number of iterations where the permutated anthroponotic/zoonotic ratio was greater than or equal to the observed ratio.

To assess the relationship between host range and adaptative signals (Fig. 4 ), we used Poisson regressions to model the expected number of host genera observed in each viral clique, λ host range . We corrected for the number of genomes in each clique, g , as a measure of sampling effort, and viral family membership, v , by including them as fixed effects in these models. These models can be formalized for mutational distance or dN/dS, d , with some p number of viral families and residual error, ε , as:

We tested whether the parameter estimates were non-zero by performing two-tailed Z -tests implemented within the ‘summary’ function in R.

To estimate the strength of adaptive signals for coronaviruses, paramyxoviruses, rhabdoviruses and circoviruses (Fig. 5 ) by gene type, we implemented two linear regression models for each viral family. Since the overall adaptive signal may differ for each viral clique, we corrected for this effect by using an initial linear model where the number of viral cliques, viral clique membership and residual are given by q , c and ε , respectively, as follows:

Subsequently, we used the corrected log(dN/dS) estimates represented by the residuals of model 1, ε model 1 , in a second linear model partitioning the effects of gene type by host jump status, j . Given r number of gene types, this model can be formalized as follows:

The estimated effects shown in Fig. 5 , representative of the difference in adaptive signals associated with jump and non-host jump lineages for each gene type, were then computed as:

To test whether this effect is statistically significant, we used a one-tailed t -test, with the t statistic computed using the standard error of the parameter estimates in model 2:

The residuals of model 2 were confirmed to be approximately normal by visual inspection (Supplementary Fig. 6 ).

Data analysis and visualization

All data analyses were performed using R v.4.3.1. All visualizations were performed using ggplot (v.3.4.2) 66 or ggtree (v.3.8.2) 67 . UpSet plots were created using the R package, UpSetR (v.1.4.0) 68 .

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

The full list of accessions considered in this study is provided in Supplementary Data 1 . The data used for the main analyses are provided in Supplementary Tables 2–7 . All reconstructed maximum-likelihood trees and ancestral sequences used for the analyses are hosted on Zenodo ( https://doi.org/10.5281/zenodo.10214868 ) 69 .

Code availability

All custom code used to perform the analyses reported here are hosted on GitHub ( https://github.com/cednotsed/vertebrate_host_jumps ).

Jones, K. E. et al. Global trends in emerging infectious diseases. Nature 451 , 990–993 (2008).

Article   ADS   CAS   PubMed   PubMed Central   Google Scholar  

Shaw, L. P. et al. The phylogenetic range of bacterial and viral pathogens of vertebrates. Mol. Ecol. 29 , 3361–3379 (2020).

Article   PubMed   Google Scholar  

Olival, K. J. et al. Host and viral traits predict zoonotic spillover from mammals. Nature 546 , 646–650 (2017).

Gibb, R. et al. Zoonotic host diversity increases in human-dominated ecosystems. Nature 584 , 398–402 (2020).

Article   ADS   CAS   PubMed   Google Scholar  

Woolhouse, M. E. J. & Gowtage-Sequeria, S. Host range and emerging and reemerging pathogens. Emerg. Infect. Dis. 11 , 1842–1847 (2005).

Article   PubMed   PubMed Central   Google Scholar  

Albery, G. F., Eskew, E. A., Ross, N. & Olival, K. J. Predicting the global mammalian viral sharing network using phylogeography. Nat. Commun. 11 , 2260 (2020).

Taylor, L. H., Latham, S. M. & Woolhouse, M. E. J. Risk factors for human disease emergence. Phil. Trans. R. Soc. Lond. B 356 , 983–989 (2001).

Article   CAS   Google Scholar  

Albery, G. F. et al. Urban-adapted mammal species have more known pathogens. Nat. Ecol. Evol. 6 , 794–801 (2022).

Karesh, W. B. et al. Ecology of zoonoses: natural and unnatural histories. Lancet 380 , 1936–1945 (2012).

Cleaveland, S., Laurenson, M. K. & Taylor, L. H. Diseases of humans and their domestic mammals: pathogen characteristics, host range and the risk of emergence. Phil. Trans. R. Soc. Lond. B 356 , 991–999 (2001).

Carlson, C. J. et al. The Global Virome in One Network (VIRION): an atlas of vertebrate–virus associations. mBio 13 , e02985-21 (2022).

Gibb, R. et al. Data proliferation, reconciliation, and synthesis in viral ecology. BioScience 71 , 1148–1156 (2021).

Article   Google Scholar  

Kuchipudi, S. V. et al. Coordinated surveillance is essential to monitor and mitigate the evolutionary impacts of SARS-CoV-2 spillover and circulation in animal hosts. Nat. Ecol. Evol. https://doi.org/10.1038/s41559-023-02082-0 (2023).

Watsa, M. & Wildlife Disease Surveillance Focus Group. Rigorous wildlife disease surveillance. Science 369 , 145–147 (2020).

Tan, C. C. et al. Genomic screening of 16 UK native bat species through conservationist networks uncovers coronaviruses with zoonotic potential. Nat. Commun. 14 , 3322 (2023).

Pepin, K. M., Lass, S., Pulliam, J. R. C., Read, A. F. & Lloyd-Smith, J. O. Identifying genetic markers of adaptation for surveillance of viral host jumps. Nat. Rev. Microbiol. 8 , 802–813 (2010).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Kaur, T. et al. Descriptive epidemiology of fatal respiratory outbreaks and detection of a human‐related metapneumovirus in wild chimpanzees ( Pan troglodytes ) at Mahale Mountains National Park, Western Tanzania. Am. J. Primatol. 70 , 755–765 (2008).

Simmonds, P. et al. Four principles to establish a universal virus taxonomy. PLoS Biol. 21 , e3001922 (2023).

Adams, M. J. et al. 50 years of the International Committee on Taxonomy of Viruses: progress and prospects. Arch. Virol. 162 , 1441–1446 (2017).

Article   CAS   PubMed   Google Scholar  

Walker, P. J. et al. Recent changes to virus taxonomy ratified by the International Committee on Taxonomy of Viruses (2022). Arch. Virol. 167 , 2429–2440 (2022).

Blaxter, M. et al. Defining operational taxonomic units using DNA barcode data. Phil. Trans. R. Soc. B 360 , 1935–1943 (2005).

Acman, M., van Dorp, L., Santini, J. M. & Balloux, F. Large-scale network analysis captures biological features of bacterial plasmids. Nat. Commun. 11 , 2452 (2020).

McDonald, S. M., Nelson, M. I., Turner, P. E. & Patton, J. T. Reassortment in segmented RNA viruses: mechanisms and outcomes. Nat. Rev. Microbiol. 14 , 448–460 (2016).

Tan, C. C. S. et al. Transmission of SARS-CoV-2 from humans to animals and potential host adaptation. Nat. Commun. 13 , 2988 (2022).

Kuchipudi, S. V. et al. Multiple spillovers from humans and onward transmission of SARS-CoV-2 in white-tailed deer. Proc. Natl Acad. Sci. USA 119 , e2121644119 (2022).

Munnink, B. B. O. et al. Transmission of SARS-CoV-2 on mink farms between humans and mink and back to humans. Science 371 , 172–177 (2021).

Article   ADS   Google Scholar  

McAloose, D. et al. From people to Panthera: natural SARS-CoV-2 infection in tigers and lions at the Bronx Zoo. mBio 11 , e02220-20 (2020).

Short, K. R. et al. One health, multiple challenges: the inter-species transmission of influenza A virus. One Health 1 , 1–13 (2015).

Nelson, M. I. & Vincent, A. L. Reverse zoonosis of influenza to swine: new perspectives on the human–animal interface. Trends Microbiol. 23 , 142–153 (2015).

Samara, E. M. & Abdoun, K. A. Concerns about misinterpretation of recent scientific data implicating dromedary camels in epidemiology of Middle East respiratory syndrome (MERS). mBio 5 , e01430-14 (2014).

Du, L. & Han, G.-Z. Deciphering MERS-CoV evolution in dromedary camels. Trends Microbiol. 24 , 87–89 (2016).

Zhang, Z., Shen, L. & Gu, X. Evolutionary dynamics of MERS-CoV: potential recombination, positive selection and transmission. Sci. Rep. 6 , 25049 (2016).

Huang, Y., Yang, C., Xu, X., Xu, W. & Liu, S. Structural and functional properties of SARS-CoV-2 spike protein: potential antivirus drug development for COVID-19. Acta Pharmacol. Sin. 41 , 1141–1149 (2020).

Carabelli, A. M. et al. SARS-CoV-2 variant biology: immune escape, transmission and fitness. Nat. Rev. Microbiol. 21 , 162–177 (2023).

CAS   PubMed   PubMed Central   Google Scholar  

Wang, N. et al. Structural definition of a neutralization-sensitive epitope on the MERS-CoV S1-NTD. Cell Rep. 28 , 3395–3405 (2019).

Shang, J. et al. Structural basis of receptor recognition by SARS-CoV-2. Nature 581 , 221–224 (2020).

Shang, J. et al. Cryo-EM structure of infectious bronchitis coronavirus spike protein reveals structural and functional evolution of coronavirus spike proteins. PLoS Pathog. 14 , e1007009 (2018).

Yuan, Y. et al. Cryo-EM structures of MERS-CoV and SARS-CoV spike glycoproteins reveal the dynamic receptor binding domains. Nat. Commun. 8 , 15092 (2017).

Messenger, A. M., Barnes, A. N. & Gray, G. C. Reverse zoonotic disease transmission (zooanthroponosis): a systematic review of seldom-documented human biological threats to animals. PLoS ONE 9 , e89055 (2014).

Article   ADS   PubMed   PubMed Central   Google Scholar  

Edwards, S. J., Chatterjee, H. J. & Santini, J. M. Anthroponosis and risk management: a time for ethical vaccination of wildlife? Lancet Microbe 2 , e230–e231 (2021).

Schrauwen, E. J. & Fouchier, R. A. Host adaptation and transmission of influenza A viruses in mammals. Emerg. Microbes Infect. 3 , e9 (2014).

Villordo, S. M., Carballeda, J. M., Filomatori, C. V. & Gamarnik, A. V. RNA structure duplications and flavivirus host adaptation. Trends Microbiol. 24 , 270–283 (2016).

Urbanowicz, R. A. et al. Human adaptation of Ebola virus during the West African outbreak. Cell 167 , 1079–1087 (2016).

Damas, J. et al. Broad host range of SARS-CoV-2 predicted by comparative and structural analysis of ACE2 in vertebrates. Proc. Natl Acad. Sci. USA 117 , 22311–22322 (2020).

Lam, S. D. et al. SARS-CoV-2 spike protein predicted to form complexes with host receptor protein orthologues from a broad range of mammals. Sci. Rep. 10 , 16471 (2020).

Starr, T. N. et al. ACE2 binding is an ancestral and evolvable trait of sarbecoviruses. Nature 603 , 913–918 (2022).

Wright, A. M., Lyons, K. M., Brandley, M. C. & Hillis, D. M. Which came first: the lizard or the egg? Robustness in phylogenetic reconstruction of ancestral states. J. Exp. Zool. B 324 , 504–516 (2015).

Liu, P., Song, Y., Colijn, C. & MacPherson, A. The impact of sampling bias on viral phylogeographic reconstruction. PLoS Glob. Public Health 2 , e0000577 (2022).

Schoch, C. L. et al. NCBI Taxonomy: a comprehensive update on curation, resources and tools. Database 2020 , baaa062 (2020).

Lefkowitz, E. J. et al. Virus taxonomy: the database of the International Committee on Taxonomy of Viruses (ICTV). Nucleic Acids Res. 46 , D708–D717 (2018).

Hulo, C. et al. ViralZone: a knowledge resource to understand virus diversity. Nucleic Acids Res. 39 , D576–D582 (2011).

Nayfach, S. et al. CheckV assesses the quality and completeness of metagenome-assembled viral genomes. Nat. Biotechnol. 39 , 578–585 (2021).

Ondov, B. D. et al. Mash: fast genome and metagenome distance estimation using MinHash. Genome Biol. 17 , 132 (2016).

Rosvall, M. & Bergstrom, C. T. Maps of random walks on complex networks reveal community structure. Proc. Natl Acad. Sci. USA 105 , 1118–1123 (2008).

Yang, Z., Algesheimer, R. & Tessone, C. J. A comparative analysis of community detection algorithms on artificial networks. Sci. Rep. 6 , 30750 (2016).

Saitou, N. & Nei, M. The neighbor-joining method: a new method for reconstructing phylogenetic trees. Mol. Biol. Evol. 4 , 406–425 (1987).

CAS   PubMed   Google Scholar  

Katoh, K., Misawa, K., Kuma, K. & Miyata, T. MAFFT: a novel method for rapid multiple sequence alignment based on fast Fourier transform. Nucleic Acids Res. 30 , 3059–3066 (2002).

Katoh, K. & Standley, D. M. MAFFT multiple sequence alignment software version 7: improvements in performance and usability. Mol. Biol. Evol. 30 , 772–780 (2013).

Minh, B. Q. et al. IQ-TREE 2: new models and efficient methods for phylogenetic inference in the genomic era. Mol. Biol. Evol. 37 , 1530–1534 (2020).

Minh, B. Q., Nguyen, M. A. T. & von Haeseler, A. Ultrafast approximation for phylogenetic bootstrap. Mol. Biol. Evol. 30 , 1188–1195 (2013).

Kalyaanamoorthy, S., Minh, B. Q., Wong, T. K., von Haeseler, A. & Jermiin, L. S. ModelFinder: fast model selection for accurate phylogenetic estimates. Nat. Methods 14 , 587–589 (2017).

Wolf, J. B., Künstner, A., Nam, K., Jakobsson, M. & Ellegren, H. Nonlinear dynamics of nonsynonymous (d N) and synonymous (d S) substitution rates affects inference of selection. Genome Biol. Evol. 1 , 308–319 (2009).

Louca, S. & Doebeli, M. Efficient comparative phylogenetics on large trees. Bioinformatics 34 , 1053–1055 (2018).

Li, W.-H. Unbiased estimation of the rates of synonymous and nonsynonymous substitution. J. Mol. Evol. 36 , 96–99 (1993).

Lu, G. et al. Molecular basis of binding between novel human coronavirus MERS-CoV and its receptor CD26. Nature 500 , 227–231 (2013).

Wickham, H. ggplot2. Wiley Interdiscip. Rev. Comput. Stat. 3 , 180–185 (2011).

Yu, G., Smith, D. K., Zhu, H., Guan, Y. & Lam, T. T. ggtree: an R package for visualization and annotation of phylogenetic trees with their covariates and other associated data. Methods Ecol. Evol. 8 , 28–36 (2017).

Conway, J. R., Lex, A. & Gehlenborg, N. UpSetR: an R package for the visualization of intersecting sets and their properties. Bioinformatics 33 , 2938–2940 (2017).

Tan, C. C. S. Supplementary data for ‘Crossing host boundaries: the evolutionary drivers and correlates of viral host jumps’ [Data set]. Zenodo https://doi.org/10.5281/zenodo.10497734 (2023).

Download references

Acknowledgements

We thank R. J. Gibbs, G. Murray and L. P. Shaw for helpful feedback and discussions. C.C.S.T. was funded by the National Science Scholarship from the Agency for Science, Technology and Research (A*STAR), Singapore. F.B. and L.v.D. were funded by the European Commission (Horizon 2021–2024, END-VOC Project). L.v.D. was also funded by the UCL Excellence Fellowship. Views and opinions expressed are, however, those of the authors only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency. For the purpose of open access, the corresponding author has applied a ‘Creative Commons Attribution’ (CC BY) licence to any author-accepted version of the manuscript. The authors acknowledge the use of the UCL Myriad High Performance Computing Facility (Myriad@UCL), the UCL Department of Computer Science High Performance Computing Cluster and associated support services in the completion of this work.

Author information

These authors contributed equally: Lucy van Dorp, Francois Balloux.

Authors and Affiliations

UCL Genetics Institute, University College London, London, UK

Cedric C. S. Tan, Lucy van Dorp & Francois Balloux

The Francis Crick Institute, London, UK

Cedric C. S. Tan

You can also search for this author in PubMed   Google Scholar

Contributions

C.C.S.T. performed all analyses. L.v.D. and F.B. jointly supervised the study. C.C.S.T., L.v.D. and F.B. wrote the manuscript.

Corresponding author

Correspondence to Cedric C. S. Tan .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Peer review

Peer review information.

Nature Ecology & Evolution thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended data fig. 1 host and geographical distribution of viral sequences..

( a ) Number of viral sequences, excluding SARS-CoV-2, associated with the top 50 vertebrate hosts observed in the ‘others’ category as shown in main text Fig. 1a . ( b ) Number of viral sequences stratified by the four most-sequenced non-human animals, excluding SARS-CoV-2. The number of viral sequences for the top 10 countries are shown as bar plots. The percentage of viral sequences for the top three most sequenced viral species for each host are annotated.

Extended Data Fig. 2 Distribution of missing metadata for viral sequences.

(Top) Proportion of all viral sequences associated to non-human vertebrates ( n  = 1,599,672) with missing genus information or (bottom) sample collection year, stratified by viral family or country of origin. Countries with no associated sequences are denoted ‘NA’.

Extended Data Fig. 3 Viral cliques for Coronaviridae .

Sparse networks of viral cliques identified (see Methods) and their corresponding user-submitted species names for the Coronaviridae , similar to main text Fig. 2 . Nodes, node shapes, and edges represent individual genomes, their associated host and their pairwise Mash (alignment-free) distances, respectively.

Extended Data Fig. 4 Summary of viral cliques identified.

( a ) Number of viral cliques identified stratified by viral family. Cliques with only animal-associated sequences, human-associated sequences, or both are annotated. ( b ) Percentage of viral cliques involving at least one of the 2,904 putative host jumps inferred, stratified by viral family.

Extended Data Fig. 5 Robustness of host jump inference.

( a ) UpSet plot providing the intersecting host jumps identified via ancestral reconstruction when using a two-fold, five-fold or ten-fold likelihood threshold. ( b ) Bar plot showing the number of anthroponotic and zoonotic events inferred using various likelihood thresholds, ( c ) at different ancestral node depths, and ( d ) stratified by viral family. For (b), the number of anthroponotic and zoonotic host jumps were stratified by the depth of the ancestral node in the tip-to-node traversal. Since multiple host jump lineages can involve the same ancestral node, the tip-to-node depths may vary depending on which lineage is selected. As such, we randomly selected a viral lineage for each distinct host jump event for this analysis.

Extended Data Fig. 6 Adaptation analysis for viral groups.

Analysis of relationships between host range and estimated adaptive signals, similar to Fig. 3 , but only considering ssDNA, dsDNA, +ssRNA or -ssRNA viruses. Distributions of minimum ( a ) mutational distance and ( b ) dN/dS for host jump and non-host jumps on the logarithmic scale. We corrected for the effects of sequencing effort and viral family membership using Poisson regression models. The estimated effects of patristic distance on host range after these corrections are annotated. We tested whether the estimated effects were non-zero using two-tailed Z-tests. For all panels, each data point represents the minimum distance or dN/dS across all host jump or randomly selected non-host jump lineages in a single clique. Line segments represent linear regression smooths without correction.

Extended Data Fig. 7 Adaptive signals in the Coronaviridae spike gene.

Analysis of the log10(dN/dS) estimates associated to different functional domains encoded by the coronavirus spike gene: N-terminal domain (NTD), receptor-binding domain (RBD), fusion peptide (FP), heptad repeats 1 and 2 (HR1 and HR2), central helix (CH), transmembrane (TM), C-terminal domains (CT). Estimates with dN=0 or dS=0 were removed and the remaining number of sequences for each domain and viral clique are annotated. Differences in distributions were tested for using two-sided Mann-Whitney U tests and the corresponding p-values are annotated. Boxplot elements are defined as follows: centre line, median; box limits, upper and lower quartiles; whiskers, 1.5x interquartile range.

Extended Data Fig. 8 The global viral host jump network.

Directed network of the vertebrate viral-sharing network, where nodes and edges represent host genera and the number of viral cliques shared. Edge widths and colour are indicative of the number of viral cliques shared.

Supplementary information

Supplementary information.

Supplementary Note, Methods and Figs. 1–6.

Reporting Summary

Peer review file, supplementary data 1., supplementary tables 1–9., rights and permissions.

Open 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 permissions

About this article

Cite this article.

Tan, C.C.S., van Dorp, L. & Balloux, F. The evolutionary drivers and correlates of viral host jumps. Nat Ecol Evol (2024). https://doi.org/10.1038/s41559-024-02353-4

Download citation

Received : 11 January 2024

Accepted : 29 January 2024

Published : 25 March 2024

DOI : https://doi.org/10.1038/s41559-024-02353-4

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.