Earth’s biodiversity and human societies face pollution, overconsumption of natural resources, urbanization, demographic shifts, social and economic inequalities, and habitat loss, many of which are exacerbated by climate change. Here, we review links among climate, biodiversity, and society and develop a roadmap toward sustainability. These include limiting warming to 1.5°C and effectively conserving and restoring functional ecosystems on 30 to 50% of land, freshwater, and ocean “scapes.” We envision a mosaic of interconnected protected and shared spaces, including intensively used spaces, to strengthen self-sustaining biodiversity, the capacity of people and nature to adapt to and mitigate climate change, and nature’s contributions to people. Fostering interlinked human, ecosystem, and planetary health for a livable future urgently requires bold implementation of transformative policy interventions through interconnected institutions, governance, and social systems from local to global levels.

Plasticity has been acknowledged as having a central role in organismal evolutionary responses to environments (e.g. West-Eberhard, 2003; Sultan, 2015; Sgrò et al., 2016). As the field of plasticity expands (Bradshaw, 1965; Schlichting & Pigliucci, 1998; Sultan, 2015), rich data sets of reaction norm variation across organisms, trait types and ecological conditions are ripe for syntheses. Quantitative syntheses have enhanced our understanding of other fundamental mechanisms of evolution (e.g. heritable genetic variation Mousseau & Roff, 1987) and generated new hypotheses. The field of evolutionary biology has moved forward by scholars following up with new empirical data to test these hypotheses. Several decades after Mousseau & Roff's (1987) synthesis of heritability estimates, Wood et al. (2016) expanded on this research to consider the role of population size, employing modern meta-analytical techniques. This work demonstrates how revisiting synthetic scholarship through time can offer new and confirmatory insights. Given the diverse commentaries on Morrissey (2016) in this collection, in our comments here, we focus on aspects of reaction norm evolution. Specifically, we (i) emphasize the fundamental goals of our evaluation of divergence in reaction norms among species pairs and ecotypes (Murren et al., 2014), (ii) reiterate the collection procedure we employed in developing the database of studies and some biological implications of those search methods in interpretation of our results, (iii) highlight the distinction between Morrissey's (2016) treatment of the environment and our method that considers environmental novelty, (iv) discuss variation in reaction norm estimates and (v) conclude by advocating for further syntheses in reaction norm biology. Work evaluating the diversification of reaction norms between pairs of populations single clades or pairs of species raised in contrasting environments (e.g. Gilchrist & Huey, 2004; Wund et al., 2008; Pfennig et al. 2010; Griffith and Sultan 2012) demonstrates that variation in environment types and trait types results in evolution of diversity of forms. Inspired by this literature, in 2014, we described a mean-standardized approach to evaluating the evolutionary divergence in closely related species or in ecotypes of the same species within a study for attributes of reaction norm including differences in trait offset (overall difference in trait value across all environments) slope and curvature (Murren et al., 2014). We reported that all three attributes of divergence were detected. Further, we advocated that, taken together, slope and curvature differences warrant further consideration as important aspects of reaction norm evolution. Our sampling was broad in scope with respect to the types of reaction norms included (i.e. quadratic (e.g. thermal performance curves) to sigmoidal (e.g. developmental switches) shapes were part of the sample). In general, we found important roles for moderators of observed evolutionary divergence in reaction norms, including the environment type, trait type, organism type and environmental range. More fine-scale analysis of particular factor levels within moderators prompted us to call for further investigations into the history of environmental heterogeneity, novelty of the environment and the potential to diverge. Since publication, other authors have examined our database and used it as a basis for additional empirical and synthetic work. For example, Stamps (2016) suggested that particular behavioural traits need to be more explicitly considered. Others have employed our divergence approach to examine thermal reaction norms of closely related species to evaluate generalist vs. specialist strategies (Berger et al., 2014) or to contextualize adaptive responses (While et al. 2016) or model results (Scheiner et al., 2015). We ended our discussion with a call for further investigation of reaction norms with refined models as theory and methods of meta-analyses are swiftly developing. Thus, we applaud Morrissey's (2016) efforts in advancing the field. Nonetheless, a clarification of the methods for inclusion of an empirical study into our database highlights the differences in the evolutionary ecological context of our work and that of Morrissey (2016). Specifically, we did not include studies investigating whether genetic variation for plasticity could be uncovered; we only included those studies which the authors of the study had ecological reasons for anticipated divergence. Thus, we did not include any study with pairs of populations; we only included ones considered ‘ecotypes’ with an explicit ecological basis for anticipated phenotypic divergence. Therefore, Morrissey's simulation example includes a greater element of noise than our empirical biological sample. We note that Morrissey (2016) does not present a formal meta-analysis of our reaction norm data set for the same reason Murren et al. (2014) did not; standard errors and variances were not reported in most studies of reaction norms included in our data set. As an alternative, Morrissey (2016) proposed a quadratic random regression mixed model (QRRMM). In this model, any variation not fit by the quadratic regression is considered part of the residual variation; thus, any shape variation not captured by a simple quadratic represents residual variation. In addition, because most studies in our database only considered three environmental levels for measuring the reaction norms, only approx. 30% of the data provide information about the variation in the (quadratic term) of the reaction norm. As a result, using the quadratic model therefore is distinct from Murren et al.'s (2014) approach as we did not explicitly evaluate a particular reaction norm shape type. Rather, we employed an inclusive perspective on shape variation. When evaluating a dataset that is drawn from a range of shape types and varies in sample size for the reaction norm attributes, it is not surprising that this method favours detection of mean differences. While the QRRMM approach offers promising new directions to explore particularly under conditions where a specific quadratic shape is assumed, it is not surprising that applying this model results in a different conclusion than Murren et al. (2014). The inclusion of the treatment interval is another key difference between our approach to the analysis of reaction norms and the approach of by Morrissey (2016). His interpretation of how cross-environment studies are designed is ‘most studies are designed to span the relevant range of environmental conditions’ (Morrissey 2016). This is in contrast with our data collection, which included a clear focus on environments that were both inside and outside the native range (Table 1, in Murren et al., 2014). The bounding and ‘meaningful limits’ proposed by Morrissey (2016) have dual effects; doing so substantially reduces the number of data records potentially used in the analysis and second eliminates any evaluation of responses to novel environments. We agree that for a narrow subset of environments (e.g. temperature reaction norms) bounding is biologically appropriate as thresholds exist at both the low and high end of environments for biological function – yet the biological context (environmental novelty) for such selections remains a meaningful distinction. While we may make these nuanced evaluations for well-studied systems with clear expectations, environment types such as ordered predator severity or multivariate changes such as in transplant studies may not easily lend themselves to a bounding approach. Moreover, novel environmental conditions do not necessarily have clear phenotypic expectations regarding whether slopes or curvature might be exaggerated or reduced (e.g. Ghalambor et al., 2007; Schlichting, 2008). Therefore, we currently do not know if the ‘centre of the ranges of the reaction norms’ represents the most frequently encountered environment or how to interpret whether novel environments are included as part of the gradient. Meta-analytic approaches and quantitative syntheses are fundamental to evaluating emerging processes, contributing to hypothesis generation and complementing traditional reviews. Until all databases are curated and publically available, some syntheses will be incomplete. In reaction norm literature, most studies report means but many do not report measures of variation around means. Even with such incomplete information, such syntheses are still valuable (Kingsolver et al., 2016). We encourage scholars to consider implementing diverse analytical methods and develop new ones. In doing so, however, it is important to recognize that reanalyses that apply specific assumptions and employ a particular subset of data that meet these assumptions can unsurprisingly result in different conclusions. New methods that consider range shifts and other elements of diverse reaction norms may be especially useful in this regard. We close by advocating for authors of reaction norm studies (including syntheses) to make data publically available, at a minimum, by reporting both estimates of central tendency and errors, but ideally at the level of individual records rather than summary statistics. We acknowledge the contributions to understanding reaction norm evolution by our colleagues, H. Maclean, C. Handelsman, M. Heskel, C. Ghalambor, H. Callahan, D. Pfennig and C. Schlichting, and the National Evolution Synthesis Center.

Given the multiple commentaries on Morrissey (2016) published here, we focus our comments on Morrissey's re-analyses of previous synthetic analyses of phenotypic selection gradients (β and γ). Morrissey argues that formal meta-analyses of selection that account for sampling error are important; we agree and have conducted such analyses (Kingsolver et al., 2012). We are therefore surprised at Morrissey's focus here on the Kingsolver et al. (2001) study and its associated nonparametric analyses, given the greatly expanded data sets, additional analyses and new statistical tools that have become available in the past 15 years (Kingsolver et al., 2001; Hereford et al., 2004; Siepielski et al., 2009, 2011; Hadfield, 2010; Kingsolver & Diamond, 2011; Morrissey & Hadfield, 2012; Nakagawa & Santos, 2012; Koricheva et al., 2013). In Kingsolver et al. (2012), we conducted mixed-effects modelling of β in a Bayesian framework using MCMCglmm (Hadfield, 2010). This framework accounted for potential effects of both sampling errors and study- and species-level autocorrelation. We also incorporated moderator variables (e.g. trait type, fitness component or taxonomic group) in the models to evaluate whether moderators influenced β. [The model statements used in these analyses were provided in the Supplement of Kingsolver et al. (2012).] We applied these analyses to the data set of Kingsolver and Diamond (2011) [updated and much larger than that of Kingsolver et al. (2001)], to estimate the posterior distribution (and mean) of β; a folded normal distribution was then used to make inferences about |β| (Hereford et al., 2004; Morrissey & Hadfield, 2012). Contrary to Morrissey's (2016) brief summary of Kingsolver et al. (2012), we reported in that paper that accounting for sampling error substantially reduced the estimated mean magnitude of selection (|β|), but that differences in average β or |β| among different trait types, fitness components or taxonomic groups were similar for analyses that do (Kingsolver et al., 2012: Fig. 1 and Table S1) or do not (Kingsolver & Diamond, 2011: Fig. 2 and Table 1; Kingsolver et al., 2001: Figs 4 and 5) formally account for sampling error. It is unclear to us whether the quantitative differences in estimates between Kingsolver et al. (2012: Table S1) and Morrissey (2016: Table 1) are due to the use of different data sets or differences in the models employed. Morrissey (2016) also asserts that the difference between Kingsolver et al. (2012)'s and Morrissey (2016)'s estimate of the mean absolute value of selection represents a mathematical error in Kingsolver et al. (2012)'s analysis. However, this criticism is misdirected. Morrissey (2016) incorrectly interprets the difference between ‘corrected’ and ‘uncorrected’ values of the mean absolute value of selection from Kingsolver et al. (2012) (Fig. 2) as reflecting estimates from models which do and do not incorporate standard errors. In fact, the distinction between uncorrected and corrected values reflects a much broader difference between estimates from simple summary statistics of the kernel density distribution of selection coefficients versus those from a formal meta-analytical model that incorporates standard errors and random effects structure to account for study and species-level autocorrelation (see Kingsolver et al., 2012: page 1104 for a description of the comparisons made, and Fig. 2, Table S2 for results). A common problem for conducting formal meta-analyses is that estimates of standard errors (or equivalent) may not be available. For example, despite repeated calls for such information, more than half of the estimates of β or γ reported in the literature do not include values for standard errors (Kingsolver et al., 2001; Siepielski et al., 2009). Morrissey (2016) is largely silent on the issue of missing data, but his re-analyses of phenotypic selection appear to only include information from data records where standard errors are available. We argue that excluding more than half of the data throws out valuable and hard-won information and has the potential to generate biases in the resulting estimates. The goal of meta-analyses is to summarize the available evidence (Gurevitch et al., 1992): informal meta-analysis of all available data may be more valuable than analyses of a fraction of the data that includes sampling error. In addition, there are many methods for imputing unknown standard errors (reviewed in Lajeunese, 2013). For example, unknown standard errors may be imputed using: the mean or median of the known standard errors (Wiebe et al., 2006); the relationship between known standard errors and sample size, with sample size being a more commonly reported metric (Ma et al., 2008); or using multiple imputation techniques (Rubin & Schenker, 1991). Whereas none of these methods perfectly solves the issue of missing standard errors, they can provide an important bridge between informal and formal meta-analysis until better reporting practices are adopted in the literature. Indeed, although imputation methods have the potential to introduce bias, comparisons of meta-analyses that use only data with available variances versus those that use data sets with imputed variances have yielded similar results (Philbrook et al., 2007). At a minimum, a comparison of results from reduced data sets (i.e. with standard errors, e.g. Kingsolver et al., 2012) and the full data set (i.e. with and without standard errors, e.g. Kingsolver & Diamond, 2011) can provide insight into potential bias associated with excluding those studies that do not report standard errors. The focus in Morrissey (2016) on sampling error alone ignores a variety of other important issues in meta-analysis including nonrandom sampling of populations, reporting bias and spatial, temporal and phylogenetic autocorrelation. These are issues that will not be solved by reporting and incorporation of standard errors into meta-analytic models (Koricheva et al., 2013; Murren et al., 2016). With the increased effort of providing raw data (Reichman et al., 2011; Vision & Cranston, 2014; Mills et al., 2015) through data repositories such as Dryad (datadryad.org), we may soon be in the position to use full data sets, rather than summary statistics, to conduct meta-analyses. The availability of such complete data sets may open the door for novel explorations of the major features characterizing evolutionary patterns and processes. Despite these criticisms, we applaud Morrissey's (2016) efforts to describe and advocate the use of formal meta-analytic models in ecology and evolution. But specifying and implementing formal mixed-effects meta-analyses for heterogeneous biological data can be challenging. As a personal example, in the late stages of the publication process of Kingsolver et al. (2012), Jarrod Hadfield recognized several errors in our model implementation and generously gave us advice and suggestions on our R code to correct these; these errors had not been identified by other reviewers of the manuscript, including several with expertise in meta-analysis. (Of course as authors we are solely responsible for any remaining errors of analysis or interpretation in the paper.) We strongly believe that making detailed code associated with meta-analyses available would be most valuable in moving this field forward. Freely available, compiled data sets on phenotypic selection and many other topics in ecology and evolution have been instrumental in motivating new studies and analyses; freely available code and workflows could greatly expand the use of such models and allow the community to explore extensions and alternatives to these models. Of course both code and data can be used inappropriately, but we are convinced that their free availability leads to less confusion and more rapid understanding.

Ecological communities are being reshaped by climatic change. Losses and gains of species will alter community composition and diversity but these effects are likely to vary geographically and may be hard to predict from uncontrolled "natural experiments''. In this study, we used open-top warming chambers to simulate a range of warming scenarios for ground-nesting ant communities at a northern (Harvard Forest, MA) and southern ( Duke Forest, NC) study site in the eastern US. After 2.5 years of experimental warming, we found no significant effects of accumulated growing degree days or soil moisture on ant diversity or community composition at the northern site, but a decrease in asymptotic species richness and changes in community composition at the southern site. However, fewer than 10% of the species at either site responded significantly to the warming treatments. Our results contrast with those of a comparable natural experiment conducted along a nearby elevational gradient, in which species richness and composition responded strongly to changes in temperature and other correlated variables. Together, our findings provide some support for the prediction that warming will have a larger negative effect on ecological communities in warmer locales at lower latitudes and suggest that predicted responses to warming may differ between controlled field experiments and unmanipulated thermal gradients.

Summary Climate change will increase both average temperatures and extreme summer temperatures. Analyses of the fitness consequences of climate change have generally omitted negative fitness and population declines associated with heat stress. Here, we examine how seasonal and interannual temperature variability will impact fitness shifts of ectotherms from the past (1961–1990) to future (2071–2100), by modelling thermal performance curves (TPCs) for insect species across latitudes. In temperate regions, climate change increased the length of the growing season (increasing fitness) and increased the frequency of heat stress (decreasing fitness). Consequently, species at mid‐latitudes (20–40°) showed pronounced but heterogeneous responses to climate change. Fitness decreases for these species were accompanied by greater interannual variation in fitness. An alternative TPC model and a larger data set gave qualitatively similar results. How close maximum summer temperatures are to the critical thermal maximum of a species – the thermal buffer – is a good predictor of the change in mean fitness expected under climate change. Thermal buffers will decrease to near or below zero by 2100 for many tropical and mid‐latitude species. Our forecasts suggest that mid‐latitude species will be particularly susceptible to heat stress associated with climate change due to temperature variation.

The research field of animal and plant symbioses is advancing from studying interactions between two species to whole communities of associates. High-throughput sequencing of microbial communities supports multiplexed sampling for statistically robust tests of hypotheses about symbiotic associations. We focus on ambrosia beetles, the increasingly damaging insects primarily associated with fungal symbionts, which have also been reported to support bacteria. To analyze the diversity, composition, and specificity of the beetles’ prokaryotic associates, we combine global sampling, insect anatomy, 454 sequencing of bacterial rDNA, and multivariate statistics to analyze prokaryotic communities in ambrosia beetle mycangia, organs mostly known for transporting symbiotic fungi. We analyze six beetle species that represent three types of mycangia and include several globally distributed species, some with major economic importance (Dendroctonus frontalis, Xyleborus affinis, Xyleborus bispinatus–ferrugineus, Xyleborus glabratus, Xylosandrus crassiusculus, and Xylosandrus germanus). Ninety-six beetle mycangia yielded 1,546 bacterial phylotypes. Several phylotypes appear to form the core microbiome of the mycangium. Three Mycoplasma (originally thought restricted to vertebrates), two Burkholderiales, and two Pseudomonadales are repeatedly present worldwide in multiple beetle species. However, no bacterial phylotypes were universally present, suggesting that ambrosia beetles are not obligately dependent on bacterial symbionts. The composition of bacterial communities is structured by the host beetle species more than by the locality of origin, which suggests that more bacteria are vertically transmitted than acquired from the environment. The invasive X. glabratus and the globally distributed X. crassiusculus have unique sets of bacteria, different from species native to North America. We conclude that the mycangium hosts in multiple vertically transmitted bacteria such as Mycoplasma, most of which are likely facultative commensals or parasites.