Local adaptation has a role in reducing vulnerability to climate change in a widespread Amazonian forest lizard

Abstract

The extant genetic variation within and among taxa reflects a long history of diversification and adaptive mechanisms in response to climate change and landscape alterations. However, the velocity of current anthropogenic changes poses an imminent threat to global biodiversity. Understanding how species and populations might respond to global climate change provides valuable information for conservation in the face of these impacts. Here, we use genomic data to observe candidate loci under climate selection and test for genetic vulnerability to climate change in a widespread Amazonian ombrophilous lizard population. We found nine populations across Amazonia with a considerable amount of admixture among them. Distinct approaches of genome-environment association analyses revealed 56 candidate single-nucleotide polymorphisms (SNPs) under climatic selection, showing an east–west gradient in the adaptive landscape and a signal of local climate adaptation across the species range. According to our results, signals of local adaptation indicate that the species may not respond equally throughout its range, with some populations facing higher extinction risks. Genomic offset analysis predicts the southern and central portions of Amazonia to have a higher vulnerability to future climate change. Our findings highlight the importance of considering spatially explicit contexts with a large sampling coverage to evaluate how local adaptation and climatic vulnerability affect Amazonian forest ectothermic fauna.

Introduction

The standing genetic diversity of current taxa illustrates a long history of diversification and adaptive processes to climate change and landscape shifts (Riddle 2019), representing the consequences of both historical and contemporary processes shaping biodiversity. Yet, the acceleration of human-induced global climate change (GCC) is now widely recognized as causing an ongoing and forthcoming catastrophe to global biodiversity (Scheffers et al. 2016). Rapid shifts in climate can cause new selective pressures upon the organisms (Hoffmann and Sgro 2011), which they may or may not be able to survive. In fact, local extinctions due to climate and environmental changes have already been documented in several taxa across the globe (e.g., Wiens 2016), particularly those that rely upon local climates to thermoregulate (i.e., ectothermic species). As global temperatures are projected to rise in the coming decades, combined with escalating land-use changes and deforestation (IPCC 2023; Nobre et al. 2016; Longobardi et al. 2016), the occurrence of extinction events may intensify.

Given the above scenario, there are three main ways in which natural populations from a given species can escape from the local extinction caused by GCC: (i) shifting their geographic distribution, tracking suitable habitats according to their physiological demands via dispersal corridors (Lenoir and Svenning 2015; Waldvogel et al. 2020); (ii) through phenotypic plasticity, that is, the phenotypic variation created by one genotype in different environmental conditions, as long as the environmental changes do not exceed the species’ physiological limits (Riddell et al. 2018; Waldvogel et al. 2020); or (iii) through adaptive evolution, driven by natural selection (Razgour et al. 2019; Waldvogel et al. 2020). Adaptive evolution is especially important for species distributed across strong environmental gradients and varying ecological pressures, as environmental changes can lead to local adaptation within populations (Hoffmann and Sgro 2011; Waldvogel et al. 2020). However, predicting how populations can or cannot adapt to climate change is still a huge scientific endeavor (Urban et al. 2024). Although such different responses can buy time and eventually avoid local extinctions (Fox et al. 2019), species and populations, particularly lizards, might not have enough time and genetic variation to achieve either possibility (Diniz-Filho and Bini 2019). Given such predictions, environmental changes would require rapid adaptive responses in the coming decades. Consequently, populations without the ability to cope with environmental change might face severe decline or even local extinction (Sinervo et al. 2010; Hoffmann and Sgro 2011; Waldvogel et al. 2020). Thus, understanding past evolutionary processes can help us predict future adaptive responses (Sosa and Pilot 2023).

Usually, forecasts of climate vulnerability are inferred using niche modeling (Helmuth et al. 2014; Pacifici et al. 2015; Foden et al. 2019). With analytical advances, genomic data have become crucial to infer local adaptation and climate vulnerability (Kardos et al. 2021; Forester et al. 2022) and therefore better inform GCC forecasts of non-model organisms (e.g., Bay et al. 2018; Razgour et al. 2019). Genome–environment association (GEA) analyses have become an important tool in local adaptation studies in populations occupying heterogeneous habitats (e.g., Rellstab et al. 2015; Gugger et al. 2021). GEA methods use associations between genetic and environmental data to identify potential adaptive variation among populations (Forester et al. 2016), thereby highlighting populations potentially capable of avoiding local extinction due to GCC and key geographic regions for conservation planning (Razgour et al. 2019). For example, local adaptation, which can be inferred by the match between adaptive genetic and environmental variation, has an important role in conferring higher survival in local environments and is key to preparing populations for challenging climates (Meek et al. 2023). Altogether, understanding how species and populations might be affected or respond to global climate change can provide valuable information for conservation strategies in the face of impending impacts (Urban 2015; Meek et al. 2023). For instance, genomic offset, defined as the amount of mismatch (dissimilarity) between current and future genotype–environmental relationships (Bay et al. 2018), can help identify populations that may experience loss of population fitness due to environmental or human pressures. Genomic offset reflects a potential shift in the adaptive optimum induced by climate change (Capblancq et al. 2020; Capblancq and Forester 2021).

Across the tropics, biodiversity crises due to climate change can be even more severe, since this region harbors most of the global biodiversity (e.g., Deutsch et al. 2008; Huey et al. 2009; Wiens 2016). In Amazonia, both climate change and associated events, such as deforestation, habitat loss, and wildfires, are the greatest threats to the persistence of local species (Malhi et al. 2008; Pereira et al. 2019), with studies predicting a decrease in precipitation levels and changes in rainfall patterns and river dynamics in the region in the coming decades (Nobre et al. 2016; Barichivich et al. 2018; Alves de Oliveira et al. 2021). Ectothermic organisms across the tropics often figure as having high vulnerability and extinction risks, since small increases in mean temperature may lead to a reduction in their daily activity and subsequent extinction of their populations (Sinervo et al. 2010; Urban 2015; Pontes-da-Silva et al. 2018). In this context, our focal species is the widely distributed Amazonian gecko Gonatodes humeralis (Sphaerodactylidae). The species inhabits forested environments across an extensive climatic gradient (Ribeiro-Júnior, 2015). Further, G. humeralis is a thermoconformer forest lizard, meaning that individuals do not actively thermoregulate and, as such, the species is part of a functional group predicted to be under high vulnerability due to GCC (Huey et al. 2009; Sinervo et al. 2010). Despite finding signals of genetic structure across Amazonia for the species (Pinto et al. 2019; Pirani et al. 2019), former studies did not take into consideration adaptive processes at the populational level, nor integrated those into future forecasts.

Herein we employ an integrative eco-evolutionary approach using the genomic data from natural populations of G. humeralis to evaluate the role of potentially adaptive genomic diversity in driving local climate adaptation and shaping extinction risks under climate change. We have addressed two main questions: 1) How is the genetic variation (both neutral and adaptive) distributed and arranged across the distribution range of Gonatodes humeralis? 2) Will climate change lead to a mismatch between current and future genome–environmental relationships in G. humeralis, impacting its vulnerability? We expect to find a more refined population structure (i.e., a higher number of populations) than previous studies, since we employ a broader geographic and genomic sampling. Also, we expect to encounter loci associated with environmental variation in the adaptive landscape for G. humeralis, following a climatic gradient across Amazonia, since previous research has demonstrated this gradient’s influence on the diversification and climate vulnerability of other taxa in the region (e.g. Silva et al. 2019; Azevedo et al. 2024; Dalapicolla et al. 2024). We predict more admixture in the eastern portion of Amazonia due to the more climatically unstable nature of this region (Ávila-Pires et al. 2012; Pinto et al. 2019), but GCC impacts may be more severe in the western portion of Amazonia since it has a more stable climate (Cheng et al. 2013). By using a genome–environmental association approach coupled with the inference of genomic offset under future scenarios of climate change, we expect to map and accomplish an accurate assessment of climatic vulnerability for G. humeralis across its distribution range.

Methods

Geographic and genomic sampling

We sequenced genomic data from 194 specimens of Gonatodes humeralis sampled from several localities across Amazonia (Fig. S1) in order to obtain the greatest possible geographic coverage. The complete list of samples and scientific collection vouchers is provided in Supplementary Table S1. For each individual, genomic DNA was extracted from muscle or liver fragments previously preserved in 100% ethanol, using the Qiagen DNeasy Blood and Tissue Kit. Two reduced representation libraries were constructed with the samples through the ddRADSeq (double digest restriction-associated DNA sequencing) approach, following the protocol established by Peterson et al. (2012). In summary, double-digested DNA using the restriction enzymes EcoR1 and MseI, and with ligated unique barcodes for each individual, was pooled and 350–450 bp fragment sizes were selected using Pippin Prep (Sage Science). Libraries were sequenced in the Illumina 2500 platform at the Center for Applied Genomics (Toronto, Canada) to generate 150-bp single-end reads. For a complete description of the genomic library preparation and DNA sequencing process, see Pirani et al. (2019).

In order to de-multiplex and quality-filter the genomic data, we used Ipyrad v.0.9.68 pipeline (Eaton and Overcast 2020), performing a de novo genome assembly and calling for SNPs. The main filtering steps in Ipyrad were adjusted for the following parameters: 5% of maximum low-quality bases in a read, minimum read length of 100 bp, minimum samples per locus for output of 108, representing 66% or two-thirds of all individuals, and maximum number of indels per locus of 5. For the remaining parameters we used the program defaults. We removed from the dataset 22 individuals with less than 600,000 reads and another three that did not pass the filtering steps, leaving behind 169 individuals. Before all downstream analyses (Sparse Nonnegative Matrix Factorization (sNMF), GEA, and genomic offset), we used VCFtools v. 0.1.13 (Danecek et al. 2011) to filter out singletons and include SNPs present in at least 95% of individuals by minor allele frequency (MAF = 0.05), to avoid count sequencing errors (Ahrens et al. 2018), and to select one SNP per read to reduce problems of linkage disequilibrium. We also filtered our dataset to remove four samples with more than 20% of missing data per individual and per locus. After all of the aforementioned steps, our final dataset was composed of 165 individuals of G. humeralis, with 36,915 unlinked SNPs and 17.25% missing data. This is the dataset used in all downstream analyses (see Supplementary Table S2 for details).

Population genetic structure

For the calculation of genome–environmental association (see below), it is first necessary to investigate the neutral genetic structure, as this is needed to disentangle the effects of neutral and adaptive genetic variation. So, to evaluate the underlying neutral population structure of Gonatodes humeralis, we estimated the number of genetic clusters (k) along with their distributions using the sNMF v. 1.2 (Frichot et al. 2014), which defines the number of clusters and admixture coefficients based on SNP data. We tested k = 1–15, with 100 replicates (α = 100) for each k; the best k was the one with the lowest cross-entropy value inferred by the sNMF (Frichot et al. 2014). The robustness of the results was verified by testing four different values of the alpha regularization parameter (10, 100, 1000, and 10,000) as recommended by the program. The sNMF analysis was performed in R v. 4.1.3 (R Core Team 2022) with the package “LEA” v. 2 (Frichot and François 2015). We acknowledge that for broadly distributed and structured populations, a larger number of samples per locality can enhance the statistical power of the analysis and reduce false discovery rates (FDR) in case of landscape genomic studies (Selmoni et al. 2020). However, given the broad geographic scope of our study and the challenges associated with obtaining samples across the entire Amazonian region (Nazareno et al. 2017), we consider our sampling coverage quite representative; we also employed alternative strategies to control the FDR (see below). In this context, our sampling design is suitable for landscape genomic studies involving non-model species in natural environments (Rellstab et al. 2015; Balkenhol et al. 2017; Selmoni et al. 2020).

Genome–environmental association analyses

To identify possible loci under climatic selection (candidate SNPs), we performed genome scans using GEA approaches. These analyses test correlations between genomic (allele frequencies in multiple loci) and environmental variables along the distribution of our samples. As environmental predictors in the GEA analyses, we used bioclimatic variables from the Climond database (available at www.climond.org – Kriticos et al. 2012), with a grid resolution of 0.167 × 0.167, which corresponds to the higher resolution of the database. We first chose the variables based on their importance for our biological system—Gonatodes humeralis—and the expectations concerning environmental pressures upon natural populations (e.g., minimum, maximum, and seasonality of temperatures and precipitation—Razgour et al. 2019; Román-Palacios and Wiens 2020) and geographic variation across Amazonia. Afterwards, we tested the variance inflation factor to estimate how much of the variance was inflated due to multicollinearity between the selected environmental variables (Dormann et al. 2013). We removed two variables that presented multicollinearity. Subsequently, we performed a model selection based on a redundancy analysis (RDA) (Capblancq and Forester 2021). We retained the following six bioclimatic variables: temperature seasonality (Bio_4), mean daily maximum air temperature of the warmest month (Bio_5), mean daily minimum air temperature of the coldest month (Bio_6), precipitation amount of the driest month (Bio_14), precipitation seasonality (Bio_15), and mean monthly precipitation amount of the warmest quarter (Bio_18). We then extracted the values of each bioclimatic variable from the collection sites of our samples used herein.

We ran two different GEA analyses to infer associations between environmental predictors and allele frequencies in our dataset. We chose to run two distinct complementary methods to minimize false positives and false negatives in our genome scan, as well as to achieve a more powerful selection of candidate SNPs (Forester et al. 2018; Capblancq and Forester 2021; Azevedo et al. 2024). First, we ran a RDA (Lasky et al. 2012; Capblancq and Forester 2021), which investigates how groups of loci covary in response to the multivariate environment, while accounting for population structure by incorporating it as a covariate (e.g., Razgour et al. 2019; Capblancq et al. 2018; Varas-Myrik et al. 2022). RDA can detect processes resulting from weak signs of selection (Forester et al. 2016 and 2018) and is a reasonable tool to evaluate GEAs (Capblancq and Forester 2021). We performed a partial redundancy analysis (pRDA) of the SNPs in relation to the environmental variables, with a PCA of all SNPs as conditioning variables to disentangle the effects of neutral and adaptive genetic variation. We used a pRDA-based variance partitioning in order to decompose the contributions of climate, neutral population structure, and geography in explaining genetic variation (allele frequencies). We used the functions rda() and rdadapt() (Capblancq et al. 2018) to perform RDA genome scans, and to extract the Q-values and select SNPs with false discovery rate <0.05, respectively. The Q-value is an adjusted measure of statistical significance derived from P-values to control the FDR, which is particularly useful in genome scans involving multiple tests, as it accounts for the proportion of false positives among the identified significant associations (Storey and Tibshirani 2003; François et al. 2016).

Second, we ran a Latent Factor Mixed Model (LFMM—Frichot et al. 2013), which investigates the correlations between environmental and genomic variation across the landscape, while levels of neutral population structure, from the previous sNMF analysis, are considered to define the best number of latent factors in the analysis. LFMM computes z-scores and p-values to quantify the strength of associations, which are also informative when compared among environmental factors, while accounting for the low rates of false positives and false negatives (Rellstab et al. 2015). Further, LFMM is good at detecting weak selection and works very well with complex hierarchical neutral genetic structures and polygenic selection (Frichot et al. 2013; de Villemereuil et al. 2014). It also has been extensively used in recent empirical studies (e.g., Prates et al. 2018; Ruegg et al. 2018; Razgour et al. 2019; Gugger et al. 2021). After running, we checked the LFMM histograms and corrected the p-values using the genomic inflation factor (GIF), extracted the Q-values, and identified the outlier loci using a threshold of FDR < 0.05.

Last, we looked for any overlap between SNPs detected by both RDA and LFMM methods and combined them into a single dataset of candidate SNPs. For the downstream analysis, we adopted a conservative approach, where we used only SNPs detected by both methods (Forester et al. 2018). Although we acknowledge that this approach reduces the overall number of candidate loci (Ahrens et al. 2018), several studies have shown that the differences in results are generally not substantial, and that using overlapping SNPs can provide greater confidence by mitigating potential false positives unique to one method (Forester et al. 2018; Razgour et al. 2019; Azevedo et al. 2024).

Adaptive landscape

Once we found the set of candidate adaptive loci, we calculated the adaptive genetic similarity across the landscape using the adaptive index (Capblancq and Forester 2021). To do so, we first ran a new RDA with only the candidate SNPs (outlier loci) and used the scores of the RDA axes to predict the adaptive index along the landscape for each environmental pixel, using the following formula:

$$\mathrm{AdaptiveIndex}\sum _{i=1}^n{a}_i{b}_i,$$

where a is the climatic variable score (RDA loading), b is the standardized value for this particular variable at the focal pixel, and i refers to one of the n different variables used in our RDA model (Capblancq and Forester 2021). The adaptive index provides an estimate of genetic dissimilarity in all pixels on the landscape as a function of the environmental predictors’ values at that exact site (Capblancq and Forester 2021). We limited our predictive landscape geographically using a minimum convex polygon, based on our sampling points, plus a buffer of three degrees, thus mapping the adaptive index across the species’ geographic range.

Genomic offset and climate vulnerability

We applied the genomic offset to investigate which populations of G. humeralis are under higher vulnerability to future GCC, using spatial adaptive gradients (GEA analysis) and their temporal shifts in response to future environmental change. We assumed that populations with a greater genomic offset index would be the most vulnerable because they would be under strong selective pressure and, possibly, would not be able to perform shifts in their adaptive optimum by changing their genetic composition rapidly enough to keep up with the environmental change in the future, possibly driving maladaptation and reducing the population fitness (Fitzpatrick and Keller 2015; Bay et al. 2018; Capblancq and Forester 2021; Gain et al. 2023). We measured the genomic offset through Euclidean distances between the current genetic dissimilarity, estimated from the Adaptive Index, and the ones predicted in the future, for each pixel. In a nutshell, higher values of genomic offset indicate higher local vulnerability to environmental change, while lower values indicate lower risk (Capblancq and Forester 2021).

For the future forecasts needed to measure the genomic offset, we used the A1B and A2 climate scenarios, which differ in their greenhouse gas emissions and socioeconomic assumptions (IPCC 2023). The A1B scenario represents a balanced approach, estimating a temperature increase of 1.7–4.4 °C by the end of the 21st century, with greenhouse gas emissions peaking in the mid-21st century. The A2 scenario portrays a future with higher greenhouse gas emissions and fewer efforts to mitigate them, forecasting a 2.0–5.5 °C temperature increase. Although not directly equivalent, the A1B scenario aligns with the moderate-emission RCP4.5 scenario, while the A2 scenario is more comparable to the high-emission RCP8.5 scenario. These scenarios provide contrasting outlooks for future climate conditions and allow for a comprehensive assessment of the potential impacts on species’ adaptation and vulnerability. We extrapolated the current adaptive index (previous section) to future climatic scenarios in order to predict a potential shift induced by climate change (genomic offset). After calculating the genomic offset for 2050 and 2100, we mapped them in the geographic space across the species range. We did not include a milder climatic scenario because of its virtual impossibility, as it would require that we had broken off greenhouse gas emissions a couple of years ago (Raftery et al. 2017).

Results

Population genetic structure

The sNMF analysis recovered, as the best-fit model, nine geographically structured populations (K = 9) across the distribution range of Gonatodes humeralis (Fig. 1). The analysis also showed high levels of admixture among several of these populations (Fig. 1), mostly along the northeastern ones. However, there is low or non-admixture in the western and southwestern Amazonian populations. Likewise, there is one isolated population (single localities) in northern Amazonia (pink cluster in Fig. 1B), and another two isolated nearby the Amazonia-Cerrado ecotone.

Fig. 1. Population structure and assignment of Gonatodes humeralis across Amazonia and transitional biomes.

A Barplot from sNMF analysis, showing high admixture and nine population clusters. B Pie chart with admixture proportions among clusters (K = 9), where multiple slices of the same color represent different individuals from the same locality and each pie chart corresponds to one locality. Maps are displayed on an equal-area Behrman projection. Photo: Rodrigo Tinoco.

Genomic environmental association

The result of the RDA variable importance indicated that the six bioclimatic variables retained in the analyses were important in explaining the genetic variation structure in G. humeralis (Table 1). In general, three precipitation and three temperature variables were significantly correlated with genetic variation in Gonatodes humeralis across its distribution range. However, the most important variable was the mean monthly precipitation amount of the warmest quarter (Bio 18 – F = 18.55), followed by the mean daily minimum air temperature of the coldest month (Bio 06 – F = 4.3). Further, the pRDA results showed that climate, geographic distance, and neutral population structure explained 4, 1, and 10% of the genomic variation in our data, respectively.

Table 1.

Selection results of RDA climatic variables.

	R2.adj	Df	AIC	F-value	p-value
bio_18	0.10	1.00	1506.74	18.55	0.002
bio_6	0.11	1.00	1504.42	4.30	0.002
bio_5	0.13	1.00	1503.25	3.12	0.002
bio_4	0.14	1.00	1502.27	2.91	0.002
bio_15	0.15	1.00	1501.56	2.64	0.002
bio_14	0.15	1.00	1500.83	2.64	0.002
All variables	0.15	-	-	-	-

Bioclimatic abbreviations: bio_18 mean monthly precipitation amount of the warmest quarter, bio_6 mean daily minimum air temperature of the coldest month, bio_05 mean daily maximum air temperature of the warmest month, bio_04 temperature seasonality, bio_15 precipitation seasonality, bio_14 precipitation amount of the driest month. Df degrees of freedom, AIC Akaike Information Criteria.

RDA-only results (GEA) recovered 451 candidate SNPs under environmental selection, while LFMM-only results recovered 492 candidate SNPs. The overlap between RDA and LFMM results identified 56 candidate SNPs under environmental selection associated with bioclimatic gradients. The new RDA that was run only with the overlapped candidate SNPs revealed putative signals of local adaptation across the Amazonian climatic gradient for G. humeralis, as well as how variables correlate among themselves (Fig. S2 – R2 = 0.28). The first two axes accounted for 80% of the explained variation. Most of the variation is accumulated in the first axes (RDA1 = 54% of variance), where allele frequencies were most associated with low precipitation in the warmest months (Bio18 and negative RDA1 scores), or with elevated precipitation seasonality (Bio15 and positive RDA1 scores). Positive scores in the second axes (RDA2 = 26% of variance) were related to temperature seasonality across the landscape.

Adaptive landscape

With the adaptively enriched genetic space—RDA with candidate SNPs only (Fig. S2), we were able to extrapolate the RDA axes scores across the landscape in the entire range of G. humeralis (Fig. 2A) using the Adaptive Index. The adaptive landscape shows how the RDA1 predicted scores, which account for most of the accumulated variation (54% of variance), contrast in an east–west gradient from the positive RDA scores in eastern Amazonia to the negative RDA1 scores in western Amazonia. There is an increase in the adaptive–genetic dissimilarity from the western to eastern Amazonia, highlighting distinct adaptive landscapes across climate gradients in Amazonia. Likewise, the adaptive gradient shown by the RDA2 index (26% of variance) differentiated the western Amazonian areas, characterized by more precipitation and annual climatic stability, from the southern and eastern portions of this domain, generally characterized by a drier climate. Together, both RDA axes reveal adaptive dissimilarities across the range of Gonatodes humeralis in Amazonia related to the bioclimatic variables of precipitation and temperature (Fig. 2A). Most of the explanation in the first axes relies on variables of precipitation while in the second axes it relies on variables of temperature.

Fig. 2. Adaptive landscape and genomic offset for Gonatodes humeralis in Amazonia.

A Adaptive landscape inferred by the predicted genetic similarity derived from an extrapolation of RDA axes 1 and 2, with the adaptive index scores as a spatial projection across the range of Gonatodes humeralis. Both RDA windows represent dissimilarities regarding the adaptive optimum in G. humeralis mostly related to precipitation (Bio_18—precipitation amount in the summer) for RDA1 and temperature (Bio_5—mean daily maximum air temperature in the warmest month and Bio_6—mean daily minimum air temperature in the coldest month) for RDA2. Similar colors across the landscape represent regions with populations adapted to similar conditions. Predicted genomic offset for G. humeralis across its distribution range in Amazonia using the projections for the years 2050 and 2100 under (B) an intermediate emission scenario (A1B) and (C) an extreme emission scenario (A2). Dark red colors represent more vulnerability.

Genomic offset

The genomic offset, or the estimated mismatch between current and predicted genome–environment relationships derived from the adaptive index, showed the vulnerability under climate change across the G. humeralis species range. Accordingly, our model predicts disparate patterns, both for 2050 and 2100, with low to moderate values of genomic offset in the southern and eastern portions of Amazonia and elevated values, so higher vulnerability, in the central and western portions of Amazonia (Fig. 2B, C). We can certainly observe an increase in vulnerability for 2100 compared to 2050, with several areas presenting the highest levels of genomic offset in central and western Amazonia.

For the extreme scenario (Fig. 2C—2100 maps), our model predicts an overall higher level of genomic offset, as expected, increasing the vulnerability across Amazonia, particularly in the aforementioned areas. Conversely, southern and eastern populations seem to be at lower risk in the future since they present lower levels of genomic offset (light colors on the map).

Discussion

Our findings elucidate the genomic–environment associations for a widespread Amazonian ombrophilous gecko, providing a risk assessment in the face of future climate change as well as an estimate of the current adaptive landscape across the most biodiverse and complex tropical rainforest. The genome–environment analyses revealed a set of putative SNPs under climate selection, using six selected bioclimatic variables as predictors. To account for the underlying neutral population structure and also better understand the population history of G. humeralis, we identified nine genetically distinct populations across different regions of Amazonia. Eastern Amazonia populations harbor distinct levels of genetic admixture, while populations located in western and southwestern Amazonia show a clearer pattern of isolation. Although some population genomics studies have recently assessed the extent of local adaptation in the Neotropics, most of them have addressed tree species as the main system (e.g., Collevatti et al. 2019; Leal et al. 2021; Vieira et al. 2022). This is the first time that the role of local adaptation is accounted for in forecasts of genomic offset under changing climates in ectothermic Amazonian organisms and particularly thermoconformer lizard species, supposedly under higher vulnerability to climate change. Accordingly, our results on local adaptation show a distinctive pattern across Amazonia, highlighting how different climatic gradients have shaped the adaptive landscape of G. humeralis across its range. Likewise, the predictive genomic offset showed that, in both 2050 and 2100 scenarios, populations from central and western Amazonia would be under a higher risk of extinction events due to future climate change. The distinctive pattern we found of genomic offset shows how different populations within the same species might be more or less vulnerable to future environmental changes. Incorporating this information into risk assessments can enhance the prospects of species and population survival.

Population structure

The east–west population structure pattern we observed for Gonatodes humeralis resembles the findings of previous studies with different molecular and geographic sampling designs for the species (e.g., Ávila-Pires et al. 2012; Pinto et al. 2019; Pirani et al. 2019). However, the nine populations we recovered depict a much more detailed degree of population differentiation and structure than former studies (Pinto et al. 2019), which used a sparse sampling across the species range and recovered only two populations within the Amazon Basin (one in ‘eastern Amazonia’ and one in ‘western Amazonia’). We believe that differences in genetic markers used (sanger vs ddRADSeq) and the reduced geographic sampling coverage across Amazonia in previous studies may account for such differences. Our denser sampling, in terms of both individuals and localities, may have resulted in more geographically structured populations, avoiding the so-called K = 2 conundrum (Janes et al. 2017), and is more in line with what would be expected from the species’ biological traits of low vagility and poor dispersal (Vitt et al. 1997; Ávila-Pires et al. 2012). The general northeast to southwest pattern with increased structure across the species’ range and decreasing levels of admixture in the same direction (Fig. 1B) can be attributed to the stability dynamics of Amazonia, with the ecoregion historically being more stable in the southwest and more unstable in the northeast (Cheng et al. 2013; Silva et al. 2019; Bonvicino et al. 2022; Miranda et al. 2022).

We found a strong genetic differentiation between populations geographically close to each other (Fig. 1B), which is consistent with the species’ low vagility (Ávila-Pires et al. 2012), but also a weak signal of isolation. For example, populations at the extreme northeastern side of the species distribution in the lower Tocantins River (blue and purple, Fig. 1B) are close to each other but were assigned to very distinct clusters. Furthermore, there are isolated single populations with little or no admixture at all (pink and green in Fig. 1B). Population structure may change over time, with multiple historical events leaving signals currently detectable in the data (Lawson et al. 2018).

Explaining admixture is not a trivial task. Several historical demographic events, even discordant ones, may cause current-day admixture patterns and many of those might have occurred in Amazonia. Different regions went through distinct historical processes and generated a landscape complexity that ultimately shaped the current Amazonian biodiversity (Hoorn et al. 2022; Cracraft et al. 2020). For example, historical expansion and retraction events may have favored the presence of strong admixture signals across eastern Amazonia, a region closer to the transition zone with Cerrado and Caatinga ecoregions and historically more unstable regarding the climate (Cheng et al. 2013; Silva et al. 2019). Likewise, the role played by rivers as primary or secondary barriers to gene flow might also have influenced the pattern of admixture we see in our results (e.g., Alfaro et al. 2015; Godinho and da Silva, 2018; Pirani et al. 2019). The geographic contact in headwater regions might have prevented isolation between populations, since the rivers may no longer act as effective barriers to gene flow (Weir et al. 2015), favoring admixture in those regions. Further, the broad admixture zones in Amazonia could have resulted from the secondary contact following lineage divergence (Barrera-Guzmán et al. 2022), where the disappearance or attenuation of physical barriers that hinder gene flow can promote admixture between genetically distinct populations (Musher et al. 2022; Weir et al. 2024).

Local adaptation

Our results showed that genetic variation was significantly correlated with the six selected environmental variables (Table 1). Regarding candidate SNPs, the potential adaptive genetic variation was strongly correlated with precipitation amount in the warmest season (Bio_18, Fig. S2). Moreover, we can assume that precipitation and temperature are important and relevant features to explain the genetic variation (neutral and adaptive) and probably resulted in patterns of local adaptation across the range of G. humeralis. The 56 putative SNPs under climatic selection that we identified through both GEA approaches reveal the likely existence of local adaptation in G. humeralis across Amazonian landscapes or a spatial variation in fitness. The historical establishment of environmental gradients within Amazonia and at its transition to other biomes (e.g., the Amazonia-Cerrado ecotone) can account for the local adaptation and adaptive landscape patterns that we found (Azevedo et al. 2024). Further, the more recent environmental changes caused by human action could act now as synergetic climatic selective pressures upon the organisms. For example, southern Amazonia presents a relatively long dry-season length (DSL), a pattern that has a trend to increase further in the late 21st century (Fu et al. 2013). When combined with human-induced fires and land use (Salazar et al. 2007; De Faria et al. 2017), this increasing DSL can have an influence on the adaptive index of G. humeralis across that region. Besides, changes in the Amazonian vegetation during the Pleistocene, especially in the southern region (Mayle et al. 2000; Sato and Cowling 2017), may play a role in explaining the adaptive landscape pattern found from the south to north. This occurred when forest retractions toward the equator took place due to climatic fluctuations (Ayres and Clutton-Brock 1992; Haffer and Prance, 2001; Haffer, 2008), which is also supported by the diversification timing found by Pinto et al. (2019) for the same focal species.

The adaptive-enriched RDA captured 80% of the variation across two axes, a reasonably high value, which indicates that the predictive environmental variables selected are effective in explaining the adaptive genetic variation in our data. Precipitation seasonality appears to be a strong predictor of the allele frequencies of the candidate SNPs across the gradient of this variable (Figs. S2 and 2A). This pattern can be carefully interpreted as a sign of local adaptation, where different alleles may predominate at opposite ends of the seasonality gradient, suggesting an adaptive response to regions with more constant rainfall versus those with marked seasonality. These are the characteristics of western Amazonia, which was precisely the region with less admixture in clustering analysis (Fig. 1B), highlighting low levels of gene flow between these populations and their eastern counterparts, and possibly explaining the local adaptation results. In conclusion, we see that accounting for both neutral and potential adaptive genetic diversity can improve our interpretation of the underlying evolutionary processes on a geographic scale across Amazonia.

Adaptive landscape and future genomic offset

As we mapped the adaptive index across the landscape, we were able to assess the scale and pattern of local adaptation of G. humeralis in Amazonia (Fig. 2A). The spatial projection of the adaptive variation provides a better understanding of the potential climate adaptation to unsampled areas within the species distribution (Gugger et al. 2021), revealing regions where patterns of local adaptation differ from the background genetic variation. Our results show a consistent gradient pattern in the adaptive optimum from areas close to the Cerrado-Amazonia transition (eastern to southern) to central and western Amazonia, with remarkable differences in the adaptive index (positive to negative scores). Interestingly, our results present an east–west difference in the adaptive index for Amazonia, which is a common direction of biodiversity patterns’ turnover in this biome (e.g., Pinto et al. 2019; Dalapicolla et al. 2024).

The adaptive landscape of G. humeralis provides valuable information to include in conservation action plans and management strategies for the survival of species (and potentially other taxa from similar functional groups—i.e., thermoconformer and ombrophilous tropical ectotherms) facing climate change (Aitken and Whitlock 2013; Fitzpatrick and Keller 2015; Gugger et al. 2021). According to future scenarios, not only will the temperature rise but also rainfall patterns are expected to change (Dai 2013; Marengo et al. 2011; IPCC 2023). As we have shown, precipitation and temperature variables were significantly important in explaining the possible patterns of local adaptation in G. humeralis across Amazonia. Consequently, the aforementioned changes will likely have a negative effect on populations that are currently adapted to a wetter climate, imposing new selective pressures upon individuals that may not be able to cope with under future scenarios. One way to avoid local climate-driven extinctions and preserve both neutral genetic variation and genomic regions putatively under selection by enhancing population resilience would be the use of adaptive index across the landscape to propose connectivity via assisted gene flow between populations (Hoffmann et al. 2021; Chen et al. 2022; Meek et al. 2023; Fitzpatrick and Funk 2019; Hohenlohe et al. 2020). Alternatively, the establishment of ecological forested corridors could be considered, taking into account distributional areas that facilitate the species’ natural dispersal (Littlefield et al. 2019).

According to our results, and accounting for the existence of an east–west gradient in the adaptive landscape index, it is reasonable to suggest that some G. humeralis populations are less vulnerable to climate change impacts, such as those in eastern Amazonia. Nonetheless, if the identified candidate SNPs were indeed associated with the climatic gradient, then we could expect that the populations in western Amazonia could be more affected by climate change and may not be able to survive like their eastern counterparts, unless such adapted SNPs come into the western population allele pool through evolutionary rescue (Azevedo et al. 2024) or novel adaptive variations arise through de novo mutation or migration (Bay et al. 2018; Razgour et al. 2019). The main reason for this is the significant disparity in adaptive optimum in the west (Fig. 2A), coupled with the potential impact of future changes in temperature and precipitation patterns. Moreover, as expected, we see an increase in vulnerability for 2100, when carbon emissions are not likely to be reduced, with more areas presenting prominent genomic offset levels.

Populations of Gonatodes humeralis in central and southwestern Amazonia are potentially under greater threat, considering the higher values of genomic offset for those regions. Although the populations in these regions might be potentially adapted to the current local conditions (Fig. 2A), the predicted environmental change across the landscape will possibly disrupt the genome–environment relationships in the future. This disruption would require more genomic change than the species could afford, which might possibly lead to local extinctions in case our results really correspond to the actual adaptive variation. Western Amazonia, as a whole, has historically been a climatically complex region. Consequently, unlike southwestern populations, those nested in northwestern Amazonia may be at a lower risk, considering the levels of genomic offset observed in this region (yellowish color in Fig. 2B, C). One explanation would be that this region will experience less change in its environmental conditions in the future, reducing the need for genomic change and thus buffering the struggle of populations to persist there. We are aware of the limitations of our findings, primarily stemming from the need to validate the results regarding adaptive variation in G. humeralis. This validation could be achieved through genome annotation, quantification of individual fitness, or common garden experiments. However, such validation was not the focus of our study, and we used the available information for our non-model system. We acknowledge these recommendations for future investigations.

Our results predict a high value of genomic offset coinciding with the distribution range of another forest gecko, Gonatodes tapajonicus, which is currently the only threatened lizard species in Amazonia (Vulnerable in the IUCN Red List and Endangered in the Brazilian Red List of Threatened Species) (Silveira et al. 2021 and ICMBio 2018, respectively). Gonatodes tapajonicus is known only from some records along the east bank of Rio Tapajós basin and faces several human-induced threats (Rodrigues 1980; Sturaro and Avila-Pires 2011; Silveira et al. 2021). Our study with G. humeralis provides a baseline for future studies regarding climate change vulnerability and adaptive potential with G. tapajonicus, which has a more restricted distribution and higher perceived conservation threat and vulnerability.

The genetic parameters estimated across populations and species (e.g., neutral and adaptive genetic diversity, gene flow, genomic offset, heterozygosity, inbreeding, and hybridization) may help build important strategies for conservation, providing valuable sources of information for biodiversity maintenance (Nielsen et al. 2022). Considering the threats to biodiversity in the Anthropocene, forest lizards (particularly thermoconformers, ombrophilous) stand out as the most endangered organisms (Huey et al. 2009; Sinervo et al. 2010; Urban 2015; Teixeira et al. 2022). Therefore, studies addressing the processes involved in population structuring and local adaptation at multiple scales become important tools for improving risk assessments and strategies for conservation of these organisms under the current global changing scenarios (Urban et al. 2016; Forester et al. 2022; Nielsen et al. 2022).

Conclusions and perspectives

This is the first population genomics investigation of a Neotropical ombrophilous lizard species that uses a multidisciplinary approach to understand how local adaptation can affect climate change vulnerability on a large Amazonian spatial scale. Our findings highlight the importance of considering spatially explicit contexts with large sampling coverage to evaluate the role played by environmental gradients in potentially selecting patterns of local climate adaptation. Besides, the focus on a fairly wide-range species, in opposition to extremely narrow ones, is of biological interest given that processes that operate at small populations, such as genetic drift and mutation load, may prevent local adaptation from buffering the effects of climate change (Meek et al. 2023). Genomic climate vulnerability is an important component of risk assessments as it can help construct new opportunities for investigation and push forward the frontier in eco-evolutionary research and conservation genetics (Forester et al. 2022). Populations showing true signs of local adaptation may deal better with environmental change when facing future climate change through natural selection and genetic rescue (Razgour et al. 2019; Capblancq et al. 2020). However, despite being an experimental field of research in evolutionary ecology and population genomics, the detection of local adaptation is presently a trending topic that is likely to undergo further methodological and interpretational advancements in the coming years (Csilléry et al. 2018; Capblancq and Forester 2021; Urban et al. 2024). Future investigation should focus on providing functional validation for our results, and on evaluating the potential of using eco-evolutionary data in creation of new protected areas and how much of the adaptive potential is covered by the extant protected areas and indigenous lands across the biome. The potential of using lizards as a model for eco-evolutionary questions is supported by our study, and Amazonia harbors hundreds of them, living in the most different sorts of environments throughout the biome.

Author contributions

AY and FPW designed the research. RMP generated the genomic samples and completed the library preparation. AY led the data analysis with assistance and supervision from JARA. AY wrote the original draft. RMP, JARA, and FPW reviewed and edited the manuscript. All authors discussed the results and implications. FPW raised the funding for this research.

Data availability

The raw data associated with this study are already published within another paper (10.1111/jbi.13676) and deposited in the Dryad Digital Repository (10.5061/dryad.ck223mq). Ipyrad output files and input files that were used in downstream analysis, which support the findings of this study, have also been deposited in the Dryad Digital Repository (10.5061/dryad.cvdncjtfn). For inquiries regarding data access, please contact [andreyves7@gmail.com].

Competing interests

The authors declare no competing interests.

Research ethics statement

This study did not require any ethical approval or collection permits, as it was based exclusively on a previously published dataset.

Supplementary information

The online version contains supplementary material available at 10.1038/s41437-025-00765-x.