Soil metabarcoding reveals the distinct microbial signature of a remnant forest patch at a local restoration initiative in the Ecuadorian Chocó

The Tumbes-Chocó-Magdalena hotspot is a globally recognized center of endemism that is undergoing severe deforestation. While conservation and restoration efforts typically prioritize the recovery of aboveground macroscopic biodiversity, the response of soil microbial communities remains largely uncharacterized in this region. To address this data gap, we assessed soil bacterial communities in three land-use types (reference forest, early-stage restored parcel, and degraded land) within the "La Esperanza Forest" reserve in the Ecuadorian Chocó. Using 16S rRNA gene metabarcoding and robust compositional data analysis (e.g., ALDEx2, PERMANOVA), we evaluated shifts in taxonomic diversity and community structure. Sequencing revealed 1,208 genera and dominant genera (e.g., Bacillus, Paenibacillus) that persisted across all habitats. However, despite the shared presence of these dominant taxa, the sampled land-use plots exhibited distinct structural differences in their microbial communities. Reference remnant forest's soils exhibited significantly higher community evenness and distinct multivariate partitioning driven by strongly associated structural taxa (e.g., Dinghuibacter, Phyllobacterium). Conversely, degraded lands were significantly enriched in stress-associated bacteria, notably Rhodanobacter. Early-stage restored parcel soil clustered structurally closer to degraded soil than to the reference forest, indicating a pronounced spatial compositional similarity to disturbed soils rather than the reference state. Our findings from this localized study indicate that the fundamental microbial scaffolding of this remnant Chocó Forest soil hosts a distinct bacterial community structure, and that recovering soil microbial diversity in the adjacent old agricultural land may require extended timeframes and additional restoration approaches. We highlight remnant forest patches as critical reservoirs of belowground biodiversity and emphasize the necessity of integrating soil microbiology into continuous, long-term conservation monitoring frameworks.

The Chocó bioregion, also referred to as the Tumbes–Chocó–Magdalena biodiversity hotspot (Hrdina and Romportl, 2017), is globally ranked as the ninth most biodiverse hotspot, harboring approximately 3% of all known plant species (~11,000 species) within an area representing less than 0.2% of Earth’s terrestrial surface (Pérez-Escobar et al., 2019). This bioregion spans from south of Panama, crossing the Pacific coast of Colombia and Ecuador, to northwestern Peru (Rodríguez-Mahecha et al., 2004). A recent assessment of the recorded data, for less than 1% of this bioregion, reported an endemism rate of 8% for the 6,821 species of plants and animals (Yánez-Muñoz et al., 2024). Despite its notable biological richness, it remains among the least studied biodiversity hotspots in the American tropics (Pérez-Escobar et al., 2019). The Ecuadorian Chocó has undergone more than 85% habitat loss over the past decades, primarily driven by anthropogenic activities such as livestock, agriculture, forest fires, and urbanization (Geist and Lambin, 2002; Cuadrado et al., 2023). Although there are remaining tracts of well-preserved forest, these areas are enduring continuous reduction in favor of agricultural land, resulting in an increasingly fragmented landscape of isolated rainforest patches (Couvreur et al., 2022; Pitman et al., 2022). Under this scenario, conservation and landscape restoration are particularly challenging activities, as rural human communities rely on agricultural land for their livelihoods (Bautista et al., 2024). Hence, effective conservation strategies must integrate restoration with sustainable productive practices, especially in areas that serve as functional corridors between the remaining forest patches.

To optimize restoration strategies and improve the resilience of key ecosystems such as the Ecuadorian Chocó, comprehensive and integrative ecosystem assessments are needed. Thus, effective restoration requires evaluations that incorporate indicators and measurements of biodiversity across flora, fauna, and microorganisms (Rawat et al., 2023). Worldwide and particularly in tropical forests, restoration indicators have focused on the traditionally visible markers of ecosystem integrity: canopy cover, vertebrate biodiversity, and aboveground biomass; therefore, conservation policies have been heavily biased toward these various macroscopic components (Caldwell et al., 2024; Kemppinen et al., 2020; Venturini et al., 2023). However, an often-ignored aspect in terrestrial restoration practices is soil health (Gatica-Saavedra et al., 2023; Raupp et al., 2024; Peddle et al., 2026). The UN Intergovernmental Technical Panel on Soils (ITPS) defines soil health as “the ability of the soil to sustain the productivity, diversity, and environmental services of terrestrial ecosystems” (FAO, 2020). As the agricultural frontier expands in the Ecuadorian Chocó, driven by the global demand for commodities such as oil palm Elaeis guineensis, cocoa Theobroma cacao, and banana (Musa spp.), the fundamental architecture of the soil ecosystem is being rewritten, potentially irreversibly (Gómez et al., 2023; Perlin and Leguizamón, 2024; Ochoa-Brito et al., 2023). Hence, to determine the current conservation status of remnant forest patches and degraded soils, there is an urgent need to assess their microscopic biodiversity.

With the development of high-throughput DNA sequencing technologies, evaluating bacterial community structure is gaining importance in ecological assessment (Köninger et al., 2026). Growing evidence suggests that soil microbial communities are highly sensitive to land-use change and can provide valuable information on ecosystem degradation and recovery. For instance, soil metabarcoding studies have shown that native plant revegetation effectively restructures native bacterial communities in former grazing fields (Gellie et al., 2017). However, belowground microbial recovery does not necessarily follow aboveground vegetation recovery in parallel. Recent studies across tropical regions demonstrate that while successional transitions produce substantial shifts in soil bacterial community composition and network assembly mechanisms (Feng et al., 2025), overall microbial diversity may remain static despite rapidly increasing plant diversity (Saltonstall et al., 2025). Likewise, in an Andean temperate forest, surface soil microbial communities were found to be strongly altered following deforestation (Navarrete et al., 2023). Also, Romero et al. (2025) found that specific microbiome taxa are strong predictors in soil monitoring of environmental gradients, with the microbiome explaining more than a quarter of the variation in soil enzymatic activity and primary productivity. These findings highlight the soil microbiome as a key indicator of ecosystem multi-functionality and support the establishment of microbial baseline conditions in areas targeted for ecological restoration and sustainable agriculture, where they can help to evaluate the long-term effectiveness of management interventions.

Within the fragmented landscape of the Ecuadorian Chocó, several restoration initiatives have applied revegetation with the enrichment of native tree species (Bautista et al., 2024). In most restoration contexts, properties were previously used for grazing or agriculture and have often been exposed to pesticides for weed control. In addition, their initial soil condition is often poorly characterized, particularly with respect to nutrient levels, potential contaminants, and, most importantly, the existing microbial diversity. Other latitudes have benefited from evidence showing that poor farming practices and changes in land use influence soil microbial communities (Azarbad, 2022), and that restoration efforts can improve and recover microbial communities (Gellie et al., 2017). This baseline information is fundamental for guiding effective restoration processes on a landscape scale (Peddle et al., 2026); however, little is known about the effects of restoration practices or changes in land use in the soil microbiota of the Chocó bioregion, as we have not found previously published studies on the subject. Consequently, developing tools for long-term monitoring of soil microbiomes in the endangered forests of the Chocó bioregion, particularly through the evaluation of microbial diversity, will be critical to improving our understanding of soil ecology in remnant forest patches, restoration initiatives, and degraded land.

One of the restoration initiatives in the Ecuadorian Chocó is the “Biocorredor Tenka—La Esperanza,” a private initiative developed in the northwestern region of the Manabí province in Ecuador. The initiative protects the "La Esperanza Forest,” a private reserve where agrochemical-free restoration efforts have been implemented since 2022, with more than 100 native tree species planted. These efforts aim to use a mature forest patch protected by the reserve as a reference system, with the long-term goal of conserving and restoring biodiversity to the reference state, providing a local framework for ecological restoration. Previous evidence has shown the effects of changes in land-use on the soil microbiome of the Amazon region (Pedrinho et al., 2023); yet, comparable information for the Chocó bioregion remains scarce. To address this gap, our study assessed the taxonomic diversity and compositional structure of soil bacterial communities across the three habitat conditions managed within the La Esperanza Forest: reference forest, early-stage restored parcel, and old cattle grazing degraded land. Specifically, within the La Esperanza Forest, this research asks whether active restoration using native Chocó plant species is associated with localized shifts in the taxonomic diversity and compositional structure of the reference forest soil bacterial communities. We hypothesize that the sampled degraded agricultural plots will exhibit an unbalanced soil microbial architecture. Instead, we predict that reference forest plots will preserve a balanced bacterial community, while the recovering soils will harbor a transitional microbial community, reflecting a structural decoupling between aboveground revegetation and belowground bacterial community shifts.

Area of study

The private reserve “La Esperanza Forest” is located in El Carmen canton, Manabí province in Ecuador (0.3752°S, 79.6043°W; Figure 1). It corresponds to the Western Ecuador Moist Forests ecoregion (Eco ID: 516) (Dinerstein et al., 2017; Merloti et al., 2022; Pedrinho et al., 2023; Venturini et al., 2025), which has suffered an accelerated and intensive land transformation throughout the last decades and is evaluated as Critically Endangered by the IUCN (Ferrer-Paris et al., 2019). The reserve spans two hillsides draining into adjacent streams and displays structural attributes typical of a primary rainforest, with a well-developed canopy 20 to 30 m tall, including large emergent trees such as Eschweilera awaensis and prominent palms like Attalea colenda and Oenocarpus bataua (Couvreur et al., 2022). Its understory is open yet botanically diverse, supporting a range of shrubs, lianas, herbs, and epiphytes (Couvreur et al., 2022; Moreira Corral, 2023). A preliminary census documented approximately 200 botanical species within a single hectare, with many still unidentified, underscoring the site’s high floristic richness (Moreira Corral, 2023). In addition, its conservation value is determined by the high diversity of vertebrate species it supports within a relatively small area (15.22 ha), including 105 bird species, 13 amphibians, 10 reptiles, and 10 large mammal species, highlighting its critical role as a biodiversity refuge.

Of the reserve’s total area, 11.22 ha (71.87%) consist of mature forest, whereas the remaining 4 ha were formerly used for agriculture and pasture. The area has not been fumigated since 2013. Between 2022 and 2024 these 4 ha underwent annual revegetation enrichment, during which a mixture of native tree species was planted each year for a total of 118 species. Livestock access was excluded from the restoration area, invasive species were removed annually, and supplementary irrigation was provided during the dry season. No soil amendments or chemical inputs have been applied since restoration began. At the time of sampling, canopy height in the revegetated area ranged from 5 to 10 m.

The degraded zone is not part of the reserve; however, sampling was allowed by the landlord. This area was fumigated prior to 2017 with atrazine and Tordon 101 (2,4-d + Picloram), a systemic herbicide used to control weeds, in order to clear vegetation for livestock production. Since then, the site has been maintained under cattle grazing and has been subject to manual slashing. The surrounding pastures, adjacent to the reserve, are subject to continuous fumigation for livestock grazing using Quemax (Paraquat Dichloride) and other herbicides.

The reserve exists as a unique island of biodiversity completely encircled by a highly modified, agricultural and pastoral matrix (Supplementary Image 1). Because this remnant is one of the few surviving mature patches with access for scientific research in the immediate region, establishing true independent spatial replicates—which would necessitate locating multiple pristine forest patches adjacent to identically aged early-stage restoration parcels and identically managed degraded pastures within the same microclimatic envelope—is practically unfeasible. Consequently, the study site represents a fragile and ecologically invaluable refugium, providing a rare but necessarily spatially constrained opportunity to establish a localized proof-of-concept for soil microbiome monitoring and restoration in the Ecuadorian Chocó.

Soil collection

Fieldwork was conducted in 2024 between the 27th and 29th of September. Samples were collected from three distinct habitat conditions at the La Esperanza Forest reserve: remnant reference forest, early-stage restored parcel and old pasture degraded land, hereafter referred to as Forest (F), Restored (R) and Degraded (D) habitats (Figure 1). The R habitat underwent restoration practices between 2022 and 2024, while the D site corresponds to an area scheduled for revegetation in 2025. To capture the most representative ecological conditions of each land-use type, a purposive sampling strategy (Etikan, 2016; Campbell et al., 2020) was employed for plot placement. Within each habitat, two 10 × 10 m plots were established in locations representative of the core habitat conditions, avoiding atypical micro-topography. Distances between plots were 90 m and distances between habitat areas ranged from 221 to 455 m. Each plot was then subdivided into a grid of twenty-five 2 × 2 m subplots. To ensure a representative spatial distribution across the plot area, eight subplots were spatially dispersed and haphazardly selected for sampling (Supplementary Images 2–4). Then, four pooled aliquots of soil at the corners of each of the eight subplots were obtained and considered as sequencing samples (ca. 5 g), for a total of 48. However, two samples from the reference forest did not yield viable DNA, leaving a final total sequencing sample size of 46 soil samples. Because true landscape-level replication was unfeasible, it is explicitly noted that the collected soil samples in each habitat represent spatial subsamples nested within their respective plots, rather than independent habitat-level replicates.

The first 15 cm of the surface soil layer were collected using a sterile spatula and placed it in 50 mL sterile plastic centrifuge tubes (Griffin et al., 2014). Tube caps were sealed with parafilm, and sample tubes were stored in a cooler box containing dry ice and transported within a day to the laboratory. Samples were kept at −20 °C until DNA extraction. Soil DNA sequencing was authorized by the government of Ecuador with the Genetic Resource Access Contract MAATE-DBI-CM-2024-0413. The source soil samples are permanently stored in the National Bank of Genetic Resources at the National Institute of Biodiversity (INABIO) in Ecuador.

Soil DNA extraction

DNA was extracted from approximately 250 mg of soil samples using the Thermo Fisher Scientific PureLink™ Microbiome DNA Purification Kit (Cat. No. A29790), following the manufacturer’s protocol for soil samples. The purified DNA was quantified using the Promega QuantiFluor® dsDNA System (Cat. No. E2670) in a Promega Quantus™ Fluorometer (Cat. No. E6150).

Bacterial 16S library preparation and metabarcoding sequencing

Polymerase Chain Reactions (PCR) amplified the 16S rRNA gene region, using primers 27F and 1492R, supplied in the Oxford Nanopore Technologies Rapid Sequencing DNA—16S Barcoding Kit 24 V14 (Cat. No. SQK-16S114.24). Each 25 μL reaction contained a final concentration of 0.16 μM of 16S barcode primers, 1X LongAmp® Hot Start Taq DNA Polymerase cocktail (Cat. No. NEB M053), and 10 ng of template DNA. Thermocycling conditions consisted of an initial denaturation step at 95 °C for 1 min, followed by 30 cycles of 95 °C for 20 s, 55 °C for 30 s, and 65 °C for 2 min, with a final extension step at 65 °C for 5 min. PCR products were visualized on a 2% agarose gel and subsequently quantified using the Promega QuantiFluor® dsDNA System, following the manufacturer’s instructions.

Sequencing libraries were prepared by pooling the barcoded PCR products at equimolar concentrations according to the manufacturer’s protocol (Oxford Nanopore Technologies Cat. No. SQK-16S114.24). Finally, 50–60 fmol of the pooled library was loaded onto Oxford Nanopore Flow Cells (Cat. No. R10.4.1), and sequencing was performed in a GridION (Oxford Nanopore Technologies, UK) device at the Biotechnology Laboratory of the Instituto Nacional de Biodiversidad—INABIO.

Sequence processing, quality control, and taxonomic classification

Bioinformatic processing of the 16S rRNA gene sequences was conducted using the wf-16s workflow (v1.4.0) via the EPI2ME Desktop application. Raw Oxford Nanopore FASTQ reads, previously basecalled and demultiplexed natively on the GridION device using the MinKNOW software, were preprocessed using the fastcat tool (v0.20.0). To eliminate truncated sequences, potential chimeras, and off-target fragments, length filtering was strictly enforced, retaining only reads between 800 and 2000 base pairs.

Taxonomic classification was performed utilizing the Kraken2 (v2.1.2) k-mer-based algorithm mapped against the NCBI targeted loci 16S/18S reference database (ncbi_16s_18s). Because Kraken2 utilizes a Lowest Common Ancestor (LCA) algorithm for ambiguous sequences, it inherently mitigates the misclassification of reads that map equally well to multiple references. Following classification, Bayesian abundance estimation was executed using Bracken to accurately re-estimate taxonomic abundances at the Genus rank (taxonomic_rank: G). Lineages that resolved to distinct phylogenetic nodes lacking a formal Linnaean genus name (annotated by the database as “uncultured” or “unclassified” within a known family) were retained and treated as independent genus-level features for all downstream quantitative analyses. Unclassified reads were explicitly discarded from the downstream analysis (analyse_unclassified: false), and a minimum abundance threshold of 1 was enforced prior to the generation of the final count matrix. Complete per-sample sequencing statistics, including read retention counts and classification proportions, are provided in Supplementary Table 1.

Ecological and differential abundance analyses

All data wrangling, statistical modeling, and visualizations were conducted within the R programming environment (R Core Team, 2025). The raw taxonomic abundance table generated by the 16S rRNA workflow was aggregated at the genus level, and unclassified reads or metadata columns were removed prior to quantitative assessment.

Taxonomic coverage and sampling sufficiency

To assess the adequacy of the sequencing depth and ensure sufficient taxonomic coverage across all samples, read-based rarefaction curves were generated prior to compositional normalization. The rarecurve function from the vegan package (Oksanen et al., 2026) was applied to the raw, untransformed genus-level count matrix to visualize the accumulation of observed genera as a function of the number of sequenced reads (Supplementary Image 5).

Alpha diversity analysis

To account for variations in sequencing depth across samples, the raw genus-level count matrix was rarefied to an even sampling depth (corresponding to the minimum library size of the dataset = 12.786 reads per sample) using the rrarefy function in the vegan package (Oksanen et al., 2026). Within-sample diversity was then assessed on this standardized matrix by calculating observed genus richness, the Shannon Diversity Index, and Pielou’s Evenness. To evaluate statistical differences across the three habitat conditions while explicitly accounting for our spatial sampling design, we fitted Linear Mixed-Effects Models (LMM) using the lme4 package (Bates et al., 2015). Habitat condition was modeled as a fixed effect, and the spatially distinct sampling plots were modeled as random intercepts to prevent spatial pseudoreplication. The significance of the habitat effect was assessed via analysis of variance (ANOVA) on the mixed models.

Community structure (beta diversity) and hypothesis testing

To statistically evaluate overall differences in bacterial community structure across the three habitat treatments, an omnibus Permutational Multivariate Analysis of Variance (PERMANOVA) was conducted. Because the read counts were previously normalized using the CLR transformation, Euclidean distances were calculated to effectively compute the Aitchison distance, a mathematically robust metric for compositional sequencing data. To explicitly account for spatial non-independence and correct structural grouping vectors that artificially inflate plot dimensions by treating subsamples as independent, the CLR-transformed abundance matrix was aggregated at the plot level by calculating the mean abundance for each taxon. An omnibus Permutational Multivariate Analysis of Variance (PERMANOVA) was then performed on this aggregated, plot-level dataset (n = 6) based on the Aitchison distance using the adonis2 function in the vegan package (Oksanen et al., 2026). To verify that the observed differences across habitats were not influenced by unequal variances, the homogeneity of multivariate dispersions was evaluated using the betadisper function (Oksanen et al., 2026). To account for the unbalanced nested spatial design, significance was subsequently tested by fitting the resulting sample-to-centroid distances to a Linear Mixed-Effects Model (LMM), modeling habitat as a fixed effect and sampling plots as random intercepts.

Compositional data normalization and differential abundance

To account for the inherently compositional nature of high-throughput microbiome sequencing data, raw read counts were transformed using the Centered Log-Ratio (CLR) approach implemented in the ALDEx2 package (Fernandes et al., 2014). ALDEx2 utilizes Monte Carlo sampling (mc.samples = 1000) drawn from a Dirichlet distribution to infer true relative abundances and estimate technical variation. To identify specific bacterial genera driving the differences between the remnant forest and the altered habitats (D and R), differential abundance testing was performed using a General Linear Model (aldex.glm). Significance was determined using false discovery rate (FDR) adjusted p -values (q- values) using the Holm method. Taxa were considered significantly differentially abundant at an FDR-adjusted threshold of q < 0.05.

Differences in overall bacterial community structure across habitats were visualized using two primary approaches based on the expected CLR-transformed values (defined as the median of the Dirichlet Monte Carlo instances generated for each individual sample). (1) A clustered heatmap of the median CLR abundances was generated using the pheatmap package (Kolde, 2025) to visualize broader patterns of community composition and group similarity. (2) A correlation biplot was constructed using the rda function from the vegan package (Oksanen et al., 2026). To clearly highlight the specific taxa most responsible for the community differentiation across habitats, species scores (loadings) were extracted and scaled. Only the most strongly associated genera, defined as those in the top 1% (99 th percentile) of vector magnitudes, were projected onto the final PCA biplot. Finally, volcano plots visualizing the relationship between differential abundance effect sizes (Median Log 2 Difference) and statistical significance (−Log 10 q -value) were constructed using ggplot2 (Wickham, 2016), with a significance cutoff plotted at 1.3 (equivalent to q < 0.05). Because standard ALDEx2 implementations do not natively accommodate nested random effects, individual samples were treated as independent observations. Consequently, to acknowledge the potential for spatial pseudoreplication, these specific differential abundance results are explicitly interpreted as exploratory.

To elucidate the underlying ecological mechanisms driving bacterial community assembly, overall beta diversity was partitioned into species turnover (replacement) and nestedness-resultant (subset-loss) components (Baselga, 2010, 2012). To explicitly account for spatial non-independence, the genus-level read count matrix was first aggregated at the plot level (n = 6) by calculating the total sequence abundance per plot and subsequently converting this to a binary presence/absence matrix. The partitioning was then computed based on the Sørensen dissimilarity index using the betapart package (Baselga and Orme, 2012) in R (R Core Team, 2025). To statistically evaluate the influence of habitat type on each distinct beta diversity component, the resulting turnover and nestedness distance matrices were independently subjected to Permutational Multivariate Analysis of Variance (PERMANOVA) using the adonis2 function from the vegan package (Oksanen et al., 2026). Post-hoc pairwise PERMANOVA tests were subsequently performed utilizing the same aggregated design, with p -values adjusted for multiple comparisons via the Holm method. Finally, the partitioned pairwise dissimilarities for all spatial samples were visualized using the ggstatsplot package (Patil, 2021) to illustrate the full distribution of multivariate dispersion.

Taxonomic composition and alpha diversity

The Read-based rarefaction curves indicated that the sequencing depth was adequate to capture the taxonomic diversity of the soil microbiomes, as the accumulation of observed genera approached an asymptote across all samples from the three habitats (Supplementary Image 5). The soil bacterial communities, assessed via 16S rRNA gene metabarcoding, revealed a total of 1208 taxa classified at the genus level across all sampled habitats. The overall community structure, based on the presence and abundance of each individual taxon, clearly discriminates the samples from the Forest (F) habitat in contrast to both the degraded (D) and restored (R) habitats (Figure 2). Analysis of the relative taxonomic abundances revealed a consistent set of dominant genera shared among the sampled habitats. The ten most dominant taxa at the genus level, including Bacillus, Haliangium, and Paenibacillus, alongside several uncultured lineages, maintained a highly consistent presence and stable relative abundance across the F, R, and D habitats (Figure 3).

Ecological indices calculated from the rarefied genus counts revealed structural shifts in within-sample diversity among the experimental treatments, although the strict spatial constraints of the nested models inherently limited statistical power (Figure 4). The standardized observed genus richness was highly variable, with the F habitat exhibiting the highest mean (381.36 ± 29.24; Median: 387.50), compared to the D (361.88 ± 23.15) and R (355.75 ± 34.37) habitats. Linear Mixed-Effects Models (LMM) accounting for the nested spatial plots indicated no significant differences in overall observed genus richness across the habitats (LMM ANOVA; F = 1.69, df = 2, p = 0.32).

Descriptively, variations were also structurally evident in community evenness and overall diversity. The reference forest soil maintained a more evenly distributed bacterial community (Shannon Mean: 3.94 ± 0.28; Pielou Mean: 0.66 ± 0.04) compared to the Degraded (Shannon: 3.73 ± 0.17; Pielou: 0.64 ± 0.03) and Restored (Shannon: 3.74 ± 0.19; Pielou: 0.65 ± 0.03) environmental states. However, due to the mathematically conservative nature of the plot-level nesting (n = 6 experimental units), these structural shifts did not reach traditional statistical significance for the Shannon Diversity Index (LMM ANOVA; F = 1.85, df = 2, p = 0.30) or Pielou’s Evenness Index (LMM ANOVA; F = 1.41, df = 2, p = 0.37). These data indicate that while descriptive trends suggest a more diverse and evenly distributed microbiome in remnant forests, the variance between isolated spatial plots limits the inferential power of univariate comparisons.

Community structure and beta diversity

The omnibus PERMANOVA test, based on the Aitchison distance (Euclidean distance of CLR-transformed abundances) aggregated at the plot level, evaluated the overall bacterial community structure. Due to the highly conservative nature of this strictly constrained spatial model (n = 6 experimental units), the test inherently lacked the statistical power to achieve traditional significance thresholds (PERMANOVA; R 2 = 0.665, F = 2.97, p = 0.20). However, the descriptive effect size (R 2) suggests that the sampled habitat condition explains over 66% of the total structural variance in the soil communities (see Supplementary file 1 for the complete PERMANOVA table).

Post-hoc pairwise PERMANOVA tests with Holm adjustment for multiple comparisons were conducted. Consistent with the omnibus model, the mathematically constrained permutations restricted to four aggregated plots per test inherently resulted in non-significant adjusted p -values (p = 1.00 for all contrasts). Nevertheless, the corresponding pairwise effect sizes (R 2) suggest substantial structural divergence among the sampled habitats. The difference in bacterial community structure between the F and R habitats yielded a large high effect size (R 2 = 0.71), as did the comparison between the F and D habitats (R 2 = 0.65). In contrast, the structural divergence between the R and D habitats was comparatively much lower (R 2 = 0.33).

Furthermore, evaluation of multivariate dispersion via betadisper coupled with the nested LMM confirmed that within-habitat structural variance did not differ significantly across the land-use types (LMM ANOVA; F = 0.04, df = 2, p = 0.96) (see Supplementary file 1 for the complete ANOVA table). This indicates that the observed multivariate separation is driven by distinct shifts in compositional centroids rather than artifacts of unequal dispersion.

These descriptive structural patterns are also visually apparent in the compositional clustering. Hierarchical clustering of the median CLR abundances showed a clear pattern of differentiation, primarily separating the F habitat from the R and D habitats (Figure 2). Similarly, the Principal Component Analysis (PCA) biplot distinctly partitions the F habitat samples completely away from the D and R habitats along the primary axis of variation (PC1: 21.7%), illustrating a highly distinct microbial architecture (Figure 5). Samples belonging to the D habitat clearly separated according to their sampling plots in the space formed by the PCA (Figure 5). The extraction of the scaled scores from the PCA identified the specific genera driving this multivariate separation. The taxa contributing strongly to the ordination axes for the separation of the F soil included Dinghuibacter, Ilumatobacter, Amaricoccus, Aminobacter, Phyllobacterium and two uncultured genera belonging to the Saprospiraceae and Nitrosomonadaceae families. In contrast, genera contributing strongly to the separation of the D and R soils along the PCA axes included Sedimentibacter, Methylocystis, Chungangia, Fervidicella, Rhodanobacter, and an unclassified genus from the Caloramatoraceae family (Figure 5, vectors).

Differential abundance analysis (DAA)

To formally test which taxa were potentially driving these compositional shifts, a differential abundance analysis was performed using ALDEx2. Contrasting the F baseline against the D soils revealed that the genus Rhodanobacter was significantly more abundant in the D environment (FDR-adjusted q < 0.05; Figure 6A). Notably, no individual genera were found to be statistically more abundant in F when directly compared to D soils under these strict threshold conditions. Active restoration was associated with the significant enrichment of the genus Fictibacillus and an uncultured genus from the family Nitrosomonadaceae (MND1) in the R soils. Conversely, the F soils maintained a significantly higher abundance of an uncultured genus belonging to the family Saprospiraceae when compared to the R soils (Figure 6B).

Beta diversity partitioning: turnover and nestedness

To elucidate the ecological mechanisms driving the observed compositional shifts, overall beta diversity was partitioned into genera turnover (replacement) and nestedness (subset-loss) components using the incidence-based Sørensen dissimilarity index.

The partitioning analysis revealed that genera turnover was the overwhelmingly dominant process structuring the soil bacterial communities across the three habitats (Figure 7). The median Sørensen dissimilarity for the turnover fraction was markedly higher (ranging approximately between 0.15 and 0.30) than the nestedness fraction, which remained consistently low (below 0.10) across all comparisons.

Notably, the turnover component exhibited a substantial increase in all between-habitat contrasts relative to the within-habitat baseline dispersions. The dissimilarity was particularly pronounced in the D vs. F and R vs. F comparisons, visually confirming a high degree of genera replacement (Figure 7).

As observed with the overall community structure, the strict spatial constraints of the aggregated PERMANOVA models (n = 6) inherently limited statistical power for traditional significance thresholds (omnibus p = 0.07 for turnover and p = 1.00 for nestedness, with Holm-adjusted p = 1.00 for all pairwise contrasts). However, evaluation of the effect sizes (R 2) suggests that within our study site, habitat condition is strongly associated with species turnover, explaining over 64% of the variance in species replacement (PERMANOVA; R 2 = 0.64, F = 2.71, p = 0.07). Post-hoc pairwise tests further supported this microbial replacement, revealing large effect sizes between the distinct environmental states (Forest vs. Restored: R 2 = 0.740; Forest vs. Degraded: R 2 = 0.570; Restored vs. Degraded: R 2 = 0.39). In stark contrast, the nestedness-resultant fraction contributed minimally to the overall structural differentiation among the soil communities, showing virtually zero variance explained by habitat type (PERMANOVA; R 2 = −0.08, F = −0.10, p = 1.00). Pairwise comparisons further confirmed the complete absence of substantial subset-loss dynamics between any of the tested habitats (maximum variance explained R 2 = 0.16 for Degraded vs. Restored, with all other R 2 ≤ 0). Together, these localized results indicate that the structural shifts associated with agricultural conversion and early-stage restoration are driven primarily by a replacement of the bacterial architecture, rather than a simple sequential loss of the native reference forest microbiome (see Supplementary file 1 for the complete ANOVA tables).

The Tumbes-Chocó-Magdalena hotspot is globally recognized for its biological richness, yet it remains one of the least studied and most heavily deforested tropical regions on Earth (Geist and Lambin, 2002; Cuadrado et al., 2023). Within this highly fragmented landscape, our localized study at La Esperanza Forest reserve provides an empirical glimpse for soil microbial communities. Our data revealed a foundational set of dominant bacterial genera shared among all sampled habitats, regardless of land-use history. The consistent dominance of generalist taxa (e.g., Bacillus and Paenibacillus) suggests that the fundamental biological scaffolding of these soils persists despite intense aboveground disturbance. This occurs even under conditions of continued use of pesticides in the surrounding area, which may leach into water bodies and contribute to contamination even within mature forest habitats. Therefore, the persistence of these taxa should not be taken as an indication of ecological integrity or soil health. However, these taxa seem to be more abundant in the D and R habitats, which may account for the descriptive reduction in community evenness observed in these disturbed habitats.

The descriptive reduction in community evenness and the massive variance explained by the distinct structural shifts observed in beta diversity suggest that the sampled agricultural plots exhibit a markedly different soil microbiome architecture compared to the reference forest plots. By partitioning this beta diversity, we observed that this architectural reorganization is driven by species turnover (replacement) rather than nestedness (subset-loss). Ecologically, within our localized study site, this indicates that the degradation of primary forest seems to create novel niches that select for replacement of the bacterial community. Consequently, reference forest patches maintain a unique microbial structure, strongly driven along the PCA ordination axes by taxa such as Dinghuibacter and Phyllobacterium that are notably reduced in the sampled agricultural and restored lands. These genera correspond to Gram-negative bacteria that have been isolated in forest soils and associated with tree leaves and roots (Lv et al., 2016; Negi et al., 2022). Particularly, some species of the Phyllobacterium genus are symbiotic nitrogen-fixing bacteria and have been considered in microbial fertilizers (Dasgupta et al., 2021). This possibly implies the ecological distinctiveness of this remnant of Chocoan tropical forest, highlighting the role of even small extant patches, as a refugia not only for above-ground fauna and flora, but also as a reservoir of soil bacterial diversity.

In addition, the mature remnant forest soil microbiome is enriched in uncultured groups such as two undescribed lineages in Saprospiraceae and Nitrosomonadaceae. The former corresponds to a family involved in the breakdown of complex organic compounds (McIlroy and Nielsen, 2014), while the latter comprise a monophyletic group of ammonia oxidizers with important roles in nitrification processes (Prosser et al., 2014). Other relevant taxa enriched in the forest are Amaricoccus and Aminobacter, both Gram negative bacteria found previously in active sludge and soils (Maszenan et al., 1997; Hördt et al., 2020; Urakami et al., 1992). On the other hand, we speculate that the strong association of Sedimentibacter and Chungangia, two genera associated with marine sediments (Imachi et al., 2016; Kim et al., 2012), and Rhodanobacter, a genus associated with denitrification in contaminated and agricultural soils (Green et al., 2012; Cho et al., 2017), with the D and R habitats could be related to old agricultural practices, which are difficult to identify. Despite these strong associations, we have not found an established consensus or referenced literature that can unequivocally characterize the metabolic and ecological functions of these various genera, much less in the context of our study. Thus, our study is currently limited to the description of those taxa that contribute to the observed differences of the three tested habitats, but we are unable to propose a functional mechanism of such observed differences.

Our localized data show that early active restoration efforts in this parcel do not immediately return soil bacterial communities to a state resembling the adjacent reference forest. This decoupling of aboveground and belowground trajectories is increasingly recognized in tropical restoration. Robinson et al. (2025) reported disparities in microbial assembly when comparing intact forests with logged or actively restored forests, and Saltonstall et al. (2025) observed that while community composition shifts distinctly during secondary succession, overall microbial diversity does not necessarily parallel the recovery of plant diversity. Furthermore, tropical forest restoration processes are driven by complex assembly mechanisms that only gradually shift as soil environments stabilize (Feng et al., 2025). While Miyawaki-style afforestation can induce rapid taxonomic and functional reorganization (Bansal et al., 2026), the structural transition to a mature, forest-like metabolic network spans decades rather than years (Sansupa et al., 2021). Thus, the restoration parcel in our study clustered structurally closer to the degraded land than to the remnant forest, lacking the high-loading structural drivers present in the reference forest soils. When analysing the specific taxa driving this spatial separation (PCA) alongside the differential abundance models (DAA), we observe a distinct taxonomic transition from a forest-associated microbiome toward a stress-associated agricultural community. These findings highlight the importance of long-term microbiome monitoring in reforested areas, with periodic assessments beginning at the earliest stages to better understand belowground recovery over time.

Bacterial communities appear to be strongly structured according to our sampling design. A rather striking pattern of this structuring was observed within the D habitat, which formed two clearly distinct clusters of samples in our ordination analysis (PCA), corresponding to each of the sampled plots. The environmental elements in our studied region seem to be strong determinants of microbial community composition, at very small scales of less than one hundred meters of separation and within apparently homogenous habitats in terms of human use. Advancements in high-throughput DNA sequencing, high-resolution spatial modeling, and geostatistical analysis have increasingly demonstrated microbial structures can be considerably predictive of environmental effects (Bay et al., 2020; Averill et al., 2021; Du et al., 2021).

Because our cross-sectional design captures these habitats at a single point in time, we cannot rule out alternative or confounding explanations for the observed microbial associations. Critically, the absence of corresponding soil physicochemical measurements (such as pH, organic carbon, moisture, nitrogen, phosphorus, and texture) in our study limits mechanistic ecological interpretation. Without these quantitative variables, it is difficult to definitively distinguish the effects associated with restoration and land-use history from underlying baseline edaphic heterogeneity. The unique clustering of samples—particularly the distinct separation between plots within the Degraded habitat—strongly suggests that unmeasured environmental heterogeneity plays a critical role in structuring these communities. However, despite this fine-scale spatial variance within the disturbed soils, the primary axis of compositional variation consistently isolates the reference forest. Factors such as localized soil chemistry, micro-topography, drainage dynamics, specific vegetation composition, spatial distance, and variations in the historical intensity of pesticide application likely interact with the broader land-use categories. Therefore, the structural differences reported here should be interpreted as complex associations driven by an interplay of historical management and fine-scale environmental gradients, rather than being solely dictated by the overarching habitat labels.

Rather than a simple return to the baseline forest microbiome, which remained uniquely enriched in uncultured Saprospiraceae as the sole statistically significant group in the contrast between the reference forest and the restored habitat, we propose that within this reserve, the active restoration parcel correlates with a distinct microbial compositional state. Exploratory differential abundance analysis identified the restored soils as being significantly enriched with Fictibacillus and Nitrosomonadaceae. This bidirectional shift suggests that early-stage restoration is associated with unique ecological conditions, hosting a community distinct from both the degraded pasture and the reference forest. Species in Nitrosomonadaceae are obligate ammonia-oxidizing bacteria (Chain et al., 2003; Prosser et al., 2014) and their significant abundance suggests a highly active or altered nitrogen cycle. While taxonomic abundance alone cannot establish metabolic activity, we hypothesize that the observed enrichment of this group is probably driven by localized nitrogen fluxes as a legacy effect of previous agriculture/pasture use. Furthermore, the significant enrichment of Rhodanobacter in the degraded plots, a genus frequently associated with environmental stress and altered biogeochemical cycling (Lin et al., 2021; Prakash et al., 2012; Zhang et al., 2021; Peng et al., 2022), suggests a potential link to the persistence of anthropogenic disturbance. Crucially, we emphasize that these functional and ecological interpretations based solely on genus-level 16S rRNA data remain speculative hypotheses. Definitively establishing biogeochemical functioning will require future validation through targeted soil chemistry, functional gene quantification, transcriptomics, shotgun metagenomics, or direct biochemical measurements (Jarrín-V et al., 2026).

Our interpretation of the ecological and biochemical roles of the particular genera found as important or significant in the applied contrasts between habitats is speculative, as we will require direct biochemical tests and experiments, as well as genomic and metabolomic data, to understand how these various taxa could behave in the context of the specific environments analyzed (Jarrín-V et al., 2026). Our findings align with emerging evidence from other tropical systems suggesting that soil microbiomes do not necessarily shift in tandem with aboveground tree biomass. While restoration through the planting of native trees is a vital first step (Toro et al., 2025), achieving a soil microbiome structure comparable to the reference state in the Ecuadorian Chocó may require extended, long-term timeframes (Vivian et al., 2025). Forest remnants like La Esperanza Forest act as critical reservoirs of native soil biodiversity that are not quickly mirrored in early-stage restoration efforts. These patches are therefore critical for conservation, not only as reservoirs of microbial diversity, but also as reference ecosystems toward which restored areas should converge over the long-term.

Our study is a localized effort within a single 15.22 ha reserve. Because our experimental design relies on spatially distinct patches for each habitat condition, true independent landscape-level replication was not feasible within this highly fragmented matrix. Consequently, these findings represent a localized rather than a universal model for the highly heterogeneous Chocó bioregion. However, by establishing this molecular proof-of-concept and demonstrating significant microbial shifts even when statistically accounting for spatial clustering, this work provides a rigorous and valuable first baseline for long-term ecological monitoring at the La Esperanza restoration initiative.

We propose moving beyond static baselines to longitudinal plots across time and to expand our study to the landscape scale. This will allow us to track bacterial community composition trajectories over the coming decades and evaluate the efficacy of regional biological corridors. Crucially, future sampling must couple DNA sequencing with standard biogeochemical metrics and transition to shotgun metagenomics or direct enzymatic assays. This methodological shift will allow us to measure other essential aspects of ecosystem recovery, as a complex, multidimensional and multifunctional process.