Drivers of Genetic Diversity in Keystone Forest Species: Contrasting Responses of Rare and Common Saproxylic Beetles to Habitat Location and Naturalness

ABSTRACT

Understanding how forest management shapes genetic diversity is pivotal for the conservation of saproxylic beetles, key indicators of ecosystem integrity. However, the relative influence of spatial isolation versus anthropogenic modification on their genetic structure remains poorly understood across Central Europe. We investigated whether geographic location (forests spanned over 600 km W‐to‐E and 500 km S‐to‐N) and forest management type (primaeval vs. commercial) drive genetic differentiation across 12 beetle species with varying abundance (common/rare) and pest status (pest/not‐pest). Using a multi‐species double digest restriction‐site associated DNA (ddRAD) approach, we analysed single nucleotide polymorphism (SNP) variation in taxa differing in abundance and mobility. We used generalized linear mixed models (GLMMs), distance‐based redundancy analysis (dbRDA) and analysis of molecular variance (AMOVA) to evaluate the landscape and ecological factors driving genetic diversity and differentiation. Furthermore, population structure, isolation‐by‐distance (IBD) and genetic admixture were explicitly resolved using ADMIXTURE and discriminant analysis of principal components (DAPC) frameworks. Analyses revealed geography, likely driven by post‐glacial recolonization and IBD, as the main determinant of differentiation. Forest management exerted a secondary but significant effect, especially among rare specialists. In these sensitive species, populations from primaeval protected sites exhibited distinct genetic signatures compared to commercial stands, indicating that intensive management isolates ancient forest specialists, while pests remained unaffected. Our results highlight that regional connectivity and the preservation of primaeval refugia must be prioritized to maintain gene flow. By integrating high‐resolution genomic data with management context, this study offers actionable insights for incorporating genetic monitoring into sustainable forest biodiversity management.

1. Introduction

Forests are among the largest terrestrial ecosystems on the planet (Feng et al. 2016; Watson et al. 2018) and have been shaped by long‐term historical processes that influence their diversity and ecological structure (Giesecke and Brewer 2018). Forests are major reservoirs of global biodiversity (Gillespie et al. 2012; Bohn and Huth 2017), making their conservation critical, especially in areas under intense anthropogenic pressure. In primaeval forests, the diversity of organisms associated with wood is typically high; however, this diversity is significantly reduced in intensively managed forests, where only a limited number of species thrive (Kuuluvainen 2009; Paillet et al. 2010). While some organisms inhabit living trees, many are specialized to colonize dead trees, often referred to as deadwood. For years, this ecological niche and its inhabitants were neglected in forest studies (Siitonen 2001), but growing awareness has led to increased research on wood‐dependent taxa (Ogle et al. 2014; Kozák et al. 2021). As primary agents of nutrient cycling and wood decomposition, these taxa represent a critical biotic component of forest health. Understanding their population dynamics is essential, as the overall functioning of forest ecosystems is shaped by a complex interplay of abiotic and biotic factors, including anthropogenic disturbances (Rasche et al. 2013; Bayat et al. 2021).

The current diversity of wood‐dwelling organisms is the result of past environmental and climatic conditions, as well as recent changes caused by human activity. Forest development changed during Pleistocene glacial cycles. In Europe, this forced forest‐dwelling organisms to contract their ranges to the southernmost refugia or to move eastward into Asia, while a minority survived in local cryptic refugia in Central or Western Europe (Taberlet et al. 1998; Hewitt 2000). These shifts in species ranges had consequences for their biogeography (the distribution of distinct lineages), with important transient zones identified among many species or subspecies in Central Europe.

Later, in the late Holocene, human activity, such as forest fragmentation, deforestation and intensive forest management, has altered woodland ecosystems (Abdullah and Nakagoshi 2007; Taylor et al. 2009). This has particularly affected species dependent on old‐growth microhabitats such as deadwood. Among saproxylic organisms, the most diverse (likely, except for microorganisms, which are poorly known) are arboreal fungi and insects, which are threatened by the depletion of these habitats in commercial woodlands (De Vivo et al. 2023). While some generalist species have adapted to managed environments and even become pests, specialized species often experience reduced populations or local extinctions. These changes are now increasingly intensified by climate change, which continues to affect biodiversity and forest composition (Pawson et al. 2013; Murray et al. 2017).

Saproxylic beetles are the most diverse and well‐studied group of organisms associated with deadwood and play key roles in forest ecosystems through decomposition, food web interactions and nutrient cycling (Grove 2002; Hjältén et al. 2012; Thorn et al. 2018). Due to their strong association with deadwood habitats, they are highly sensitive indicators of forest biodiversity and ecological continuity (Hjältén et al. 2012). However, forest fragmentation and intensive logging have led to the loss of deadwood microhabitats and extensive declines in saproxylic beetle populations (de Groot et al. 2019; Mazur et al. 2021). Conversely, many cambio‐ and xylophagous beetles (e.g., bark beetles, longhorn beetles, etc.) that can undergo population outbreaks are regarded as pests in timber production, although they remain natural elements in old‐growth forests (Singh et al. 2024).

The genetic consequences of habitat fragmentation are largely mediated by a species' dispersal capability and ecological specialization (Nolte et al. 2017). While some species are highly vagile colonizers (e.g., many bark beetle ‘pests’) capable of maintaining high gene flow across managed landscapes (Singh et al. 2024), others are sedentary specialists restricted to stable, old‐growth microhabitats. For these dispersal‐limited species, even moderate forest fragmentation can act as a significant barrier, leading to increased isolation‐by‐distance (IBD) and the formation of distinct regional lineages (Sgarlata et al. 2025). Understanding these trait‐dependent responses is essential for predicting how different ecological groups, from rare relicts to common specialists, will persist in increasingly fragmented European forests.

Conservation efforts for these beetles require extensive knowledge of their population structure, genetic diversity and gene flow, which are often difficult to resolve using traditional ecological and morphological approaches due to cryptic genetic diversity (Gibson and Reed 2008). Molecular methods provide powerful tools for determining genetic variability and population connectivity, enhancing conservation and management strategies (Audisio et al. 2009; Eberle et al. 2021). Recently, knowledge of the population genetics of saproxylic beetles has been summarized (Kajtoch et al. 2022; Krovi et al. 2024), supporting their complex genetic variability across Europe, with Central European forests as a transitional zone between various evolutionary lineages. Although previous studies relied on traditional molecular markers (e.g., mitochondrial DNA or microsatellites), genomic data have only recently begun to be implemented for saproxylic beetles in Europe.

High‐throughput sequencing (HTS) techniques, such as double‐digest restriction‐site associated DNA sequencing (ddRADseq), have revolutionized population genetics by enabling cost‐effective, genome‐wide discovery of single‐nucleotide polymorphisms (SNPs) in non‐model organisms (Aguirre et al. 2019; Peterson et al. 2012). These genomic markers ‘SNPs’ are powerful tools for assessing genetic diversity and demographic history (Funk et al. 2012; Wenne 2023), providing clear management insights in beetles ranging from rare Lucanus lineages (De Vivo et al. 2023) to widespread pests like Ips typographus (Linnaeus 1758; Ellerstrand et al. 2022). Such data are essential for developing strategies that maintain adaptive capacity across diverse forest‐dependent taxa (Palsbøll et al. 2007; Casacci et al. 2014).

In this study, we used ddRADseq to assess the genetic diversity and population genetic structure of several saproxylic beetle species in Central European forests. These species were sampled across a wide range of forest types in Poland.

Our objectives are: (i) to quantify genetic diversity between beetle populations collected from forests located in a presumed transient zone in Central Europe to test the hypothesis that westernmost and easternmost forests are inhabited by significantly differentiated populations; (ii) to determine whether forest type (commercial vs. primaeval) affects saproxylic beetles' genetic diversity, specifically testing the hypothesis that populations from primaeval forests are more diverse than those in commercial forests; (iii) to assess whether observed patterns are consistent among various species (e.g., rare vs. common) hypothesizing that rare, habitat‐specialized species exhibit stronger genetic structure than common, mobile generalists; and (iv) to provide data that can support conservation efforts by identifying populations with low genetic variability and/or high isolation, with the hypothesis that isolated primaeval fragments will show signatures of genetic bottlenecks or restricted gene flow, requiring targeted conservation measures. This research contributes to the growing fields of population and conservation genetics, highlighting the importance of integrating molecular tools into biodiversity monitoring programmes, especially for ecologically significant but often overlooked taxa, such as saproxylic beetles.

2. Materials and Methods

2.1. Study Sites

Study sites were established in eight forest complexes across Poland, covering lowland, upland and mountainous regions and representing the geographic extremes of ranges of the studied species (Figure 1, Table S1). Beetles were collected in 2022 and 2023. The sites encompassed a gradient of forest types, from protected primaeval forests (Białowieża Forest, Holy Cross Mountains and the Carpathians) to commercial forests (Augustów Forest, Knyszyn Forest, Silesian Forest, Oder Forest, Barycz Forest) (Figure 1). In commercial forests, sampling included typical logged sites (without protection), referred to as ‘logged’ throughout the manuscript, and nature reserves (old‐growth forests resembling natural ones, though of limited area and isolated within a commercial woodland matrix), referred to as ‘nature reserve’ throughout the manuscript. In primaeval forests, beetles were collected in strictly protected stands (national parks and nature reserves) referred to as ‘primaeval protected’, and in managed stands (with much lower logging intensity than in other managed forests) referred to as ‘primaeval not protected’ throughout the manuscript. For some analyses, forests were grouped by geography: NE (Białowieża Forest, Knyszyn Forest, Augustów Forest—all from the Podlasie lowlands); SW (Silesian Forest, Oder Forest, Barycz Forest—all from the Silesia lowlands); and SE (Holy Cross Mountains and the Carpathians—both from mountainous regions). This grouping reflects both physiographic conditions and similarities in forest habitats. In the case of the Holy Cross Mountains and the Carpathian foothills, we decided to group them together despite their distinct orography, due to similarities in altitude (300–700 m a.s.l.), the structure of old‐growth, mixed, multispecies forests and the presence of species‐rich communities of saproxylic beetles. These two regions represent the only montane refugia for relict taxa in Poland. In total, thousands of sites were visited, but only 421 sites hosted the selected beetles, with some sites hosting more than one taxon (Table S1).

FIGURE 1.

Map of sampling sites of 12 saproxylic beetles species across Poland. BF—Białowieża Forest, KF—Knyszyn Forest, AF—Augustów Forest, HM—Holy Cross Mts, CM—Carpathian Mts, OF—Oder Forest, DB—Barycz Forest, SF—Silesia Forest. NE, SE, SW—three regions considered in the study. Diagram refined and compiled using CorelDraw.

2.2. Beetle Sampling

Beetles were collected in 2022 and 2023 as non‐random, standardized sampling using complementary collection methods: (i) inspection of microhabitats under the bark of dying or dead trees, (ii) cutting decayed deadwood, (iii) inspection of fungi (using a sieve) and (iv) funnel and barrier traps with pheromone lures for selected taxa.

Sampling targeted ecologically important saproxylic beetle species representing various trophic guilds (saprophages, xylophages, mycetophages, predators) and conservation statuses (Table 1). In this study, species frequently present across all sites are referred to as ‘common’ throughout the manuscript. A subset of these species known to cause damage to timber during population outbreaks are referred to as ‘pests’, while species occurring infrequently or in low numbers are referred to as ‘rare’. Among selected beetles, 12 species were chosen based on the above criteria and sampling availability (with at least several sites available across all examined forests). While sampling effort was uniform across all sites, the resulting differences in sample size reflect the natural abundance and inherent rarity of the target taxa. These rarer habitat specialists were intentionally included to test the hypothesis that specialized species are more sensitive to management‐induced isolation than generalists.

TABLE 1.

Summary of species taxonomic classification (family), ecological characteristics (abundance, pest, status and trophy), sampling details (collection details) and the number of examined sites (L, logged; Pn, primaeval not protected sites; Pp, primaeval protected sites; R, natural reserve sites) used in this study.

Species	Family	Abundance	Pest	Status	Trophy	Collection stage	No. of examined sites
							Total	Pp	Pn	R	L
Neomida haemorrhoidalis	Tenebrionidae	Rare	No	Unprotected	Mycetophagous	Adults	40	14	4	13	9
Boros schneideri	Boridae	Rare	No	Protected	Predator	Larvae*	32	13	5	4	10
Cucujus cinnaberinus	Cucujidae	Rare	No	Protected	Saprophagous	Larvae*	39	8	9	10	12
Elater ferrugineus	Elateridae	Rare	No	Protected	Predator	Adults**	21	7	3	5	6
Bolitophagus reticulatus	Tenebrionidae	Common	No	Unprotected	Mycetophagous	Adults	38	10	8	9	11
Stenurella melanura	Cerambycidae	Common	No	Unprotected	Xylophagous	Adults	43	10	5	11	17
Schizotus pectinicornis	Pyrochroidae	Common	No	Unprotected	Saprophagous	Larvae	37	7	8	11	11
Thanasimus formicarius	Cleridae	Common	No	Unprotected	Predator	Adult*	37	2	13	6	16
Ips acuminatus	Curculionidae	Common	Yes	Unprotected	Xylophagous	Adults	28	4	4	0	20
Ips typographus	Curculionidae	Common	Yes	Unprotected	Xylophagous	Adults	44	5	11	9	19
Rhagium inquisitor	Cerambycidae	Common	Yes	Unprotected	Xylophagous	Larvae*	34	4	11	9	10
Phaenops cyanea	Buprestidae	Common	Yes	Unprotected	Xylophagous	Adult*	28	2	11	0	15

Note: Tissue codes indicate head and thorax (*), leg (**) or whole body (no asterisk). For a full list of geographic coordinates and regional classifications for each site, see Table S1.

Selected rare taxa included relics of primaeval forests (Eckelt et al. 2018) (Boros schneideri (Panzer, 1796) [n = 154], Cucujus cinnaberinus (Scopoli, 1763) [n = 190], Neomida haemorrhoidalis (Fabricius, 1787) [n = 199] and Elater ferrugineu s (Linnaeus, 1758) [n = 84]), common taxa (e.g., Thanasimus formicarius (Linnaeus 1758) [n = 178], Bolitophagus reticulatus (Linnaeus, 1767) [n = 185], Schizotus pectinicornis (Linnaeus, 1758) [n = 167], Stenurella melanura (Linnaeus, 1758) [n = 199]) and pests (e.g., Ips acuminatus (Gyllenhal, 1827) [n = 125], Ips typographus [n = 201], Rhagium inquisitor (Linnaeus, 1758) [n = 164] and Phaenops cyanea (Fabricius, 1775) [n = 140]). All selected species are native to Poland. The list of these species, along with their characteristics, is presented in Table 1. Detailed information on sample sizes is provided in Table S2.

Each specimen (larva or adult, Table 1) was stored in 95%–99% ethanol and preserved at −20°C until DNA extraction. Each individual was identified to species by specialists (coleopterologists) on the research team. No selected beetle taxa required special techniques for identification (e.g., preparation of reproductive organs).

2.3. Laboratory Procedures

2.3.1. DNA Extraction

Prior to DNA isolation, each sample was washed with distilled water to remove external contaminants. In total, 1986 beetles were used for DNA extraction. Each specimen was processed separately under sterile conditions.

For small beetles (< 1 cm in length), the entire body was used for DNA extraction. In larger beetles, only the head, thorax and legs were used, while the abdomen was retained (Table 1). For large larvae, only the head and thorax were used for DNA isolation (Table 1). For the largest species ( E. ferrugineus ), only one leg was collected due to restrictions imposed by nature protection authorities, and the living beetle was released.

DNA was extracted using Sherlock AXE (A&A Biotechnology) according to the manufacturer's protocol.

2.3.2. Library Preparation and Sequencing

ddRad‐seq libraries were prepared following the method described by Bayona‐Vásquez et al. (2019). Briefly, DNA concentration was normalized to 20 ng/μL and 5 μL of each sample was transferred to 96‐well plate. Next, iTru Fusion Primers with Internal Tags were added to identify each sample within the plate (iTru_NheI_R1_stub_A—H and iTru_EcoRI_R2_RC_stub_1–12), while appropriate restriction enzymes were added for digestion. After 1 h incubation at 37°C, ligation was performed with T4 ligase. Ligation products were pooled (96 samples into two tubes of 240 μL each) and, after purification with AMPure XP magnetic beads (Beckman Coulter), PCR was performed with i5 and i7 primers to identify each pool during sequencing. PCR products were again purified with Agencourt AMPure XP beads and quantified using a Qubit. Finally, the library range was analysed on the TapeStation, and fragments between 400 and 600 bp were selected using the Pippin Prep instrument (Sage Science). Libraries from four plates (384 samples) were pooled, normalized to 2 nM and sent to the sequencing facility. Pooled libraries were sequenced at Admera Health Company (New Jersey, USA) on an Illumina NovaSeq X Plus—10B—PE 150, yielding approximately 4 million reads per sample.

2.4. Bioinformatic Analyses

2.4.1. Quality Control of Raw Sequencing Data

Quality assessment of the raw FASTQ files was performed using FastQC v0.11.9 (Andrews 2010), a standard tool for evaluating the quality of HTS data. As this is a ddRADseq dataset generated with 141 bp paired‐end reads, all sequences have a uniform length determined by the sequencing protocol. Therefore, no explicit minimum read length filter was applied at this stage. Each sequencing file was assessed individually for key metrics, including per‐base sequence quality, GC content, sequence length distribution, adapter contamination and duplication levels. Summarized quality reports were generated and reviewed using MultiQC v1.32 (Ewels et al. 2016) to identify samples with abnormal patterns or quality concerns. Reads with low‐quality scores (Phred < 30) or uncalled bases were removed. Only reads that passed quality filtering and retained their full restriction‐site overhangs were kept for further processing using the Stacks pipeline. This rigorous quality control ensured high‐confidence SNP calling and minimized downstream biases in population genetic analyses.

2.4.2. Alignment and Variant Calling (Stacks Pipeline)

Raw reads were demultiplexed, quality‐filtered and trimmed using the process_radtags module in Stacks v2.68 (Rochette et al. 2019). De novo locus assembly was performed using the ustacks, cstacks and sstacks modules. Key parameters included a minimum stack depth (−m = 3) and a maximum number of mismatches (−M = 2) per individual. The catalogue was built using cstacks across all individuals with a mismatch allowance (−n = 1) between loci.

Detection and genotyping of all loci from the aligned reads were performed using the gstacks programme included in Stacks v2.68. The detected loci were filtered using the ‘populations’ software to call SNPs and compute population‐level statistics, which is included in Stacks v2.68. It was run with (‐‐r = 0.80) to retain only loci present in at least 80% of the population. The following filters were also applied to ensure high genotyping quality:

1. Minimum minor allele frequency (MAF) of 0.03.

2. Exclusion of samples with an average coverage lower than 5× from further analysis (‐min‐gt‐depth 5).

3. Only positions informative in at least 70% of individuals were retained for further analysis (Geno of 0.3).

Output files were exported in VCF and STRUCTURE formats for downstream analyses.

2.5. Population Genetic and Statistical Analysis

2.5.1. Genetic Data Preparation for Downstream Analysis

All ddRADseq datasets were processed using the same SNP filtering pipeline applied separately to each species (MAF = 0.03, GENO = 0.3). The identical filtering criteria ensured methodological consistency across species. Differences in the final number of retained SNPs and loci among species (Table S2) reflect variation in genome characteristics (e.g., heterozygosity, restriction site distribution and sequence depth) rather than differences in data processing. To account for linkage disequilibrium, SNP datasets for each species were LD‐pruned prior to ADMIXTURE v.1.3 (Alexander and Lange 2011) analysis using a threshold of r ² < 0.1. For other analyses (distance‐based redundancy analysis (dbRDA), analysis of molecular variance (AMOVA), discriminant analysis of principal components (DAPC) and heatmaps), the full filtered SNP sets were used without LD pruning to retain maximum genetic information, as these multivariate and distance‐based approaches are less affected by linkage (Excoffier et al. 1992; Jombart et al. 2010). The number of SNPs retained after LD pruning for each species is provided in Table S2.

2.5.2. Spatial Genetic Patterns

2.5.2.1. Drivers of Genetic Differentiation

We evaluated the significance of geographic distribution, management, pest status (pest/non‐pest) and abundance (common/rare) (Objectives 1 and 2) using generalized linear mixed models (GLMMs) with nested taxonomic effects as a random factor paired with dbRDA to partition unique versus shared variance between geography and management. We employed GLMM implemented via the glmmTMB package (Brooks et al. 2017). The response variable was the linearized pairwise genetic differentiation (F _ST/ (1—F _ST)). Our raw dataset contained a massive pairwise matrix (N = 6883 comparisons) which displayed a highly right‐skewed distribution characterized by a distinct mass of pairs with zero genetic differentiation (Figure S1). The fixed effects included four primary predictors:

1. Forest management comparison (MGMT_comp): a categorical variable tracking whether a given population pair belongs to the same management types or different types (Same_management vs. Different_management), evaluating the impact of management types on gene flow.

2. Geographic Comparison (Geo_comp): a categorical variable identifying whether the population pair resides within the same geographic region or different regions (Same_region vs. Different_region), controlling for large‐scale spatial isolation

3. Pest status (Pest_status): a binary fixed factor with two mutually exclusive levels: pest versus non‐pest

4. Abundance (Abundance): a categorical fixed factor with two strictly mutually exclusive levels assigned at the species level: Common versus Rare.

Prior to model evaluation, list‐wise deletion of all missing observations (NAs) was performed to ensure that the sample size remained perfectly consistent across all evaluated candidate models. To determine the optimal balance between goodness‐of‐fit and model parsimony, model selection was performed using the Akaike Information Criterion (AIC) across a candidate set of four GLMMs. These four models evaluated all combinations of variable complexity for our main predictors while keeping pest status and abundance constant in all models: (1) Two management types (primaeval vs. commercial) + 3 sub‐regional categories (NE, SE and SW) (M1), (2) Four management types + 3 sub‐regional categories (NE, CM, HM, SW; splitting SE following Kondracki 2002) (M1a), (3) Two management types (primaeval‐protected, primaeval‐not‐protected, reserve and logged) + 4 geographic regions (M1b) and (4) Four management types + 4 sub‐regional categories (M1c) (Table S3). The model displaying the lowest AIC value was selected as our final global architecture for downstream hypothesis testing.

Analyses of pairwise matrix data present inherent spatial dependencies that violate the structural independence assumptions of standard regression models, as individual populations appear in multiple distinct population pairs. To control for this non‐independence without losing statistical power, we applied a crossed random intercepts structure consisting of (1 | pop_i) + (1 | pop_j) into our global model: F_ST_linearized ~ MGMT_comp + Geo_comp + Pest_status + Abundance (1 | Genus/species) + (1|pop_i) + (1|pop_j), fitted under zero‐inflated Gamma distribution family with a log link function for the conditional model component. To account for the structural point mass of exact zero‐differentiation pairs across the landscape matrix, we specified a constant baseline probability for the zero‐inflation component using an intercept‐only formula (ziformula = ~1).

This framework accounts for population‐specific baseline variations in genetic drift by isolating the unique variance contribution of each individual locality. Furthermore, to account for phylogenetic relatedness and taxonomic hierarchies among the multi‐species community, we included a nested random intercept of (1|Genus/species). Although several genera were monotypic within our sampling pool, each distinct species was highly replicated across landscape matrices, comprising > 25 separate populations (ranging from 28 to 44 populations; see Table S1) (yielding > 378 pairwise F _ST observation rows per species). This density ensures highly robust variance component partitioning across the nested taxonomic levels.

To ensure structural stability and confirm that our fixed predictions (Management, Geography, Pest_status and Abundance) were completely independent, we evaluated multicollinearity by calculating Generalized Variance Inflation Factors (GVIF) and adjusted GVIF (adj GVIF) using the performance package (Lüdecke et al. 2021). To quantify the explanatory power of our fixed effects, we calculated Nakagawa's marginal R ² value. The conditional R ² value could not be analytically isolated due to mathematical constraints inherent to extracting theoretical variance from multi‐level mixed architectures (nested taxonomic blRocks and crossed population blocks) fitted under a zero‐inflated distribution family. Potential model overfitting was controlled structurally by evaluating model parsimony via AIC selection and by implementing a crossed random intercepts architecture to absorb localized spatial noise. Finally, model specification and residual calibration were validated using randomized quantile residuals (1000 simulations) via the DHARMa package (Hartig 2022).

To validate the robustness of our fixed‐effect inferences, statistical significance was assessed using two complementary testing frameworks. First, we conducted Wald Type II Chi‐square tests (chi²) utilizing the Anova function from the car package (Fox and Weisberg 2019). Second, we performed single‐term deletion Likelihood Ratio Tests (LRT) using the drop1 function from the core stats package (R Development Core Team 2022). The LRT framework systematically evaluated each predictor by dropping individual terms from the finalized global model and comparing nested candidate model likelihoods to verify the stability and strength of our fixed‐effect conclusions.

We utilized dbRDA with linearized pairwise genetic differentiation F _ST (F _ST/[1−F _ST]) as the response variable (Legendre and Andersson 1999). These methods allowed us to partition variance and assess the relative strength of environmental drivers for each taxon independently. Management was encoded as a categorical predictor with two management types (commercial and primaeval). Spatial structure was modelled using dbMEM eigenfunctions derived from site coordinates (Peres‐Neto and Legendre 2010). Collinearity between the two predictors was assessed using the chi‐squared test of independence, and the strength of association was quantified using Cramér's V (Cramér 1946). For each species, we calculated pairwise Weir‐Cockerham F _ST among populations/sites and used the linearized F _ST (F _ST/[1−F_ST]) as the response variable to meet the distance‐based model assumptions (Weir and Clark 1984; Rousset 1997). The geography predictor was calculated as distance (km, log‐transformed), and the management predictor was estimated using categorical dummies for the two types. Because geography and management are partially collinear, we quantified for each species the unique (conditional) contribution of management after controlling for geographic distance and of geographic distance after controlling for management using partial R ²‐ proportion of variation in the outcome explained uniquely by one predictor term. Significance p‐values and partial R ² were obtained using a permutation‐based model (permutations = 9999) for each predictor.

2.5.2.2. Spatial and Hierarchical Genetic Structure

To assess consistency across various species (Objective 3), we used AMOVA and Distance Decay to determine if the ‘threshold’ of management impact varies according to species‐specific abundance (common/rare) and pest status (pest/non‐pest). The AMOVA test (Excoffier et al. 1992) was conducted separately for populations grouped by management type and by geographic region using the poppr package (Kamvar et al. 2014). Variance was partitioned among and within populations for two management types (commercial and primaeval) and three geographic groups (NE, SE and SW). Significance was assessed by permutation tests (9999 permutations) for each species separately.

For each species, we evaluated the relationship between genetic differentiation and geographic distance under different management types using distance decay analysis. Pairwise linearized F _ST was used as the response variable and merged with geographic coordinates and management categories (metadata file). Geographic distances (km) were calculated using the geosphere package (Hijmans 2025). Isolation by distance and management effect for each species was quantified using multiple matrix regression with randomization (MMRR) implemented in the ecodist package (Goslee and Urban 2007), with 9.999 permutations. The response variable was linearized F _ST, and the predictors were log‐transformed geographic distance and management.

2.5.3. Genetic Structure and Population Clustering

Finally, to support conservation efforts by identifying populations with high isolation or low variability (Objective 4), we examined fine‐scale genetic structure using a multi‐faceted approach. We employed ADMIXTURE analysis to estimate individual ancestry proportions and identify potential hybridization or gene flow between sites, and DAPC to identify distinct genetic clusters of population models. To complement these clustering methods, we calculated pairwise F _ST values, visualized through heatmaps using four distinct management types: (i) Primaeval‐protected, (ii) Primaeval‐not‐protected, (iii) Nature reserves and (iv) logged. This four‐types classification allowed us to detect distinct genetic signatures that might be masked in a broader two‐type model and provided the multi‐dimensional space necessary for DAPC to visualize relative genetic distances and overlapping management regimes. Together, these methods allow us to distinguish between broad regional shifts in ancestry and the localized effects of isolation in specific forest remnants.

The most probable number of ancestral clusters (K) in ADMIXTURE was determined using a combination of cross‐validation and low linkage disequilibrium (LD) values (< 0.1). For each individual, the admixture algorithm returns the probability Q (Q₁, Q₂, …, Q₁₀) that it belongs to each of the K populations. The resulting individual ancestry coefficients (Q‐matrix) were imported into R and visualized as stacked barplots with individuals ordered by populations based on four management types (logged, nature reserve, primaeval‐not‐protected and primaeval‐protected) and three geographic groups (northeast [NE], southeast [SE] and southwest [SW]) using a metadata file.

DAPC was implemented using the adegenet package (Jombart 2008; Jombart et al. 2010). Genotype data were converted into a standard genotype‐individual (genind) object and grouped by the same management and geographic categories. The number of principal components retained for the analysis (n.pca = 50) was determined using α‐score optimization to balance discrimination power and overfitting, and the number of discriminant functions (n.da = 3 for management, n.da = 2 for geography, i.e., n.da = groups−1). Individual scores in DA1 and DA2 were visualized with 95% confidence intervals based on normal data.

Pairwise F _ST was quantified separately for populations grouped by management type and by geographic region. It was estimated for all population pairs with permutation‐based p‐value and BH‐adjusted false discovery control; the resulting F _ST matrix was visualized as a heatmap (rows/columns ordered by hierarchical clustering) to highlight clusters and gradients of differentiation.

To ensure that varying sample sizes among species and management types did not bias our estimates of genetic differentiation, we conducted a sensitivity analysis by down sampling management categories to a uniform size of n = 10 for a subset of species. We then compared the resulting F _ST values to those obtained from the full dataset.

For statistical analyses, we used R version 4.2.2 (R Development Core Team 2022). All visualizations have been done using the ggplot2 package (Wickham 2016).

3. Results

The number of SNPs ranged from 1484 in I. typographus to 27,583 in P. cyanea , with a mean of SNPs per species 7226.2 SNPs. This variation in SNP yield reflects interspecific differences in genome size, level of polymorphism and the density of restriction sites across the diverse beetle families studied. Within each species, Mean F _ST values across geographic regions ranged from 0.004 to 0.404, with the highest differentiation observed in common species (S. pectinicornis, T. formicarius and B. reticulatus ) (Figure S6). In contrast, mean F _ST values between forest management types (commercial vs. primaeval) were generally lower, ranging from −0.034 to 0.206. However, significant differentiation between management types was still detected in specific taxa (e.g., E. ferrugineus and S. pectinicornis) (Figure S7).

3.1. Drivers of Genetic Differentiation

Based on AIC selection, the finalized zero‐inflated GLMM model was identified as the most parsimonious and well‐supported candidate architecture, safely accounting for continuous boundary point masses via an intercept‐only zero‐inflation component (ziformula = ~1) (Table S3). Downstream collinearity evaluations confirmed the complete absence of predictor redundancy (All adj GVIF values < 1.10). Comprehensive DHARMa residual diagnostics confirmed that the model was structurally well‐specified and stable. The simulated quantile residuals exhibited a well‐calibrated error structure with no evidence of overdispersion or underdispersion (Nonparametric dispersion ratio measures observed variance of residuals vs. expected variance from simulations computed using rank = 2.33, p = 0.22; see Figure S2), while a bootstrap outlier test proved that extreme tail frequencies fell safely within null expectations (p = 0.17). Noteworthy, a robust zero‐inflation simulation test (p = 0.98) demonstrated an almost perfect convergence between empirical and simulated zero frequencies (ratioObsSim = 1.00). The model selection process strongly favoured a simplified two management types (primaeval and commercial) over a four‐types classification (primaeval‐protected, primaeval‐not‐protected, nature reserve and logged) and three geographic sub‐regions (NE, SE, SW) over four (NE, HM, CM, SW) with Δ AIC = 19.11 (Table S3). Model fit comparisons strongly supported the inclusion of a nested random effects structure (1|Genus/species) over a simplified (1|Genus) structure (Δ AIC = 324.61) (Table S3), demonstrating the critical necessity of partitioning background evolutionary and life‐history variance at both taxonomic levels.

Fixed‐effects inference from the finalized model revealed that spatial and ecological predictors strongly structure continuous genetic distances across the saproxylic community (Table S4). Specifically, population pairs sharing the same geographic region (β = −0.515, p < 0.001) and the same forest management framework (β = −0.193, p < 0.001) exhibited significantly reduced pairwise genetic differentiation. On an ecological scale, species classified as pests exhibited a highly significant decrease in pairwise genetic differentiation compared to the non‐pest baseline reference group (β = −1.374, p = 0.01), whereas no significant baseline differentiation was resolved between rare and common taxa (β = −0.576, p = 0.285) (Table S4). LRT revealed that all evaluated fixed predictors significantly explained variation in pairwise genetic differentiation (F _ST). Geographic isolation exerted the most pronounced influence on genetic structure (LRT ${\chi }^2$ = 287.52, Wald ${\chi }^2$ = 299.20, df = 1, p < 0.001), with population pairs situated in different geographic regions displaying highly elevated genetic differentiation compared to those residing within the same region. Forest management comparison also had a highly significant independent impact on genetic connectivity (LRT ${\chi }^2$ = 19.13, Wald ${\chi }^2$ = 19.01, df = 1, p < 0.001), with populations located across differing forest management regimes showing significantly altered genetic differentiation relative to populations experiencing similar management type. Furthermore, functional pest status emerged as a powerful driver of genetic structure (LRT ${\chi }^2$ = 5.27, Wald ${\chi }^2$ = 6.58, df = 1, p = 0.01). Pest species exhibited the lowest baseline differentiation, pointing to widespread gene flow. Conversely, baseline population abundance didn't significantly affect patterns of genetic differentiation (LRT ${\chi }^2$ = 1.10, Wald ${\chi }^2$ = 1.14, df = 1, p = 0.28). Overall, these fixed landscape and ecological predictors accounted for a substantial portion of the continuous variation in genetic structure, yielding a Nakagawa's marginal R ² value of 0.44.

3.2. Distance‐Based Redundancy Analysis

dbRDA showed variable contributions of geography and management factors to genetic diversity among 12 beetle species (Table 2). Due to significant collinearity between the two factors (X ² = 264.17, df = 6, p < 0.001; Cramér's V = 0.49), dbRDA results are interpreted in terms of unique (conditional) and shared contributions rather than independent effects. Overall, dbRDA results suggest that genetic differentiation was more strongly associated with geographic distance than with forest management after accounting for shared variance between predictors (Table 2). Geographic distance explained a larger unique proportion of genetic differentiation in several species, with the highest R ² values in the common species ( S. melanura , S. pectinicornis, T. formicarius and B. reticulatus ), followed by pests ( I. acuminatus , I. typographus and P. cyanea ), indicating a clear spatial genetic structure consistent with isolation by distance (Figure 2, Table 2). Rare species showed little to no genetic differentiation, except for N. haemorrhoidalis, which showed a relatively moderate R ² proportion (R ² = 0.228). In contrast, management effects contributed less to overall variation, with low and mostly non‐significant unique R ² values across all species, except for a weak trend (p_management = 0.09) in R. inquisitor and P. cyanea (Table 2).

TABLE 2.

Summary of distance‐based redundancy analysis (dbRDA) results showing the variance in genetic diversity explained by geographic distance and forest management for 12 saproxylic beetle species.

Species	Abundance	No. sites	No. pairs	No. of SNPs	Geography levels	Managment levels	R 2 geography	p geography	R 2 management	p management	R 2 full
Neomida haemorrhoidalis	Rare	34	561	10,501	3	4	0.228	< 0.001	0.007	0.518	0.264
Boros schneideri	Rare	31	465	6574	3	3	0.000	0.826	0.000	0.714	0.006
Cucujus cinnaberinus	Rare	39	741	2480	3	4	0.085	< 0.001	0.009	0.382	0.114
Elater ferrugineus	Rare	21	207	10,210	3	4	0.000	0.555	0.000	0.872	0.000
Bolitophagus reticulatus	Common	38	703	2841	3	4	0.398	< 0.001	0.234	0.346	0.486
Stenurella melanura	Common	42	861	12,063	3	4	0.707	< 0.001	0.011	0.956	0.786
Schizotus pectinicornis	Common	35	595	2512	3	4	0.552	< 0.001	0.010	0.593	0.716
Thanasimus formicarius	Common	37	666	2426	3	4	0.436	< 0.001	0.059	0.135	0.454
Ips acuminatus	Common	26	325	3957	3	4	0.251	0.031	0.118	0.149	0.266
Ips typographus	Common	41	820	1484	3	4	0.113	0.001	0.000	0.999	0.094
Rhagium inquisitor	Common	34	561	4083	3	4	0.099	< 0.001	0.046	0.090	0.093
Phaenops cyanea	Common	28	378	27,583	3	3	0.079	0.097	0.088	0.09	0.125

Note: Reported are partial (conditional) adjusted R ² values, permutation‐based p‐values (9.999 permutations) for each predictor (geography and management), and adjusted R ² values for the full model. Significance was determined using a threshold of p‐value < 0.05.

FIGURE 2.

Isolation by distance (IBD) across the two forest management types. Genetic differentiation was modelled using multiple matrix regression with randomization (MMRR). Significance (p‐values) for the relationship between geographic and genetic distance was determined using 10,000 permutations.

3.3. Genetic Structure—AMOVA Test—Geography

AMOVA revealed greater genetic structure among populations categorized by geographic regions than among those categorized by forest management types, with significant geographic structure in all species (Table 3). The percentages of variance among populations ranged from 1.5% to 50.7%, with the strongest differentiation observed in common taxa (e.g., S. pectinicornis, T. formicarius, B. reticulatus and S. melanura ) (Table 3). In contrast, the ‘within‐population’ component accounted for the majority of the variance in several species (up to 98.5% in P. cyanea ), highlighting high levels of local genetic diversity. However, this within‐population variance was strikingly lower in taxa with strong spatial structure, such as S. pectinicornis (49.3%), suggesting that for these species, regional isolation is as significant a driver of genetic architecture as individual‐level variation. Intermediate differentiation was detected in rare taxa and was comparatively weak in a subset of pest species (except for I. acuminatus ) (Table 3).

TABLE 3.

Analysis of molecular variance (AMOVA) results for the 12 studied beetle partitioned by geographic regions and forest management types: Column abbreviations are defined as follows: No. of ind., number of individuals analysed; df, degrees of freedom; σ ² among, variance component among groups (regions or management types); % among, percentage of total genetic variation explained by differences among groups; σ ² within, variance component within groups; % within, percentage of variation explained by differences within groups; p, statistical significance based on 1000 permutations (threshold < 0.05); and Φ_ΣΤ, fixation index representing the level of genetic differentiation.

Species	Abundance	No. of ind.	Geography							Management
			df	σ 2 among	% among	σ 2 within	% within	p	Φ_ΣΤ	df	σ 2 among	% among	σ 2 within	% within	p	Φ_ΣΤ
Neomida haemorrhoidalis	Rare	199	2	14.371	5.814	233.150	94.191	0.0001	0.058	3	6.760	2.762	237.960	97.240	0.0001	0.027
Boros schneideri	Rare	158	2	1.921	3.191	58.430	96.810	0.0001	0.032	3	1.563	2.621	58.231	97.381	0.0001	0.026
Cucujus cinnaberinus	Rare	190	2	2.091	5.991	32.711	94.012	0.0001	0.059	3	1.241	3.631	33.011	96.372	0.0001	0.036
Elater ferrugineus	Rare	84	2	45.342	8.173	509.570	91.821	0.0001	0.081	3	38.321	6.990	509.340	93.000	0.0001	0.069
Bolitophagus reticulatus	Common	185	2	35.571	23.421	116.280	76.572	0.0001	0.234	3	7.710	5.491	132.711	94.513	0.0001	0.054
Stenurella melanura	Common	199	2	43.592	12.311	310.543	87.681	0.0001	0.123	3	5.871	1.721	334.205	98.274	0.0001	0.017
Schizotus pectinicornis	Common	167	2	50.240	50.662	48.910	49.333	0.0001	0.506	3	10.941	13.073	72.770	86.921	0.0001	0.131
Thanasimus formicarius	Common	178	2	129.491	46.331	150.008	53.660	0.0001	0.046	3	23.351	9.940	211.532	90.053	0.0007	0.099
Ips acuminatus	Common	125	2	6.434	9.582	60.631	90.413	0.0001	0.095	2	1.221	1.891	63.232	98.120	0.051	0.018
Ips typographus	Common	201	2	0.411	4.180	9.442	95.823	0.0001	0.041	3	0.112	1.223	9.560	98.773	0.003	0.012
Rhagium inquisitor	Common	164	2	0.703	2.411	28.562	97.594	0.0001	0.02	3	0.311	1.070	28.731	98.931	0.0002	0.011
Phaenops cyanea	Common	140	2	15.660	1.463	1052.651	98.534	0.0002	0.014	2	11.272	1.060	1056.071	98.942	0.0006	0.011

3.4. Genetic Structure—AMOVA Test—Management

AMOVA partitioning by management showed that most genetic variation occurred within populations, while only a small proportion was explained by differences among populations categorized by management type (Table 3). The percentages of variance between populations ranged from 1.2% to 13.1%, with the highest differentiation observed in the common species S. pectinicornis and T. formicarius (Table 3). Moderate differentiation was observed in the rare‐threatened species E. ferrugineus (6.99%) and the common microhabitat‐specialized species B. reticulatus (5.49%) (Table 3). Only I. acuminatus showed a marginally non‐significant result (p = 0.05) (Table 3).

3.5. Distance—Decay MMRR

Across the 12 species, distance decay plots showed that genetic diversity generally increased with increasing geographic distance, with a significant geographic distance effect in all species except R. inquisitor and P. cyanea (Figure 2). Management effects were weaker and more idiosyncratic, with a significant management effect in four species (three common: S. melanura , S. pectinicornis and B. reticulatus ; and one pest: I. typographus).

3.6. Genetic Structure—DAPC—Geography

Across the 12 species for which we ran DAPC by region (NE, SE and SW), we observed heterogeneous but overall geography‐driven structure (Figure 3). DAPC showed strong genetic structure in common species (S. pectinicornis, B. reticulatus and S. melanura ), each forming well‐separated (east—west) regional clusters. In contrast, the remaining common and pest taxa (T. formicarius, I. typographus, I. acuminatus and R. inquisitor ) exhibited broad overlap among regions, consistent with higher connectivity. The rare species (C. cinnaberinus, N. haemorrhoidalis and E. ferrugineus exhibited) intermediate genetic structure across populations. Some taxa ( R. inquisitor , E. ferrugineus and I. typographus) showed variable regional clustering separation (north—south) (Figure 3). One species, B. schneideri , lacking a SW population, showed differences between NE and SE populations (Figure 3).

FIGURE 3.

Genetic structure represented by discriminant analysis of principal components (DAPC). Clusters are colour‐coded by geographic regions (NE, north east; SE, south east; SW, south west).

3.7. Genetic Structure—DAPC—Management

Across the 12 beetle species for which we ran DAPC by management type (logged, nature reserve, primaeval not protected and primaeval protected), we observed a moderately heterogeneous management‐driven structure (Figure 4). DAPC panels of species showed the most pronounced separation between the primaeval protected and logged sites in four species (S. pectinicornis, B. reticulatus , E. ferrugineus and C. cinnaberinus), with partial weak overlap in rare ( N. haemorrhoidalis , B. schneideri ), common ( S. melanura ) and pest ( P. cyanea ) species. In contrast, the remaining common (T. formicarius), and pest taxa ( R. inquisitor , I. typographus and I. acuminatus ) exhibited wide overlap among management types (Figure 4).

FIGURE 4.

Genetic structure represented by discriminant analysis of principal components (DAPC) based on four forest management types (logged, nature reserve, primaeval not protected and primaeval protected).

3.8. Genetic Structure—Admixture ‐Geography

Ancestry assignments were consistent with DAPC ranking. The common species ( B. reticulatus and S. pectinicornis), the pest ( I. acuminatus ) and the rare taxa ( N. haemorrhoidalis and E. ferrugineus ) displayed regionally coherent ancestry blocks with limited cross‐assignment among populations (NE, SE, SW) (Figure 5). In contrast, the species S. melanura , T. formicarius, R. inquisitor , I. typographus, P. cyanea , C. cinnaberinus and B. schneideri showed more mixed ancestry across regions, indicating weaker geographic partitioning of gene pools (Figure 5).

FIGURE 5.

Genetic structure represented by ADMIXTURE based on forests' geographic regions (NE, north east; SE, south east; SW, south west) and four management types (logged, nature reserve, primaeval not protected and primaeval protected).

3.9. Genetic Structure—Admixture ‐Management

Ancestry estimates aligned with DAPC proportions. The common species ( B. reticulatus and S. pectinicornis), the pest ( I. acuminatus ) and the rare species ( N. haemorrhoidalis ) exhibited the strongest structure, with coherent ancestry blocks and limited cross‐assignment across management types, especially between primaeval protected and logged sites (Figure 4; Figures S3–S5). The rare species ( E. ferrugineus ) showed moderate genetic structure (Figure 4; Figure S4). In contrast, the remaining common (T. formicarius), pest taxa ( R. inquisitor and I. typographus) (Figure 4, Figure S5) and the other rare taxa (C. cinnaberinus and B. schneideri ) displayed more mixed ancestry across management types (Figure 4; Figure S4).

3.10. Pairwise F _ST—Heatmap—Geography

The pairwise F _ST heatmap revealed pronounced east–west contrasts in several species, while a few taxa displayed north–south separation (e.g., R. inquisitor and E. ferrugineus ) (Figure S6). The east—west pattern was most evident in some common beetles (e.g., S. pectinicornis, T. formicarius and B. reticulatus ) and in rare species (e.g., N. haemorrhoidalis and E. ferrugineus ) (Figure S6). The remaining taxa showed weak variance among geographic regions (Figure S6).

3.11. Pairwise F _ST—Heatmap—Management

In the pairwise F _ST heatmap, several taxa displayed detectable differences among management types, with the strongest contrasts typically occurring between primaeval protected and logged sites (Figure S7). This pattern was most evident in some common beetles (e.g., S. pectinicornis and B. reticulatus ) and in rare species (e.g., E. ferrugineus and C. cinnaberinus) (Figure S7). By contrast, conifer taxa (I. typographus, I. acuminatus ) and the conifer‐oligophage R. inquisitor mostly showed weak variance among management types (Figure S7).

3.12. Robustness of F _ST Estimates

To verify that varying sample sizes did not bias our estimates of genetic differentiation, we conducted a sensitivity analysis on three representative species: S. pectinicornis (common), P. cyanea (pest) and E. ferrugineus (rare). Comparison of F _ST heatmap generated from the full dataset versus those down‐sampled to a uniform n = 10 individuals per category revealed highly consistent patterns of differentiation (Figure S8). In all cases, the relative magnitude of genetic partitioning between primaeval and commercial forests remained stable, confirming that our findings are biologically robust and independent of sampling effort (Table 4).

TABLE 4.

Summary of dbRDA and AMOVA results assessing the effects of geography and management on genetic variation and gene flow across saproxylic beetle species. Significance levels are indicated by symbols: ***p < 0.0001, **p < 0.001, *p < 0.01, –, p < 0.1 (weak trend); —, not significant, and NA populations from west not present. Patterns of spatial and management gradients are abbreviated as E–W, east–west, N–S, north–south and Pp‐L, primeval protected‐logged.

Species	Abundance	Pest	Geography effect		Managment effect		E–W gradient	N–S gradient	High	Pp‐L gradients	High
									Gene flow		Gene flow
			dbRDA	AMOVA	dbRDA	AMOVA			Geography		Management
Neomida haemorrhoidalis	Rare	No	***	***	—	***	✔			✔
Boros schneideri	Rare	No	—	***	—	***	NA	✔		✔
Cucujus cinnaberinus	Rare	No	***	***	—	***	✔			✔
Elater ferrugineus	Rare	No	—	***	—	***		✔		✔
Bolitophagus reticulatus	Common	No	***	***	—	***	✔			✔
Stenurella melanura	Common	No	***	***	—	***	✔			✔
Schizotus pectinicornis	Common	No	***	***	—	***	✔			✔
Thanasimus formicarius	Common	No	***	***	—	***			✔		✔
Ips acuminatus	Common	Yes	*	***	—	—			✔		✔
Ips typographus	Common	Yes	**	***	—	**			✔	✔
Rhagium inquisitor	Common	Yes	***	***	—	***		✔			✔
Phaenops cyanea	Common	Yes	—	***	—	***	✔				✔

4. Discussion

The main finding of this study is that the genetic structure of saproxylic beetle populations is significantly shaped by both geography and forest management. By utilizing GLMMs that account for taxonomic non‐independence, alongside dbRDA, we demonstrated that forest management significantly shapes genetic differentiation (F _ST) across species. While geographic distance emerged as the primary driver of genetic structure across both modelling frameworks, forest management exerted a distinct, independent effect on connectivity. While we initially hypothesized that rare old‐growth specialists would exhibit heightened genetic distinctiveness due to habitat fragmentation, our model indicated that baseline abundance alone does not significantly alter population structure. Instead, the starkest contrast in regional connectivity is driven by functional dynamics: rare specialists and common generalists alike maintain significantly higher levels of genetic distinctiveness (differentiation) across forest management types compared to eruptive pest species, whose cyclical population explosions (outbreaks) prompt widespread gene flow and erode regional isolation.

4.1. West‐To‐East Differentiation of Populations

The findings of this study indicate that geographic distance explains a substantial proportion of genetic variation in many taxa, especially common species. Genetic structure, as shown by DAPC and admixture (by region), revealed strong east–west clustering in six species (three common: S. pectinicornis, B. reticulatus and S. melanura ; one pest: P. cyanea ; two rare: C. cinnaberinus and N. haemorrhoidalis ), supporting the first hypothesis that westernmost and easternmost forests are inhabited by significantly differentiated populations. This geographic structuring is consistent with classic IBD patterns (van Strien et al. 2015; Hancock et al. 2024), reflecting limited gene flow over long distances. This aligns with previous work on saproxylic beetles, which revealed strong spatial genetic structure, especially among niche‐specialized and less vagile species (Drag et al. 2015; Eberle et al. 2021). For example, Oleksa et al. (2013, 2015) found that two cavity‐associated beetles ( E. ferrugineus and Osmoderma barnabita Motschulsky, 1845) showed strong kin‐based spatial autocorrelation. Although Oleksa et al. (2013) examined spatial genetics on a local scale, this pattern parallels our results, where, the effect is stronger in these common taxa rather than in rare species (except N. haemorrhoidalis ). Moreover, the degree of geographic structuring observed appears to be closely linked to the species' dispersal capabilities and ecological requirements. For example, the weak genetic structure in pest species like I. typographus and I. acuminatus is consistent with their high mobility and ability to colonize widely distributed resources. In contrast, common species such as S. pectinicornis and B. reticulatus , despite being abundant, showed strong IBD. This suggests that these species may be limited to mature forest patches (e.g., due to limited flight range or specific feeding habitats), making even common taxa vulnerable to regional fragmentation. The rare species (C. cinnaberinus, N. haemorrhoidalis ) exhibited intermediate structure, which may reflect a balance between their specialized habitat needs and the long‐term stability of their remaining populations. Reviews of saproxylic beetle genetics emphasize that limited dispersal and patchy deadwood habitats favour IBD, especially in older forests (Komonen and Müller 2018; Kajtoch et al. 2022). For example, across Europe, Rosalia alpina (Linnaeus 1758) exhibited geographically coherent clades and a significant IBD pattern consistent with restricted dispersal and dependence on older forest deadwood during and after post‐glacial dynamics (Drag et al. 2015). Additionally, populations of Osmoderma eremita, show limited inter‐tree movement and metapopulation dynamics over decades, conditions that promote fine‐scale genetic structure and IBD in a stable substrate (Ranius 2000). While most species in our study exhibited east–west isolation, R. inquisitor and E. ferrugineus showed a different pattern, with north–south clustering and partial mixing across latitude. This anisotropic structure suggests that gene flow is directionally constrained, likely by a combination of current and historical ecological processes that align with latitude rather than longitude. Several mechanisms could explain the emergence of this (north–south) genetic pattern. First, a post‐glacial recolonization legacy can plausibly generate north–south isolation if lineages expanded along distinct latitudinal routes after the last glacial maximum, then met in secondary contact zones (Drag et al. 2015; Cox et al. 2019). Such histories can leave shallow but persistent genomic clusters that run, north–south even when contemporary dispersal is not overtly limited. Second, isolation by environment where latitude covaries with season length, temperature and humidity, shapes wood‐decay dynamics, host‐tree composition and phenology; key factors for saproxylic beetles (Oleksa et al. 2013; Kajtoch et al. 2022). For R. inquisitor , a conifer‐associated longhorn, latitudinal shifts in conifer dominance and stand age (and thus decay stage of deadwood) could similarly structure populations along latitude more than longitude (Victorsson 2012; Skrzecz et al. 2016). For E. ferrugineus, well‐documented latitudinal gradients in veteran tree continuity can restrict north–south connectivity even across modest geographic distances (Musa et al. 2013; Andersson et al. 2014), with moderate management‐driven structure between primaeval protected and logged habitats, suggesting additive effects (latitude‐linked habitat quality/continuity and local stand history). Some species showed weak IBD, likely because our sampling covered only a limited part of their ranges. Previous studies that detected IBD (e.g., C. cinnaberinus; Sikora et al. 2023) examined much larger spatial scales, where geographic distance has a stronger effect on genetic differentiation.

Overall, our results confirm that geographic distance and functional connectivity (the degree of regional isolation) are the main factors shaping population genetic differentiation, with limited but detectable anisotropies in certain taxa. The observed east–west and north–south genetic clustering indicates that peripheral populations represent distinct genetic units shaped by both geographic distance and historical landscape processes. Central European forests may serve as a transient zone in the genetic structuring of saproxylic beetle populations.

4.2. The Impact of Forest Status on Genetics

The contribution of forest management to genetic differentiation was ambiguous. An effect of management was detected, at least in some species, by the MMRR, DAPC, and admixture analyses, but was generally not supported by dbRDA. This suggests that forest management explains only a small proportion of the variation in genetic differentiation. Nevertheless, this conclusion is strongly supported by our pairwise GLMM, which demonstrated that forest management has a highly significant, independent effect on genetic connectivity. This divergence highlights the complementary, scale‐dependent strengths of the two statistical architectures. The dbRDA operates globally across the entire multivariate distance matrix, making it highly sensitive to broad‐scale landscape configurations like regional geography, but potentially less capable of isolating subtle environmental signals when species exhibit massive, compounding variations in baseline genetic drift. In contrast, the pairwise GLMM utilizes a robust nested taxonomic random intercept structure paired with crossed population intercepts. By explicitly partitioning out population‐specific sampling noise and background species‐specific life‐history discrepancies such as the massive baseline variance seen in T. formicarius, the GLMM successfully unmasked a subtle but persistent signature of forest management that was otherwise overwhelmed by geographical noise in the global ordination space. Thus, the overall second hypothesis that populations from primaeval forests are more diverse than those in commercial stands receives partial support, with a subset of species—particularly rare and common old‐growth forest specialists—showing clear genetic structure across management types, especially between primaeval protected and logged. Yet in many taxa—notably common species and pests that are either generalists or adapted to tree species favoured in commercial forests—management differences do not result in strong genetic structuring.

This pattern highlights that even common species can exhibit management‐driven genetic structure when they are ecologically specialized on microhabitats widely represented in primaeval habitats, such as late‐decay substrates or large‐diameter deadwood (Seibold et al. 2016; Komonen and Müller 2018). Although these species are broadly distributed, they often depend on old‐growth or continuity‐dependent resources, such as hollow trees and coarse woody debris, which are scarce in intensively managed forests (Stokland et al. 2012; Parisi et al. 2018). Consequently, their effective dispersal may be limited by habitat connectivity rather than geographic range. Similar patterns have been documented for other nominally widespread saproxylic beetles showing fine‐scale genetic differentiation linked to habitat continuity (Oleksa et al. 2013; Seibold et al. 2016).

Therefore, the observed management effect in these common species is not paradoxical but reflects ecological specialization within widespread distribution; a scenario in which general geographic ubiquity masks dependence on rare microhabitats (Seibold et al. 2016). In contrast, the more disturbance‐tolerant conifer taxa such as R. inquisitor (conifer‐generalist), I. typographus (spruce specialist), I. acuminatus and P. cyanea (pine specialists) showed wide overlap across management regimes, consistent with their ability to exploit younger or disturbed habitats and disperse effectively across landscapes (Netherer et al. 2021; Papek et al. 2024). Thus, our results align with ecological expectations once differences in habitat specialization and effective dispersal are considered. This raises the question of why the management effect might be muted in some taxa.

First, while it is possible that the limited sample sizes for certain rare specialists or the geographic clustering of forest types (as discussed in ‘Study Constraints’) reduced our power to detect subtle management‐driven differentiation, the overall pattern remains compelling. Even in species with robust sampling, the geographic signal consistently overshadowed management. This suggests that for many saproxylic beetles, the historical and regional context of the landscape plays a more significant role in shaping genetic structure than contemporary forest use. Second, if dispersal is high, gene flow may rapidly homogenize populations across management types. For example, the pest I. typographus has likely gain connectivity across forest landscapes (Netherer et al. 2021), and the rare C. cinnaberinus shows low genetic differentiation across its European range (Sikora et al. 2023), probably due to recent expansion. Third, if commercial and primaeval forests (as well as protected and unprotected stands), are relatively close in space and not completely isolated, landscape connectivity can maintain gene flow. In this case, management may act more as a fine‐scale environmental filter rather than a genetic barrier, consistent with landscape genetic theory, which suggests that the permeability of nearby habitat mosaics often outweighs categorical habitat labels (Lowe et al. 2015; Winiger et al. 2023). Fourth, regarding the timescale of differentiation: genetic structure develops gradually, particularly in long‐lived organisms or those with overlapping generations; if management changes are recent relative to generation times, the genetic signal may lag (Landguth et al. 2010; Lowe et al. 2015). Fifth, if deadwood continuity and structural complexity are maintained, even in commercial forests, the genetic consequences may be reduced (Seibold et al. 2015, 2016). While a detailed analysis of specific landscape corridors and structural connectivity is beyond the scope of this study, our results suggest that the permeability of the managed forest matrix is a key determinant of regional genetic structure. Future research focusing on high‐resolution forest structure and specific movement paths will further clarify how local management decisions translate into landscape‐level connectivity. Sixth, our sampling design contrasted four discrete forest‐management types, whereas both environmental conditions (e.g., deadwood availability) and genetic structure are continuous. Therefore, detecting relationships between forest naturalness and genetic diversity may require a different analytical strategy. Practical constraints also limited sampling; even common species were largely absent from commercial forests lacking deadwood (typical in Central Europe; Vítková et al. 2018), while conifer‐specialized pests were scarce in broadleaved protected forests, as deadwood quality and connectivity strongly influence saproxylic populations (Stokland et al. 2012; Parisi et al. 2018). Commercial forests retaining sufficient structural complexity may still avoid severe genetic isolation.

In rare species, the moderate management‐driven structure suggests that these taxa may be more sensitive to differences in forest history. For all studied rare beetle taxa ( B. schneideri, C. cinnaberinus, E. ferrugineus, N. haemorrhoidalis ), some analyses showed the greatest distinctiveness in populations from primaeval stands and from other forests (logged stands and nature reserves located within commercial forests). Moreover, in all these species, populations from protected and non‐protected primaeval forests are at least partially differentiated. These findings are consistent with expectations that populations of rare (threatened, relict) species are strongly affected by forest naturalness, reflected by, for example, the history of management or conservation. This also applies to some still common species that are often more associated with old‐growth forests (e.g., S. pectinicornis, B. reticulatus ). However, small sample sizes, lower connectivity (limited gene flow) and potentially historical declines (caused by past forest management and a deficiency of deadwood in non‐protected forests), weaken the signal (Andersson et al. 2014; Sikora et al. 2023).

In our results, the strongest management‐driven structure appeared in those species that already showed strong geographic structuring, suggesting that management reinforces existing isolation by further limiting connectivity. For example, in S. pectinicornis and B. reticulatus , primaeval sites may act as refugia with a continuous supply of deadwood, while commercial forests increase genetic drift (Jonsson et al. 2003; Parisi et al. 2018). Thus, management effects are present but modest and context dependent, with geography and dispersal limitation remaining the dominant drivers of genetic structure (Lowe et al. 2015).

Interestingly, the AMOVA indicated significant differentiation between management types, while the dbRDA showed only weak or non‐significant trends. This discrepancy is likely explained by the spatial confounding of our study design. Because management types are geographically clustered (e.g., primaeval in the SE vs. commercial in the SW, while both types are present in the NE), the AMOVA may be capturing broader regional differences as a ‘management’ effect. In contrast, our partial dbRDA specifically partitions out the variance explained by geographic distance first (Legendre and Legendre 2012). The lack of significance in the dbRDA suggests that once the dominant signal of geography is statistically controlled, management itself explains a negligible portion of the remaining genetic variance consistent with findings that landscape configuration often overrides local habitat effects in highly mobile or diverse taxa (Wang 2013).

4.3. Implications for Conservation and Management

Strong geographic structuring in multiple species (common: S. pectinicornis, B. reticulatus and S. melanura ; and rare: E. ferrugineus , N. haemorrhoidalis ) suggests that regional populations may function as distinct management or even evolutionarily significant units (Knutsen et al. 1999; Zamoroka et al. 2022). In contrast, C. cinnaberinus likely shows weaker differentiation due to recent recovery (Bełcik et al. 2019; Sikora et al. 2023), while B. schneideri remains geographically restricted. Species with management‐driven structure highlight that commercial forest populations may experience reduced gene flow and smaller effective population sizes, whereas primaeval forests act as genetic sources due to higher connectivity and continuity of deadwood habitats (Paillet et al. 2010; Haeler et al. 2024). Such patterns align with concerns about genetic drift and sink dynamics in fragmented managed landscapes (Pulliam 1988; Aravanopoulos 2018). Therefore, conservation should: (1) maintain connectivity across regions, (2) protect and restore old‐growth forest stands within commercial stands, (3) prioritize species showing both strong genetic structure and management sensitivity (Kajtoch et al. 2022; Krovi et al. 2024).

Finally, distinct ancestry blocks defined by geography and management indicate cryptic subdivisions that warrant recognition in planning. Integrating geography and forest management into management units can guide monitoring and management, protect source populations and identify gene‐flow bottlenecks. Our dataset thus provides actionable genetic evidence to support targeted deadwood retention and improved forest management for saproxylic beetle conservation.

4.4. Study Constraints

Our study design faced certain constraints common in large‐scale ecological genomics. First, sample sizes varied across the studied species and management types, reflecting the inherent rarity of certain saproxylic specialists (e.g., E. ferrugineus ). However, we used two distinct approaches to ensure the robustness of our conclusions. Our sensitivity analyses demonstrated that patterns of genetic differentiation (pairwise F _ST) remained highly consistent with the full dataset. Second, the historical distribution of old‐growth forests in the study region led to a degree of spatial clustering of management types, with primaeval forests concentrated in the SE and commercial forests primarily in the SW, while both types were present in the NE. This spatial imbalance could potentially confound the effects of forest management with geographic distance. However, by employing partial dbRDA to partition unique variance, we were able to statistically isolate the geographic signal from management effects. The fact that the unique contribution of management remained low across most species, despite the above‐mentioned geographic clustering, suggests that regional isolation is the overriding force shaping the genetic architecture of these beetles in the studied landscape rather than sampling artefacts or spatial bias.

5. Summary and Conclusion

Our genomic study of 12 saproxylic beetle species shows that geographic distance is the primary driver of genetic differentiation, while forest management has secondary species‐specific effects (Table 4). East–west structuring supports our first hypothesis, whereas management‐related divergence is apparent only in certain taxa. Nevertheless, identifying isolated or distinct populations offers valuable conservation guidance. Conservation efforts should priortize maintaining regional connectivity and continuous deadwood resources. Primaeval and old‐growth forests remain crucial refugia for low‐dispersal specialists, while generalists maintain gene flow across managed landscapes. These findings are consistent with broader evidence that spatial processes dominate genetic structuring in saproxylic beetles (Komonen and Müller 2018; Kajtoch et al. 2022). Our integrated genomic approach underscores the importance of incorporating genetic monitoring into forest biodiversity management.

Author Contributions

N.R.A., R.S.K.: conception and design, data analysis, interpretation of data, writing. A.W.: laboratory works, approval of manuscript. R.P., M.K., T.J., A.S.: conception and design, funding acquisition, sampling, interpretation of data, approval of manuscript. A.O.: sampling, interpretation of data, approval of manuscript. T.S.: data analysis, interpretation of data, approval of manuscript. M.O.: conception and design, funding acquisition, laboratory works, interpretation of data, approval of manuscript. Ł.K.: conception and design, funding acquisition, sampling, interpretation of data, writing.

Funding

This study was funded by a grant from the National Science Centre, Poland (UMO‐2021/43/B/NZ9/00991, PI—Łukasz Kajtoch) and the grant (UMO 2011/03/B/NZ9/03103, PI—Andrzej Oleksa).

Disclosure

Benefit‐sharing section: The whole sampling was conducted in Poland; therefore, the Nagoya Protocol is not applicable. Benefits from this research accrue from the sharing of our data and results on public databases as described above.

Ethics Statement

Collection of protected beetle species and in protected areas was executed according to permissions granted by Polish Ministry of Environment (DOP‐1.61.61.2021.TP 1516071.4994962.3977511 and DOP‐1.61.61.2021.TP 1516071.4994586.3977515), General Directorate of Environmental Protection (DZP‐WF.6401.68.2021.AS) and Regional Directorates of Environmental Protection in Białystok (WPN.6205.74.2021.MM), Rzeszów (WPN.6205.116.2021.MP.2 and WPN.6205.114.2021.ŁL.2), Kielce (WPN.I.6205.1.49.2021.EJP.2) and Wrocław (WPN.6205.143.2021.AR).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Empirical distribution of linearized F _ST raw dataset across all 6883 population pairs.

Figure S2: DHARMa Residual Diagnostics for the Final Zero‐Inflated Gamma GLMM (Linearized F _ST). Randomized quantile residual diagnostic plots generated via the DHARMa package for the final zero‐inflated Gamma generalized linear mixed model (GLMM). Top‐Left: Uniform Q‐Q plot comparing observed versus expected residuals. Top‐Right: Scaled residuals plotted against rank‐transformed model predictions, with empirical quantile lines tracking the ideal horizontal 0.50 baseline. Bottom‐Left: Frequency histogram of standardized residuals mapping overall uniformity. Bottom‐Right: DHARMa zero‐inflation simulation test based on 1000 bootstrap replicates (p = 0.984).

Figure S3: Genetic structure represented by ADMIXTURE based on four management types (logged, nature reserve, primaeval not protected and primaeval protected), presented for four beetle species assigned to the category ‘Common (Non‐Pest)’.

Figure S4: Genetic structure represented by ADMIXTURE based on four management types (logged, nature reserve, primaeval not protected and primaeval protected), presented for four beetle species assigned to the category ‘Common (Pest)’.

Figure S5: Genetic structure represented by ADMIXTURE based on four management types (logged, nature reserve, primaeval not protected and primaeval protected), presented for four beetle species assigned to the category ‘Rare (Non‐Pest)’.

Figure S6: Genetic structure represented by pairwise F _ST heatmap among forests' geographic regions (NE, north east; SE, south east; SW, south west).

Figure S7: Genetic structure represented by pairwise F _ST heatmap among four forests' management types (logged, nature reserve, primaeval not protected and primaeval protected).

Figure S8: Sensitivity analysis of genetic differentiation patterns in three species; Phaenops cyanea (pest), Schizotus pectinicornis (common) and Elater ferrugineus (rare). Heatmaps of pairwise F _ST values comparing results obtained from the full dataset and a subsampled dataset restricted to a maximum of n = 10 individuals per site. This comparison was performed to evaluate the impact of uneven sample sizes on genetic structure estimates.

Table S1: Sampling sites of the studied 12 beetle species with corresponding geographic coordinates, forest locations and management type.

Table S2: Sample sizes (number of collected beetles) across the experimental groups, based on geography regions (NE, north east; SE, south east; SW, south west) and based on management types (L, logged, Pn, primaeval not protected; Pp, primaeval protected; R, reserve).

Table S3: Evaluated candidate mixed‐effect model architectures and AIC‐driven parsimony selection rankings. Predictors include Management (MGMT_comp), Geography (GEO_comp), Pest status (Pest_status [pest/non‐pest]) and Abundance (common/rare).

Table S4: Fixed effect estimates and 95% confidence intervals from the optimal zero‐inflated Gamma GLMM predicting pairwise genetic differentiation (F _ST) framework (N = 6883). Parameter estimates (β), standard errors (SE), Wald z‐values and corresponding p‐values for the conditional (Gamma) and zero‐inflation components of the top‐performing model (M1). Confidence intervals (95% CI) are equal‐tailed and derived using a Wald z‐distribution approximation. Predictor coefficients are presented on the native log‐link scale; negative estimates in the conditional model indicate a reduction in pairwise genetic differentiation (F _ST).

Data S1: Executable R code script for the optimal zero‐inflated Gamma generalized linear mixed model layout using glmmTMB, multicollinearity diagnostics, and validation using DHARMa.

Data Availability Statement

Raw sequencing data (FASTQ files) and associated sample metadata have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession PRJNA1399704 (https://www.ncbi.nlm.nih.gov/sra/PRJNA1399704).

Filtered single‐nucleotide polymorphism (SNP) datasets generated from double‐digest RAD sequencing (ddRAD‐seq) for multiple saproxylic beetle species are available as species‐specific Variant Call Format (VCF) files in RepoOD of the Institute of Systematics and Animal Evolution PAS (https://doi.org/10.18150/MYUQJG).