🧬 biology

Detectability of landscape-genetic signal in fragmented landscapes: habitat area, configuration, dispersal, and the dual role of corridors

Using individual-based simulations, this study demonstrates that the detectability of landscape-genetic signals is governed by complex interactions between habitat area, configuration, and dispersal ability, revealing that while corridors consistently increase genetic diversity, they often reduce the statistical detectability of landscape resistance by eroding genetic differentiation gradients.

Original authors: Samuel A. Cushman

Published 2026-08-31

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Samuel A. Cushman

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

In the quiet science of landscape genetics, researchers try to read the history of a species written in its DNA. They look for patterns in how genetic differences spread across a map, hoping to see how the land itself—forests, fields, rivers, and roads—shapes the movement of animals and the flow of their genes. For decades, a central debate has raged over what matters most: the sheer amount of habitat available, or the way that habitat is arranged. Does a landscape with the same total amount of forest but broken into many small, scattered pieces affect animals differently than one with a few large, connected blocks? A related, yet less explored question asks what happens when we try to fix broken landscapes. If we build narrow strips of habitat, known as corridors, to link isolated patches, do we simply help animals move, or do we also change the very genetic signature that scientists use to study them?

This question matters because conservationists often rely on genetic studies to decide where to protect land or build corridors. If a study fails to find a link between the landscape and the animals' genes, it might mean the animals don't care about the landscape, or it might mean the landscape is so connected that the signal is hidden. Understanding when and why these genetic signals appear or disappear is crucial for making accurate decisions about the natural world.

To untangle these complex interactions, a researcher named Samuel Cushman ran a massive series of computer simulations. Instead of tracking real animals in the wild, which would take decades and be impossible to control, he created a virtual world. He built 1,500 different digital landscapes, each with a specific mix of features. In every single simulation, the total amount of habitat was kept exactly the same, but the way it was arranged changed. Some landscapes had habitat scattered in tiny, isolated islands; others had it clumped into large, solid blocks. He also varied how far the virtual animals could travel in a single generation, from short hops to long journeys. Finally, in half of the simulations, he added narrow corridors connecting the nearest patches of habitat, carefully removing an equal amount of habitat from the edges of the existing patches so the total area never changed. This strict control allowed him to isolate the effect of the corridors themselves, separate from simply adding more habitat.

The results revealed a landscape of genetic signals that is far more nuanced than a simple "more is better" or "corridors always help" story. The simulations showed that the ability to detect a relationship between the landscape and the animals' genetics is not constant; it rises and falls depending on the specific mix of habitat amount, arrangement, and animal movement. The strongest genetic signals appeared in the middle ground: landscapes with a moderate amount of habitat, arranged in a moderately clumped pattern, where the animals had limited ability to travel. In these conditions, the barriers of the landscape were just strong enough to create distinct genetic differences between groups, but not so strong that the populations died out or became completely isolated.

At the extremes, the signal faded. When the habitat was very scarce and scattered, the small populations became so isolated that random chance, rather than the landscape structure, dictated their genetic makeup. When the habitat was abundant or the animals could travel very far, the groups mixed so thoroughly that the landscape barriers disappeared from the genetic record, leaving no trace of the terrain's influence. This creates a "sweet spot" for detection, a window where the landscape's shape is clearly visible in the DNA.

The introduction of corridors produced a surprising and dual effect. On one hand, the corridors acted as a lifeline. In landscapes where the patches were large but isolated, the corridors allowed animals to move between them, boosting the genetic diversity within each patch and preventing the loss of genetic variation. This is the good news for conservation: the corridors worked to save the populations from genetic decline.

On the other hand, these same corridors made it harder for scientists to see the landscape's influence in the genetic data. By connecting the patches, the corridors smoothed out the differences in how easily animals could move across the terrain. The distinct barriers that once created a clear pattern of genetic separation were eroded. In the simulations, adding corridors reduced the strength of the landscape-genetic signal in the majority of cases, especially when the animals were capable of traveling longer distances. The more the animals could move, the more the corridors homogenized the landscape, effectively hiding the very resistance the scientists were trying to measure.

Perhaps the most critical finding is that these two effects—the rescue of genetic diversity and the erosion of the landscape signal—do not happen in the same places at the same time. The boost in genetic diversity happened most strongly in landscapes with large, clumped patches where populations were struggling with isolation. The weakening of the landscape signal happened most strongly in landscapes with a moderate amount of habitat and moderate clumping, where the original barriers were most distinct. Because these two processes peak in different parts of the landscape, they do not cancel each other out in a simple, predictable way. A corridor might save a population's genetic health in one scenario while simultaneously making the landscape's influence invisible in another.

This research suggests that when a scientist looks at a landscape and finds no genetic signal, it does not necessarily mean the landscape has no effect. It might mean the landscape is so well connected, perhaps by natural features or human-made corridors, that the animals move freely enough to erase the genetic boundaries. Conversely, a strong signal might not mean the landscape is the only thing shaping the animals, but rather that the populations are in a specific state of balance where the landscape's limits are just right to be seen.

The study concludes that understanding landscape genetics requires looking at the whole picture: the amount of habitat, how it is arranged, how far the animals can go, and whether corridors exist. It is not enough to ask if a corridor helps an animal move; one must also understand how that movement changes the genetic story the animal tells. The corridors are doing real work, saving populations and mixing genes, but in doing so, they can also blur the map that scientists use to understand the world. Recognizing this dual nature is essential for interpreting genetic data correctly and for designing conservation strategies that truly reflect the complex reality of the natural world.

Technical Summary: Detectability of Landscape-Genetic Signal in Fragmented Landscapes

Problem Statement
The paper addresses a long-standing, contested question in landscape ecology: whether habitat fragmentation exerts effects on genetic differentiation beyond the simple reduction of habitat area. A complementary, less-developed question is how the introduction of corridors connecting isolated patches alters the strength and statistical detectability of landscape-genetic signals, and how this depends on dispersal ability. While previous literature has documented isolation-by-resistance effects, it has largely treated the detectability of these effects as a fixed property rather than a function of landscape pattern, dispersal, and population size. The study seeks to operationalize a multi-factor interaction framework (habitat amount × configuration × dispersal) to determine if and when landscape-genetic signals are recoverable, specifically examining the dual role of corridors in both demographic rescue and inferential signal erosion.

Methodology
The study utilizes a four-factor, area-conserved factorial design based on individual-based simulations using CDPOP (Cushman, 2006; Landguth & Cushman, 2010).

Experimental Design: 1,500 total simulations comprising 150 unique combinations replicated 10 times.
- Habitat Area: 5 levels (10% to 50% of a 100×100 landscape).
- Habitat Clumping: 5 levels (Gaussian smoothing $\sigma$ = 3 to 7).
- Connectivity: 2 levels ("base" vs. "linked" where nearest-neighbour corridors are added). Crucially, total habitat area is held strictly constant; corridor pixels are added by stochastically eroding edge pixels of existing patches.
- Dispersal Threshold: 3 levels (50, 100, and 150 cost units).
Simulation Parameters: 100 individuals initialized at sample points, simulated for 200 generations with a carrying capacity ( $K$ ) of 100. Genotypes consist of 10 unlinked loci with 5 alleles each, no mutation, and density-dependent exponential growth.
Genetic Analysis:
- Pairwise Bray–Curtis genetic distance, Euclidean distance, and cost-distance matrices were computed for surviving individuals.
- Statistical Tests:
  1. Mantel Tests: Correlations between genetic distance and cost-distance ( $r_{cost}$ ) vs. Euclidean distance ( $r_{Euclidean}$ ). The metric $\Delta r = r_{cost} - r_{Euclidean}$ measures the gain of the resistance hypothesis over the geographic null.
  2. MLPE Mixed Models: Linear mixed models regressing genetic distance on cost-distance with random effects for population structure. Metrics extracted include the cost coefficient ( $\beta$ ), marginal $R^2$ (variance explained), and likelihood-ratio test $p$ -values for detection power.
Analysis: Four-way factorial ANOVA was used to analyze effects on $\Delta r$ , $\beta$ , and marginal $R^2$ . Inference relies on partial $\eta^2$ effect sizes and interaction structures rather than $p$ -values, given the large simulation ensemble.

Key Contributions

Non-Monotonic Detectability: The study demonstrates that landscape-genetic detectability is not a linear function of fragmentation but follows a "two-sided window." Signal strength peaks at intermediate habitat area, intermediate clumping, and low-to-moderate dispersal.
The Dual-Channel Corridor Effect: The paper identifies and mechanistically separates two opposing effects of adding corridors:
- Variance-Supply Channel (Positive): Corridors reduce drift in isolated patches, increasing within-population heterozygosity ( $H_o$ ). This effect is strongest in highly clumped landscapes with large, compact patches.
- Spatial-Heterogeneity Channel (Negative): Corridors equalize gene flow across the resistance gradient, eroding the spatial heterogeneity required to generate a cost-distance signal. This effect is strongest in landscapes with intermediate clumping and area where the resistance gradient was previously strongest.
Dispersal-Dependent Erosion: The reduction in landscape-genetic signal strength ( $R^2$ ) caused by corridors is dependent on dispersal ability. Higher dispersal thresholds result in greater signal erosion (34–35% reduction at 100–150 cost units vs. 16% at 50 cost units) because corridors become functionally shorter relative to movement capabilities, more effectively homogenizing the landscape.
Decoupling of Biological and Inferential Outcomes: The study shows that while corridors simultaneously increase genetic diversity ( $H_o$ ) and decrease landscape-genetic signal strength ( $R^2$ ), these two effects are uncorrelated at the combination level ( $|r| < 0.13$ ). They operate in different regions of the area $\times$ clumping parameter space.

Results

Detectability: MLPE detected a significant cost effect in 96.7% of runs. However, the magnitude of the effect varied systematically. The dominant driver of variation in both $\beta$ and marginal $R^2$ was the interaction between habitat area and clumping ( $\eta^2_p = 0.239$ and $0.226$, respectively).
Signal Strength: Maximum signal strength occurred at intermediate area (20–40%), intermediate clumping ( $\sigma=4–5$ ), and low dispersal (50 cost units). Signal collapsed at extremes: high fragmentation led to drift-dominated isolation (Case IV), while high area/high dispersal led to panmixia.
Corridor Impact: Adding corridors reduced marginal $R^2$ in 58 of 75 area $\times$ clump $\times$ dispersal combinations (average reduction of 16–35% depending on dispersal). Conversely, corridors increased $H_o$ in 52 of 75 combinations.
Interaction Structure: The linkage $\times$ dispersal interaction was significant for both $\beta$ and $R^2$ , confirming that the inferential impact of corridors cannot be separated from the species' dispersal ability. The four-way interaction (Area $\times$ Clump $\times$ Linkage $\times$ Dispersal) was marginally significant for $\beta$ , reinforcing that no single factor summarizes the system.

Significance and Claims
The paper claims that the detectability of landscape-genetic relationships is governed by a "limiting factors" framework extended to include a two-channel mechanism.

Inferential Implications: A failure to detect a landscape-genetic signal does not necessarily imply a lack of biological response to the landscape. Instead, it may indicate that the landscape is either too fragmented (drift dominates) or too connected (panmixia dominates) to generate a detectable resistance gradient.
Corridor Interpretation: The "erosion" of the landscape-genetic signal by corridors should not be interpreted as a failure of the corridor to function biologically. On the contrary, the signal erosion is a direct inferential consequence of the corridor successfully homogenizing gene flow and rescuing genetic diversity. The two outcomes (higher diversity, lower signal) are concurrent but mechanistically distinct and spatially dissociated.
Methodological Shift: The study argues for moving beyond site-specific inferences to a multi-factor interaction framework. The effect of habitat configuration on genetic differentiation is real but is irreducibly an interaction effect dependent on habitat amount, dispersal ability, and connectivity architecture.
Practical Application: Researchers designing landscape-genetic studies must calibrate power analyses to the specific location of their study system within the area $\times$ clumping $\times$ dispersal space. A species may show strong signals in transitional fragmentation regimes but weak signals in highly intact or severely fragmented landscapes, regardless of its biological traits.

The paper concludes that landscape-genetic tests are not fixed instruments but are coupled to the ecological context; the "map" of detectability across the parameter space is the answer to understanding when and why these signals are recoverable.

Technical Summary: Detectability of Landscape-Genetic Signal in Fragmented Landscapes

More like this