Runs of Homozygosity and Inferences in Wild Populations

ABSTRACT

Long homozygous chromosome segments are known as runs of homozygosity (ROH); these reflect patterns of identity by descent and can be used to measure individual inbreeding, map recessive traits, and reconstruct demographic histories. Here, we review some key considerations with ROH detection and the inferences pertaining to inbreeding and demographic analyses in wild populations.

1. What Are Runs of Homozygosity?

When relatives mate, identity by descent (IBD) segments of the genome arise in their offspring. The proportion of an individual's genome that is IBD was traditionally estimated with pedigree data as the inbreeding coefficient F (Wright 1922). Genome‐scale data permit identifying IBD parts of an individual's genome as large homozygous chromosome segments known as runs of homozygosity (ROH; Figure 1a). The summed length of ROH divided by the sequenced autosomal genome size, or F _ROH, has become a popular genomic measure of inbreeding (Table 1). ROH have been studied extensively in humans and domestic animals (e.g. Bosse et al. 2012; Broman and Weber 1999), most often with SNP arrays (Gonzales et al. 2022; Gorssen et al. 2021). Domestic and model species have some inherent advantages when assessing ROH, notably, (i) relatively high levels of IBD and (ii) chromosome‐scale reference assemblies and high‐density SNP panels.

FIGURE 1.

Generation of runs of homozygosity (ROH) in a hypothetical pedigree of a female whose parents are first cousins (a). Examples of how ROH can be used to compare inbreeding values (b) and ROH length distributions (c) by population.

TABLE 1.

Population genetic metrics derived from runs of homozygosity (ROH). Included are the equation, required parameters, and key considerations.

Population genetic metrics	Equation	Variables	Key considerations	Reference
Inbreeding coefficient (F ROH)	$F_{\mathrm{ROH}}=\frac{\sum \mathrm{ROH}}{\text{Genome size}}$	Genome size is total length of autosomal scaffolds screened for ROH (in base pairs)	ROH detection and length is sensitive to method and marker density	McQuillan et al. 2008
Mutational load per ROH	$\frac{{\sum }_1^{n}s}{{\sum }_1^m\mathrm{ROH}\text{length}\left(\mathrm{cM}\right)}$	Selection coefficients (s) for n deleterious mutations of m ROH segments.	ROH segments are broken into size classes. Requires distribution of fitness effects (DFE)	Stoffel et al. 2021b
Time to most recent common ancestor (TMRCA)	$L=\frac{100}{2\times g}$	L is length of ROH in centimorgans, g is no. of generations to TMRCA	ROH detection and length is sensitive to method and marker density. Low precision and potential bias in estimates	Thompson 2013
Effective population size (N e)	$F_{\mathit{ROH},t}=1-{\left(1-\frac{1}{N_e}\right)}^t$	Coalescent time (t) generations for a ROH	Dependent on estimating TMRCA of ROH	Khan et al. 2021

With the accessibility of reduced‐representation and whole genome sequencing (WGS), ROH studies have become more commonplace in the assessment of wild populations, often targeting species of conservation concern (e.g. Dedato et al. 2022; Humble et al. 2023; Khan et al. 2021). ROH detection methods can be grouped into two categories: rule‐based (sometimes referred to as observation‐based) and model‐based approaches. Rule‐based methods, most notably implemented in PLINK (Purcell et al. 2007), identify ROH as consecutive windows of called genotypes that satisfy user‐defined criteria. Users define the number of loci included per window, the degree of window overlap, the number of missing genotypes allowed in a window and an upper limit to the proportion of heterozygous loci allowed in a presumably IBD window (McQuillan et al. 2008). In contrast, model‐based methods typically rely on hidden Markov models, do not use defined windows and can incorporate genotype likelihoods (Bertrand et al. 2019; Narasimhan et al. 2016; Renaud et al. 2019). Some model‐based approach use a logarithm of odds or LOD score for the hypothesis that a contiguous set of loci in an individual is IBD given the allele frequencies and assumed mutation and genotyping error rates (Broman and Weber 1999; Kardos et al. 2023; T. J. Pemberton et al. 2012; Wang et al. 2009). In all approaches, markers must be mapped (i.e., ordered) and be of sufficient density to capture ROH. This is an important consideration in natural populations, as, for example, cross‐amplification of SNP arrays is likely not suitable (Shafer, Miller, and Kardos 2016) and low marker density can result in missed or even artificially merged ROH (Ceballos et al. 2018). Differing SNP densities have been shown to produce different F _ROH values (Samuels et al. 2016), and the required densities vary by ROH detection method (Lavanchy and Goudet 2023).

Lavanchy and Goudet (2023) found using simulations that 22 SNPs/MB were required to reliably detect ROH using the common rule‐based methods. This density is likely to be achieved only via WGS in many natural populations, although some systems and sequencing methods might generate enough variable loci. Identifying ROH is also often dependent on user‐defined parameters (Meyermans et al. 2020) and typically requires moderate to high depth of coverage (Duntsch et al. 2021), but alternative algorithms (Bertrand et al. 2019) and strategies (Cars et al. 2024) for the identification of ROH in low‐coverage sequence data are emerging. Lavanchy and Goudet (2023) found that ROH > 2 MB were usually identified by both rule‐ and model‐based approaches, but variation in depth was not factored into the SNP and ROH calls. Human clinical studies have set some ROH minimum length thresholds (Gonzales et al. 2022) and best practices for SNP array‐based analyses of domestic animals exist (Meyermans et al. 2020), but there lacks a standard accepted approach to detect ROH with WGS data in natural systems. Still, high‐quality WGS data should make it possible to detect ROH and measure F with very little error because an individual can in principle be scored as heterozygous or homozygous at the great majority of positions in the genome (Kardos et al. 2016).

Further, ROH are also being used to map partially recessive deleterious alleles affecting particular phenotypes and fitness components (Quinodoz et al. 2021; Stoffel et al. 2021a). The length of the ROH can be connected to the coalescence time of IBD haplotypes (Thompson 2013), and populations with larger historical effective population size (N _e) typically have shorter ROH, whereas isolated or bottlenecked populations should have longer ROH (Ceballos et al. 2018). Given the connection between genomic health and demographic history, there is growing interest in assaying ROH in natural systems. Here, we discuss how ROH are being used to assess patterns of inbreeding and demographic history in wild populations.

2. ROH, Inbreeding, and Inbreeding Depression

Quantifying inbreeding (F) in free‐ranging populations is a crucial component of many molecular ecology studies, with results potentially informing management and conservation strategies (Hedrick and Kalinowski 2000; Kardos et al. 2016; Keller and Waller 2002). Microsatellite‐based measures of F or those derived from pedigrees were long considered state‐of‐the‐art (Coltman and Slate 2003; Pemberton 2004). However, inbreeding estimates from pedigrees are imprecise and underestimate F because pedigrees do not account for the effects of Mendelian segregation, recombination, and earlier ancestry on estimates of F. Microsatellite‐based measures of F are imprecise because they are typically based on a small number of loci (Kardos et al. 2016; Knief et al. 2015). Summing ROH over the assayed autosomal genome produces the inbreeding coefficient F _ROH (Table 1), which can be compared between populations (Figure 1b), and simulations showed that this value has the strongest relationship with the true F (Forutan et al. 2018). Additional advantages are that ROH estimates can be performed on one individual and that F _ROH appears correlated to higher mutational load (Kardos et al. 2023; Keller, Visscher, and Goddard 2011; Szpiech et al. 2013; von Seth et al. 2021), although this is dependent on demographic history (Wootton et al. 2023). Clear links of F _ROH to inbreeding depression further expand potential applications (Paul, D'Ambrosio, and Phocas 2022; Stoffel et al. 2021a). In killer whales, for example, Kardos et al. (2023) used ROH as a covariate in survival and fecundity models to show support for inbreeding depression impacting population dynamics (see also example in hihi—Duntsch et al. 2021). Additional impacts were shown by Stoffel et al. (2021b) where simulations captured the relationship between ROH size and density of homozygous deleterious alleles (Table 1), with longer ROH being enriched for deleterious alleles (Szpiech et al. 2013). Such analyses empirically connect genetic load and purging to ROH.

Homozygosity mapping is a tool that has been employed in medicine and agriculture to identify loci responsible for recessive disease phenotypes (Charlier et al. 2008; Kijas 2013; Lander and Botstein 1987). The principle of this approach is that the causative alleles underlying recessive phenotypes are expected to be confined to a ROH. Few studies of non‐model organisms have employed homozygosity mapping. In white‐tailed deer, Cars et al. (2024) found ROH restricted to animals manifesting known recessive traits (leucism and jaw malocclusions), with genes located in the ROH the same as known causative genes in model organisms. Similarly, studies of island feral horses (Colpitts, McLoughlin, and Poissant 2022) and wild tigers (Zhang et al. 2023) detected population‐specific ROH with potential phenotypic impacts inferred via gene ontology analysis. This latter approach can be applied when phenotypic data sets are absent, although genetic drift, purifying selection, and story‐telling (Pavlidis et al. 2012) are important to consider when interpreting such ROH. There is also interest in using ROH to identify deleterious alleles affecting fitness components. In feral Soay sheep, Stoffel et al. (2021a) used an ROH‐based association mapping analysis of ~6000 samples and identified loci, with both large and small effects, associated with lamb survival. In a study of ~1400 hatchery trout, Paul, D'Ambrosio, and Phocas (2022) detected ROH impacting reproduction that also overlapped known quantitative trait loci. Similarly, Duntsch et al. (2021) screened ~1500 individuals and found ROH in three genomic regions were associated with lifetime reproductive success. ROH‐based mapping of complex traits requires large data sets, in terms of both sample size and marker density, to obtain sufficient power (see Clark et al. 2019).

Concerns have emerged that ROH might not accurately reflect IBD and that standard rule‐ based methods often omit ROH < 1 MB (Yengo et al. 2018); here, users should be aware that user‐defined size thresholds do not apply to model‐based inferences of ROH (Bertrand et al. 2019). When gene mapping, shared regions of ROH among individuals are most often because of positive selection in populations with high N _e (Hewett et al. 2023). There is also a negative correlation between recombination rate and ROH density, with power to detect ROH higher in areas of low recombination; (Kardos, Qvarnström, and Ellegren 2017). Accordingly, smaller ROH will persist in outbred populations, which is why human clinical work often sets a minimum threshold of 3 MB of detection (Gonzales et al. 2022); most assessments in free‐ranging populations have not set such a large threshold. F _ROH is clearly a useful metric for natural populations, and if model‐based approaches or high‐depth sequencing is used, ROH biases should be minimised. Isolated and low N _e populations should be the focus of ROH‐based gene mapping studies to identify loci involved in inbreeding depression in the wild.

3. ROH and Demographic Inference

Historical demographic events leave signatures of IBD in the genomes of contemporary individuals. Relatively longer ROH arise from more recent common ancestors of parents and are thus informative of relatively contemporary demographic history. Deeper historical events such as population bottlenecks produce more and shorter ROH in contemporary individuals (Ceballos et al. 2018; Foote et al. 2021). Regions of IBD can also be used as a summary statistic in Approximate Bayesian Computation models aimed at recovering detailed demographic histories (Sanchez et al. 2021). The length of the ROH has a direct empirical connection to the Time to Most Recent Common Ancestor (TMRCA; Thompson 2013—see Table 1). Similarly, Khan et al. (2021) derived N _e estimates from F _ROH (Table 1). Summarising and comparing the abundance and ROH distributions by length thus have a link to demographic histories and can be readily visualised alongside F _ROH estimates (Figure 1c).

The ROH patterns can help to determine whether contemporary inbreeding is due to recent versus deeper historical demography. For example, the TMRCA estimates derived from ROH largely corresponded to the colonisation of wolves into Scandinavia in the early 1980s (Kardos et al. 2018). In contrast, Cars et al. (2024) found that the TMRCA of the ROH underlying two recessive traits appeared to predate island colonisation, suggesting a founder effect driving the prevalence of the trait. Higher proportions of shared ROH (> 10 MB) between populations corresponded to lower F _ST and migration estimates in alligators (Yang et al. 2023). In killer whales, longer ROH (> 1.5 MB) were observed in populations in lower latitude, less productive waters, which is consistent with conservation concerns for lower‐latitude populations (Foote et al. 2021). Foote et al. (2021) also observed this ROH pattern despite increasing nucleotide diversity in lower latitudes. Similarly, F _ROH values of lizard populations differed by > 20%, suggestive of divergent demographic trajectories, whereas alternative methods produced nearly identical demographic reconstructions (Xie et al. 2022). ROH patterns thus can capture demographic dynamics not picked up by other models or summary statistics.

The link between demography and ROH can be nuanced and is heavily impacted by recent admixture, with, for example, immigration into a re‐established population of wolves leading to individual values of F _ROH ranging from 0.01 to 0.54 (Kardos et al. 2018). Patterns of ROH can be further confounded by method and data. Silva et al. (2024) used simulations to show that the accuracy of ROH detection varied across populations with different demographic histories; specifically, declining populations exhibited the highest error in F _ROH estimates relative to the known value of F. Hewett et al. (2023) also showed that demographic impacts on ROH distributions are stronger than on recombination and selection. Direct estimates of TMCRA (Table 1) can be imprecise because of the stochasticity of recombination and Mendelian segregation (Thompson 2013; Kardos et al. 2016) and may be biased in some cases because of apparent long ROH arising via the conflation of multiple short adjacent IBD segments (Chiang, Ralph, and Novembre 2016).

4. Key Considerations and Recommendations

Quantification and interpretation of ROH in wild populations is becoming more common, and there is clearly valuable information to be gained regarding inbreeding and demographic history in wild populations. Given the variation in ROH detection due to SNP number and analytical approach, we recommend using model‐based inferences of ROH with whole‐genome sequencing data sets. If using rule‐based approaches, altering parameters will help understand variance in ROH detection, but increased ROH detection often comes with increased false positives (Meyermans et al. 2020). Overall, broad inbreeding and demographic patterns will be captured by the ROH distributions, but inferences of TMRCA and demographic history based on ROH distributions should be interpreted with caution. Linkage information also provides an important insight into the ROH landscape, but such information is uncommon in wild populations. Importantly, all ROH reflect IBD, even if they result from positive selection and low recombination. Integrating information on how ROH contribute to genetic load (Stoffel et al. 2021b; Szpiech et al. 2013) and affect fitness components (Duntsch et al. 2021; Kardos et al. 2023; Stoffel et al. 2021a, 2021b) will better link ROH dynamics to ecological importance and conservation decisions.

Author Contributions

Aaron B. A. Shafer and Marty Kardos conceived and wrote the study.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

No data has been generated.