Population Decline, Inbreeding and Hybridization Shape the Genetic Vulnerability of a Critically Endangered Seabird

ABSTRACT

Many endangered species have been rescued from the brink of extinction, yet their long‐term viability remains uncertain due to the prolonged genetic consequences of severe bottlenecks. In particular, the combined effects of inbreeding and hybridization with closely related species remain poorly understood. Using a comparative conservation genomic framework, we investigated these issues in the Chinese crested tern ( Thalasseus bernsteini ), a critically endangered seabird once presumed extinct but which has increased to approximately 100 individuals over the past two decades and its close relative, the great crested tern ( T. bergii ). We demonstrate that the Chinese crested tern has experienced ongoing genetic erosion. Its prolonged population decline has intensified genetic drift and reduced the efficiency of purifying selection, resulting in genome‐wide accumulation of highly deleterious mutations. Highly inbred individuals also carry significantly more homozygous deleterious mutations than less inbred individuals. In both species, runs of homozygosity (ROH) show a relative enrichment of moderately and weakly deleterious mutations. Even more concerning, we found an enrichment of moderately and weakly deleterious mutations in putatively introgressed regions in both species, suggesting that occasional hybridization may have facilitated the spread of deleterious variants and could potentially compromise long‐term fitness. These findings highlight that severe bottlenecks have lasting genomic consequences and that both inbreeding and hybridization can increase the genetic vulnerability of small populations—underscoring the importance of management strategies that promote population growth and sustained genetic monitoring.

1. Introduction

Many species have experienced significant population declines and severe bottlenecks due to human activities and climate change (Cowie et al. 2022). While conservation efforts have saved some species from the brink of extinction, it is often difficult for these populations to recover to historical levels (Dussex et al. 2021; Femerling et al. 2023; Li et al. 2014; Robinson et al. 2022). To effectively boost populations of endangered species and ensure their long‐term viability, conservation managers must understand the genetic factors that drive small populations toward extinction (Chen et al. 2025). Various demographic processes—such as bottlenecks, inbreeding and hybridization—can lead to loss of genetic diversity or influence the dynamics of genetic load, thereby impairing their fitness and survival (Bertorelle et al. 2022; Dussex et al. 2023; Moran et al. 2021; Van Oosterhout 2020; van Oosterhout et al. 2022).

Increased inbreeding can expose highly deleterious mutations, potentially reducing population fitness and causing inbreeding depression (Bertorelle et al. 2022; Dussex et al. 2023). Efficiently purging these highly deleterious mutations is therefore vital for the survival of small populations, as documented in the island fox and other endangered species (Dussex et al. 2023; Grossen et al. 2020; Robinson et al. 2018, 2022). Failure to purge such mutations could drive species toward extinction (Hedrick and Garcia‐Dorado 2016; Kardos et al. 2023; Robinson et al. 2019).

Purging in small populations is shaped by contrasting forces (Dussex et al. 2023). On one hand, increased homozygosity due to inbreeding can expose deleterious mutations to natural selection, potentially facilitating their removal (Bertorelle et al. 2022; Hedrick and Garcia‐Dorado 2016). On the other hand, intensified genetic drift reduces the efficacy of purifying selection, making it more difficult to eliminate deleterious mutations (Dussex et al. 2023; Glemin 2003). As a result, while highly deleterious mutations are often effectively purged, moderately and weakly deleterious mutations—which are less efficiently purged—may persist in the genome and potentially reduce individual fitness (Grossen et al. 2020).

In addition to the consequences of inbreeding and drift, hybridization can further shape the genetic load of small populations. However, the genetic effects of hybridization are complex and can vary among species. Some studies suggest that hybridization benefits small, inbred populations by enhancing evolutionary potential through increased genetic diversity and the introduction of adaptive variations (Moran et al. 2021; Suarez‐Gonzalez et al. 2018; Vedder et al. 2022). However, hybridization may threaten rare taxa through genetic swamping, especially when hybrids between an endangered and a common species have higher fitness than the parental lineages and hybridization occurs frequently, potentially leading to gradual replacement of the endangered species (Todesco et al. 2016). Even when hybrids lack a fitness advantage and fail to replace the endangered species, hybridization can still introduce deleterious mutations from common species into rare taxa, raising concerns when these mutations affect reproduction, adaptation and fitness, especially in vulnerable small populations (Harris and Nielsen 2016; Juric et al. 2016; Liu et al. 2022).

Understanding the genetic consequences of severe bottlenecks in species that persist at chronically small population sizes is crucial for evaluating long‐term viability and deciding whether additional genetic management is warranted (Ahrens et al. 2026; Hedrick and Garcia‐Dorado 2016; Robinson et al. 2018, 2019). Hybridization represents a particularly important and contentious process in this context. Yet few studies have directly examined how hybridization influences genetic load in endangered species, even though such knowledge is crucial for assessing its genetic consequences and for informing future conservation strategies (Fitzpatrick et al. 2015; Moran et al. 2021; Todesco et al. 2016; Wayne and Shaffer 2016). For instance, it can help determine whether to promote gene flow into rare taxa or instead prioritize the maintenance of isolated breeding populations during conservation management (Wayne and Shaffer 2016).

The crested tern system provides a rare opportunity to investigate how hybridization and demographic history jointly shape genetic diversity, inbreeding and genetic load in chronically small populations. The Chinese crested tern ( Thalasseus bernsteini ), classified as ‘Critically Endangered’ on the IUCN Red List (IUCN 2021), was believed extinct after its last sighting in 1937 (Chen et al. 2010). In 2000, however, eight adults and four chicks were discovered breeding on Taiwan's Matsu Islands (Liang et al. 2000). Since the launch of a restoration project in 2013, the global population has increased but remains very small, with just over 100 individuals (Lu et al. 2020). Breeding colonies are now limited to a few islets off Zhejiang and Fujian coast, and one island near the Korean Peninsula (Lu et al. 2020; Song et al. 2017). The existing population on Wuzhishan Island and Jiushan Island off Zhejiang has been monitored, and their breeding habitats are actively managed. By contrast, its close relative, the great crested tern ( T. bergii ), shares some breeding sites but has a much broader range spanning the Pacific and Indian Oceans and a much larger population (estimated at c.150,000–1,100,000 individuals) (Gochfeld et al. 2018). Heterospecific pairing between the two species has been observed at breeding colonies (Chen and He 2011; Ottenburghs and Nisbet 2025) and hybridization has been confirmed by genetic analyses using mitochondrial and sex‐linked markers, together with microsatellite data (Gu et al. 2021; Yang et al. 2018). However, although hybridization has only been documented in a few cases, its actual frequency remains unknown, and current marker‐based studies provide only limited insight into the genome‐wide consequences of bottlenecks, inbreeding and hybridization in these two tern species.

Using whole‐genome resequencing data, we first reconstructed the ancient and recent demographic histories of both tern species. We then investigated the genomic consequences of the recent population decline. Finally, we assessed the genetic impacts of hybridization on genetic load in both species. This work provides a comprehensive analysis of how the combined effects of population decline, inbreeding and hybridization shape the dynamics of genetic load, offering new insights into conservation strategies for long‐term persistence of small, highly threatened populations.

2. Materials and Methods

2.1. Sampling and DNA Extraction

We analysed 37 samples from three tern species: 13 Chinese crested terns ( Thalasseus bernsteini ), 21 great crested terns ( Thalasseus bergii ) and three little terns ( Sternula albifrons ) as one of the outgroup species (Table S1).

We initially attempted to use abandoned eggs, museum toe pads and dead individuals to minimize disturbance to the Chinese crested tern. However, > 90% of these samples were heavily contaminated, likely due to rapid decomposition of abandoned eggs and dead individuals during the hot summer and were unsuitable for sequencing. Ultimately, we obtained high‐quality DNA from 13 Chinese crested terns and 21 great crested terns, including one Chinese crested tern and six great crested terns that were collected as deceased specimens; the remaining blood samples were collected from live nestlings during annual banding activities via minimally invasive brachial venipuncture using capillary tubes. The deceased Chinese crested tern individual was used for de novo genome assembly and was not included in population‐level analyses. Sample collection was approved by the Zhejiang Forestry and Grassland Administration and the Wuzhishan and Jiushan Archipelago authorities (Zhejiang, China) and conducted in accordance with the Chinese Animal Welfare Act (20090606).

Genomic DNA was extracted from all samples using the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany) and quantified with a NanoDrop ND‐2000. In addition, for transcriptome‐assisted gene annotation, we collected RNA samples from two Chinese crested terns and stored them in Trizol reagent (Invitrogen/Thermo Fisher Scientific, Waltham, MA, USA). RNA was extracted using an Ambion RiboPure Kit (Life Technologies, Carlsbad, CA, USA) following the manufacturer's protocol.

We conducted a phylogenetic analysis to confirm species identification and rule out misidentification, which is a concern due to the overlapping breeding colonies and morphological similarity between the Chinese crested tern and great crested tern nestlings. We amplified the partial cytochrome b gene (cytb) from all blood samples and reconstructed a maximum‐likelihood tree using RAxML v.7.3.1 (Stamatakis 2006), incorporating sequences from this study and published data (Yang et al. 2018).

Furthermore, based on our ADMIXTURE results, we identified one great crested tern individual (GCT8) with a notable proportion of mixed ancestry (10.11%). To minimize bias from potential hybridization, we excluded this individual from all downstream analyses except those related to population structure (Admixture, PCA, phylogenetic tree) and hybridization.

2.2. Reference Genome Assembly and Annotation

We performed de novo sequencing on one Chinese crested tern to assemble its reference genome. To accomplish this, we constructed ten libraries with insert sizes of 250 bp, 450 bp, 2 kb, 5 kb and 10 kb and sequenced them on an Illumina HiSeq X Ten at Novogene (Beijing, China), following the manufacturer's protocol.

We removed adapters and low‐quality reads from the raw data using Trimmomatic v.0.36 (Bolger et al. 2014) and removed duplicated reads with FastUniq v.1.1 (Xu et al. 2012). After processing, we retained 185.23 GB of clean reads. We assembled the genome using SOAPdenovo v.2 (Li et al. 2010) and reconstructed scaffolds with SSPACE v.3.0 (Boetzer et al. 2011). Gaps in the assembly were closed using GapCloser v.1.12 (Li et al. 2010). Assembly completeness was assessed using BUSCO v.5.8.2 (Simao et al. 2015) with the aves_odb10 lineage dataset. To generate pseudo‐chromosomes, we performed a whole‐genome synteny alignment between the Chinese crested tern genome and the chicken ( Gallus gallus ) genome using Satsuma v.3.1.0 (Grabherr et al. 2010).

We annotated the reference genome of the Chinese crested tern using an integrated workflow. First, we identified repetitive elements by combining homology‐based and de novo predictions, then scanned the genome for tandem repeats and transposable elements using RepeatMasker v.4.0.5 (Smit et al. 2000). Next, we predicted protein‐coding genes through a combination of homology mapping, de novo prediction and transcriptome evidence. For the homology‐based prediction, we collected avian protein sequences of 12 species from the NCBI database and processed them with Maker v.2.31.9 (Holt and Yandell 2011) to generate a gene set. For the transcriptome analysis, we mapped RNA‐seq reads with TopHat v.1.3.1 (Trapnell et al. 2009), reconstructed transcripts using Cufflinks v.1.3.0 (Trapnell et al. 2010) and merged transcripts with Cuffmerge (Trapnell et al. 2010). Finally, we combined gene models from all three methods into a non‐redundant gene set.

2.3. Whole‐Genome Resequencing, Variant Calling and Filtering

We performed whole‐genome resequencing on 12 Chinese crested terns, 21 great crested terns and three little terns. DNA libraries were sequenced on Illumina HiSeq X Ten and NovaSeq 6000 platforms with 150 bp paired‐end reads, following the manufacturer's protocols. We processed the raw data by removing low‐quality reads and adapters, then aligned the clean reads to the reference genome using BWA v.0.5.17 (Li and Durbin 2009) with default settings. After removing PCR duplicates with Picard v.1.91 (Picard 2019), we called single‐nucleotide polymorphisms (SNPs) using GATK v.3.8 (McKenna et al. 2010). We filtered the SNPs through a two‐step approach: initial GATK hard filtering followed by additional quality control using VCFtools v.0.1.14 (Danecek et al. 2011) with specific parameters (‐‐minQ 30, ‐‐minDP 5, ‐‐maxDP 70, ‐‐max‐missing 0.95, ‐‐max‐alleles 2).

For analyses where rare variants could potentially affect the results, such as population structure, we additionally applied a minor allele frequency filter (‐‐maf 0.05) and required a minor allele count of at least 3 (‐‐mac 3). We calculated the pairwise relatedness using KING v.2.2.7 (Manichaikul et al. 2010) and no individuals were excluded based on close relatedness.

2.4. Population Structure and Phylogenetic Tree

To minimize potential bias from background linkage disequilibrium (LD) in population structure analyses, we pruned SNPs using PLINK v.1.91 (Falush et al. 2003; Purcell et al. 2007). This resulted in a final set of 1,494,952 bi‐allelic autosomal SNPs for subsequent population structure analyses.

We employed three methods to investigate population clustering among individuals of the two crested tern species. First, we reconstructed a maximum‐likelihood phylogenetic tree for all the individuals using SNPhylo v.20180901 (Lee et al. 2014). Second, we performed a principal component analysis (PCA) for both the entire dataset and separately for each species using PLINK v.1.91. Finally, ancestry components were inferred with ADMIXTURE v.1.3 (Alexander and Lange 2011) by testing K values from 1 to 5 with 200 replicates each. The optimal K was determined based on the lowest cross‐validation error.

2.5. Genetic Diversity and Inbreeding Statistics

We estimated genetic diversity through two metrics. First, we calculated genome‐wide heterozygosity for each individual by dividing the number of heterozygous SNPs by the total length of autosomal chromosomes (excluding Ns and gaps). Second, we computed the nucleotide diversity (π) for each species using a 50‐kb non‐overlapping window approach in VCFtools v.0.1.14. To rule out potential bias from unequal sample sizes between species (12 CCT vs. 21 GCT), we calculated π for the great crested tern using two different group sizes (20 and 12 individuals).

To evaluate genetic diversity in a broader avian context, we compared the genome‐wide heterozygosity of the two terns with published whole‐genome resequencing data of 62 avian species (Cavill et al. 2024; Dierickx et al. 2020; Dussex et al. 2021; Ellegren et al. 2012; Femerling et al. 2023; Hung et al. 2014; Lai et al. 2019; Li et al. 2014, 2022; Murray et al. 2017; Wang et al. 2021, 2022; Zhan et al. 2013) (Table S5). As these published studies used different pipelines, we did not perform a formal statistical comparison across all species; instead, these data serve as a general reference.

To estimate the genome‐wide inbreeding level, we identified the runs of homozygosity (ROH) for all individuals using two methods: BCFtools v.1.14 (a hidden Markov model approach) (Danecek et al. 2021; Li 2011) and PLINK v.1.9 (a scanning window approach). We used a constant recombination rate of 1.5 cM/Mb (Backstrom et al. 2010) in the BCFtools analysis. We performed PLINK v.1.9 with the following parameters: ‐‐homozyg‐window‐snp 50, ‐‐homozyg‐snp 50, ‐‐homozyg‐window‐missing 3, ‐‐homozyg‐kb 100, ‐‐homozyg‐density 50 and ‐‐homozyg‐window‐het 3. Previous studies have indicated that high missing rates may influence the accuracy of detecting ROHs (Meyermans et al. 2020; Narasimhan et al. 2016). To rule out this potential confounding effect, we compared the F _ROH between species using two datasets: one with all individuals (n = 32, 12CCT and 20 GCT) and another excluding five individuals with high missing rates (1%–10%; n = 27). The results confirmed that while ROH detection in the five high‐missing‐rate individuals was indeed affected differently by the two methods, their exclusion did not change the overall pattern of F _ROH differences between the two species (Figure S2B). To ensure robustness, all subsequent ROH‐related analyses were performed using both the BCFtools and PLINK ROH sets. The results were highly consistent between the two sets of ROHs. Therefore, the main text presents results based on BCFtools ROH sets, while the results from PLINK ROH sets are provided in Appendix S1.

We estimated the mean age of ROHs (i.e., the number of generations since the coalescence of the haplotype, denoted as g) based on their length using the formula g = 100/(2rL), where r is the recombination rate (1.5 cM/Mb, the average recombination rate of the zebra finch ( Taeniopygia guttata ) genome (Backstrom et al. 2010)) and L is the ROH length in Mb (Thompson 2013).

We calculated the genome‐wide inbreeding coefficient (F _ROH) for each individual in both species, using ROHs longer than 0.5 Mb. We defined F _ROH as the total length of ROH divided by the total length of autosomal chromosomes (Ceballos et al. 2018). A higher F _ROH value indicates higher inbreeding in a population.

2.6. Demographic Histories

We reconstructed the demographic histories using three methods. First, we employed the pairwise sequential Markovian coalescent (PSMC) model (Li and Durbin 2011) to infer the ancient demographic history. This included seven Chinese crested terns and 16 great crested terns with sequencing depth ≥ 18×. Autosomal SNPs were called using SAMtools v.1.2.1 (Li et al. 2009) and PSMC was run with parameters ‘N30 –t5 –r5 –p 4+30*2+4+6+10.’ We assumed a mutation rate of 4.8 × 10⁻⁹ substitutions per site per generation (Zhang et al. 2014) and a generation length of 11 years (IUCN 2021). Since PSMC has limited resolution in recent history (Li and Durbin 2011), we used SMC++ v.1.15.3 (Terhorst et al. 2017) to reconstruct the more recent demographic trajectories for all the individuals. We did not apply mirror allele frequency filtering to the SNPs used for SMC++ and GONE.

Finally, to extend our inference to the very recent past (~350 generations), we applied GONE for both species (Santiago et al. 2020). As the recombination rate for the Chinese crested tern and closely related species is unknown, we used a rate of 1.5 cM/Mb from the zebra finch ( Taeniopygia guttata ) (Backstrom et al. 2010). In addition, to assess the potential impact of the recombination rate, we also performed GONE analyses using the default recombination rate (1 cM/Mb) (Figure S3B). Our findings indicate that the fluctuation in effective population size over the past 300 generations is consistent across different recombination rates.

2.7. Identification of Deleterious Mutations

We estimated genetic load by examining derived alleles in the Chinese crested tern and the great crested tern. To determine the ancestral state at each focal site, we used a four‐species outgroup panel consisting of the little tern sequenced in this study and three additional outgroup species with publicly available genome data: common tern ( Sterna hirundo ; SRA: SRR27847731), gull‐billed tern ( Gelochelidon nilotica ; SRA: ERR15107121), large‐billed tern ( Phaetusa simplex ; SRA: SRR9946722, SRR9946724, SRR9947238). Only sites where at least three outgroup species shared an identical homozygous genotype were considered to carry the ancestral allele; all other sites were excluded. We then classified the derived variants in coding regions into synonymous, nonsynonymous and loss‐of‐function (LoF) categories using snpEff v.4.3 (Cingolani et al. 2012). Nonsynonymous variants were further categorized as benign or deleterious based on Grantham's score (Grantham 1974), with scores < 150 considered benign and those > 150 considered deleterious. Variations with splice donor, splice acceptor, start‐lost, stop‐lost, stop‐gained or stop‐retained mutations were classified as LoF variants.

2.8. Population Statistics on the Deleterious Mutations of Two Tern Species

To evaluate the mutation load in the two tern species, we quantified both the absolute number and relative frequency of deleterious mutations. The absolute number of deleterious mutations (including total, heterozygous and homozygous counts) was categorized into LoF, deleterious nonsynonymous and benign nonsynonymous types. To control for technical biases, these raw counts were standardized by dividing by the total number of intergenic‐derived alleles per individual.

We then compared the relative frequency of total deleterious mutations (R _A/B) between the Chinese crested tern and the great crested tern using a custom script, following the method of Xue et al. (2015). We estimated variances using a block jackknife approach with 0.2 million consecutive SNPs per block. To ensure robustness against unequal sample size, we performed the R _A/B analysis using both the full dataset (Figure 2C, 12 CCT and 20 GCT) and a subset dataset (Figure S4B, 12 CCT and 12 GCT).

FIGURE 2.

Accumulation of deleterious mutations in the Chinese crested tern and the great crested tern. (A) Total and (B) homozygous deleterious mutation counts (standardized by intergenic mutation counts). (C) Relative frequency (R _A/B) of deleterious mutations in 12 Chinese crested terns and 20 great crested terns. R _A/B > 1 indicates more derived alleles in the Chinese crested tern, and R _A/B < 1 indicates more derived alleles in the great crested tern. Data are shown as mean ± standard deviation. (D) Ratios of LoF, deleterious, and benign nonsynonymous homozygous mutations to synonymous homozygous mutations in BCFtools‐inferred ROH versus non‐ROH regions. Corresponding results based on PLINK‐inferred ROH are shown in Figure S6. CCT, Chinese crested tern; GCT, Great crested tern; LoF, Loss‐of‐function; NS, Non‐significant; *p < 0.05; **p < 0.01; ***p < 0.001.

A previous study suggested that runs of homozygosity (ROH) might be under stronger purifying selection because they can carry more homozygous deleterious mutations (Szpiech et al. 2013). To explore the effects of inbreeding through ROH on purifying selection, we compared the ratios of homozygous LoF, deleterious, and benign nonsynonymous mutations to homozygous synonymous mutations in ROH regions versus non‐ROH regions.

2.9. The Genetic Consequence of Inbreeding in the Chinese Crested Tern Population

We identified three Chinese crested terns with relatively high inbreeding and classified the remaining individuals as a low‐inbreeding group. We first examined the correlation between genome‐wide heterozygosity and F _ROH for these individuals. We then compared both heterozygosity and F _ROH between the high‐ and low‐inbreeding groups.

To evaluate how inbreeding affects the accumulation of deleterious mutations, we first examined the correlation between the homozygous deleterious mutations and F _ROH. Second, we compared the relative frequency (R _A/B) of deleterious mutations between the high‐ and low‐inbreeding groups. Third, we compared the number of total, heterozygous and homozygous deleterious mutations between the two groups. We also calculated the proportion of homozygous deleterious mutations at both the genome‐wide level and within ROH regions. Finally, we calculated the ratio of homozygous LoF, deleterious and benign nonsynonymous mutations to homozygous synonymous mutations in ROH versus non‐ROH regions for both groups.

2.10. Forward Genetic Simulations

2.10.1. Wright‐Fisher Simulation

We performed Wright‐Fisher simulations using SLiM v.4.0.1 (Haller and Messer 2023) to evaluate how recent population decline affects the accumulation of deleterious mutations. Using the GONE‐inferred demographic history over the past 200 generations, we simulated 50 replicates for both species.

Each simulation modelled 6000 genes (representing approximately 35% of the reference genome) distributed across 11 chromosomes, with a gene length of 1500 bp. We set the mutation rate to 4.8 × 10⁻⁹, and the recombination rate to 1 × 10⁻⁸, assuming no recombination within genes but free recombination between chromosomes. Selection coefficients and the ratios of deleterious to neutral mutations were derived from species‐specific distribution of fitness effects (DFE) estimated with polyDFE v.2.0 (Tataru and Bataillon 2019). Dominance coefficients were set using the ‘hmix’ model, following previous studies (Kyriazis et al. 2023, 2021).

Each simulation included a burn‐in period of 10‐fold N _e generations to allow the population to reach equilibrium. We collected data every 1000 generations during the burn‐in period and every two generations thereafter. Outputs included mean heterozygosity, F _ROH, total mutation load, realized load and the counts of strongly (s ≤ −0.01) and weakly deleterious (−0.01 < s ≤ −0.00001) mutations from 40 sampled individuals per replicate.

2.10.2. Non‐Wright‐Fisher Simulations

To assess whether the severe bottleneck in the Chinese crested tern increases its extinction risk relative to the great crested tern, we conducted non‐Wright‐Fisher simulations following Kyriazis' methods (Kyriazis et al. 2021). Simulations covered the last 200 generations based on GONE‐inferred demography, followed by future projections until extinction (N = 1). We tested bottleneck carrying capacities (K _bottleneck) of 70, 500, 1000, 2000 and 4500 (the latter approximating the great crested tern's bottleneck population size), while keeping the ancestral carrying capacity (K _ancestral) constant.

Genomic parameters (gene/chromosome numbers, mutation/recombination rates, selection coefficients, burn‐in and sampling frequency) matched those of our Wright‐Fisher simulations. To assess the impact of the distribution of fitness effects (DFE), we ran simulations using DFEs estimated for the Chinese crested tern, the great crested tern and human (Kim et al. 2017).

We collected data every two generations after the burn‐in. Outputs included population carrying capacity (K), population size (N), mean heterozygosity, total and realized load and the average counts of strongly (s ≤ −0.01) and weakly deleterious (−0.01 < s ≤ −0.00001) mutations.

2.11. Test the Introgression Between Two Tern Species

To detect genome‐wide gene flow between the Chinese crested tern and the great crested tern, we calculated Patterson's D (ABBA‐BABA test) and f ₄‐ratio statistics using Dsuite (Malinsky et al. 2021; Patterson et al. 2012). Because D‐statistics are not directly applicable between sister lineages, we split the GCT into two geographic populations: Wuzhishan (GCTa) and Jiushan (GCTb) (Figure 1A). This allowed us to use the topology (((GCTa, GCTb), CCT), Outgroup), thereby enabling the detection of gene flow between GCTb (P2) and CCT (P3). Although this approach relies on treating GCTa and GCTb as separate populations and may not fully capture the GCT population structure, it provides a practical framework for testing for potential gene flow between GCT and CCT.

FIGURE 1.

Sampling locations, population structure, genetic diversity, inbreeding level and demographic histories of the Chinese crested tern and the great crested tern. (A) Geographic range and sampling sites of the Chinese crested tern. The yellow rectangle marks the study area; arrows indicate the sampling sites; pie charts indicate sample sizes. Yellow and green dots indicate current and historical breeding sites, respectively. (B) ADMIXTURE clustering (K = 2) of the two species. Each bar indicates one individual. Individual GCT8 (#) shows the highest admixture proportion. (C) Genome‐wide heterozygosity. (D) Inbreeding coefficient (F _ROH) inferred using BCFtools. The three highly inbred Chinese crested terns are marked with ^. Corresponding results based on PLINK‐inferred ROH are shown in Figure S2B. (E) Demographic history inferred with SMC++. The light grey, dark grey and orange blocks indicate the Last Glacial Period (LGP), the Last Glacial Maximum (LGM) and Mid‐Holocene (MH) interglacial periods, respectively. (F) Recent demographic history inferred with GONE. Grey block indicate the estimated timing of ROH events (approximately the last 67 generations). ROH, Runs of homozygosity; CCT, Chinese crested tern; GCT, Great crested tern; **p < 0.01; ***p < 0.001.

To test the direction of historical gene flow between the two species, we used Migrate‐N v.3.5.1 (Beerli 2006) to estimate migration rates under three models: bidirectional migration, unidirectional migration in either direction. Because migrate‐N is not designed to handle whole‐genome sequences as a single locus, we randomly selected ten independent 1‐Mb genomic regions and, for each region, defined multiple loci from evenly spaced SNPs. We ran Migrate‐N separately for each region and compared model likelihoods to infer the predominant direction of gene flow. Furthermore, we estimated admixture time using a Hidden Markov Model approach (Corbett‐Detig and Nielsen 2017) with 100 bootstraps to generate confidence intervals.

We employed two methods to identify potential introgressed regions (or mixed ancestry regions) in both species. First, we used HybridCheck v.1.0.1, which identifies introgressed blocks by comparing nucleotide similarity across chromosomes in triplets of aligned sequences (Ward and van Oosterhout 2016). We analysed 278 scaffolds longer than 0.5 Mb (representing > 80% of the reference genome) using a triplet design. Each triplet consisted of one Chinese crested tern, one great crested tern and one little tern. We set the window size to 1000 bp with a step size of 1 bp. The Chinese crested tern–little tern comparison served as the baseline (non‐introgressed) reference; we defined putative introgressed blocks in the Chinese crested tern–great crested tern comparison as windows with sequence similarity exceeding the reference pair. For each individual, we calculated the proportion of such blocks across all comparisons.

Second, we used RFMix v.1.5.4 (Maples et al. 2013), which infers local ancestry using a conditional random field parameterized by random forests trained on reference panels. We phased LD‐pruned (r ² < 0.2) autosomal SNPs using BEAGLE v.5.1 (Browning et al. 2018) and inferred local ancestry on the same 278 scaffolds, specifying eight admixture generations (default). For each individual, we then calculated the proportion of mixed‐ancestry regions.

2.12. The Genetic Impact of Introgression

To evaluate the potential impact of introgression, we first calculated the number of LoF, deleterious and benign nonsynonymous mutations within putative introgressed regions and non‐introgressed genomic regions for each individual. We then calculated the ratio of LoF, deleterious and benign nonsynonymous mutations to synonymous mutations. To validate the robustness of our findings, we repeated these comparisons using mixed‐ancestry and non‐mixed regions identified by RFMix v.1.5.4.

3. Results

3.1. Genome Assembly and SNP Calling

We generated a high‐quality reference genome for the Chinese crested tern, with an assembled genome size of 1.26 Gb and a scaffold N50 of 6.8 Mb. The assembly showed high completeness (BUSCO: 94.2%; Tables S2 and S3). Repeat elements accounted for 11.64% of the genome, and we annotated 17,922 protein‐coding genes (Table S4).

We performed whole‐genome resequencing on 36 samples, including 12 Chinese crested terns, 21 great crested terns, and three little terns ( Sternula albifrons ) (Figure 1A). The average sequencing depth was 26.6‐fold (Table S1). We identified a total of 4,872,896 biallelic single‐nucleotide polymorphisms (SNPs) for downstream analysis. Results from ADMIXTURE (Alexander and Lange 2011), the phylogenetic tree and the PCA revealed distinct genetic lineages between the Chinese crested tern and the great crested tern, although we detected no subdivisions within each species (Figure 1B and Figure S1).

3.2. Genetic Diversity and Inbreeding

The Chinese crested terns exhibited a mean nucleotide diversity (π) of 1.14 × 10⁻³ and genome‐wide heterozygosity of 1.11 × 10⁻³. These values were slightly higher than the great crested tern, which were 1.06 × 10⁻³ and 1.01 × 10⁻³, respectively (Figure 1C, Figure S2A). This pattern was consistently observed across subsampled datasets of the great crested tern, indicating that the results were robust to variation in sample size (Figure S2A). The genetic diversity of the Chinese crested tern was lower than that observed in 73.7% of 61 other bird species included for comparison (Table S5).

The Chinese crested tern has experienced higher inbreeding compared to the great crested tern, as indicated by their runs of homozygosity (ROH). The genome‐wide inbreeding coefficient for the Chinese crested tern (F _ROH, inferred using BCFtools), reflecting inbreeding over approximately the last 67 generations, was significantly higher than that of the great crested tern (Figure 1D). A similar pattern was observed using PLINK‐inferred F _ROH, although the difference was not significant (Figure S2B). Three Chinese crested tern individuals (CCT8, CCT10 and CCT12) exhibited more extensive and longer ROHs than the remaining samples (Figure 1D). Notably, one individual, CCT10, carried the longest ROH at 6.8 Mb, corresponding to inbreeding events within approximately the last five generations.

3.3. Demographic History

The PSMC results indicated that the Chinese crested tern maintained a relatively stable population size before the Last Glacial Period (LGP), then experienced a population expansion at the onset of the LGP followed by a sharp decline (Figure S3A). These demographic trends were broadly consistent with the SMC++ results, although SMC++ revealed additional fine‐scale fluctuations in more recent times (Figure 1E). SMC++ further showed that this LGP bottleneck persisted until the end of the period, followed by a gradual recovery during the warmer Mid‐Holocene (MH). In contrast, PSMC trajectories for the great crested tern suggested stronger population fluctuations over the past 100,000–1000,000 years (Figure S3A). Both PSMC and SMC++ indicated that the great crested tern, like the Chinese crested tern, underwent a decline during the LGP followed by a gradual recovery during the MH (Figure 1E, Figure S3A).

We further inferred recent demographic changes over the past 300 generations using GONE. Both species exhibited similar trajectories over this period, with population declines occurring within the last 100 generations (Figure 1F, Figure S3B). However, the Chinese crested tern experienced a markedly steeper decline, with the effective population size (N _e) collapsing to approximately 70, whereas N _e in the great crested tern declined to around 4500.

3.4. Accumulation of Deleterious Mutations in Two Tern Species

The Chinese crested tern carried significantly fewer nonsynonymous mutations (both deleterious and benign) than the great crested tern, a pattern consistent across both homozygous and heterozygous forms (Figure 2A,B, Figure S4A). However, no significant differences were observed between the two species in the number of loss‐of‐function (LoF) mutations.

To test for evidence of purging on the deleterious mutations, we calculated the relative frequency of deleterious mutations (R _A/B) using all available individuals for each species (Figure 2C; 12 Chinese crested terns and 20 great crested terns). This analysis revealed a pronounced excess of LoF mutations in the Chinese crested tern, whereas deleterious and benign mutations showed a slight excess in the great crested tern. To rule out potential bias from unequal sample size, we repeated the R _A/B calculation using all 12 Chinese crested terns and a random subset of 12 great crested terns and found highly consistent results (Figure S4B).

To further evaluate the efficacy of purifying selection, we analysed the allele frequency spectrum of derived alleles. In both species, derived deleterious alleles were overrepresented at low frequencies compared with putative neutral variants (synonymous), indicating the signal of purifying selection (Figure S5). However, at intermediate to high frequencies, LoF mutations were markedly depleted relative to neutral mutations in the great crested tern, whereas the Chinese crested tern showed similar frequencies of LoF and neutral mutations in these classes. This pattern suggests that purifying selection against highly deleterious mutations is more effective in the great crested tern than the Chinese crested tern.

In addition, we assessed how inbreeding influenced the distribution of deleterious mutations by examining the accumulation of homozygous deleterious mutations within and outside ROH regions. In both species, the ratios of homozygous deleterious and benign nonsynonymous mutations to homozygous synonymous mutations were significantly higher inside ROH regions than outside them (Figure 2D, ROHs inferred with BCFtools; Figure S6, ROHs inferred with PLINK). These two ROH‐calling approaches yielded consistent patterns. In contrast, the ratios of homozygous LoF mutations were significantly lower in ROH regions in both species, indicating stronger selection against highly deleterious mutations inside ROH segments (Figure 2D, Figure S6).

3.5. The Genetic Impacts of Inbreeding in the Chinese Crested Tern

Based on their F _ROH, three Chinese crested terns (CCT8, CCT10 and CCT12) were identified as a high‐inbreeding group, with the remaining nine individuals classified as a low‐inbreeding group (Figure S7A). Genome‐wide heterozygosity was negatively correlated with F _ROH, and the high‐inbreeding group exhibited significantly lower heterozygosity than the low‐inbreeding group (Figure S7B,C). In contrast, the total number of homozygous deleterious mutations showed a positive correlation with F _ROH (Figure 3A).

FIGURE 3.

The genetic consequences of inbreeding in high‐ and low‐inbreeding groups of the Chinese crested tern. (A) Genome‐wide inbreeding coefficient (F _ROH, inferred using BCFtools) positively correlated with the number of homozygous deleterious mutations. (B) Number of homozygous deleterious mutations. (C) Relative frequency (R _A/B) of deleterious mutations in high‐ and low‐inbreeding groups of Chinese crested terns. R _A/B > 1 means more derived alleles in the high‐inbreeding group, and R _A/B < 1 means more derived alleles in the low‐inbreeding group. Data are shown as mean ± standard deviation (SD). (D) The proportion of homozygous deleterious mutations in ROH regions (inferred using BCFtools). Corresponding results based on PLINK‐inferred ROHs are shown in Figure S9. High, High‐inbreeding group; Low, Low‐inbreeding group; LoF, Loss‐of‐function; NS, Non‐significant; *p < 0.05; **p < 0.01.

The high‐inbreeding group had significantly more homozygous deleterious mutations than the low‐inbreeding group, whereas the numbers of total and heterozygous deleterious mutations were similar in both groups (Figure 3B, Figure S8). Similarly, the R _A/B analysis revealed an excess of LoF mutations in the high‐inbreeding group, while the nonsynonymous mutations showed no significant difference between the two groups (Figure 3C), indicating an accumulation of highly deleterious mutations in the high‐inbreeding group.

The high‐inbreeding group exhibited a significantly higher proportion of homozygous deleterious mutations among all deleterious mutations than the low‐inbreeding group (Figure S9A). Moreover, a significantly higher proportion of these homozygous deleterious mutations occurred within ROH regions in the high‐inbreeding group than in the low‐inbreeding group (Figure 3D, Figure S9B).

3.6. Simulated Impacts of Recent Population Decline on Extinction Risk

We simulated demographic scenarios over the past 200 generations for both species using SLiM (Figure 4A). Across the five sampling time points, the Chinese crested tern population showed a marked increase in F _ROH and a minor decline in heterozygosity by generation one, the end of the bottleneck (corresponding to the current population) (Figure S10). Meanwhile, the Chinese crested tern accumulated significantly more homozygous deleterious mutations and a higher realized load, whereas the total mutation load remained unchanged by generation one (Figure 4B–D). In contrast, these parameters showed little change across the five sampling time points in the great crested tern (Figure 4B–D, Figure S10).

FIGURE 4.

Simulated demographic histories and their effects on genetic load in the Chinese crested tern and the great crested tern. (A) Simulated demographic scenarios. Dotted lines indicate sampling time points (200, 150, 100, 50 and 1 generation ago, with 1 generation ago marking the end of the bottleneck). (B) Total mutation load. (C) Realized load. (D) The number of homozygous strongly deleterious mutations. Colours in (B–D) correspond to the five sampling time points shown in (A). CCT, Chinese crested tern; GCT, Great crested tern; ***p < 0.001; NS in (B–D) indicates non‐significant differences in all pairwise comparisons (p > 0.05).

To further assess how bottleneck population size affects future extinction risk, we simulated forward demographic scenarios for the Chinese crested tern under varying bottleneck sizes (Figure S11A). Populations experiencing smaller bottlenecks tended to go extinct more quickly, whereas this pattern was no longer significant once bottleneck sizes exceeded 1000 individuals (Figure S11B). Under similar ancestral population sizes, smaller bottlenecks led to higher realized load in the surviving populations (Figure S11C–F). Nevertheless, there was no significant difference in time to extinction across different distributions of fitness effects, suggesting that the proportion of deleterious mutations in the genome alone may not be the primary driver of extinction (Figure S11G).

3.7. The Genetic Impacts of Hybridization

To evaluate whether hybridization between the two tern species might affect persistence of the Chinese crested tern, we first tested for evidence of introgression between them using three methods. Both Patterson's D and the f4‐ratio statistics revealed significant gene flow between the two tern species (Table S6). Moreover, the Migrate‐N test (Beerli 2006) indicated that bidirectional migration was the most probable scenario, although the gene flow appeared limited (Table S7).

We also estimated the genome‐wide admixture time for each individual using a Hidden Markov Model approach (Corbett‐Detig and Nielsen 2017). One individual (GCT8) showed a relatively recent admixture time of 154 generations, whereas admixture times for the remaining individuals reached 10,000 generations or longer (Table S8).

To further characterize the extent of introgression, we estimated the proportion of potential introgressed regions using two methods. HybridCheck (Ward and van Oosterhout 2016) identified similar mean proportions of putative introgressed regions in both species (12.11% in the great crested tern and 11.97% in the Chinese crested tern), with the estimated age of these blocks dating back tens of thousands of years (Figure 5A). In contrast, RFMix (Maples et al. 2013), which detects more recent admixed ancestry (within the past eight generations), inferred a low proportion of mixed ancestry in both species. The great crested tern had a relatively higher proportion (1.37%) of such regions than the Chinese crested tern (0.90%) (Figure 5B). This difference was attributable to a single great crested tern individual (GCT8) displaying 10.11% admixed ancestry, consistent with the ADMIXTURE analysis (Figure 1B).

FIGURE 5.

Introgression and its impact on deleterious mutations in the Chinese crested tern and the great crested tern. (A) Proportion of putative introgressed regions (HybridCheck‐inferred). (B) Proportion of putative mixed ancestry regions (RFMix‐inferred). The blue‐highlighted individual (GCT8) corresponds to the admixed individual in Figure 1B. (C) The number of deleterious mutations per Mb in the putative introgressed versus non‐introgressed regions (HybridCheck‐inferred). (D) Ratios of LoF, deleterious and benign nonsynonymous mutations to synonymous mutations in the same regions. Corresponding results based on the RFMix‐inferred mixed ancestry regions are shown in Figure S13. CCT, Chinese crested tern; GCT, Great crested tern; LoF, Loss‐of‐function; NS, Non‐significant; *p < 0.05; ***p < 0.001.

To assess the genetic consequences of introgression, we compared the burden of deleterious mutations between putative introgressed and non‐introgressed regions. In both species, putative introgressed regions exhibited a significantly higher density of nonsynonymous mutations, including both deleterious and benign variants (Figure 5C). A similar pattern was observed when considering only heterozygous nonsynonymous mutations (Figure S12). Furthermore, ratios of deleterious and benign mutations to synonymous mutations were higher in introgressed regions than in non‐introgressed regions (Figure 5D). By contrast, no significant difference was observed in LoF mutations. We also examined the density of deleterious mutations and their ratios to synonymous mutations in mixed‐ancestry regions identified with RFMix. In these regions, only benign mutations showed significant differences, whereas deleterious variants did not (Figure S13), which may reflect limited statistical power because mixed‐ancestry tracts are typically short and contain few variants.

4. Discussion

4.1. Demographic History and Its Impacts on Genetic Diversity and Genetic Load

Our results indicate that the genetic diversity of the two tern species was influenced by their demographic history. Although the Chinese crested tern exhibited slightly higher genetic diversity than the great crested tern—seemingly inconsistent with its more severe recent decline—this pattern may be explained by their more recent demographic trajectories. SMC++ and GONE analyses revealed that the great crested tern had a lower effective population size over the past 100–300 generations, which may have contributed to the reduced genetic diversity in our samples. Importantly, these samples were collected from less than 10% of the great crested tern's breeding range—which spans the Pacific and Indian Oceans—and thus may not fully reflect species‐wide genetic diversity. In contrast, the higher inbreeding level in the Chinese crested tern is consistent with its more severe population decline, a pattern further supported by our forward simulations.

Genetic load in these populations is jointly influenced by demographic history, purifying selection and genetic drift (Bertorelle et al. 2022; Dussex et al. 2023; Hedrick and Garcia‐Dorado 2016). R _A/B analysis revealed an excess of highly deleterious (LoF) mutations in the Chinese crested tern, despite its slightly lower total number of deleterious mutations compared to the great crested tern. This apparent paradox likely reflects the reduced efficacy of purifying selection and enhanced genetic drift resulting from its more pronounced population decline and lower N _e (Dussex et al. 2023; Glemin 2003; Kardos et al. 2023; Kennedy et al. 2014). Consistent with this, the derived allele frequency spectrum showed that highly deleterious mutations are skewed toward low frequencies in both species, indicating ongoing purifying selection. However, the Chinese crested tern harbours a relatively higher proportion of highly deleterious mutations at intermediate and high frequencies, suggesting either less efficient or insufficient time for the removal of these alleles compared to the great crested tern. Taken together, these patterns support the view that the more severe recent decline in the Chinese crested tern has weakened purging against highly deleterious mutations, allowing some to drift to intermediate frequencies despite ongoing purifying selection.

Within the Chinese crested tern population, individuals with higher inbreeding coefficients accumulated more highly deleterious mutations, predominantly in the homozygous state and within ROH regions. This pattern suggests that purifying selection may not have sufficient time or strength to remove these exposed mutations efficiently, which may compromise individual fitness and further elevate extinction risk (Bertorelle et al. 2022; Dussex et al. 2023; Szpiech et al. 2013).

4.2. Genetic Consequences of Introgression on the Chinese Crested Tern

An important genetic consequence of introgression is its effect on the distribution of deleterious mutations in our study system. We found that introgressed regions in both tern species are enriched for moderately and weakly deleterious mutations, but not for highly deleterious ones (LoF mutations). Similar considerations have been raised in humans, where studies of Neanderthal introgression suggest that gene flow from an inbred population—in which weakly deleterious mutations accumulated due to reduced purifying selection—can introduce such mutations into the recipient population (Harris and Nielsen 2016; Juric et al. 2016).

Three non‐mutually exclusive mechanisms are likely to contribute to this pattern in terns. First, our genome‐wide analyses indicate that gene flow between the two species is bidirectional, so each species can act as both donor and recipient of introgressed haplotypes. Besides, the two species share similar demographic histories and comparable genetic load; although the great crested tern carries slightly more deleterious mutations, the difference accounts for less than 2% of the total burden. Previous work has suggested that effective population size, purifying selection efficiency and existing genetic load can influence the frequency of deleterious mutations across the genome via introgression (Harris and Nielsen 2016; Juric et al. 2016; Kim et al. 2018; Moran et al. 2021). Hence, the broadly similar demographic and genetic load of these two species likely facilitates the bidirectional exchange and persistence of deleterious mutations between them. Finally, highly deleterious mutations are expected to experience strong purifying selection regardless of genomic context, and it is therefore unsurprising that we do not observe their enrichment within introgressed regions (Kim et al. 2018). Taken together, these observations suggest that introgressed regions tend to carry weakly deleterious mutations that are inefficiently purged, whereas highly deleterious mutations are more easily removed by purifying selection.

4.3. Genetic Health of the Chinese Crested Tern and Its Conservation Implications

Our findings show that the Chinese crested tern remains genetically vulnerable following severe population decline and chronically small population size. The combination of low genetic diversity, recent inbreeding, elevated genetic load and hybridization highlights the need for integrated conservation strategies. However, we caution that our genomic analyses do not indicate that immediate genetic rescue should be a primary management priority at this stage. Instead, we argue for actions that increase genetic diversity and reduce inbreeding by increasing population size, while using genetic monitoring to carefully evaluate any future gene flow‐based interventions.

Building on these results, we therefore suggest several priority actions aligned with this strategy. First and most urgently, reducing anthropogenic threats and enhancing habitat protection to facilitate population growth should be the primary goal, as a larger population can reduce inbreeding and enhance the efficacy of purifying selection, thereby limiting the accumulation of deleterious mutations (Bertorelle et al. 2022; Hedrick and Garcia‐Dorado 2016). Second, given that external pressures such as typhoons and predators affect hatching success (Chen et al. 2015; Martínez‐Abraín et al. 2001; Sutherland et al. 2020; Wingate 1972), well‐managed captive breeding programs, such as rearing abandoned eggs or chicks for release, may help increase population size. Third, long‐term monitoring of life history traits, breeding ecology, pedigrees and hybridization is essential to evaluate the success of these actions and to inform future genetic management. With respect to hybridization with the great crested tern, our results highlight the need for cautious monitoring rather than immediate management interventions. Systematic recording of heterospecific pairs and genetic screening of chicks to detect hybrids, followed by tracking their survival and reproductive success, would allow direct assessment of fitness consequences of introgression (Moran et al. 2021; Wayne and Shaffer 2016).

In conclusion, this study demonstrates that the genetic vulnerability of the Chinese crested tern is shaped by the compounding effects of prolonged population decline, ongoing inbreeding and occasional hybridization. Severe bottlenecks leave lasting genomic legacies that cannot be rapidly reversed even as census sizes stabilize or modestly increase. Critically, two decades of conservation management represent only about two generations for this long‐lived seabird, given an estimated generation time of 11 years. This interval is far too short for meaningful purging or restoration of genetic diversity through natural processes alone. Together, these findings reinforce the view that conservation management must extend beyond demographic recovery to encompass active genomic monitoring and, where necessary, genetic intervention—an increasingly important principle for securing the long‐term viability of critically endangered species worldwide.

Author Contributions

Conceptualization: Yang Liu, Shuihua Chen, Gang Song. Sampling and fieldwork: Yiwei Lu, Siyu Wang, Zhongyong Fan, Guoling Chen, Peng Ding, Jian Zhang. Wet lab work: Guoling Chen. Data submission: Yinan Fu. Analysis: Guoling Chen, Chenqing Zheng, Lanhui Peng, Hongzhou Lin, Feng Dong, Shou‐Hsien Li. Visualization: Guoling Chen, Chenqing Zheng. Writing – original draft: Guoling Chen. Writing – review and editing: All authors.

Funding

This work was supported by the National Natural Science Foundation of China (No. 31572291 and No. 32370545), the Biodiversity Investigation, Conservation and Restoration of Rare and Endangered Animal Resources in Zhejiang (No. 2021C02044), the Zhejiang Rare and Endangered Wildlife Rescue and Conservation Project (2021–2025) (ZYF), the Observation and Assessment Program (2019–2023) of the Ministry of Ecology and Environment of China (ZYF, SHC), the Zhejiang Provincial Natural Science Foundation of China (No. LGN18C030001 and No. LGN19C040002) and the National Key Research and Development Program of China (No. 2017YFC1403500).

Disclosure

Benefit‐Sharing Statement: Benefits Generated: As detailed in our Methods, the samples for this study were sourced from wild animals and museum collections. Our work provides crucial genetic data for a critically endangered species, information that is vital for its future conservation. Furthermore, all data from this study are publicly available via open‐access databases.

Ethics Statement

Fieldwork and sample collection for this study were approved by the Zhejiang Forestry and Grassland Administration and the Wuzhishan and Jiushan Archipelago authorities (Zhejiang, China) and were conducted in accordance with the Chinese Animal Welfare Act (20090606). No additional institutional animal care and use committee approval was required beyond these permits. All procedures were designed to minimize disturbance and harm to individuals and breeding colonies.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Population structure of the Chinese crested tern and the great crested tern. (A) Cross‐validation errors across K values in ADMIXTURE. (B) Principal component analysis (PCA) of the three species. (C) PCA of 12 Chinese crested terns. (D) PCA of 20 great crested terns (GCT8 excluded). (E) Maximum likelihood tree of the three tern species. CCT, Chinese crested tern; GCT, great crested tern; LT, little tern.

Figure S2: Genetic diversity and inbreeding estimates under varying sample sizes and inference methods. (A) Nucleotide diversity. (B) Genome‐wide inbreeding coefficients (F _ROH). CCT, Chinese crested tern; GCT, great crested tern; NS, non‐significant; **p < 0.01; ***p < 0.001.

Figure S3: Inferred demographic histories of the Chinese crested tern and the great crested tern. (A) PSMC‐inferred demographic history of high‐coverage individuals (> 18×). Light and dark grey blocks indicate the Last Glacial Period (LGP) and the Last Glacial Maximum (LGM), respectively. Generation time: 11 years; mutation rate: 4.8 × 10⁻⁹ per base pair per generation. (B) GONE‐inferred demographic history with a recombination rate of 1 cM/Mb. CCT, Chinese crested tern; GCT, great crested tern.

Figure S4: Accumulation of deleterious mutations in the Chinese crested tern and the great crested tern. (A) Standardized count of heterozygous mutations. (B) Relative frequency (R _A/B) of deleterious mutations in 12 Chinese crested terns and 12 great crested terns. R _A/B > 1 indicates more derived alleles in the Chinese crested tern, and R _A/B < 1 indicates more derived alleles in the great crested tern. Data are shown as mean ± standard deviation (SD). CCT, Chinese crested tern; GCT, great crested tern; LoF, loss‐of‐function; NS, non‐significant; *p < 0.05; ***p < 0.001.

Figure S5: Derived allele frequency spectra for 12 Chinese crested terns and 20 great crested terns. CCT, Chinese crested tern; GCT, great crested tern; LoF, loss‐of‐function.

Figure S6: Ratios of homozygous LoF, deleterious and benign nonsynonymous mutations to homozygous synonymous mutations in PLINK‐inferred ROH and non‐ROH regions. CCT, Chinese crested tern; GCT, great crested tern; LoF, loss‐of‐function; NS, non‐significant; *p < 0.05; **p < 0.01; ***p < 0.001.

Figure S7: Genome‐wide heterozygosity and inbreeding coefficients for high‐ and low‐inbreeding groups of the Chinese crested tern. (A) Inbreeding coefficients (F _ROH) based on ROH inferred from (left) BCFtools and (right) PLINK. (B) Genome‐wide inbreeding coefficient (F _ROH, inferred using BCFtools) negatively correlated with the heterozygosity. (C) Genome‐wide heterozygosity. High, high‐inbreeding group; Low, low‐inbreeding group; *p < 0.05, **p < 0.01.

Figure S8: Accumulation of deleterious mutations in high‐ and low‐inbreeding groups of the Chinese crested tern. (A) Number of total mutations. (B) Number of heterozygous mutations. High, high‐inbreeding group; Low, low‐inbreeding group; LoF, loss‐of‐function; NS, non‐significant; *p < 0.05.

Figure S9: Proportion of homozygous mutations in the two Chinese crested tern groups. (A) Proportion of homozygous mutations. (B) Proportion of homozygous mutations in PLINK‐inferred ROH regions. High, high‐inbreeding group; Low, low‐inbreeding group; LoF, loss‐of‐function; *p < 0.05; **p < 0.01.

Figure S10: Inbreeding coefficients and heterozygosity under simulated demographic scenarios. (A) Inbreeding coefficient. (B) Heterozygosity. Colours in (A) and (B) correspond to the five sampling time points (200, 150, 100, 50 and 1 generation ago, with generation 1 marking the end of the bottleneck) shown in Figure 4A. CCT, Chinese crested tern; GCT, great crested tern; ***p < 0.001. NS in (A) and (B) indicates non‐significant differences in all pairs comparisons (p > 0.05).

Figure S11: Simulated extinction risk under varying bottleneck sizes. (A) Simulated demographic scenarios of the Chinese crested tern. Impact of K _bottleneck on (B) generations to extinction, (C) total mutation load and (D) realized load. Impact of K _bottleneck on the number of (E) strongly and (F) weakly deleterious mutations. (G) Impact of the distribution of fitness effect (DFE) on generations to extinction. CCT: the DFE of the Chinese crested tern; GCT: the DFE of the great crested tern; human: the DFE of Homo sapiens . **p < 0.01; ***p < 0.001; NS: non‐significant. NS in (E) and (F) indicates non‐significant differences in all pairs comparisons (p > 0.05).

Figure S12: Number of heterozygous deleterious mutations per Mb in the putative introgressed regions and non‐introgressed regions in two tern species. The putative introgressed regions were identified using HybridCheck. CCT, Chinese crested tern; GCT, great crested tern; LoF, loss‐of‐function; NS, non‐significant; ***p < 0.001.

Figure S13: Accumulation of deleterious mutations in putative mixed ancestry versus non‐mixed ancestry regions of the Chinese crested tern and the great crested tern. The mixed ancestry regions were identified using RFMix. (A) Number of total deleterious mutations per Mb. (B) Ratios of LoF, deleterious and benign nonsynonymous mutations to synonymous mutations. CCT, Chinese crested tern; GCT, great crested tern; LoF, loss‐of‐function; NS, non‐significant; **p < 0.01.

Table S1: Sampling information of the three tern species in this study.

Table S2: Genome assembly statistics for the Chinese crested tern.

Table S3: Genome assembly quality comparison between the Chinese crested tern and three other avian species.

Table S4: Genome completeness of the Chinese crested tern and chicken ( Gallus gallus ).

Table S5: Genome‐wide diversity and IUCN conservation status of 62 published avian species.

Table S6: The results of Patterson's D statistic and f‐4 ratio test between Chinese crested tern and great crested tern.

Table S7: Log marginal likelihoods, log Bayes factors and migration model probabilities from ten Migrate‐N runs.

Table S8: Estimated admixture times for individual of the two tern species.

Chen, G. , Zheng C., Peng L., et al. 2026. “Population Decline, Inbreeding and Hybridization Shape the Genetic Vulnerability of a Critically Endangered Seabird.” Molecular Ecology 35, no. 16: e70519. 10.1111/mec.70519.

Data Availability Statement

The data that support the findings of this study are openly available in National Center for Biotechnology Information at https://www‐ncbi‐nlm‐nih‐gov.eproxy.lib.hku.hk/bioproject/?term=PRJNA1480142.