Unravelling Bulk Ichthyoplankton Diversity in Vietnam: Metabarcoding Validation With Controlled Mock Samples

ABSTRACT

The sustainability of Southeast Asian fisheries hinges on high‐throughput tools for monitoring early life‐stage fish biodiversity. However, applying DNA metabarcoding to hyper‐diverse tropical ichthyoplankton requires rigorous calibration to ensure quantitative reliability. We systematically evaluated the metabarcoding workflow using controlled mock communities, revealing that taxonomic recovery is governed by a stochastic limit of detection at a normalised proxy biomass threshold of ≤ 0.05. Quantitative analysis confirmed a significant linear relationship between specimen size and read abundance (R ² up to 0.817), demonstrating that biomass‐driven template competition induces frequent false negatives for low‐biomass taxa, a phenomenon exacerbated by increasing community complexity (ANOVA: p < 0.001). To mitigate these systemic biases, we applied a size‐stratified specimen‐balancing strategy intended to increase the representation of small‐bodied components in natural bulk samples. Applying this optimised workflow to field samples from Khanh Hoa, Vietnam, we identified 139 species and unmasked a North–South biogeographic dichotomy (PERMANOVA: R ² = 53%, p = 0.001) driven by transect‐scale environmental gradients and local hydrography. Notably, we identified diversity hotspots requiring > 200,000 reads for saturation, suggesting these sites act as critical larval retention zones. The contrast between functional management zones was highly significant (p = 0.002), with the conservation area (Zone B) exhibiting higher alpha richness and a nine‐fold increase in unique indicator species compared to high‐activity areas (18 vs. 2). Our work demonstrates that comprehensive validation is vital for accurate metabarcoding, offering a robust framework to understand how ecological gradients and localised human pressures shape Vietnam's critical marine spawning sites and nursery grounds.

1. Introduction

Precise biodiversity assessment serves as the bedrock for ecological research, crucially informing our understanding of ecosystem function, facilitating environmental change tracking and enabling the formulation of science–based conservation policies. Conventional morphological identification, particularly when applied to highly delicate or early life‐stage organisms such as ichthyoplankton (fish eggs and larvae), presents significant operational hurdles; these methods are notoriously labour–intensive, time–consuming and demand highly specialised taxonomic expertise, frequently resulting in a substantial ‘identification deficit’ that constrains the scope and detail of species distribution knowledge (Jiang et al. 2022; Mateos‐Rivera et al. 2020).

The convergence of DNA metabarcoding with Next‐generation sequencing (NGS) technologies has fundamentally transformed this landscape, marking a molecular revolution in ecological surveying. This advanced method enables rapid, cost‐efficient and exceptionally high‐throughput taxonomic resolution from bulk environmental samples (Deiner et al. 2017; Taberlet et al. 2012), furnishing larger community composition data for diverse aquatic taxa, including morphologically challenging groups like zooplankton (Machida et al. 2009), phytoplankton (Marinchel et al. 2023) and benthic meiofauna (Gielings et al. 2021; Zizka et al. 2018). Specifically for ichthyoplankton, metabarcoding fundamentally augments our capacity to map the distribution of early life stages, providing empirical substantiation for the existence of critical spawning areas (Govender et al. 2023) and offering a powerful tool for evaluating the functional success of Marine Protected Areas (MPAs) by quantifying resource recruitment inside versus outside protected zones (Gold et al. 2021).

Despite its transformative potential for biodiversity monitoring, the transition of DNA metabarcoding from a qualitative to a quantitative tool remains obstructed by systemic biases inherent at every stage of the molecular workflow (Shaffer et al. 2025; Shelton et al. 2023). The accuracy of ecological metrics derived from high‐throughput sequencing is frequently compromised by technical distortions that decouple sequencing read abundance from the actual biological state of the sample. To achieve reliable quantitative inference, particularly in complex assemblages such as ichthyoplankton or zooplankton, research must critically address three interconnected challenges: biomass bias, amplification bias and the detection limit.

Biomass bias represents a fundamental disconnect where the physical mass of an organism does not linearly correlate with its DNA contribution (Elbrecht et al. 2017). In ichthyoplankton samples, this challenge is intensified by drastic physiological shifts across developmental stages; eggs and larvae exhibit substantial variation in body size, tissue density and cellular integrity (Carvalho 2022). Furthermore, taxon‐specific differences in mitochondrial DNA (mtDNA) copy number and varying DNA extraction efficiencies further confound the relationship between input biomass and read counts, leading to severe quantitative inaccuracies (Moutinho et al. 2024; Teixeira et al. 2023). The lack of species‐specific correction factors for different life stages remains a primary source of uncertainty in bulk sample analysis (Duke and Burton 2020).

Simultaneously, amplification bias remains a pervasive issue during the PCR phase. Variations in primer‐to‐template affinity across different taxa often lead to preferential amplification, where certain species are disproportionately over‐represented while others—especially those with minor primer mismatches—are underestimated or entirely missed (Bylemans et al. 2019). In high‐diversity marine assemblages, this bias can significantly skew relative abundance profiles, complicating the assessment of community structure (Lamb et al. 2019; Moutinho et al. 2024).

Finally, the detection limit of a metabarcoding assay determines its capacity to identify rare or small‐bodied taxa within a complex mixture. In heterogeneous samples, the genetic signals of low‐abundance species are frequently obscured by the ‘masking effect’ of dominant taxa or lost due to stochasticity during early PCR cycles (Bylemans et al. 2019; Evans et al. 2016; Gold et al. 2023). Identifying the minimum biomass threshold for consistent detection is critical for monitoring programmes, as the failure to detect rare but ecologically significant larvae (false negatives) or the inclusion of sequencing artefacts (false positives) can lead to profound misinterpretations of population dynamics and ecosystem functioning (Carvalho 2022; Duke and Burton 2020; González et al. 2023; Shaffer et al. 2025).

Addressing these three factors through systematically designed mock communities—controlled synthetic assemblages of taxonomically verified specimens—is an indispensable step for assessing technical variance and optimising parameters towards quantitative metabarcoding (Duke and Burton 2020; González et al. 2023; Shelton et al. 2023). Without a comprehensive understanding of how biological heterogeneity interacts with these molecular biases, the ecological interpretation of metabarcoding data remains limited to presence–absence observations, precluding the accurate assessment of biomass–based community dynamics (Carvalho 2022; Lamb et al. 2019).

The South Central coast of Vietnam, encompassing the biologically rich waters of Khanh Hoa Province (a globally significant marine biodiversity hotspot), is ecologically vital, sustained by a robust coastal upwelling system that supports substantial regional fish stocks (Bui and Phan 2022; Vo et al. 2014), yet severely imperilled by intense and unsustainable fishing pressure (Nguyen and Nguyen 2025). The Nui Chua MPA, situated within this ecological nexus, acts as an essential biological refuge (Le 2024; Vo et al. 2014; Vu 2012) whose functional significance necessitates precise and repeatable ichthyoplankton community assessment (Gold et al. 2021). To address these fundamental methodological and quantitative knowledge gaps, this research focuses on the comprehensive validation of the metabarcoding workflow for bulk ichthyoplankton samples. Subsequently, the calibrated protocol will be applied to assess the ecological integrity of the marine area within Khanh Hoa Province.

The central objectives of this study are: (i) To construct ichthyoplankton mock communities that progressively increase in complexity (e.g., from low–diversity assemblages of 10 taxa to highly diverse assemblages of up to 33 taxa) in order to simulate natural bulk samples; (ii) To conduct a systematic validation of a comprehensive DNA metabarcoding workflow, assessing its accuracy and quantitative reliability using these controlled mock samples; and (iii) To apply the calibrated workflow to in situ bulk ichthyoplankton samples from critical spawning/nursery grounds in Khanh Hoa Province to assess species diversity and community composition, contrasting the Nui Chua MPA impact with adjacent fishing grounds.

2. Materials and Methods

This study details the application and validation of a DNA metabarcoding workflow for characterising ichthyoplankton communities. The methodologies encompass a comprehensive approach from sample collection and mock community construction to DNA extraction, metabarcoding library preparation and a detailed bioinformatics and statistical analysis (Figure 1).

FIGURE 1.

Ichthyoplankton metabarcoding workflow: From sample collection and mock community validation to bulk analysis.

2.1. Sampling Design and Identification

Plankton samples were collected at critical spawning/nursery grounds in 2023 during two distinct reproductive seasons: the primary spawning season (March to May) and the secondary spawning season (July to September) (Nguyen 2016). Sampling was conducted at seven representative locations across four major ecozones along the Vietnamese coast: Hai Phong and Quang Ninh (Northern—Gulf of Tonkin), Khanh Hoa and Lam Dong (Central), Ca Mau and Vinh Long (Southeast) and An Giang (Southwest). At each location, a total of 15 sampling stations were designed across five cross‐shore transects. Each transect comprised three stations located in nearshore, middle‐shore and offshore areas. These stations were specifically arranged by distance from the coast, with the inter‐station distance ranging from 3.675 to 7.5 nautical miles.

Ichthyoplankton sampling was performed using standardised plankton nets (350 μm mesh size) across two distinct strata: the surface and the entire water column (vertical hauls). Each net was equipped with a calibrated flowmeter to quantify the volume of filtered water. Surface samples were collected via horizontal tows using a rectangular net (90 × 56 cm) deployed at a constant vessel speed of 2–3 knots for a duration of 10–12 min. To characterise the vertical distribution of the community, vertical hauls were conducted using a conical net (37 cm in diameter) at an ascent rate of 0.2–0.3 m/s (Moser and Watson 2006; Smith and Johnson 1996). For these hauls, the precise bottom depth at each station was determined in real time using a hull‐mounted sonar depth finder (echo sounder) at the exact sampling coordinates. The net was then lowered to a depth approximately 1–2 m above the seafloor—monitored via a marked deployment line—before being hauled vertically to the surface. This depth–integrated approach ensured a representative capture of both benthic‐associated and epipelagic larvae throughout the 42.5–131 m bathymetric range of the study area.

Specimens collected from these natural environments served a dual purpose: providing high‐quality genomic material for the regional DNA reference library and constructing tissue‐based mock communities (all locations). For the field application, bulk ichthyoplankton analysis was focused specifically on the Khanh Hoa area. All samples were immediately preserved in 96% molecular‐grade ethanol with a preservation ratio of at least 1:3 (sample volume to ethanol); the ethanol was replaced after 24 h to ensure long‐term DNA integrity. No permits or ethical approvals were required for this study.

In the laboratory, bulk plankton samples were carefully sorted under a Euromex StereoBlue stereoscope (Euromex, Arnhem, the Netherlands) to isolate ichthyoplankton (fish eggs and larvae). Each specimen underwent preliminary morphological identification based on diagnostic characteristics such as egg shape and diameter, as well as larval body length and pigmentation (Leis and Carson‐Ewart 2004; Shadrin et al. 2003).

To establish a robust and comprehensive reference library, we integrated these field collections with complementary specimens from non‐natural sources, including aquaculture hatcheries and local markets. In aquaculture hatcheries in Nha Trang, eggs and early‐stage larvae were sourced directly from tanks with known broodstock. These specimens provided a precise identification for the reference library, as their taxonomic identity was definitively established at the source, bypassing the inherent ambiguity of morphological identification for early life stages. Additionally, adult specimens were collected from local markets; muscle tissues were sampled and morphologically validated (Nelson et al. 2016) to provide high‐quality reference sequences for commercially significant species. In total, 20 species were integrated into the reference set (Table S1).

To optimise sample utilisation and ensure consistency across validation experiments, a size‐based specimen selection strategy was employed:

• Hatchery‐derived specimens (Eggs and early larvae): Due to their minimal individual biomass, multiple individuals of the same species were occasionally pooled within a single replicate to meet the DNA yield threshold required for sequencing. This pooling was strictly controlled; even with known broodstock, taxonomic identity was independently verified through DNA barcoding of representative individuals to ensure 100% certainty.

• Market‐derived specimens (Adults): Following morphological and molecular confirmation of the source fish, the remaining tissue was partitioned to construct mock community replicates.

• Natural larvae specimens: For field‐collected larvae, accurate taxonomic assignment relied on the DNA barcoding of each individual. Large larvae (> 15 mm) were subdivided to generate replicates, while medium (5–15 mm) and small (2–5 mm) larvae were utilised as distinct individuals per replicate.

The critical rationale for prioritising individually barcoded specimens (or genetically verified pools for hatchery samples) was to eliminate uncertainty arising from cryptic genetic variation and to avoid the inclusion of closely related sequences from potentially different species that could confound species‐level resolution. By linking each specific tissue input in the mock community to its own verified DNA barcode, we ensured that technical variance in the metabarcoding pipeline was isolated from taxonomic misidentification. This rigorous control guarantees the reliability of the biomass–to–read abundance relationship without the interference of hidden genetic diversity.

Finally, specimen dimensions—specifically egg diameter (spherical eggs), major axis (ovoid eggs) and total length of tissue (from larvae or body portions)—were measured using ImageJ software (Schneider et al. 2012). These measurements served as a reproducible proxy for relative biomass, a necessary methodological choice given that direct weighing (in grams) of microscopic ichthyoplankton is prone to significant error due to rapid evaporative loss. By using these standardised physical metrics, we established a quantitative baseline to assess the correlation between organismal size and sequencing read abundance. This integrated approach ensured that each mock community replicate reflected the biological and structural complexity of natural bulk samples while maintaining a strictly DNA‐verified taxonomic composition.

2.2. Mock Community Design and Construction

To rigorously validate the metabarcoding workflow, 14 tissue‐based mock communities were constructed, strategically designed to evaluate the method's performance across three key areas: biomass bias, amplification bias and the detection limit, all of which are influenced by community complexity. To ensure a transparent and robust evaluation, the taxonomic composition and physical characteristics of each mock community were detailed, with taxa categorised based on developmental stage (egg or larvae), specimen size (diameter or total length) and the availability of reference sequences in public databases. A comprehensive summary of these attributes, including the classification of ‘small‐sized’ specimens and those ‘without a reference’—underrepresented or lacking high‐quality sequences in public databases, is provided in Table S1.

To mimic the natural variability of ichthyoplankton, the mock communities were structured with varying relative biomasses. Due to the technical difficulty of weighing microscopic larvae and eggs, individual specimen dimensions—specifically egg diameter, larval total length and the size of adult fish tissue pieces—were recorded as practical proxies for biomass. A critical focus was placed on establishing a lower size limit of 0.3–1 mm to strictly challenge the method's sensitivity. The distribution of specimen sizes within each mock community was visualised and statistically analysed using Kruskal‐Wallis (Kruskal and Wallis 1952) tests to confirm that no significant bias existed in the sampling design across replicates (Figure S1). This deliberate focus on biomass distribution allowed for a robust assessment of how disparities in DNA input influence sequencing outcomes and detection accuracy (True Positives vs. False Positives/Negatives). This framework further enabled a precise correlation analysis between the known input biomass proportions (based on specimen dimensions) and their resulting sequence read proportions.

Amplification bias was assessed using replicated samples within the Mock 3 strategy, a low‐complexity approach including two sets of four replicates: one with 10 taxa (Mock 3a–d) and another with 14 taxa (Mock 3e–h). This design enabled a direct comparison of amplification success for specific taxa, such as Sciaenops ocellatus (Linnaeus, 1766), Epinephelus fuscoguttatus (Forsskål, 1775) and Caranx ignobilis (Forsskål, 1775).

To evaluate the detection limit, low‐biomass taxa were integrated into Mock 3 either randomly (e.g., Callionymus meridionalis (Suwardji, 1965) in replicates 3a, 3b, 3e and 3h) or consistently across a group (e.g., Epinephelus bleekeri (Vaillant, 1878) and unidentified specimens belonging to the family Engraulidae in all 3e–h samples). This setup assessed detection consistency across different community backgrounds.

Subsequently, the Mock 4 strategy—a medium‐complexity design with four identical replicates of 33 taxa—further challenged the detection limit. Specimens with minimal DNA contributions, such as Selar boops (Cuvier, 1833) (in 4c) and Abudefduf vaigiensis (Quoy & Gaimard, 1825) (in 4b), were included to evaluate how increased taxonomic richness and the resulting competitive environment affect detectability.

Finally, the Mock 5 strategy provided two independent tests: Mock 5a (10 taxa) evaluated performance with species lacking reference library matches, while Mock 5b (33 taxa) included five taxa at the 1 mm threshold to validate the detection limit within a highly diverse environment.

2.3. Bulk Ichthyoplankton Construction

Field application was conducted using bulk samples from Khanh Hoa province. Bulk samples from 15 sampling stations (designated from KH1 to KH15) were collected during the secondary spawning season. Detailed station coordinates, sampling depths and the number of fish eggs and larvae collected at each station are provided in Table S2 and Figure 2. Eggs and larvae collected at each station, which were pooled from both surface and vertical hauls, were systematically stratified into three spatial classification systems to enable comprehensive ecological and management‐based analysis:

FIGURE 2.

Ichthyoplankton sampling stations in Khanh Hoa province, Vietnam. The 15 sampling stations (KH1–KH15) are shown across nearshore, middle‐shore and offshore areas, with bathymetric contour lines used to delineate the depth‐based coastal zones. Red circles indicate sampling stations, with circle size proportional to the total number of fish eggs and larvae collected. The black diagonal line separates Zone A (outside the designated conservation area; KH1–KH9) from Zone B (Conservation Zone; KH10–KH15), which includes the core sites and buffer zone of Nui Chua Marine Protected Area. Hatched polygons indicate the core sites and buffer zone of Nui Chua MPA.

Stations were first grouped into three distance‐based coastal zones reflecting established patterns of hydrographic and environmental gradients across the shelf: the nearshore zone (KH1, KH6, KH7, KH12, KH13), characterised by shallow water (40–65 m) and high coastal influence; the middle‐shore zone (KH2, KH5, KH8, KH11, KH14), representing the transition zone (65–80 m); and the offshore zone (KH3, KH4, KH9, KH10, KH15), predominantly influenced by oceanic currents and deeper water (> 80 m).

Secondly, the stations were organised into ecological zones with five contiguous sampling units, cross‐shore continuous transects (T1 to T5), specifically T1 (KH1–KH3), T2 (KH4–KH6), T3 (KH7–KH9), T4 (KH10–KH12) and T5 (KH13–KH15)—designed to capture fine–scale community shifts along the continuous onshore–offshore gradient.

Finally, stations were categorised into Functional and Management Zones based on their designated management status and level of anthropogenic activity, facilitating the evaluation of conservation efficacy provided by the Nui Chua MPA (Le 2024; Vo et al. 2014; Vu 2012). This classification includes Zone A (High Activity Area/Exploitation Zone) (KH1–KH9), located outside the core/buffer zones and subject to intensive fishing and Zone B (Conservation Zone) (KH10–KH15), encompassing the Nui Chua MPA core and buffer zones (Figure 2), intended as a biological refuge. This functional classification enables three critical comparative analyses of community structure.

2.4. Metabarcoding Workflow

2.4.1. DNA Extraction and Library Preparation

Total genomic DNA was extracted from the mock community and bulk samples using the DNeasy Blood and Tissue kit (Qiagen, Hilden, Germany). To enhance extraction efficiency, the standard manufacturer's protocol was optimised by adjusting the ratio of Proteinase K (20 mg/mL) to ATL buffer to a range of 1.5:10 to 2.0:10. This modification was crucial for ensuring the complete digestion of tissues, which were incubated at 55°C until fully digested. Following the initial digestion, the remaining steps of the protocol were performed as specified by the manufacturer. The elution volume of AE buffer was adjusted based on the sample's biomass, varying between 50 and 100 μL to maximise the final DNA yield. Negative controls were included throughout the entire procedure to monitor contamination. Finally, the extracted DNA was evaluated for quality via gel electrophoresis and quantified using a Qubit v2.0 fluorometer (Thermo Fisher) with a 1× dsDNA HS Assay Kit (Thermo Fisher).

A two‐step PCR protocol (Chen et al. 2021) was used to amplify the mitochondrial cytochrome c oxidase subunit I (COI) region, enabling the construction of metabarcoding libraries for subsequent sequencing. The first PCR amplified the COI region from Leray et al. (2013) with Illumina TruSeq tails. Three replicates were performed for each mock and bulk sample. Each 10 μL reaction contained 2 μL of DNA template, 5 μL of 2× Hotstart Go Taq mix, 0.3 μL of the iTru_COI_F primer (10 μM), 0.3 μL of the iTru_COI_R primer (10 μM), 0.1 μL of BSA (20 mg/mL) and 3.3 μL of molecular biology grade water. The thermal cycling conditions were as follows: an initial denaturation at 95°C for 7 min; 35 cycles of denaturation at 95°C for 30 s, annealing at 46°C for 30 s and extension at 72°C for 30 s; and a final extension at 72°C for 5 min. PCR products were subsequently verified by gel electrophoresis to confirm the presence of fragments at the expected size of 430 bp. Triplicate products were then pooled, purified with KAPA beads and quantified using a Qubit fluorometer.

The second PCR attached unique i5 and i7 indices from Integrated DNA Technologies to the pooled COI amplicons to enable multiplexed sequencing. Each 25 μL reaction included 2 μL of the purified COI product, 12.5 μL of 2X Hotstart Go Taq mix, 0.75 μL of the i7 index (10 μM), 0.75 μL of the i5 index (10 μM), 0.25 μL of BSA (20 mg/mL) and 8.75 μL of molecular biology grade water. The thermal cycling programme consisted of an initial denaturation at 95°C for 5 min; 8 cycles of denaturation at 95°C for 30 s, annealing at 55°C for 30 s and extension at 72°C for 30 s; and a final extension at 72°C for 5 min. The indexed libraries were verified by gel electrophoresis for fragments at the expected size of 499 bp, purified with KAPA beads and quantified using a Qubit fluorometer.

The final metabarcoding libraries were then sequenced on the Illumina MiSeq platform at the Genomic Core Lab, Texas A&M University Corpus Christi, employing paired‐end 250 bp sequencing.

2.4.2. Sequence Analysis

Advanced bioinformatics tools were applied to compare the obtained molecular operational taxonomic units (MOTUs) with available bioinformatics databases. First, raw forward and reverse reads were processed using Trimmomatic v0.32 (Bolger et al. 2014) to remove sequences with a quality score below Q30. The filtered reads were then merged into contigs using Pear v0.9.6 (Zhang et al. 2014). These contigs were subsequently clustered with Vsearch v2.27 (Rognes et al. 2016), grouping those with more than 97% similarity.

For taxonomic assignment, the representative clustered sequences were compared against the COI mtDNA reference database on GenBank using the BLAST+ command‐line tool (Camacho et al. 2009). The classification was performed using a custom automated script based on a threshold‐based approach: a similarity of 80%–85% assigned sequences to the Class level, 85%–90% to Order, 90%–95% to Family, 95%–97% to Genus and over 97% to Species. To resolve cases where a query sequence matched multiple taxa within the same similarity threshold, a conservative lowest common ancestor approach was implemented. In such instances, the assignment was automatically rolled up to the next higher taxonomic rank (e.g., to the genus or family level) to avoid ambiguous or false‐positive species identification.

Filtering was performed to remove sequences identified in negative controls or recognised as non‐target amplification artefacts. Quantitative analysis for mock samples involved comparing observed sequence counts/proportions to known input quantities based on specimen size and biomass from individually barcoded components for tissue‐based mocks to assess quantitative accuracy, biases and detection limits.

2.5. Mock Statistical Analysis and Metabarcoding Validation

2.5.1. Data Transformation and Normalisation

All statistical analyses were conducted in the R statistical environment (R Core Team 2024), utilising a multi‐layered validation framework. For replicated communities (Mock 3 and Mock 4), specimen dimensions—specifically egg diameter, larval total length and adult tissue size—were utilised as practical proxies for organismal biomass. To ensure a robust quantitative analysis, these raw measurements were processed using Min‐Max scaling to normalise values to a 0–1 scale. This normalisation procedure (Legendre and Legendre 2012) was specifically employed to mitigate morphological biases arising from the use of disparate physical metrics. By transforming raw values (e.g., spherical diameters for eggs and linear lengths for larvae) into a standardised, dimensionless framework, the relative biomass contribution of each specimen could be compared on a unified scale. This approach ensures that the input ‘size‐proxy’ is mathematically consistent across different life stages and experimental replicates, regardless of the original unit or shape. For the non‐replicated Mock 5a and 5b, raw size values (mm) were retained to preserve the unique taxonomic and biomass compositions of these specific strategies. To account for variation in sequencing effort, read counts for each molecular operational taxonomic unit (MOTU) were expressed as relative abundances (% read count), providing the basis for subsequent correlation analyses with the normalised biomass proxies. Rarefaction curves were generated for each mock library using rarecurve and rareslope functions in vegan to assess sequencing‐depth adequacy and whether taxonomic recovery approached saturation (Oksanen et al. 2025). These mock‐based saturation patterns provided an empirical reference for evaluating taxonomic recovery and sequencing‐depth sufficiency in subsequent bulk field samples.

2.5.2. Statistical Analysis and Validation

Performance metrics, including precision, recall and the F1‐score, were calculated to evaluate the reliability of the metabarcoding workflow (Powers 2011). These metrics were used to summarise overall detection performance across mock communities, whereas detection thresholds were estimated from size proportion (%) and binary detection status. An empirical scan and binomial logistic regression were used to determine the relative size proportion associated with 90% detection reliability (Bustin et al. 2009; Forootan et al. 2017; Klymus et al. 2020).

To quantify biomass bias and assess quantitative accuracy, linear regression analyses were performed using normalised size values against observed sequence abundances. The coefficient of determination (R ²) derived from this regression provided a direct measure of the method's reliability; a high R ² value confirms that the sequencing data accurately reflect the initial biomass composition and that the protocol is resilient to significant biomass bias. Furthermore, this approach enabled the identification of the detection threshold—the specific point on the regression line, defined by the normalised biomass proxy, where the correlation with read abundance weakens and false negatives begin to dominate. This statistical confirmation was essential for establishing a biomass‐based confidence threshold, ensuring that the presence and proportions of taxa within the community could be reliably inferred only when their normalised biomass proxies exceed the empirically determined detection limit of the sequencing platform. Additionally, a one‐way ANOVA followed by a post hoc F‐test was applied to the normalised data to ensure reproducibility across replicates, confirming that the workflow consistently captured the designed community structure.

2.6. Bulk Sample Analysis: Protocol Refinement and Field Application

2.6.1. Specimen Selection and Complexity Management

Following the validation phase, the established biomass confidence thresholds and detection limits were used to guide refinement of the specimen‐selection and pooling procedure for natural ichthyoplankton samples. To mitigate biomass bias and increase the likelihood of detecting components with low DNA contributions, a size‐based specimen‐selection strategy was employed. Smaller eggs and larvae were included in greater numbers than larger individuals to increase their relative DNA contribution, with the aim of reducing their risk of non‐detection. Specimens were grouped into five size classes (XS, S, M, L and XL) according to body size prior to pooling. To manage overall community complexity and reduce potential competition for sequencing reads, bulk samples were categorised into three abundance‐based groups according to the total number of individuals (n), comprising both fish eggs and larvae per station: low‐abundance (n ≤ 100), medium‐abundance (100 < n ≤ 500) and high‐abundance (n > 500). In addition to abundance‐based classification, specimen size structure, size disparity and morphological heterogeneity were considered key components of sample complexity, as conceptually illustrated in Figure S2. For samples identified as medium or high abundance, a sub‐sampling strategy was implemented where the material was divided into equal parts to reduce taxonomic complexity and ensure robust species recovery within each library. Conversely, low‐abundance samples were processed in their entirety to maximise the capture of genetic material from rare or low‐biomass species. Rarefaction curves were subsequently generated for each bulk library to verify whether the targeted sequencing depth was sufficient to reach the saturation point of taxonomic detection, serving as a cross‐validation of the thresholds previously established during the mock trials.

2.6.2. Diversity and Community Structure Analysis

Alpha diversity indices, including Observed OTUs, Chao1, Shannon and Simpson, were calculated for all 15 stations to evaluate species richness and evenness. These metrics were then compared across three distinct grouping strategies (as mentioned above) to assess how each grouping affects diversity within a sample. The top 30 most abundant taxa were also identified and analysed at the species level. Beta diversity metrics (Bray‐Curtis dissimilarities) were used to analyse and compare the community structure between different samples. This was visualised using non‐metric multidimensional scaling (NMDS) to provide a spatial representation of community relationships. A PERMANOVA was then applied to formally test for statistically significant differences in community composition across the grouping strategies. A p‐value of < 0.05 was considered statistically significant. Additionally, the homogeneity of multivariate dispersions was verified using the PERMDISP (Anderson 2001, 2006) test to ensure that PERMANOVA results were not confounded by heterogeneous variance. All alpha and beta diversity metrics were calculated using the vegan package (Oksanen et al. 2025) in R. Visualisations were generated using the ggplot2 package (Wickham 2016) for enhanced clarity and aesthetic quality.

2.6.3. Functional/Management Zonation and Indicator Species

To investigate the functional zonation of the study area, we implemented a strategy to identify indicator species for two distinct functional and management zones: Zone A (High Activity Area) and Zone B (Conservation Zone) using an indicator species analysis (ISA). This method identifies species that are not only present but are also significantly more frequent or abundant in a particular zone compared to others. The analysis was performed using IndVal and a Monte Carlo permutation test in the indicspecies package (De Cáceres and Legendre 2009) to determine the statistical significance of each indicator species. This approach allowed for a robust, data‐driven classification of species that can serve as biological indicators for human impact versus conservation efforts. Finally, a Venn diagram analysis, using the VennDiagram package in R (Chen and Boutros 2011), was conducted to visualise the unique indicator species for Zone A and Zone B, as well as the species shared between the two zones.

3. Results

3.1. Sequencing Performance and Data Processing

Sequencing of the 29 libraries yielded a robust volume of raw data for both mock and bulk samples, together with two negative‐control libraries. Sequencing read summaries for all samples, including mock communities, bulk field samples and negative controls, are provided in Table S3. Negative controls showed negligible read counts below the filtering threshold, indicating no substantial contamination in downstream analyses. The systematic reduction in sequence counts across the bioinformatics pipeline—from raw reads to assembled and high‐quality contigs—reflected a rigorous filtering process designed to eliminate sequencing artefacts and chimeras. Notably, the clustering step effectively condensed redundant sequences into representative units, consistent with bioinformatic principles for converting read abundance into taxonomic data. Output analysis revealed a clear distinction in community complexity: mock samples produced a concentrated range of MOTUs (16–60) aligning with designed compositions and intraspecific haplotypes, while bulk samples exhibited significantly higher diversity (23–122 MOTUs) as expected from natural field communities.

3.2. Performance Metrics and Taxonomic Recovery in Mock Communities

The mock community analysis was conducted to delineate the quantitative scaling between biomass and sequencing output while identifying the technical boundaries of the metabarcoding workflow.

3.2.1. Quantitative Scaling Between Normalised Size and Read Abundance

Linear regression analysis across all experimental groups revealed a consistent positive correlation between the normalised biomass proxy and relative read abundance (Figure 3). In replicated trials, high coefficients of determination were recorded for Mock–3 (a–d) (R ² = 0.817, p < 0.001), Mock–3 (e–h) (R ² = 0.742, p < 0.001) and the high‐complexity Mock–4 (a–d) (R ² = 0.696, p < 0.001). These results provide empirical evidence for biomass bias, while simultaneously highlighting the impact of amplification bias. In the Mock–3 (a–d) community, three large‐size taxa (> 4 mm) consistently dominated the sequencing output, yet their read abundances exhibited significant stochasticity relative to their normalised sizes. For instance, Caranx ignobilis (Forsskål, 1775) reached a normalised size of 1.00 in both Mock–3a and Mock–3b, yet its read abundance increased from 26.96% in Mock–3a to 37.15% in Mock–3b. A similar inconsistency was observed for Sciaenops ocellatus (Linnaeus, 1766); although it achieved a normalised size of 1.00 in Mock–3c, its read abundance (14.79%) was lower than the 18.73% recorded for itself in Mock–3b when its normalised size was only 0.93. Furthermore, Protonibea diacanthus (Lacepède, 1802) in Mock–3d also reached a normalised size of 1.00 but yielded only 28.00% of total reads (Table S4). These data confirm that while a larger biomass proxy generally leads to higher read counts, the PCR process introduces a level of stochasticity that can decouple the final read abundance from the initial DNA concentration. This trend remained robust in the non–replicated series, with Mock–5a (n = 10) and Mock–5b (n = 33) exhibiting R ² values of 0.714 (p = 0.008) and 0.724 (p < 0.001) (Figure 3).

FIGURE 3.

A linear relationship between relative size and read abundance in mock community analysis. (A) Mock 3 (a–d) community, with an R ² value of 0.817, p < 0.001. The linear trend is described by the equation y = −10.41 + 34.24x. (B) Mock 3 (e–h) community, with an R ² value of 0.742, p < 0.001. The linear trend is described by the equation y = −5.13 + 27.73x. (C) Mock 4 (a–d) community, with an R ² value of 0.696, p < 0.001. The linear trend is described by the equation y = −3.14 + 15.4x. (D) Mock 5a community, with an R ² value of 0.714, p = 0.008. The linear trend is described by the equation y = −36.46 + 16.88x. (E) Mock 5b community, with an R ² value of 0.724, p < 0.001. The linear trend is described by the equation y = −11.29 + 5.28x.

Statistical validation via ANOVA (Table 1) confirmed a highly significant effect of relative size on read abundance (p < 0.001). While the low‐complexity Mock–3 showed consistency across replicates (p > 0.05), the high‐complexity Mock–4 exhibited significant inter‐replicate variability (p < 0.001), indicating that amplification bias intensifies with increasing taxonomic complexity. Beyond biomass‐driven trends, the results highlighted the influence of PCR stochasticity; even when taxa possessed consistent normalised sizes, their relative read counts fluctuated significantly. For example, Protonibea diacanthus (Lacepède, 1802) showed a wide read range from 14.50% to 28.76% and Epinephelus fuscoguttatus (Forsskål, 1775) varied from 1.49% to 4.75% (Table S4), suggesting that differential amplification efficiency can partially decouple read abundance from initial biomass.

TABLE 1.

ANOVA analysis of the effect of relative size and %Read on the mock communities.

Factor	df	Sum of squares	Mean of squares	F value	p
Mock (3a–d)
Relative size versus %Read count	1	3299	3299	172.730	< 0.0001
%Read count	3	65	22	1.138	0.349
Relative size	3	31	10	0.535	0.661
Mock (3e–h)
Relative size versus % Read count	1	3576	3576	232.492	< 0.0001
%Read count	3	48	16	1.043	0.382
Relative size	3	19	6	0.417	0.742
Mock (4a–d)
Relative size versus % Read count	1	1501.5	1501.5	360.106	< 0.0001
%Read count	3	83	27.7	6.633	0.0003
Relative size	3	52.5	17.5	4.198	0.007

3.2.2. Taxonomic Recovery and Detection Thresholds

Taxonomic recovery accuracy, measured by F1‐scores, ranged from 0.778 to 0.970 (Table 2). Lower performance in Mock–3f–h (F1: 0.783–0.800) was driven by 4–5 false negatives (FNs) per sample, reducing recall to 0.64–0.71. These FNs define the empirical limit of detection (LOD) at the lower end of the normalised scale (normalised size ≤ 0.05) (Table S4), where detection becomes stochastic (Figure 3). In Mock–3g and 3h, Sillago sihama (Fabricius, 1775) (real size: 1.05–1.2 mm) was recorded as an FN in 50% of the replicates, whereas the smaller Callionymus meridionalis (Suwardji, 1965) (0.81 mm) was successfully detected at a 0.11% frequency. In the high‐diversity Mock–5b (n = 33), six taxa remained undetected, including mid‐sized individuals (1.5–2.5 mm) such as Coilia rebentischii (Bleeker, 1858), Encrasicholina punctifer (Fowler, 1938) and Sphyraena jello (Cuvier, 1829). This confirms that the LOD is a statistical threshold where biomass dilution in complex samples triggers detection failure. Conversely, amplification bias was evidenced by false positives (FPs) in samples like Mock–3a (4 FPs) and Mock–4c (5 FPs), which compromised precision.

TABLE 2.

F1‐scores and performance metrics for mock sample analysis.

Sample	Pooled taxa	Meta‐output	True positive	False positive	False negative	Precision	Recall	F1‐score
Mock–3a	10	14	10	4	0	0.714	1.00	0.833
Mock–3b	10	8	7	1	3	0.875	0.70	0.778
Mock–3c	10	10	9	1	1	0.900	0.90	0.900
Mock–3d	10	10	9	1	1	0.900	0.90	0.900
Mock–3e	14	14	13	1	0	0.929	0.929	0.929
Mock–3f	14	11	10	1	4	0.909	0.71	0.800
Mock–3 g	14	9	9	0	5	1.000	0.64	0.783
Mock–3 h	14	9	9	0	5	1.000	0.64	0.783
Mock–4a	33	31	29	2	4	0.935	0.879	0.906
Mock–4b	33	36	32	4	1	0.889	0.970	0.928
Mock–4c	33	38	33	5	0	0.868	1.000	0.930
Mock–4d	33	33	32	1	1	0.970	0.970	0.970
Mock–5a	10	10	8	2	2	0.800	0.800	0.800
Mock–5b	33	27	27	0	6	1.000	0.818	0.900

Note: Performance metrics for mock community analysis based on DNA metabarcoding. Pooled taxa indicate the total number of taxa included in each mock community, and Meta–output represents the number of taxa detected. True positive (TP) refers to taxa correctly detected, false positive (FP) to taxa incorrectly detected (not present in the mock) and false negative (FN) to taxa present but not detected. Precision (TP/(TP + FP)) reflects the proportion of correct detections among all detected taxa. Recall (TP/(TP + FN)) indicates the proportion of detected taxa among those actually present, and F1‐score represents the harmonic mean of Precision and Recall. Overall, higher F1‐scores indicate better detection performance, whereas discrepancies between Precision and Recall highlight potential biases such as false positives or false negatives.

In addition to the normalised scale, size proportion (%) was used to express the relative contribution of each mock‐community component to the total size proxy. A complementary threshold analysis based on size proportion (%) and binary detection status confirmed this stochastic detection boundary (Figure S3). For the ≥ 90% detection criterion, the empirical scan identified a threshold of 1.30%, whereas binomial logistic regression estimated a more conservative threshold of 3.46% (p < 0.001). Thus, 1.30% represents the empirical lower boundary for reliable detection, while 3.46% provides a conservative probability‐based threshold. Based on these two values, an approximate 2% size proportion was identified as a practical benchmark in the controlled mock communities to reduce false‐negative risk. However, because the taxonomic identities and input proportions of organisms in natural bulk samples cannot be accurately determined, direct taxon‐specific benchmarking is not feasible for field samples. To reduce size‐related biomass imbalances, the composition of field samples was therefore adjusted based on specimen morphology, particularly body shape and size, by including a greater number of small specimens than large specimens.

To evaluate sequencing adequacy, rarefaction curves were generated (Figure S4A). The curves reached a definitive plateau between 30,000 and 50,000 reads for all communities. This early saturation, regardless of taxonomic richness (10 to 33 taxa), confirms that the majority of diversity above the LOD was captured. Consequently, a standard depth of 100,000 reads per library was determined to be more than sufficient to ensure comprehensive taxonomic coverage while recovering rare taxa.

3.3. Methodological Refinement for Natural Bulk Samples

3.3.1. Mitigating Biomass Bias Through Size‐Stratified Balancing

During mock community validation (Section 3.2.2), normalised size and size proportion were used to quantify biomass bias, characterise stochastic detection patterns and establish empirical detection thresholds. These validation findings then guided a practical size‐stratified balancing procedure in natural bulk samples to reduce, but not eliminate, biomass‐related bias. Because mock trials indicated that low‐biomass taxa became statistically vulnerable to omission below a normalised LOD of 0.05, the procedure was implemented to reduce this masking effect. This approach aimed to increase the detection probability of low‐biomass components rather than guarantee that all taxa would exceed the detection threshold.

The fish eggs and larvae within each bulk sample were partitioned into five discrete size classes: XS (< 1 mm), S (1–2 mm), M (2–3 mm), L (3–4 mm) and XL (> 4 mm). To reduce the disproportionate contribution of large specimens to the pooled biomass and DNA template, specimen numbers were adjusted according to size class: smaller individuals (XS and S) were included at frequencies 1.5–3 times higher than larger specimens, whereas larger specimens (L and XL) were physically subdivided before pooling. This workflow increased the cumulative representation of small‐bodied organisms while limiting the biomass contribution of larger individuals. Consequently, the procedure was intended to improve the likelihood of detecting low‐biomass components in complex assemblages, although it could not eliminate biomass‐related biases or ensure the detection of every taxon.

3.3.2. Evaluation of Sequencing Saturation and Technical Redundancy

While mock communities achieved a definitive plateau at 30,000–50,000 reads (Figure S4A), natural bulk samples exhibited much higher complexity. To ensure representativeness, we implemented triple‐independent PCR replicates and a targeted depth of 100,000 reads per library.

However, rarefaction analysis (Figure S4B) revealed that this redundancy was not universally sufficient for total saturation. While low‐diversity samples reached a plateau, hyper‐diverse assemblages such as KH11, KH12, KH13 and KH14 (Observed OTUs ≥ 50) maintained positive slopes even at 200,000 reads. This indicates that in real‐world tropical ichthyoplankton samples, the ‘long‐tail’ of rare taxa remains partially undetected at standard depths. Consequently, while our protocol provides a robust semi‐quantitative framework, these results establish a new empirical benchmark, suggesting that depths exceeding 200,000 reads are essential for full taxonomic recovery in high‐diversity marine monitoring.

3.4. Application to Real‐World Ichthyoplankton Samples From Vietnamese Waters

3.4.1. Ichthyoplankton Community Structure and Spatial Turnover

The high‐resolution taxonomic inventory successfully identified a diverse ichthyoplankton assemblage comprising 166 OTUs, with 139 resolved to the species level (83.7%) (Table S5). Spatial analysis of the top 30 dominant taxa revealed a pronounced spatial taxonomic turnover, delineating a fundamental biogeographic dichotomy across the study area (Figure 4; Table S6). The community structure was characterised by a distinct separation between geographic sectors: the northwestern stations (KH1–KH9) were consistently dominated by the deep‐water snapper Pristipomoides multidens (Day, 1871), whereas the southeastern stations (KH10–KH15) exhibited a categorical transition towards an assemblage dominated by the ponyfish Photopectoralis bindus (Valenciennes, 1835).

FIGURE 4.

Stacked bar chart of taxonomic composition for the top 30 ichthyoplankton taxa illustrated by their relative abundance.

This macro‐scale transition was modulated by the co‐occurrence of ubiquitous generalists and habitat‐specific specialists. Nemipterus bathybius (Snyder, 1911) demonstrated the highest ecological plasticity, occurring at 13 of 15 stations, while P. bindus maintained high localised density across 6 stations. Conversely, a substantial fraction of the total richness was contributed by spatially restricted stenotopic species recorded at single locations, such as the anemonefish Amphiprion melanopus (Bleeker, 1852), the moonfish Mene maculata (Bloch & Schneider, 1801) and the monocle bream Scolopsis japonica (Bloch, 1793) (Table S5).

3.4.2. Spatial Heterogeneity in Alpha Diversity

The distribution of alpha diversity indices exhibited significant spatial heterogeneity, reflecting the complex interaction between taxonomic richness and abundance evenness (Figure 5). Species richness metrics—Observed OTUs and Chao1—identified the southeastern margin (KH11 and KH12) as localised richness hotspots (Figure 5A,B), with maximum values reaching 93 and 122 OTUs, respectively (corresponding to 55 and 50 validated species in Table S5).

FIGURE 5.

Alpha diversity metrics of the bulk ichthyoplankton community at 15 sampling stations in Khanh Hoa Province. (A) Observed OTUs (ObsOTU); (B) Chao1 richness estimator; (C) Shannon diversity index; (D) Simpson diversity index.

While richness fluctuated significantly across the sampling grid, the Simpson index values (Figure 5D) varied within a range (0.6 to 0.9). This statistical stability reflects a robust core evenness among the dominant ichthyoplankton components, suggesting that the underlying community structure remains stable despite observed taxonomic shifts. The Shannon index (Figure 5C) integrated these patterns, confirming that the peak diversity at KH11 and KH12 resulted from the synergistic co‐occurrence of high species wealth and balanced relative abundances, as exemplified by the co‐dominance of P. bindus and the monacanthid Paramonacanthus choirocephalus (Bleeker, 1851) at these sites.

3.4.3. Ecological and Functional Comparisons

Comparative analysis across three broad ecological strata (Nearshore, Middle‐shore and Offshore) yielded non‐significant differences in all alpha diversity indices (Figure S5): Observed OTUs (p = 0.75), Chao1 (p = 0.73), Shannon (p = 0.62) and Simpson (p = 0.76). In contrast, when evaluated along the five continuous cross‐shelf transects (T1–T5), a statistically significant gradient emerged for Observed OTUs (p = 0.03), Chao1 (p = 0.033) and Simpson (p = 0.033) (Figure S6A).

Functional zonation analysis demonstrated a pronounced divergence in species richness between the two designated management areas (Figure 6B; Figure S6B). Zone B (Conservation Area) exhibited significantly higher alpha diversity than Zone A (High Activity Area), as confirmed by Observed OTUs (p = 0.0095) and Chao1 (p = 0.0076). However, the similarity in Shannon (p = 0.088) and Simpson (p = 0.39) indices suggests that while the conservation zone accumulates a higher number of species, the evenness of the abundance distribution remains statistically comparable to the areas subject to human activity.

FIGURE 6.

Alpha diversity (Observed OTUs and Shannon indices) of ichthyoplankton communities from continuous transect (A) and functional/management zone (B) samples.

3.4.4. Beta Diversity and Community Differentiation

Beta diversity analysis confirmed that community differentiation is primarily governed by transect‐scale ecological gradients rather than broad coastal zones. Grouping by Nearshore/Middle‐shore/Offshore showed no significant compositional shift (PERMANOVA: p = 0.58) (Figure S7). Conversely, the five‐transect grouping yielded a highly significant structural differentiation (PERMANOVA: R ² = 53.0%, p = 0.001) (Figure 7B), with the NMDS ordination achieving an excellent fit (stress = 0.123) (Figure 7A). PERMDISP analysis showed no significant heterogeneity in multivariate dispersion among transects (F = 3.16, p = 0.0618), indicating that the observed PERMANOVA result was primarily attributable to differences in community composition rather than unequal within‐group dispersion.

FIGURE 7.

Beta diversity of ichthyoplankton communities. (A, B) Non‐metric multidimensional scaling (NMDS) ordination and PERMANOVA analysis for continuous transects: T1 (KH1–KH3), T2 (KH4–KH6), T3 (KH7–KH9), T4 (KH10–KH12) and T5 (KH13–KH15). (C, D) NMDS ordination and PERMANOVA analysis for functional zones. The inset figures show the statistical results for each analysis. Zone A (High Activity Area) and Zone B (Conservation Area).

Furthermore, a significant distinction was validated between functional Zone A and Zone B (PERMANOVA: R ² = 34.3%, p = 0.002). However, PERMDISP detected significant heterogeneity in dispersion between the two functional zones (F = 12.01, p = 0.0053), indicating that this pattern should be interpreted cautiously because both centroid separation and differences in within‐zone variability may contribute to the observed functional‐zone effect. In contrast, the transect‐based grouping explained a larger proportion of community variation (R ² = 53%) and was not confounded by significant dispersion heterogeneity (F = 3.16, p = 0.0618).

3.4.5. Indicator Species and Functional Composition

To resolve the unique taxonomic signatures of the functional zones, an indicator species analysis was integrated with a Venn diagram of species distributions (Figure 8). The analysis identified 28 characteristic indicator taxa (Figure S8) with highly asymmetric distributions between the two management regimes.

FIGURE 8.

A Venn diagram showing the number and list of species of unique and shared indicator species between the two functional zones.

The Venn diagram revealed that Zone A was characterised by a depauperate indicator set of only two unique species, including the dragonet Callionymus planus (Ochiai, 1955) and goldbanded jobfish Pristipomoides multidens . In sharp contrast, Zone B harboured 18 unique species, dominated by reef‐associated families such as Serranidae ( Cephalopholis boenak (Bloch, 1790)) and Nemipteridae (Scolopsis spp.). A shared core of eight species, including highly mobile and adaptable taxa like the barracuda Sphyraena jello (Cuvier, 1829) and the giant trevally Caranx ignobilis (Forsskål, 1775), occurred across both areas. This discrepancy underscores that while a shared pool of generalist species persists, the Conservation Zone (Zone B) functions as a critical reservoir for specialised larval diversity, particularly for reef‐dependent taxa.

4. Discussion

4.1. Methodological Validation: Advances in Semi‐Quantitative Metabarcoding

The use of mock ichthyoplankton communities was a critical first step in validating our metabarcoding workflow under controlled conditions. The high R ² values (0.714–0.817; Figure 3) unequivocally confirm the presence of biomass bias, a phenomenon where a taxon's read count is directly proportional to its tissue mass rather than its true individual abundance (Hilário et al. 2023). This is particularly confounding in ichthyoplankton samples where developmental stages—from tiny eggs to relatively large larvae or biological variations of specimen types—exhibit drastic variations in cellular density and mitochondrial DNA copy numbers.

Our findings also shed light on the stochastic LOD, particularly for low‐biomass specimens. Despite their known presence in mock communities, species such as Sillago sihama were frequently missed, resulting in false negatives. This occurs because low‐quantity DNA from tiny organisms fails to compete effectively for primer binding and amplification resources during early PCR cycles (Duke and Burton 2020; Shaffer et al. 2025; Shelton et al. 2023). Additionally, we observed amplification bias, where taxa of consistent nominal sizes exhibited significant variability in read counts, likely due to primer‐template mismatches or the inherent stochasticity of the PCR process (Braukmann et al. 2019; Shaffer et al. 2025; Teixeira et al. 2023).

An important methodological contribution of this study is the evaluation of a size‐stratified balancing strategy applied to real‐world bulk samples. By adjusting pooling ratios to increase the representation of smaller specimens and reduce the contribution of larger specimens, the strategy was intended to improve the likelihood that low‐biomass taxa would exceed the empirical detection benchmark identified in the mock communities. However, because comprehensive taxonomic identification of eggs and larvae was not possible before pooling, the approximately 2% benchmark could not be directly assigned to or verified for individual taxa in natural samples. This complexity‐reduction strategy was intended to mitigate the ‘DNA swamping’ effect identified in our mock trials, and the resulting workflow recovered 139 species (83.7% resolution). This optimised workflow provides a practical and robust tool for sensitive biodiversity assessments, although it cannot ensure the detection of every rare or small‐bodied taxon (Duke and Burton 2020; Gielings et al. 2021; Hilário et al. 2023).

4.2. Ecological Insights: Local Hotspots and Biogeographic Turnover

Applying this optimised workflow to the coastal waters of Khanh Hoa provided high‐resolution ecological insights that would be unattainable through traditional morphology. While the study area is located within a broader coastal upwelling region, the specific sampling locations functioned as complex biological catchments. The high species richness recorded across the grid highlights the utility of metabarcoding in documenting dynamic marine environments (Deiner et al. 2017; Govender et al. 2023).

Our analysis revealed a significant synergy between alpha and beta diversity patterns. While broad ecological zones (Nearshore vs. Offshore) showed no significant difference (p = 0.58), a powerful structural differentiation emerged when samples were grouped by ecological transects (PERMANOVA: R ² = 53.0%, p = 0.001). This indicates that community assembly is governed by transect‐scale environmental gradients—such as bathymetry and local wave energy—rather than simple distance from the shore. This is further evidenced by a sharp North–South biogeographic dichotomy: the northwestern cluster (KH1–KH9) was dominated by the deep‐water snapper Pristipomoides multidens , whereas the southeastern cluster (KH10–KH15) transitioned towards a community dominated by Photopectoralis bindus and Nemipterus bathybius .

Specifically, stations KH11, KH12, KH13 and KH14 emerged as local hotspots of species richness (reaching up to 122 OTUs). Rarefaction analysis for these sites maintained a positive slope even at 200,000 reads, underscoring their immense diversity. The complex bathymetry and local eddies at these sites likely facilitate larval retention, concentrating both reef‐associated specialists and pelagic taxa. These findings establish a new empirical benchmark: in hyper‐diverse tropical waters, sequencing depths exceeding 200,000 reads are essential to fully capture the ‘long‐tail’ of rare taxa (Duke and Burton 2020).

4.3. Indicator Species and Functional Connectivity

The identification of indicator species provides a ‘molecular fingerprint’ for regional environmental monitoring. In Zone A (High Activity Area), the presence of demersal fish Callionymus planus and Pristipomoides multidens signals a community adapted to disturbed, nutrient‐rich coastal‐estuarine environments (Paradis et al. 2024; Zeng et al. 2022). In sharp contrast, Zone B (Conservation Area) harboured 18 unique indicator species, including reef‐associated taxa like Cephalopholis boenak and Scolopsis spp., confirming the effectiveness of conservation efforts in maintaining ecological complexity.

Although Zone B exhibited significantly higher richness (p = 0.0095), the presence of shared species like Sphyraena jello and Caranx ignobilis across both zones suggests potential hydrographic connectivity. These mobile generalists demonstrate that the conserved and human‐impacted zones form a dynamic, interconnected system (Nguyen and Mai 2020; Pineda et al. 2007). While anthropogenic activities exert a measurable impact on community structure (PERMANOVA: R ² = 34.3%, p = 0.002), the higher explanatory power of ecological transects (53%) suggests that broader hydrographic factors remain the dominant force shaping the assemblage. This underscores the need for an integrated management approach that transcends static boundaries to protect larval sources and ensure regional stock resilience (Dugal 2023).

4.4. Achievements, Limitations and Future Directions

This study represents a significant advancement in marine molecular ecology by successfully applying a calibrated metabarcoding workflow to characterise ichthyoplankton assemblages in a complex coastal environment. The identification of specific indicator species and the documentation of community shifts in response to both natural gradients and human activities offer a powerful tool for evidence‐based fisheries management.

However, several limitations must be acknowledged. While our size‐stratified balancing strategy aimed to reduce size‐related imbalance, it highlighted a specific quantification challenge regarding unidentified ichthyoplankton specimens. Since individual eggs and many early‐stage larvae are morphologically indistinguishable or cannot be reliably identified at the species level, pooling a large number of specimens (e.g., 10–20 per sample) carries the risk of disproportionately over‐representing a single dominant species if those individuals happen to belong to the same taxon. Conversely, rare taxa represented by only one or a few individuals may remain under‐represented. Thus, this approach balances specimen size classes rather than taxa and reduces, but does not eliminate, biomass‐ and abundance‐related biases. This accidental ‘pooling bias’ could lead to increased template competition, potentially skewing relative read abundances. Furthermore, as this study did not sequence parallel non‐normalised bulk samples, we could not empirically measure the exact magnitude of improvement or the specific biases introduced by this pooling strategy.

Additionally, metabarcoding remains essentially semi‐quantitative. Future research should incorporate internal standards (spike‐ins) to calibrate species‐specific read counts against absolute DNA concentrations, enhancing quantitative precision. Our current dataset also represents a seasonal snapshot; to fully capture the dynamic nature of these waters, time‐series sampling designs are required to account for seasonal and diurnal variations. Moreover, we acknowledge that ichthyoplankton distribution is heavily influenced by pelagic larval duration and passive transport. To further substantiate our findings regarding functional zones and hotspots, future studies should integrate metabarcoding data with biophysical dispersal modelling (e.g., Lagrangian particle tracking) to differentiate between local larval retention and long‐distance connectivity. Finally, we advocate for the continued integration of this molecular protocol with traditional morphological identification. This dual approach provides a robust cross‐validation mechanism, minimising the risk of false results and building greater confidence in using metabarcoding for large‐scale marine conservation strategies.

5. Conclusion

This research successfully validated and adjusted a DNA metabarcoding workflow for bulk ichthyoplankton samples to mitigate the biomass bias and detection limitations identified using mock communities. The application of this robust method in Khanh Hoa, Vietnam, revealed that while ecological gradients are the primary drivers of community structure, localised human pressures cause measurable and distinct negative impacts, confirming the conservation success in protected marine zones. The identified indicator species offer a powerful molecular tool for environmental monitoring and resource management in the region. Our findings affirm the necessity of workflow validation and provide a critical framework for the conservation of Vietnam's vital fish spawning grounds.

Author Contributions

Cam Hong Van conducted fieldwork, designed the study, performed laboratory work, analysed the data and wrote the manuscript draft. Long Van Nguyen supervised, conducted fieldwork and revised the manuscript. Oanh Thi Truong performed laboratory work, analysed the data and revised the manuscript. Sang Quang Tran analysed the data and revised the manuscript. Huy Quoc Pham conducted fieldwork and revised the manuscript. Binh Thuy Dang supervised, acquired the funding, designed the study, analysed the data, wrote and revised the manuscript.

Funding

This research was funded by Vingroup Innovation Foundation (VINIF) under project code VINIF.2022.DA00021.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Distribution of specimen sizes across mock communities. Specimen size was measured as egg diameter, larval total length or adult tissue‐piece size and used as a practical proxy for relative biomass. Kruskal–Wallis tests were used to evaluate whether specimen‐size distributions differed significantly among mock replicates.

Figure S2: Conceptual illustration of ichthyoplankton bulk‐sample compositions used to guide specimen selection and complexity management. (A) Low‐abundance sample composed of relatively few, uniformly small individuals; (B) medium‐abundance sample with strong size disparity, where one or a few large individuals may disproportionately contribute biomass; (C) medium‐abundance sample with heterogeneous morphology and body‐size variation; (D) high‐abundance sample dominated by numerous small individuals (e.g., eggs).

Figure S3: Detection‐threshold analysis using relative size proportion and binary detection status in mock communities.

Figure S4: Rarefaction curves used to evaluate sequencing‐depth sufficiency. (A) Rarefaction curves for mock communities, showing taxonomic saturation at approximately 30,000–50,000 reads. (B) Rarefaction curves for natural bulk ichthyoplankton libraries, showing higher complexity and incomplete saturation in highly diverse samples.

Figure S5: Alpha diversity of ichthyoplankton communities across three distinct coastal zones.

Figure S6: Alpha diversity (Chao1 and Simpson Indexes) of ichthyoplankton communities from continuous transect (A) and functional/management zone (B) samples.

Figure S7: Beta diversity of ichthyoplankton communities across three distance‐based coastal zones. (A) Non‐metric multidimensional scaling (NMDS) ordination based on Bray–Curtis dissimilarity. (B) PERMANOVA analysis comparing community composition among nearshore, mid‐shore and offshore zones. Inset panels show the statistical results for each analysis.

Figure S8: Indicator species analysis of the ichthyoplankton community, showing the Indicator Value (IndVal.g) and statistical significance (p ≤ 0.05) for each species' association with two distinct functional zones.

Table S1: Taxonomic composition, specimen source, developmental stage and size of taxa used in mock communities.

Table S2: Sampling station details and Ichthyoplankton collection summary.

Table S3: Sequencing read summary and negative‐control results for each sample during data analysis.

Table S4: Species composition, relative size and read proportion in mock communities.

Table S5: Species composition and read abundance across sampling station at Khanh Hoa province.

Table S6: Relative abundance matrix of the top 30 ichthyoplankton taxa across the 15 sampling stations, corresponding to the stacked‐bar composition plot shown in Figure 4.

Data Availability Statement

Raw sequence reads are deposited in the SRA (BioProject PRJNA1347229).

Benefits Generated: Benefits from this research accrue from the sharing of our data and results on public databases as described above.