Environmental DNA identifies coastal plant community shift 1,000 years ago in Torrens Island, South Australia

Coastal marine plants such as seagrasses, mangroves, and tidal marshes provide critical ecosystem services including habitat, shoreline stabilization, carbon storage, and food-web support, yet are declining due to human impacts. Understanding historical dynamics and identifying ecosystem tipping points over geologic timescales are essential to contextualize present-day change and guide effective restoration. Restoration targets are often biased by the shifting baseline syndrome due to limited long-term data. Traditional archives (observations, fossils, pollen, isotopes) provide insights but are biased or preservation-limited in coastal settings, particularly for macrophytes like seagrasses. Environmental DNA (eDNA) recovered from soil cores offers an emerging tool for reconstructing historical plant communities; when coupled with environmental proxies a detailed picture of change can emerge. This study applies a high-resolution multi-proxy approach—targeted capture eDNA together with soil chemical analyses—to four dated soil cores from a temperate wetland on Torrens Island, South Australia, to reconstruct the past ~4,000 years and identify long-term ecosystem shifts.

The paper outlines limitations of traditional methods for reconstructing coastal vegetation history: archival observations cover only decades and can be biased; macrofossils and pollen are preservation-limited and biased toward taxa that deposit identifiable material, with seagrass pollen often poorly preserved in coastal sediments. Isotopic, fossil, and pollen proxies provide valuable but incomplete long-timescale information. Recent literature shows sedimentary eDNA can complement or enhance paleoecological reconstructions by capturing taxa not represented in pollen/macro remains and enabling finer taxonomic resolution. However, standard metabarcoding is susceptible to primer bias and degraded templates. Targeted capture of multiple chloroplast gene regions reduces amplification bias and can improve taxonomic recovery. Integrating eDNA with geochemical proxies (e.g., δ13C, %Corg, bulk density, XRF elemental ratios) can yield robust, multi-line evidence for environmental and community shifts over millennia.

Study site and sampling: Four replicate soil cores (~1 m long, 7.5 cm diameter) were collected from a mangrove-dominated wetland on Torrens Island, South Australia (−34.7929, 138.5265) in 2017 (Core 1) and 2018 (Cores 2–4). The site has high sedimentation and dense Avicennia marina root systems, providing stable stratigraphy. Cores were collected with PVC pipes by manual percussion/rotation. Core compression was measured and used to compute decompressed depths. Cores were stored at 4 °C. Core 3 was split lengthwise; one half scanned by XRF and the other processed as the other cores. Sectioning and measurements: Cores were sliced into 0.5 cm increments in the top 20 cm and 1 cm increments thereafter. For each slice, a 0.5 g subsample was taken for DNA; the remainder was dried (60 °C) to constant weight for dry bulk density (DBD) calculation. Subsamples were used for radiocarbon (14C) dating, % organic carbon (%Corg), and δ13C of organic carbon. Environmental DNA: 0.5 g soil from the center of each slice was split into two replicates (A, B). DNA was extracted with QIAGEN DNeasy PowerLyzer Soil Kit using zirconia beads under stringent contamination control; extraction blanks were included (2 per 20 samples). Sonication was tested and found to reduce unique taxa in older sediments; thus, no sonication was applied. Library preparation used NEBNext Ultra II with reduced volumes and custom stubby Y-adaptors, PCR-amplified with in-house primers. Targeted capture baits were designed from RefSeq plastid sequences (~160 taxa) to target 20 chloroplast loci (approx. 15,000 120-mer probes, 2x tiling). Hybridization at 65 °C for 48 h; captured libraries were amplified, size-selected (300–600 bp), quantified, pooled, and sequenced on Illumina HiSeq X Ten (2×150 bp). Bioinformatics: Raw reads were demultiplexed (Bcl2fastq), further demultiplexed by internal barcodes (AdapterRemoval), adapters/low-quality bases trimmed (PALEOMIX/AdapterRemoval), reads <25 bp discarded. Reads were mapped with BWA-MEM (MAPQ≥30) to a reference database combining a temperate coastal plant database with a curated local plant database. Duplicates were removed (Picard). Variants were called (SAMtools mpileup; ploidy=1; baseQ and mapQ≥30; depth≥50), normalized (BCFtools), and consensus FASTA generated. Blanks failed filtering and were discarded. Due to limited discriminatory power of some loci and database gaps, consensus sequences were clustered (CD-HIT-EST, 95% identity; coverage thresholds aL 0.1, aS 1) against the reference to assign the highest reliable taxonomic rank (order/family/genus/species). Replicates A and B were merged; taxa with fewer than four genes recovered were excluded. Taxonomic assignments were summarized at family level (species/genus retained for interpretation) and relative abundance calculated per depth per core using phyloseq. Community grouping and modeling: Detected plant families were categorized as subtidal (seagrasses: Cymodoceaceae, Posidoniaceae, Ruppiaceae, Zosteraceae), intertidal (Acanthaceae, Chenopodiaceae), and high intertidal (Scrophulariaceae, Aizoaceae, Primulaceae, Convolvulaceae). A generalized additive model (mgcv::gam) assessed relative abundance change with depth across cores: Relative abundance ~ community type + s(depth, by=community type) + s(core, bs="re"), smoothing via REML. Radiocarbon dating and chronology: Three to five bulk soil subsamples per core below 22 cm were pretreated (acid washes for carbonate removal and NaOH for humics), combusted and graphitized, and measured by AMS. Ages were calibrated in OxCal 4.4.4 using SHCal20 or mixed SHCal20/Marine20 depending on marine influence inferred from δ13C; local reservoir correction AR = −150 ± 59 was applied. A hiatus in all cores precluded full age-depth models; pre-hiatus age-depth models to 63 cm were generated for cores 1, 2, and 4 using Bacon; core 3 used absolute calibrated ages only. Oldest calibrated age 4200 cal BP. Geochemical analyses: %Corg determined after acidification on elemental analyzers; δ13C measured by IRMS, normalized to VPDB with in-house standards. DBD calculated from dried mass and volume. XRF: A half-section of Core 3 scanned by Itrax XRF (Mo tube 55 kV, 30 mA, 10 s dwell, 200 µm step), producing elemental counts/ratios. Five proxies were analyzed: Molybdenum incoherent/coherent scattering ratio (Moly ratio; proxy for organic matter), Br/Cl (marine organic matter and salinity indicator), Ti/Ca (terrigenous input), Sr (normalized to Mo inc+coh; marine indicator), Ca/Fe (carbonate productivity proxy). Multivariate and changepoint analyses: PCA was applied to DBD, %Corg, δ13C for all cores; and separately to the five XRF proxies for Core 3. PC1 scores per depth were subjected to changepoint analysis (changepoint::cpt.meanvar) to identify significant shifts in mean/variance along core profiles. Identified changepoints at 63 cm (DBD/%Corg/δ13C) and 60 cm (XRF) were used to demarcate pre- (>1000 cal BP) and post-hiatus (<1000 cal BP) intervals.

• Radiocarbon chronology: A consistent hiatus was detected in all four cores, preventing full age-depth models; reliable pre-hiatus models were constructed to 63 cm for Cores 1, 2, and 4. The oldest calibrated age was ~4200 cal BP. Pre-hiatus defined as >1000 cal BP; post-hiatus as <1000 cal BP.
• eDNA community shifts with depth: Across all cores, subtidal plant DNA increased with depth (exponential trend; edf = 2.3, P < 0.05), intertidal decreased linearly with depth (edf = 1.0, P < 0.05), and high intertidal showed no significant change (edf = 1.988, P = 0.3). Random effect of core was non-significant (P = 1), indicating consistent patterns among cores.
• Species detections: Older sections (>1000 cal BP) were dominated by subtidal seagrasses, particularly Zostera nigricaulis (max relative abundance per core: Core1 32%, Core2 40%, Core3 66%, Core4 32%). Additional subtidal detections included Posidonia australis and Ruppia maritima (Cores 2 and 4) and Amphibolis antarctica (Core 2), each <10%. Younger sections (<1000 cal BP) were dominated by the intertidal mangrove Avicennia marina, reaching 100% of the recovered community at some depths. High intertidal families (e.g., Myoporum insulare, Wilsonia humilis; with single detections of Disphyma crassifolium, Samolus repens in Core 1) were present but never exceeded 35%.
• Geochemical proxies and changepoints: PCA of DBD, δ13C, and %Corg explained 73.9% of variance on PC1; all three variables strongly correlated with PC1 (R2 > 0.6), with a significant changepoint at 63 cm indicating environmental shift. In Core 3 XRF data, PC1 explained 76.4% of variance; Sr contributed most (24%) and correlated strongly (R2 = 0.92). A changepoint at 60 cm indicated a major geochemical shift. With depth, Ca/Fe and Sr increased (marine influence), while Moly ratio, Br/Cl, and Ti/Ca decreased (lower organic matter and terrigenous input), supporting a more marine environment in older layers.
• Environmental interpretation: Older sediments (>1000 cal BP) show strong marine signals and subtidal seagrass communities; younger sediments (<1000 cal BP) show mixed marine-terrestrial influence consistent with intertidal conditions and mangrove dominance. The transition likely reflects sea-level lowering (potential tectonic uplift and/or gradual climate-driven fall) that enabled mangrove colonization and may explain the depositional hiatus.
• Methodological insight: Targeted capture of multiple chloroplast loci improved coastal plant detection over traditional metabarcoding and, when paired with geochemical proxies, provided robust multi-line evidence for millennial-scale vegetation shifts.

The study demonstrates that integrating targeted capture eDNA with soil geochemical proxies can reconstruct high-resolution, millennial-scale changes in coastal plant communities. The combined lines of evidence indicate a clear ecosystem transition at Torrens Island from a subtidal seagrass-dominated system prior to ~1000 cal BP to an intertidal mangrove-dominated system over the last millennium. This directly addresses the challenge of limited, biased traditional archives and shifting baselines by revealing pre-industrial and pre-observational ecological states. The eDNA detections of mangrove DNA at depth are interpreted as likely downward root penetration and DNA shedding rather than true past presence under subtidal conditions; this underscores the importance of pairing biological and geochemical signals to avoid misinterpretation. Geochemical proxies (δ13C, %Corg, DBD, XRF ratios) corroborate environmental changes from more marine to more intertidal/terrestrial influence, aligning with the observed biotic shift. The likely drivers include sea-level lowering due to tectonic uplift and/or gradual climatic changes, possibly accompanied by erosional events that generated the stratigraphic hiatus. The approach is broadly applicable to other dynamic vegetated coastal systems and permafrost or sea-level transition zones, offering a pathway to better contextualize contemporary change and guide restoration to historically appropriate targets.

This proof-of-concept multi-proxy study reconstructs ~4,000 years of vegetation and environmental change at Torrens Island, identifying a major ecosystem shift around 1,000 years ago from subtidal seagrasses to intertidal mangroves. Targeted capture eDNA across multiple chloroplast loci, combined with soil geochemistry and radiocarbon dating, provided robust, convergent evidence of past community composition and environmental conditions. The findings highlight the value of integrating eDNA with chemical proxies to overcome limitations of traditional archives and to counter shifting baselines. Future work should incorporate additional independent proxies (e.g., pollen, macrofossils), expand reference databases to improve taxonomic resolution, refine chronologies where hiatuses occur, and apply the approach across diverse coastal settings to inform protection, conservation, and restoration strategies.

• Chronological uncertainty: A depositional hiatus in all cores prevented full age-depth modeling; only pre-hiatus models to 63 cm were constructed for some cores, and older ages rely on absolute calibrated 14C dates, limiting temporal resolution across the hiatus.
• Potential vertical DNA movement: Deep-rooting mangroves (Avicennia marina) may introduce modern DNA into older layers, potentially generating false positives for past presence; interpretation required corroboration with geochemical proxies.
• Taxonomic resolution constraints: Some chloroplast loci cannot resolve closely related species; incomplete local reference databases may misassign reads to higher taxonomic ranks or related taxa.
• Proxy-specific variability: Ti/Ca showed a mid-core spike (~25 cm) reducing correlation with PC1, possibly reflecting event-driven sediment input and complicating interpretation.
• Spatial representativeness: eDNA signals may include detrital inputs from adjacent habitats, not solely in situ vegetation, potentially inflating detection of high intertidal taxa at the core site.