Introduction

Over the past century, biodiversity loss has increased at a rate of 30 to 120-fold higher than the natural mean extinction rates1, considerably influencing the stability and function of lake ecosystems, mainly caused by human activities and climate change. One of the most important drivers of these changes is the rise in the global surface temperature. Since the onset of industrialization, global surface temperatures have increased by approximately 1.5 °C, with projections suggesting further increases of 1.5–4.5 °C by the end of the 21st century2. This warming is particularly pronounced in the high-latitude and cold regions, where the ecosystem is highly vulnerable to accelerated warming3. High-latitude lake ecosystems are biodiversity hotspots, home to unique species adapted to extremely cold conditions4. Global warming trends have far-reaching implications for this high-latitude region, where biological processes are highly temperature-dependent5,6.

Currently, the impacts of climate warming on the biodiversity and stability of lake ecosystems remain debatable. The stability of the lake ecosystem could be conceptualized as an ecological network, a framework that has proved essential to understanding changes in ecosystem structure and function7,8. While some studies suggest that rising temperatures can reduce biodiversity and enhance ecosystem stability by altering species metabolism9,10, other evidence points to the potential destabilization of aquatic ecosystems due to the disrupted biodiversity-mediated networks and interactions within biological communities6. Climate warming increases the intensity of species interactions and community dynamics through biochemical reactions and metabolic rates, and nearly all other rates of biological activity, contributing to the more complex ecological networks11,12. Yet, how network complexity correlates with stability remains unclear, which must be explored.

Except for temperature playing a role in the ecological network, other biotic (e.g., species composition, interaction, and diversity) and abiotic factors (e.g., nutrient enrichment caused by human activities) are also important. Changes in the biotic structure are known to affect ecosystem stability13, such as interactions between species richness and biomass, which are highly correlated with network stability6. This topic regarding the relationships between biodiversity and stability is often referred to as the diversity–stability debate14. Further investigation is required to understand how these biotic structures may change under future climate change scenarios15. One promising mechanism suggests that weak trophic links can potentially limit the oscillations between consumers and resources16. Therefore, the strength of links in biotic communities plays an important role in ecological networks17. In a word, to maintain or enhance ecological stability in lake ecosystems, it is essential to integrate their multidimensional nature18 and comprehensively quantify the overall stability of ecological communities19.

Disentangling the complex influences of global warming and human activity on dynamic lake ecosystems requires a long-term perspective on biodiversity and ecosystem stability. Undisturbed sediment cores based on sedimentary DNA (sedDNA) analysis provide a continuous, high-resolution temporal record that captures decadal to multidecadal changes, which is invaluable for uncovering the underlying mechanisms driving lake ecosystem processes and predicting future changes in ecosystem stability20,21,22,23. Owing to their rapid response to environmental changes, phytoplankton and zooplankton serve as key indicators of a lake’s ecological dynamics and overall ecosystem health24. Therefore, research on the temporal biodiversity and stability of plankton communities using sedDNA techniques could identify the relative contributions of biotic and abiotic factors to changing lake ecosystem stability and reveal their underlying mechanisms. Temporal structure and interactions within plankton communities also lay the basis for understanding how changes in environmental conditions alter community structure and stability8.

In this study, we investigated the long-term biodiversity dynamic in Lake Chagan, a seasonally frozen lake in Northeastern China (Fig. 1). We hypothesized that nutrient loadings and climate variables directly affect biodiversity, and then influence the plankton network stability. To test this hypothesis, we (1) analyzed temporal alpha (α) and beta (β) diversity, including species presence (i.e., turnover and nestedness) and species abundance (i.e., balanced variation and abundance gradient), and identified their driving factors; (2) constructed a series of network models to explore changes in network stability over the past century and analyzed the effects of temperature on the network stability; (3) explored the relationships among biodiversity, network complexity, and stability. Our study uncovered the underlying mechanisms of how warming drives plankton community structure, network complexity, and stability.

Fig. 1: Geographical location of Lake Chagan and sediment coring site.
figure 1

The sediment coring site is highlighted with a yellow circle. Land use and land cover change (LUCC) data were obtained from Gong et al.81.

Full size image

Results

Nutrient loadings in Lake Chagan and climate changes over the past century

Nutrient loading and climatic variables exhibited changes since 1938, particularly since the mid-1990s (Fig. 2). The levels of TC, TN, TOC, and OP in the sediment cores of Lake Chagan increased considerably since the mid-1990s, with a particularly sharp increase between 2013 and 2021. In contrast, these variables remained low and stable before the mid-1990s (Fig. 2a–d). Specifically, the TN and TOC content nearly tripled in 2021 compared to the 2013 levels (Fig. 2b and c), while the TC and OP content doubled in 2021 relative to the mid-1990s (Fig. 2a and d). The TP content and C/N ratio showed a slight decrease after the mid-1990s, whereas precipitation increased (Fig. 2e–g). Overall, temperature has exhibited a generally increasing trend, with fluctuations since 1938 (Fig. 2h).

Fig. 2: Temporal changes in nutrients and climatic variables over the past century.
figure 2

a–f TC, TN, TOC, OP, TP, and C/N ratios, respectively. g and h Precipitation and temperature data. Light blue rectangles indicate the periods after 1995.

Full size image

Temporal changes in diversity over the past century

A total of 6,317,708 eukaryotic sequences were clustered into 757 phytoplankton OTUs, spanning 6 phyla, 13 classes, 23 orders, 85 families, 163 genera, and 255 species. Chlorophyta dominated the community, representing more than 40% of the total DNA sequences between 1938 and 2004. This dominance shifted to Dinophyta from 2004 to 2016 before reverting to Chlorophyta (Fig. 3e). A total of 5,709,248 sequences were clustered into 47 zooplankton OTUs across 23 genera belonging to the Cladocera, Copepoda, Protozoa, and Rotifera (Fig. 3d). An important shift in community structure was observed in 2013 (Fig. 3b), which was characterized by a relatively high abundance of Cladocera.

Fig. 3: Temporal changes of plankton communities and environmental variables.
figure 3

a–c Principal coordinates analyses (PCoA) of phytoplankton, zooplankton, and environmental variables from Lake Chagan since 1938. d and e The relative abundances of zooplankton and phytoplankton at the phylum level. f Alpha diversity (Shannon, Richness, and Pielou) of phytoplankton and zooplankton communities across different phases: CGPI (phytoplankton pre-1995), CGPII (phytoplankton post-1995), CGZI (zooplankton pre-1995), and CGZII (zooplankton post-1995).

Full size image

The PLR results showed that the estimated breakpoints for the environmental variables, phytoplankton, and zooplankton were 1986 (±5.9), 1994 (±2.3), and 1997 (±4.7), respectively (Table S1). The slight lag in the zooplankton response may reflect their delayed reaction to changes in available resources. Additionally, the cluster analysis for phytoplankton identified 1995 as a critical break (Fig. S1), closely coincided with the tipping points for both zooplankton and phytoplankton dynamics (Fig. 2a and c). This consistency highlights 1995 as a representative tipping point, marking an important period when environmental changes began significantly influencing the aquatic ecosystem.

Alpha (α) diversity metrics, including the Shannon–Wiener index, species richness, and Pielou’s evenness, exhibited a consistent pattern: phytoplankton diversity was notably higher than that of zooplankton (Fig. 3f). While the Shannon–Wiener diversity of phytoplankton remained similar between phases pre-1995 (CGI) and post-1995 (CGII), the richness of phytoplankton was higher in phase CGII than in phase CGI (Fig. 3f).

Dissimilarities within phytoplankton communities in phase CGI were greater than phase CGII, both in terms of species richness (Pβjaccard mean ± SD: 0.55 ± 0.05 for CGI; Pβjaccard mean ± SD: 0.37 ± 0.09 for CGII) and abundance (Pβbray mean ± SD: 0.55 ± 0.10 for CGI; Pβbray mean ± SD: 0.32 ± 0.17 for CGII) (Fig. 4a and b, Table S2). Species turnover contributed significantly to changes in phytoplankton richness, accounting for 89% (0.89) of the observed variations, whereas nestedness accounted for only 4% (0.04) (Fig. 4f). Similarly, the abundant gradient in phase CGII (Pβgradient mean ± SD: 0.19 ± 0.16) was more pronounced than in phase CGI (Pβgradient mean ± SD: 0.14 ± 0.15), indicating that changes in phytoplankton communities during CGII were driven by this gradient (Fig. 4b). In contrast, the pairwise incidence-based dissimilarities in zooplankton ranged widely from 0.01 to 0.99 (Fig. 4c). Similarly, species replacement was the dominant driver of community changes, accounting for 89% of the variation, with nestedness contributing 4% (Fig. 4f). Overall, the dissimilarities based on the Bray–Curtis distance were greater in the CGI (Zβbray mean ± SD: 0.85 ± 0.37) than in CGII (Zβbray mean ± SD: 0.82 ± 0.27) (Fig. 4c and d, Table S2).

Fig. 4: Dissimilarity components of plankton communities and their driving factors.
figure 4

a–d Dissimilarity components (β diversity) for species richness (turnover, nestedness, and similarity) and abundance (balanced variation, abundance gradient, and similarity) of phytoplankton and zooplankton community in Lake Chagan since 1938. Light blue dots represent CGI, and light green triangles represent CGII. e Relationships between environmental variables and community structure in Lake Chagan since 1938. Asterisks denote significance levels: *p < 0.05, **p < 0.01, ***p < 0.001. f Total values and explained variations of the dissimilarities and their components.

Full size image

Regarding α diversity, both phytoplankton and zooplankton species richness negatively correlated with IP (p < 0.01) and TP (p < 0.05) (Fig. 4e). The Shannon diversity index for zooplankton communities also showed a negative relationship with TP (p < 0.01) and IP (p < 0.05). For β diversity, the turnover-related components in the zooplankton were positively correlated with lake level changes (p < 0.05), while the nestedness-related components were negatively associated with these changes. In phytoplankton, Jaccard dissimilarity and turnover-related components were positively linked to changes in IP (p < 0.01) and TP (p < 0.05). At the same time, they were negatively associated with OP (p < 0.05), TC (p < 0.001), TN (p < 0.001), and TOC (p < 0.05). Dissimilarities based on Bray-Curtis distance and balanced variation components were positively related with IP (p < 0.001) and TP (p < 0.05), and negatively associated with OP (p < 0.01), TC (p < 0.01), TN (p < 0.01) and TOC (p < 0.05). Notably, apart from these environmental factors, dissimilarities in abundant species showed an important relationship with lake-level changes (p < 0.05).

Temporal changes in stability over the past century

To investigate the temporal interrelationship between phytoplankton and zooplankton, a total of 26 temporal interdomain ecological network models (Fig. S2), along with three phase-specific network models: CGI (pre-1995), CGII (post-1995), and CG (1938-2021) were constructed based on Spearman’s correlations (Fig. 5). In these models the degree represents individual species, whereas the edges (lines) indicate interactions between species. The nodes in the CGI, CGII, and CG networks were predominantly categorized into eight, nine, and ten plankton phyla, respectively (Fig. S3). The major phytoplankton phyla were Chlorophyta (>50%), Bacillariophyta (>15%), Dinophyta (>10%), and Chrysophyta (>1.9%) (Figs. 5a–c and S3). Network models revealed three primary modules, each comprising >5% of the network. In CGI, the key modules were associated with Chlorophyta and Dinophyta (Fig. 5d and g). In contrast, the M2 module in CGII contained the highest diversity of phyla compared to the other two modules (Fig. 5e and h). Additionally, the zooplankton phyla were present in the CG network, underscoring the interconnectedness across these temporal scales (Fig. 5h and i).

Fig. 5: Phytoplankton-zooplankton interdomain network models from Lake Chagan.
figure 5

a Network model for the period pre-1995; Panel b Network model for post-1995; c Network model for 1938–2021. Nodes are colored by different phyla. d–f Modules containing more than 5% of nodes are shown in distinct colors; modules with fewer nodes are shown in gray. g–i Stacked bar charts displaying the proportion of each phylum within the module networks.

Full size image

The network in CGII, consisting of 479 edges with an average degree of 5.6, was larger than that in CGI, which had 112 edges with an average degree of 2.13 (Table 1). Overall, the network models for the CGI, CGII, and CG exhibited notable differences. The networks in CGII and CG had higher values for edges, average degree, network diameter, graph density, and average path length than those in CGI (Table 1). Degree and betweenness centrality were highest in CG, followed by CGII, and were lowest in CGI (Fig. 6a and b). Conversely, closeness centrality in CGI was the highest when compared to the CGII and CG periods (Fig. 6c). The robustness of CGII was greater than that of CGI, regardless of the proportion removed; however, the robustness of CG exceeded that of CGII when >35% of the network was removed (Fig. 6d). The Wilcoxon rank-sum test with continuity correction analysis revealed significant differences in the degree between the CG (p = 0.0007) and CGII (p = 0.0016) periods (Table S3).

Fig. 6: Temporal characteristics of network stability parameters.
figure 6

a–d Degree, betweenness centrality, closeness centrality, and robustness during the CGI, CGII, and entire CG periods, e Robustness analysis based on the removal of 25% and 50% of the proportion. f Average degree and modularity over time. g Coefficient of variation (CV) of network parameters. h Negative links of network models over time.

Full size image
Table 1 Network properties in CGI, CGII, and CG periods
Full size table

Temporal robustness analysis showed that the robustness in the pre-1995 period remained relatively stable and higher than that in the post-1995 period, with an obvious increase observed in 2021 (Fig. 6e). As the temperature increased during the past 40 years, the modularity of the network models increased, whereas the average degree decreased (Fig. 6f). The CV increased over the past 40 years (Fig. 6g), and the negative links between zooplankton and phytoplankton in the post-1995 period were higher than those in the pre-1995 period (Fig. 6h). Spearman’s correlation analysis demonstrated a negative relationship between temperature and both the average degree (p < 0.001) and the average clustering coefficient (p < 0.05) (Fig. S4). In contrast, temperature positively correlated with modularity (p < 0.05). Additionally, the average degree was also negatively correlated with precipitation (p < 0.05), lake level (p < 0.05), TC (p < 0.05), TOC (p < 0.05), TN (p < 0.01), and OP (p < 0.05), whereas the average clustering coefficient was negatively associated with OP (p < 0.05) (Fig. S4). However, the robustness showed positive correlations with TN (p < 0.001), TC (p < 0.01), TOC (p < 0.01), and OP (p < 0.05).

Relationships among the environmental variables, diversity, and stability

The PLS-PM revealed direct and indirect relationships between environmental variables, species richness, diversity, network complexity, and stability (Fig. 7a and b). The PLS-PM indicates that the predictors and response variables within each latent construct of the outer model satisfy the model’s assumptions (i.e., loading > 0.70). Additionally, the model demonstrated a good fit to the data, with a goodness-of-fit (GOF) value of 0.76 (Fig. 7a). The results showed that temperature exerts a direct effect on network complexity (path coefficient = –0.55, p < 0.001), while network complexity, in turn, has a positive impact on network stability (path coefficient = 0.74, p < 0.001). Consequently, the temperature had a strong indirect influence on network stability (Fig. 7b). In contrast, nutrient loadings, including TN and TOC, directly enhanced network stability, with values of 0.67 and 0.46, respectively (Fig. 7b, Table S4). TP negatively affected Prichness (path coefficient = –0.64, p < 0.01) and Zrichness (path coefficient = –0.49, p < 0.01). Subsequently, reductions in Prichness and Zrichness lead to decreased Pβturnover (path coefficient = –0.45, p < 0.05) and Zβturnover (path coefficient = –0.68, p < 0.05). Therefore, TP indirectly and positively affects Zβturnover and Pβturnover through its influence on Prichness and Zrichness (Fig. 7b). Furthermore, Prichness and Zrichness have indirect effects on network complexity and stability (Fig. 6b).

Fig. 7: Relationships among environmental variables, community diversity, and network stability.
figure 7

a and b Partial least-squares path modeling (PLS-PM) illustrating direct and indirect relationships among environmental variables, community richness, diversity, network complexity, and stability, with a goodness-of-fit of 0.76. Red and green solid arrows indicate positive and negative relationships (p < 0.05), respectively; dashed arrows represent non-significant paths (p > 0.05). Detailed path coefficients are provided in Table S4. c Generalized additive model (GAM) based on the Gaussian family depicting the relationship between temperature and complexity. d Relative importance of stability factors assessed using the Last method. Asterisk represents significance at *p < 0.05, **p < 0.01, ***p < 0.001.

Full size image

Based on the GAMs results (Table S5, Fig. 7c), network complexity generally decreased with the increase in temperature, although it seems to increase slightly when the temperature exceeds 6.7 °C. According to the relative importance of the top five metrics for network stability, 60.6% of the variance was explained by the model, with network complexity accounting for the most important proportion of variance (43.67%), followed by TN (28.05%) and temperature (10.93%) (Table S6, Fig. 7d).

Discussion

Understanding temporal differentiation in the succession of zooplankton and phytoplankton under varying environmental conditions is vital for understanding how the lake ecosystems respond to disturbances, particularly those driven by human activities and climate change. Our study focused on variations in structures of ecological networks in Lake Chagan over the past century, highlighting the key tipping points and driving mechanisms behind these changes. We identified the mid-1990s as a pivotal period for environmental change, marked by obvious shifts in nutrient levels and climate variables. In 1995, the increase of TC, TN, IP, and OP is attributed to increased human activity, which has dramatically enhanced nutrient enrichment. The results of the TOC and C/N ratio (>10) suggest a substantial input of terrestrial organic matter, driven by increased runoff. These findings align with those of previous studies showing a general decline in the trophic state index of Lake Chagan over the past three decades, with a slight upward trend beginning in 201325. Concurrently, a notable warming trend was observed post-1995, with annual average temperatures significantly higher than those recorded in earlier decades.

Species richness and composition, including species replacement and nestedness, are crucial for understanding the effects of climate change and human activities on biodiversity26. Our findings revealed that species turnover was more influential than species nestedness in shaping plankton community composition. Specifically, species turnover in the pre-1995 period was significantly higher than that in the post-1995 period (Fig. 4). These patterns reflect nestedness in unconnected environments, supporting the hypothesis that environmental homogeneity leads to community dissimilarity through species replacement, while the nestedness component remains low27. Zooplankton at higher trophic levels exhibited nearly twice the turnover rate of those at lower trophic levels, indicating that aquatic organisms at higher trophic positions experienced stronger β-diversity turnover gradients28.

Lake-level fluctuations significantly influenced the zooplankton community structure (Fig. 4e). Species turnover accounted for most of the observed dissimilarities and was strongly correlated with lake-level changes, underscoring the critical role of these fluctuations in shaping zooplankton dynamics29. Our study demonstrated changes in lake levels may amplify species turnover in seasonally frozen lakes with low zooplankton richness and diversity. In addition, nutrient enrichment has negatively affected the diversity of phytoplankton and zooplankton communities, which usually leads to nestedness in heterogeneous communities27,30. In other words, excess nutrient loading, as the enormous anthropogenic pressure, potentially leads to a reduction of biodiversity in aquatic systems31. In our study, the similarity in the phytoplankton community in the post-1995 was obviously higher than that in the pre-1995 period (Fig. 4a), which indicates nutrient enrichment contributes to environmental homogenization27. For instance, nutrient inputs in the early stages primarily altered the phytoplankton composition, which subsequently resulted in changes to the zooplankton community, reflecting how environmental homogeneity fostered species replacement over nestedness32.

Our study demonstrated that TP content overall decreased over time, and TN content has significantly increased (Fig. 2), highlighting the complex nutrient dynamics influenced by inputs mainly from the Huolin River and irrigation areas (Qianguo and Da’an) (Table S7). Elevated TN levels can stimulate phytoplankton growth33, leading to alterations in the structure and function of aquatic ecosystems. In contrast, the reduced TP availability may limit certain phytoplankton species34, thereby resulting in the increase of species turnover rates. Our findings aligned with relevant studies in freshwater lakes, where nitrogen pollution is becoming a dominant driver of changes in aquatic biodiversity35. These results underscore the need for integrated nutrient management strategies that address both nitrogen and phosphorus to mitigate their combined effects on biodiversity.

Temperature is a key driver of ecological processes and significantly influences the stability of competitive communities, as evidenced by extensive experimental research9,36. Our study demonstrated that increasing temperatures reduce the temporal stability of planktonic networks by decreasing the network complexity (Fig. 7). The average degree and modularity of these networks reflect the strength of species interactions and habitat heterogeneity37, an indicator of network complexity, was found to be inversely related to an increase of temperature. Regarding the species interactions, the negative links between zooplankton and phytoplankton increased over time (Fig. 6h), illustrating that consumer–prey relationships are enhanced with increasing temperature. This finding shows that warming can increase species interactions between phytoplankton and zooplankton, indicating that strengthening predation interactions could reduce the network complexity.

Changes in species interactions due to warming can lead to fluctuations in species abundance. We observed that the negative impact of species richness on network stability could be attributed to the positive contribution of consumers, with zooplankton richness being markedly higher during the post-1995 period than during the compared to pre-1995 period (Fig. 3f). This phenomenon suggests that warming associated with the food web, likely due to intensified consumer–producer interactions that lead to fluctuations in species abundance38,39. Many simulation studies have also documented that warming can reduce the stability within food webs by enhancing consumer control over producers, resulting in fluctuations in species abundance40,41. Mechanistically, temperature increases accelerate predator metabolic and consumption rates42, which can result in a shift from stable equilibrium dynamics to more chaotic or cyclical patterns, further destabilizing the ecosystem43. The destabilizing caused by rising temperatures is likely a widespread phenomenon in Northern Hemisphere lakes, driven by global warming, as it accelerates the metabolic rates of aquatic organisms8,36. These changes reduce the resistance to external disturbances, particularly in thermally sensitive systems such as shallow or seasonally frozen lakes. Compared to other northern lakes, lakes with minimal anthropogenic influence were more vulnerable to destabilization from warming44.

Although temperature plays a dominant role in influencing network stability, nutrient loading also significantly influences community dynamics. Interestingly, we found no direct correlation between nutrient loading and network stability (Fig. 7), suggesting that nutrients play a more nuanced role in dynamic ecological processes. However, nutrient loading has an important influence on species richness and phytoplankton turnover and is negatively related to stability, which implies nutrient enrichment indirectly affects network stability through its impact on species richness. This finding is supported by other studies that have identified nutrient inputs as important drivers of biodiversity, influencing both species turnover and species richness45. Additionally, the important relationships between nutrient loading, species richness, phytoplankton turnover, and network stability indicate network stability is indirectly affected by nutrient loading and mediated by biodiversity, which is consistent with previous studies on lake plankton communities6,8.

As global temperatures continue to rise, the stability of biotic communities in aquatic ecosystems, such as Lake Chagan, is likely to weaken, which could lead to reductions in biodiversity and ecosystem services. Therefore, understanding the role of temperature in controlling ecological networks is essential to develop effective conservation strategies to mitigate the impacts of climate change. This study concluded that climate warming and nutrient enrichment led to dramatic changes in the structure of plankton communities, particularly between the pre-1995 and post-1995 periods. We proposed a mechanism that warming directly weakened network complexity, subsequently enhanced network stability. Whether this mechanism was observed in Lake Chagan can be applied to other Northern Hemisphere lakes mainly depends on regional environmental conditions, including climate patterns, lake morphology, and anthropogenic pressure. These findings lay the groundwork for the use of long-term data sets to study how biotic and abiotic factors affect community composition, complexity, and stability in lake ecosystems. In summary, sedDNA metabarcoding has been demonstrated as a promising technique for reconstructing depositional histories in aquatic ecosystems, although it still carries some uncertainties. Future studies should combine traditional paleolimnological techniques with advanced non-targeted metagenomic approaches to achieve higher-resolution, more accurate long-term biological monitoring data, which will offer critical insights into lake ecosystem to temperature variability and inform the development of robust ecological models to predict the potential impacts of global warming on ecological networks.

Methods

Study region

Lake Chagan (45°05′42″–45°25′50″N, 124°03′28″–124°30′59″E), one of the ten largest freshwater lakes in China, is located in a high-latitude and cold region (Fig. 1) and is renowned for its fishery industry46. However, Lake Chagan faces various environmental challenges, including salinization and eutrophication, which pose potential risks to both public and ecological health47,48. The water pollution inflows into the lake mainly include the Qianguo and Da’an irrigation areas, Huolin River, and the heavily waterlogged areas (HWA) (Table S7). This lake covers a surface area of 372 km2 (Fig. 1) with a water storage capacity of 6 × 108 m3 and an average depth of 2.5 m25. The water inflow into Lake Chagan mainly includes the Qianguo and Da’an irrigation area, rainfall, and groundwater recharge49. According to data from the Qianguo meteorological station (www.data.cma.cn), the region has an annual average temperature of 4–5 °C, an annual average precipitation of 400–500 mm, and an annual average evaporation of 1140–1270 mm47. This lake has an ice period of ~130 days, with 0.9 m of the average ice thickness50.

Sampling and dating

In September 2021, a 28-cm sediment core was collected from the center of Lake Chagan (45°14.0′N, 124°18.6′E, Fig. 1) using a gravity corer (UWITEC, Austria) with an internal diameter of 86 mm equipped with a 60 cm PVC tube. This central location reduces variability compared to other areas of the lake, which allows it to include DNA content representing the lake’s depositional changes22,51,52. The sediment core was divided into 35 samples at 8 mm intervals in the field, which was performed after core collection using sterile equipment to minimize the risk of DNA mixing between layers. Subsamples were primarily used for two purposes: ancient DNA extraction, stored at –80 °C, and chronology and physiochemical analysis, kept in centrifuge tubes at 4 °C.

A total of 35 contiguous samples were selected for the measurement of 226Ra,210Pb, and 137Cs, which were conducted at the State Key Laboratory of Cryospheric Sciences and Frozen Soil Engineering, Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences. Core dating was performed using a high-purity Ge gamma spectrometer (ORTECGWL-120-15, USA). The core dates were calculated using the constant flux: constant sedimentation (CFCS), constant initial concentration (CIC), and constant rate of supply (CRS) 210Pb dating models. These dates were further calibrated using the fallout radionuclide 137Cs.

A comparison among varve counting, artificial radionuclides, and the 210Pbex model indicated that the CRS model was the most suitable for this core (Fig. S5). There was evidence of a “too young” age error in the CFCS and CIC models, as observed in the 137Cs results (Fig. S5). The sediment accumulation rate (SAR) is well-defined (SAR = 0.46 mm/yr, r2 = 0.809), and the age model aligns well with the 137Cs markers corresponding to 1955, 1963, and 1986 AD, with somewhat lesser agreement with the Chornobyl fallout. Consequently, the CRS model provided a reliable and convincing age model based on 137Cs data. Optimal age modeling was performed using the serac package in R (version 4.4.0)53.

Environmental variables

Environmental variables including total organic carbon (TOC), total nitrogen (TN), total phosphorus (TP), inorganic phosphorus (IP), and organic phosphorus (OP) were measured to evaluate organic matter content and nutrient dynamics. TOC and C/N ratios were used to indicate the origin of organic matter, whereas TN, TP, IP, and OP were used to indicate the nutrient loading in the lake ecosystem. A total of 35 samples were freeze-dried at –40 °C and pretreated with 10% hydrochloric acid to remove inorganic carbon. TN and TOC contents were measured using a FlashSmart elemental analyzer (Thermo Fisher Scientific, Waltham, MA, USA). The concentrations of TP, IP, and OP were determined using a spectrophotometer (Shimadzu UV2700, Japan) following the phosphorous molybdenum blue method54.

Data on the annual average precipitation, annual average temperature, and annual average evaporation were obtained from the National Meteorological Center (www.data.cma.cn) and previous studies25. Lake-level data were sourced from QianGuo Meteorological Station monitoring records and previous study25.

DNA extraction, PCR amplification, and high-throughput sequencing

Approximately 5 g of sediment from each sample was extracted using the DNeasy® PowerSoil® Pro Kit (QIAGEN, USA) following the manufacturer’s instructions. Ancient DNA extraction and PCR analyses were conducted in a dedicated clean laboratory following strict anti-contamination protocols, including physical workspace separation, UV sterilization, and the use of sterile single-use equipment to avoid contamination with modern DNA55. The quality and concentration of the extracted DNA were assessed using 1% agarose gels and a NanoDrop2000 spectrophotometer (Thermo Scientific, United States), after which the DNA was stored at –80 °C for further use.

The V9 region of the 18S rDNA gene (~130 bp) in eukaryotic phytoplankton was amplified using an ABI GeneAmp 9700 PCR thermocycler with primer pairs 960F (GGCTTAATTTGACTCAACRCG)56 and NSR1438 (GGGCATCACAGACCTGTTAT)57. Negative controls were used to confirm the absence of modern DNA contamination. The PCR reaction mixture consisted of 10 μL of 2× Pro Taq, 0.8 μL of each primer (5 μM), 10 ng/μL of template DNA, and ddH2O to a final volume of 20 μL. Amplification conditions for the 960F_1438NSR primer pair were as follows: an initial denaturation at 95 °C for 10 min, followed by 35 cycles of 1 min at 94 °C, 1 min at 55 °C, and 1 min 30 s at 72 °C, with a final extension at 72 °C for 10 min.

The mitochondrial cytochrome c oxidase subunit I (COI) gene (~390 bp) in zooplankton was amplified using primer pairs mlCOIintF (GGWACWGGWTGAACWGTWTAYCCYCC)58 and jgHCO2198R (TAIACYTCIGGRTGICCRAARAAYCA)59. For the mlCOlintF_jgHCO2198R primer pair, the amplification conditions were as follows: an initial denaturation at 95 °C for 2 min, followed by 16 cycles of 10 s at 95 °C, 30 s at 62 °C, and 45 s at 72 °C, and then 27 cycles of 10 s at 95 °C, 30 s at 46 °C, and 60 s at 72 °C. A final extension was performed at 72 °C for 5 min, followed by holding at 4 °C until halted by the user.

After PCR amplification, the products were detected on a 2% agarose gel according to the manufacturer’s instructions. Non-specific products, including larger fragments, were discarded. Only fragments matching the target amplicon size were recovered using a DNA Gel Recovery and Purification Kit (PCR Clean-Up Kit, Yuhua, China). The purified PCR products were quantified using a Qubit 4.0 (Thermo Fisher Scientific) and used for library preparation with the NEXTFLEX Rapid DNA-Seq Kit. Purified amplicons were pooled in equimolar amounts and sequenced using paired-end sequencing on an Illumina MiSeq system (Illumina, Inc., San Diego, CA, USA), following the standard protocols provided by Majorbio Bio-Pharm Technology Co. Ltd. (Shanghai, China).

Bioinformatics

The raw sequences were processed using the fastp tool (version 0.20.0) for quality control (https://github.com/OpenGene/fastp) and subsequently trimmed using FLASH (version 1.2.7) (http://www.cbcb.umd.edu/software/flash) according to the following criteria: (i) Reads were filtered to remove low-quality bases (Q < 20) using 10-bp sliding window approach. Reads <50 bp and those containing ambiguous bases (N) were discarded; (ii) paired-end reads were merged with a minimum overlap of 10 bp and a maximum mismatch ratio of 20%; and (iii) barcode and primer regions were used to assign sequences to samples, with zero barcode mismatches and up to two primer mismatches allowed. Operational taxonomic units (OTUs) were clustered with 97% similarity using the UPARSE pipeline (version 7.1) [http://drive5.com/uparse/] according to the following criteria: (i) non-redundant sequences were extracted to reduce computational redundancy; (ii) singleton sequences (non-redundant sequences appearing only once) were removed; (iii) OTU clustering was performed at a 97% similarity threshold, during which chimeras were detected and removed, and then representative OTU sequences were identified; (iv) these representative sequences were then classified at 97% similarity using the RDP classifier (version 2.2) [http://rdp.cme.msu.edu/] based on the Bayesian algorithm. Phytoplankton and zooplankton were compared with the nucleotide sequences from the National Center for Biotechnology Information (NCBI) database.

Temporal network construction

Since 1938, temporal community stability has been assessed through temporal network stability, based on a series of network models exploring co-occurrence patterns and uncovering the underlying interactions within phytoplankton and zooplankton communities. Before network construction, OTUs with a relative abundance >0.01% were selected to construct species co-occurrence network models. When constructing the ecological networks, we grouped every eight samples to maintain consistency across time series. These models were constructed using Spearman’s correlation coefficient to elucidate the temporal interactions between phytoplankton and zooplankton communities. Only data with correlation coefficients >0.6 and a significance level of p < 0.001 were retained for further analysis60. The Spearman correlations between plankton communities were calculated using the psych package61, and the construction of co-occurrence models and the analysis of network topological properties were performed using the igraph packages in R62.

To examine the network properties, multiple topological indices (e.g., average degree, node, network density, betweenness centrality, closeness centrality, average clustering coefficient, and modularity) were calculated using the igraph package62. Among all topological indices, closeness and betweenness are two complementary centrality indices used to measure the proximity of a node to all other nodes in the community, while betweenness centrality represents the significance of a node as a connector63. Temporal variations in the topological characteristics of the plankton networks were analyzed using the Wilcoxon index. Modularity reflects the concentration of nodes in the network, which could have moderate stabilizing effects on parameters when anti-modularity can significantly destabilize the ecological network64. A detailed description of all the indices and calculations is provided in the Supplementary Material. All network models were visualized using Gephi software (version 0.10.1)65.

The stability of these network models was assessed through robustness, which was defined as the proportion of remaining species in the network following random node removal66. The robustness R of a network can be calculated using the following formula67(1):

$$R=1-\frac{1}{S}\quad {\sum} _{i=1}^{S}\frac{P(i)}{S}$$
(1)

where \(S\) is the total number of species (nodes) in the network and \(P(i)\) is the fraction of the network’s species that remain connected after the removal of the ith node.

The coefficient of variation (CV) of the plankton community stability over time was calculated as the temporal stability variation68, using the following formula (2):

$${{CV}}=\frac{\sigma (s(t))}{\mu (S(t))}$$
(2)

where\(s(t)\) is the stability measure of the network at time t; \(\sigma \left(s\left(t\right)\right)\) is the standard deviation of the stability measure over time; and \(\mu (s\left(t\right))\) is the mean of the stability measure over time.

Negative links between the phytoplankton and zooplankton nodes suggest trophic interactions, typically indicative of a predator–prey relationship8. By contrast, positive relationships between these nodes may indicate occurrence patterns driven by shared environmental preferences rather than direct facilitation or competition, which promotes their co-occurrence in similar habitats.

Statistical analysis

Cluster analysis was conducted using the hclust function in the pheatmap package to classify the temporal patterns of the plankton community since 193869. Principal coordinates analysis (PCoA) was used to determine differences in phytoplankton and zooplankton communities from 1938 onwards. Piecewise linear regression (PLR) with break-points based on the PCoA axis one scores of phytoplankton, zooplankton, and environmental variables were used to explore their tipping point variations over the past century. The PCoA and PLR were performed using the vegan and segmented packages in R70,71.

α diversity, an effective tool for measuring temporal differences in plankton communities, was calculated using the vegan package in R71. Components derived from β diversity including species replacement (i.e., turnover and nestedness) based on the Jaccard dissimilarity72 and abundance (i.e., balanced variation and abundance gradient) based on the Bray–Curtis dissimilarity73, were calculated using the betapart package in R74. Spearman correlation coefficients were calculated to identify the relationships between environmental variables and 16 diversity indices, with significant correlations (p < 0.05) visualized using the pheatmap package69. Similarly, Spearman correlations were applied to examine the associations between environmental variables, network complexity properties (e.g., average degree, modularity, connectance, and average clustering coefficient), and stability indicators (e.g., robustness), with results visualized using the pheatmap and corrplot packages in R69,75.

To explore the direct and indirect effects of environmental variables on phytoplankton and zooplankton community diversity, network complexity, and stability, we conducted partial least squares path modeling (PLS-PM)76 using the plspm package77. Generalized additive models (GAMs), fitted with the Gaussian family using the mgcv package78 for modeling and the gratia package79 for visualization and diagnostics in R, were applied to analyze the relationships between the most important variable and network complexity/stability. The relative contributions of important variables (top five) for network stability were assessed using the lm function based on the Last method in the relaimpo package80. All statistical analyses were performed in R (version 4.4.0).