Long-read DNA sequencing reveals hidden seasons of lake microbes

For decades, scientists studying the microscopic life of lakes have relied on a paradox: the very technology that made microbial eukaryotes visible to molecular ecology also blurred their identities. Short-read DNA sequencing, the workhorse of environmental genomics, can catalogue millions of organisms from a drop of water, but the fragments it reads are so short that close relatives often become indistinguishable. A new year-long study of Lake Aydat in the French Massif Central, published in Metabarcoding and Metagenomics, shows what is lost when sequences are cut short — and what becomes visible when they are not. By running short-read and long-read metabarcoding side by side on the same samples, the researchers provide one of the most direct comparisons yet of the two approaches, and the results suggest that long-read sequencing is ready to transform how freshwater ecosystems are monitored.

The study focused on Lake Aydat, a shallow, nutrient-rich lake that stratifies and mixes twice a year, making it a classic model system for freshwater ecology. From November 2018 to January 2020, the team collected water from the sunlit euphotic zone at the deepest point of the lake, sampling monthly and intensifying to twice weekly during late spring. Fourteen litres of water per sample were filtered to concentrate microbial biomass, which was preserved at minus 80 degrees Celsius until DNA could be extracted. This careful, time-resolved sampling design was essential: the goal was not merely to list the species present, but to track how entire communities of algae, ciliates, fungi and parasites rose and fell across a full annual cycle.

The short-read side of the comparison used Illumina sequencing of the V4 hypervariable region of the 18S ribosomal RNA gene, a fragment of roughly 380 base pairs that has become the standard marker for surveying eukaryotic diversity. The long-read side used Pacific Biosciences Sequel II technology with circular consensus sequencing, a method that reads the same DNA molecule repeatedly to correct errors, achieving accuracy comparable to Illumina. Instead of a single hypervariable region, the long-read approach amplified nearly the entire ribosomal RNA operon — approximately 4,500 base pairs spanning the 18S gene, the internal transcribed spacer regions, the 5.8S gene and part of the 28S gene. In total, the team recovered more than 153,000 filtered long reads, from which they extracted over 20,000 full-length 18S sequences averaging about 1,232 base pairs each.

Both datasets were processed with comparable pipelines, clustering sequences into operational taxonomic units at 97 percent identity and assigning taxonomy through phylogenetic placement against curated reference databases. After filtering, the short-read dataset contained 747 OTUs and the long-read dataset 546. To make a rigorous, like-for-like comparison, the researchers performed an elegant analytical trick: they extracted the V4 region computationally from the long-read sequences, creating a hybrid dataset that could be clustered together with the genuine short-read OTUs. This allowed them to ask precisely whether differences in taxonomic assignment arose from the length of the sequence itself or from the different organisms each method happened to capture.

The answer was unambiguous. At broad taxonomic levels — supergroups and divisions — the two methods agreed well, painting similar pictures of the lake’s major eukaryotic lineages. But at finer scales the approaches diverged sharply. Among clusters containing sequences from both datasets, more than a quarter showed deeper taxonomic assignment with the long-read data, in some cases resolving up to seven additional hierarchical ranks. Overall, nearly 30 percent of long-read OTUs could be classified to genus level, compared with only 11 percent of short-read OTUs. The long-read dataset also recovered 35 genera that the short-read approach missed entirely, against just 13 genera unique to the short reads — and those short-read-only sequences tended to be rare, low-abundance taxa, consistent with Illumina’s greater sequencing depth.

Certain groups benefited disproportionately from the longer sequences. Ciliates, single-celled predators that are central to aquatic food webs, were frequently assigned only to phylum by short reads, while long reads resolved 34 percent of ciliate OTUs to genus. Chytrid fungi, parasites of phytoplankton with major ecological influence, saw genus-level identification jump from 8 percent with short reads to 36 percent with long reads. Diatoms, the silica-shelled algae that underpin much of freshwater primary production, were classified to genus in 92 percent of long-read OTUs versus 56 percent of short-read OTUs. Where the short-read data suggested a lake dominated by the diatom genus Fragilaria, the long-read data revealed substantial populations of Cyclotella and Asterionella — a finding that matched earlier microscopic observations of the lake, underscoring how short-read surveys can misrepresent community composition.

The seasonal dynamics the long-read data revealed are where the study becomes genuinely vivid. Because taxa could be identified precisely, the researchers could watch individual genera succeed one another through the year. The ciliate Condylostoma peaked in June 2019, while Carchesium, a related filter-feeder with similar dietary preferences, showed a different temporal pattern — a succession that may reflect differing life-history strategies or tolerances to temperature, oxygen and nutrient fluctuations. Among the diatoms, Cyclotella flourished in spring, Fragilariforma dominated in July, and Asterionella reached its greatest abundance in the cold months of January and April. Such fine-grained temporal partitioning among close relatives is exactly the kind of ecological process that short-read data, with its blurred taxonomy, tends to obscure.

Perhaps the most ecologically compelling result concerns parasitism. Chytrid fungi infect and lyse phytoplankton, sometimes terminating algal blooms and reshaping entire food webs, yet host–parasite dynamics are notoriously difficult to observe in nature. The long-read data allowed the team to track the chytrid genus Zygorhizidium alongside its diatom hosts month by month. Zygorhizidium abundance rose sharply between March and May, coinciding with the spring diatom bloom — a pattern consistent with laboratory work showing that Zygorhizidium affluens infects the diatom Asterionella formosa, but now observable directly in a natural lake. A December peak in chytrids followed a November surge of the golden alga Mallomonas, echoing previous reports of chytrid-driven algal declines. The data also showed simultaneous peaks of Perkinsea parasites and colonial green algae, supporting emerging evidence that these parasites, better known from marine hosts, play a similar role in freshwaters.

The authors are careful to note that short-read sequencing retains real advantages: it is cheaper, generates far greater depth, benefits from mature analytical pipelines, and excels at detecting rare taxa. Long reads also demand higher-quality DNA, since degraded environmental samples can fail to amplify across 4,500 base pairs, and reference databases remain thinner for full-length sequences, which occasionally forced more conservative assignments. Indeed, in a small minority of cases the short reads achieved finer classification, likely because the V4 region is better represented in reference databases. These caveats matter for anyone designing a monitoring programme, and the study does not suggest that short reads should be abandoned.

But the trajectory is clear. As sequencing costs fall and reference databases grow, the study argues that long-read metabarcoding should move from specialist tool to routine instrument of environmental surveillance. The ability to resolve genera and species in time series opens the door to tracking bioindicator ciliates for water quality, detecting parasitic outbreaks before blooms collapse, and mapping the cryptic diversity that short fragments cannot distinguish. Combined with the 28S sequences the team plans to analyse next, and with emerging single-cell and genome-based reference collections, long-read surveys promise to close taxonomic gaps that have limited freshwater ecology for two decades. For the invisible majority of lake life, the picture is finally coming into focus — one long read at a time.

Subject of Research: Comparing short-read and long-read metabarcoding of eukaryotic diversity in a lake over one year

Article Title: Profiling the eukaryotic diversity over a year in a lake ecosystem through short- and long-read metabarcoding

Article References: Chauvet, M., Terzian, P., Moné, A., Debroas, D., Bronner, G., & Lepère, C. (2026). Profiling the eukaryotic diversity over a year in a lake ecosystem through short- and long-read metabarcoding. Metabarcoding and Metagenomics, 10, Article e184566. https://doi.org/10.3897/mbmg.10.184566

Image Credits: AI Generated

DOI: 10.3897/mbmg.10.184566

Keywords: metabarcoding, long-read sequencing, PacBio, eukaryotic diversity, Lake Aydat, 18S rRNA, chytrid fungi, diatoms, ciliates, freshwater ecology, seasonal dynamics, host-parasite interactions