Sediment Metagenomes as Time Capsules of Lake Microbiomes

Lakes are critical freshwater resources under mounting pressure from climate change and other anthropogenic stressors. The reconstruction of ecological time series from sediment archives with paleolimnological techniques has been shown to be an effective means of understanding how humans are modifying lake ecosystems over extended timescales. In this study, we combined shotgun DNA sequencing with a novel comparative analysis of surface water and sediment metagenomes to expose the diversity of microorganisms preserved in lake sediments. The detection of DNA from a broad diversity of preserved microbes serves to more fully reconstruct historical microbiomes and describe preimpact lake conditions.

mental data. Internal lake dynamics and changes driven by terrestrial and atmospheric inputs are integrated and archived through sediment deposition (3,4). As climate change and other anthropogenic stressors exert increasing pressures on lakes (5), it is of great interest to examine how lakes are being modified from their preindustrial states. In the absence of long-term monitoring, chronologies reconstructed from lake sediment records can describe departure from preimpact baselines (6) and provide insights into the myriad effects of the anthropogenic impact observed since circa 1880 or earlier (7)(8)(9). Classical microbial paleoindicators (e.g., diatom and chrysophyte assemblages) identified from subfossils, pigments, and other proxies are widely applied in paleolimnology for inferring key aspects of historical water columns (10,11).
Molecular genetic techniques have revolutionized environmental microbiology by exposing a vast, uncultured diversity of microorganisms (12). Molecular genetic analysis, advanced by PCR gene amplification and shotgun metagenomics, is continually expanding the body of knowledge on the diversity and importance of microbial communities in all habitable environments, including aquatic ecosystems (13). The development of molecular genetic methods in paleolimnology has enhanced the detection of soft-bodied microorganisms that lack preserved morphological features but that can be tracked in sediment DNA archives (14,15). The inclusion of a broader diversity of microorganisms in paleolimnological reconstructions allows access to the microbiomes which underlaid lake ecosystems and biogeochemical processes in the past. The detection of preserved microbial diversity further offers opportunities to explore the still largely untapped potential of microbes as paleoindicators, grounded in the ubiquity of microorganisms in aquatic systems, their unparalleled phylogenetic and ecological breadth (16), and their rapid responses to change (17). Microbial time series reconstructed from lake sediment DNA have so far been used to investigate the emergence of toxic and bloom-forming Cyanobacteria (18)(19)(20)(21), microeukaryotic community succession in response to eutrophication and climate change (22,23), microbial evolution in response to anthropogenic mercury loading (24), and past methane oxidation dynamics (25). The PCR-directed methods dominating molecular genetic analysis in paleolimnology are useful for reporting partial taxonomic and functional components of historical microbiomes. However, reconstructing historical diversity through PCR amplicon sequencing is limited by taxonomic blind-spots (26), restricted gene quantification value (27), and dependency on intact DNA template (28). Furthermore, many major microbial groups have never been targeted in standard single-gene surveys, leaving unexplored any new insights they might yield on past lake ecosystems.
Metagenomics has the potential to provide a powerful new perspective to probe the unconstrained diversity preserved in sediment archives. Shotgun sequencing has gained traction in other ancient DNA disciplines (e.g., studies of ancient hominins in evolutionary anthropology) to recover ultrashort DNA fragments not amplifiable by PCR and to avoid other PCR biases (29). To date, only a handful of paleolimnological studies have applied shotgun sequencing. Pedersen et al. reconstructed the postglacial succession of plants, algae, and animals from lake sediment metagenomes (30). Subsequent research demonstrated that shotgun sequencing expands the richness of catchment flora detected in lake sediments (31) and that ancient chloroplast and mitochondrial genomes can be resolved from the DNA remains of vegetation (32) and algae (33). Aside from the studies on eukaryotic algae, the only other paleolimnological reconstruction to describe autochthonous microbiomes with shotgun metagenomics profiled the succession of archaea in an ancient lake (34). In these studies, historical DNA represented tiny fractions of sediment metagenomes but still furnished new information (i.e., not overlapping with classical proxies) about past lacustrine or catchment biota. As these works explore the potential of metagenomics, they underscore the primary challenges associated with reconstructing historical records from sediment metagenomes. Sediment DNA pools derive foremost from indigenous sediment microbes, and historical DNA is subject to high turnover (35). Historical signals are difficult to detect because they are weaker than, and are distorted by, indigenous sediment populations.
In this study, whole-metagenome shotgun sequencing was used to probe the diversity of microorganisms preserved in contemporary and preindustrial sediment archives to expand the taxonomic scope of lake microbiome reconstructions. To identify historical DNA in sediment metagenomes, microbial preservation signals were captured by mapping sediment metagenome reads to surface water metagenome assemblies. This fragment recruitment strategy was used to disentangle historical DNA from the indigenous sediment background, using the contemporary diversity in the overlying water column as a reference. The detection of preserved DNA archives was subsequently expanded in the unconstrained explorations of whole-sediment metagenomes, guided by leads developed through metagenome capture. By teasing out the DNA preservation signals of historical lake microbiomes in the sediment record, we are able to harness a more complete perspective on the diversity of lakes and their ecological functioning over decades to millennia.

RESULTS
Environmental context. Whole-metagenome shotgun sequencing was performed on surface water and sediment DNA sampled from three regionally representative eastern Canadian lakes: Lac Paula, Eightmile Lake, and Grand lac Touradi ( Fig. 1; Table 1). Lac Paula is a medium-sized, deep lake set in a small, equal parts clear-cut and naturally forested watershed in central Quebec in the Boreal Shield ecozone. Eightmile Lake is a small lake of intermediate depth set in a small, ϳ38% clear-cut watershed in northern New Brunswick in the Atlantic Highlands ecozone. Grand lac Touradi is a large,  (78) and political boundaries of Canada (79). Spatial coordinates, morphological characteristics, and physicochemical profiles of study lakes are summarized in Table 1.
Sediment Metagenome Time Capsules deep lake in an expansive, naturally forested watershed in eastern Quebec in the Atlantic Highlands ecozone. Top sediments were sampled from the recently deposited, upper 1 cm of each core, and bottom sediments were sampled from preindustrial deposits (on average, 40 cm below the water-sediment interface). A conservative lead 210 Pb (lead-210) dating model placed all bottom sediment ages in the preindustrial background (i.e., pre-1880 Common Era [C.E.]) (see Fig. S1 in the supplemental material).
Preserved metagenome capture. We define captured metagenomes as products of in silico read recruitment and free metagenomes as whole metagenomes unconstrained by capture. Microbial DNA preserved in sediment metagenomes was captured by mapping top and bottom sediment metagenome reads to surface water metagenome assemblies (Fig. 2a). Some low-similarity alignment was expected to arise from distant evolutionary relationships between freshwater microorganisms and indigenous sediment populations. To disentangle historical and indigenous sediment microbiome DNA, capture of closely related sequences was performed with a stringent 90% identity threshold, empirically derived from the consistent delineation of closely (Ն90%) and distantly (Ͻ90%) related populations (Fig. 2b). Most scaffolds in captured metag-  Table S1 in the supplemental material). enomes did not contain highly conserved ribosomal or transfer RNA genes (97.9% Ϯ 1.0%) (see Table S1 in the supplemental material). Still, to avoid overinflating the estimates of DNA preservation from false-positive detections of distantly related organisms aligned at conserved sequences, scaffolds containing ribosomal or transfer RNA genes were removed (Table S1). Mapping between sediment and surface water metagenomes was detected across all lakes (0.2 to 0.8%), but it was lower than mapping between bottom and top sediment metagenomes (5.5 to 21.0%) ( Table 2). On average, 98.9% Ϯ 0.8% of scaffolds in captured metagenomes contained at least one annotated gene, and the majority of scaffolds across captured metagenomes contained a single gene (56.9% Ϯ 11.8%) (Fig. S2a). Taxonomic diversity in free and captured metagenomes. We performed taxonomic assessments of free and captured metagenomes to compare the compositions of natural surface water and sediment microbiomes with the microbial diversity preserved in sediment DNA archives ( Fig. 3; Table S2). Free surface water metagenomes in all lakes contained high relative coverage of Actinobacteria, Betaproteobacteria, Alphaproteobacteria, and Bacteroidetes. Free top sediment metagenomes also contained high relative coverage of Actinobacteria and Betaproteobacteria but were also rich in Deltaproteobacteria, Euryarchaeota, and Gammaproteobacteria. Free bottom sediment met- Analyses of the captured metagenomes considered at the phylum level demonstrated a relatively consistent assemblage of preserved microbiota, with a few differences between contemporary and preindustrial sediments ( Fig. 3; Table S2). Comparisons of the captured surface water signals in either top or bottom sediment metagenomes contained high relative coverage of viruses, Actinobacteria, Betaproteobacteria, Cyanobacteria, and Alphaproteobacteria. Captured surface water signals in bottom sediment metagenomes also contained high relative coverage of Planctomycetes. Captured top sediment signals in bottom sediment metagenomes contained high relative coverage of Euryarchaeota and Deltaproteobacteria.
To elucidate the taxonomic diversity of free and captured metagenomes at higher phylogenetic resolution, taxonomic analyses were performed at the order level ( Fig. 4a; Table S3). A principal-component analysis (PCA) of these data demonstrated a clear distinction between free and captured metagenomes along the first principal component (PC) axis, reflecting the taxonomic distinctness of surface water and sediment microbiomes (Fig. 4b). Free and captured surface water metagenomes were separated along the second PC axis, suggesting that only a subset of microbial diversity is preserved in sediments (Fig. 4b). The first two PC dimensions together explained 64.8% of the variation between metagenome taxonomic compositions. "Candidatus Nanopelagicales" and Pelagibacterales were highly covered in the free surface water metagenomes of all lakes, but they were not highly covered in captured surface water metagenomes ( Fig. 4a; Table S3). Free and captured sediment metagenomes contained high relative coverage of Methanomicrobiales, Syntrophobacterales, Desulfobacterales, Clostridiales, Desulfuromonadales, Myxococcales, Nitrosomonadales, Rhizobiales, and Nitrospirales ( Fig. 4a; Table S3).
Highlights of preservation signals in captured metagenomes. Since most freshwater Cyanobacteria are photosynthetic, Cyanobacteria DNA detected in the metagenomes of subsurface sediments can be distinguished from the indigenous sediment background and identified as the preserved DNA of Cyanobacteria that inhabited historical sunlit lake biomes. Captured surface water metagenomes exhibited higher relative coverage of Cyanobacteria than free surface water metagenomes, highlighting recruitment capture as a viable strategy for detecting well-preserved historical microorganisms ( Fig. 3; Table S2). Of the three lakes, Grand lac Touradi displayed the most sizable relative capture of preserved Cyanobacteria in contemporary and preindustrial sediments ( Fig. 3; Table S2). Synechococcales were the most highly represented Cyanobacteria in captured surface water metagenomes ( Fig. 4a; Table S3).
Captured surface water metagenomes also showed that sediments accumulate DNA from some-but not all-bacterioplankton, as well as viruses. The most highly covered bacterioplankton that were represented in captured surface water metagenomes were Corynebacteriales, Burkholderiales, Planctomycetales, Sphingomonadales, and Rhizobiales ( Fig. 4a; Table S3). Burkholderiales were highly covered in both free and captured surface water metagenomes and in free top sediment metagenomes, suggesting that some planktonic Burkholderiales are preserved in sediments, while other Burkholderiales are indigenous to sediments and may potentially confound historical signals without higher-resolution taxonomic assessments ( Fig. 4a; Table S3). Some highly abundant bacterioplankton (Ca. Nanopelagicales, Pelagibacteriales) were only marginally covered in captured surface water metagenomes ( Fig. 4a; Table S3). Viruses were highly covered in the captured surface water metagenomes of all lakes, but they were not highly covered in the free surface water metagenomes ( Fig. 3; Table S2). Caudovirales were the most highly captured viruses, especially in Eightmile Lake ( Fig. 4a; Table S3). A phylogeny inferred from the terminase large subunit protein (TerL) identified a wide diversity of Caudovirales, but host identities could not be resolved (Fig. S3).
Exploring unconstrained preservation signals in free metagenomes. While metagenome capture was practical for developing leads on preserved taxa, experimental limitations rendered this approach unsuitable for recruiting unconstrained preservation signals for historical lake microbiomes. Specifically, the seasonal, depth-discrete, and contemporary community structures of surface water snapshot samples constrained the capture of a broad microbial diversity. To expand the detection of preserved DNA to include the diversity that escaped capture, the search for preserved taxa was extended to the unconstrained DNA libraries of free sediment metagenomes. Across all lakes, free top and bottom sediment metagenomes contained traceable accumulations of genes from Cyanobacteria, bacterioplankton, and microeukaryotes (Fig. 5). Genes assigned to Cyanobacteria (Synechococcus, Aphanothece, and Cyanobium) numbered in the thousands in free surface water metagenomes and were detected to a lesser extent in top and bottom sediment metagenomes (Fig. 5). Sediment metagenomes contained genes from other bacterioplankton, including Betaproteobacteria (Rhodoferax, Polynucleobacter, and Limnohabitans), Alphaproteobacteria (Sphingomonas), Actinobacteria (Mycobacterium, Mycolicibacterium, Ilumatobacter, and Rhodoluna) (Fig. 5), and others (Fig. S4). A small number of genes from eukaryotic plankton, namely, diatoms (Phaeodactylum and Thalassiosira), were also detected in sediment metagenomes (Fig. 5). Some bacterioplankton ("Candidatus Methylopumilus," "Candidatus Fonsibacter," "Candidatus Planktophila," and "Candidatus Nanopelagicus") that were highly covered in surface water metagenomes were represented by relatively low gene counts in free sediment metagenomes (Fig. 5).
To demonstrate how preserved microbial diversity might be identified at high phylogenomic resolution, surface water and sediment metagenome reads were mapped to publicly available reference genomes from Polynucleobacter and Limnohabitans, two genera of Betaproteobacteria with well-described freshwater microdiversity and that were common in captured metagenomes. Surface water metagenomes from Eightmile Lake and Grand lac Touradi contained Polynucleobacter and Limnohabitans populations with high sequence similarity (Ն90%) to reference genomes (Fig. S5). In contrast, reference genomes were not captured in any of the top or bottom sediment metagenomes (Fig. S5).

DISCUSSION
This study illustrates that shotgun sequencing can access a broad diversity of microorganisms preserved in contemporary and preindustrial lake sediments. Preserved taxa were identified by mapping sediment metagenome reads to reference surface water metagenome assemblies with a stringent 90% sequence identity threshold derived to disentangle freshwater signals from the indigenous sediment background. Once well-preserved lineages were identified through the in silico capture of surface water signals in sediment metagenomes, unconstrained explorations for preserved taxa were conducted in free sediment metagenomes to uncover preserved diversity that escaped capture. This strategy recovered a wide diversity of freshwater microorganisms previously overlooked by more classical paleolimnological approaches. Given that microorganisms have critical functions in lake ecosystems, including carbon and other elemental cycling, studying the dynamics of lake microbiomes in the past could be quite insightful.
Betaproteobacteria were abundant in surface waters and well preserved in top and bottom sediment metagenomes. The ubiquity of Burkholderiales in water column and sediment microbiomes illustrates the importance of applying a stringent sequence identity cutoff to capture the elements of sediment metagenomes closely related to freshwater taxa as an approach to identifying historical DNA. Burkholderiales exhibit wide physiological and ecological diversity and have been isolated from a range of different sources, including soils and freshwater lakes (36,37). The sheer diversity of Burkholderiales makes it difficult to pin down specific traits which can explain their preservation in sediment metagenomes.
In general, taxa were not equally preserved in sediment metagenomes. Actinobacteria and Alphaproteobacteria were numerically important in surface water metagenomes, reflecting distribution patterns typical of freshwater lakes (38,39). Despite dominating surface water microbiomes, Ca. Nanopelagicales (Ca. Nanopelagicus and Ca. Planktophila), Pelagibacterales (Ca. Fonsibacter), and other ubiquitous bacterioplankton (Ca. Methylopumilus) were not well preserved in top or bottom sediments. Ca. Nanopelagicales, Pelagibacterales, and Ca. Methylopumilus are ultramicrobacteria characterized by free-living lifestyles and streamlined genomes with low guanine-cytosine (GC) contents (40)(41)(42), traits held in common with the well-preserved ultramicroscopic Burkholderiales. A variety of mechanisms has been put forth to explain the differential DNA preservation across microbial groups previously observed in single-gene time series, including sporulation (43), cellular architecture and membrane biochemistry (44), genome GC content (45), and autoecology. However, further work is needed to elucidate how DNA is differentially preserved across taxa and under different lake conditions.
The robust preservation of viruses in the captured metagenomes of all three lakes suggests that historical viromes are at least partially accessible to reconstruction from sediment archives. The best preserved viruses were Caudovirales, tailed bacteriophages infecting a wide range of hosts through both lytic and lysogenic replication (46). A phylogeny based on TerL appeared to resolve the Caudovirales as phages associated with several bacterial taxa; however, the tree lacked both the strong bootstrap confidence and clear delineation of clades necessary for definitive host identifications. The strong preservation of bacteriophage DNA is likely related to two mechanisms. First, prophages are widely integrated in the genomes of infected microorganisms (47), whose own DNA is variably preserved in sediments. Paleoecological time series can exploit the concurrent preservation of viral and host DNA to reconstruct virus-host evolution and illuminate the influence of environmental change on virus-host dynamics (48). The second mechanism of preservation is the hardiness of encapsidated virions, which are suited to withstand adverse conditions and remain intact under long-term cold storage (49). As abundant biological entities that exert enormous control over microbial populations in aquatic ecosystems (50,51), viruses could lend valuable insights into historical lake microbiomes.
A major issue in ancient DNA disciplines is that historical signals can be distorted or diluted by cross-contamination from contemporary DNA (14). The exposure of sediment DNA to modern contaminants can potentially occur at any stage of sediment core sampling or laboratory work. The first precaution we took to minimize crosscontamination was to retrieve the innermost sediments furthest from the core tube walls at top and bottom intervals using sterile spatulas and syringes. Another precaution was to perform DNA extractions from water filtrates and sediments in separate dedicated laboratories under positive-pressure UV hoods. Precautions aside, the historical microbial signals detected in sediment metagenomes appeared valid and not contaminated by contemporary DNA, as evidenced by the observed trends in DNA preservation across groups. If sediment metagenomes were contaminated by surface water DNA, we would expect the preferential recruitment of sediment metagenome reads to DNA from common and abundant lacustrine microorganisms (e.g., ultramicroscopic Actinobacteria and Alphaproteobacteria). Instead, preferential recruitment of sediment metagenome reads to rarer surface water taxa was observed.
Harnessing microbial ecotypes as paleoindicators. The detection of preserved DNA from a broad diversity of microorganisms expands the taxonomic scope of paleolimnology. To access the paleoindicator potential of preserved microbiota, we suggest that preimpact lake conditions can be reconstructed from records of past microbial ecologies. In the future, paleolimnological reconstructions focused on wellpreserved heterotrophic bacteria could provide key insights into past ecosystem functions. The ecologies of ubiquitous lacustrine bacterioplankton are extensively detailed in published literature (38,52). For some freshwater bacteria, ecotypic diversity has been described, i.e., genetically differentiated lineages with specific ecological requirements (53,54). We propose that the ecologies of well-studied bacterioplankton, particularly those with resolved ecotypes, can clarify the niche dimensions of historical microbiomes.
Our explorations of captured and free metagenomes detected key bacterioplankton groups for which ecotypic diversity is well described, e.g., Polynucleobacter and Limnohabitans. The microdiversity of ubiquitous Polynucleobacter and Limnohabitans has been investigated in a wide variety of freshwater lakes and has been physiologically characterized through culture, transplant experiments, and genome sequencing (55)(56)(57)(58). The niche specificities of Polynucleobacter or Limnohabitans ecotypes in DNA time series could potentially describe historical water column conditions (e.g., pH, carbon and other nutrient substrates, light penetration, temperature, and oxygen levels). Historical ecotypes might be identified by mapping sediment metagenomes to ecotype reference genomes, or a reduced approach might focus on recovering marker genes with sufficient phylogenetic information to resolve microdiversity. Toward the identification of ecotypes, we attempted to map sediment metagenomes to publicly available Polynucleobacter and Limnohabitans reference genomes. While reference genomes were captured in the surface water metagenomes of Eightmile Lake and Grand lac Touradi, they were not captured in the sediment metagenomes. The low-similarity alignment of sediment metagenomes and reference genomes representing genera detected in sediment metagenomes suggests that these contemporary reference genomes do not reflect historical microdiversity. Ecotypes could potentially be resolved if sets of genomes or metagenome-assembled genomes representative of local to regional pre-and postimpact diversity were made available, especially through largescale metagenomic lake surveys (59). Comparing the ecological dependencies reconstructed in ecotype time series could then retrace environmental change.
Preserved metagenome capture. Metagenome capture was designed to frame a novel methodology and proof of concept for the detection of historical DNA with shotgun metagenomics. However, with the approach we adopted in this pilot study, we are limited in our abilities to represent the totality of preserved microbial diversity. First, our surface water metagenomes were snapshot samples of DNA collected from the photic zone during summer stratification. These reference metagenomes comprised microbial assemblages partially representative of the vertical structure of stratified lakes, which varies seasonally. Second, surface water metagenomes represented contemporary diversity. Some degree of community turnover is expected across the more than 150 years separating preindustrial and contemporary microbiomes, and preindustrial diversity no longer occurring in contemporary microbiomes cannot be recruited. Finally, metagenome assembly only partly incorporated unassembled metagenomes, with the effect that reference assemblies further constricted the scope of sampled diversity. These limitations parametrizing the capture of snapshot surface water metagenomes effectively set up the capture of preservation signals to fail. Yet even so, we recovered partial preserved metagenomes from sediments with novel taxonomic information.
Metagenome capture was most prolific between sediment metagenomes, which displayed 20-to 30-fold-higher sequence mapping than captured surface water metagenomes. The continuity between top and bottom sediment metagenomes was observed in each lake, despite a depositional age difference of over 150 years and commonly observed steep geochemical gradients along sediment horizons (60). In contrast, the high taxonomic variation between surface water and sediment metagenomes can be explained by the stark environmental dissimilarities between photic zone and sediment subsurface habitats (61) and by the low composition of sediment DNA that is of historical origin (35). Experimental capture showed that very small fractions of sediment metagenomes originated from historical freshwater microorganisms (Ͻ1%, under the limitations of recruitment to snapshot reference surface water metagenomes). The disparity between the proportions of indigenous and historical DNA in sediments underscores the need to develop strategies to optimize historical DNA recovery while cancelling out signals from background sediment populations. One obvious path forward would be to apply the capture approach on a combined assembly of reference water column metagenomes, collected over a larger set of lakes or across multiple seasons. Nonetheless, the historical DNA detected by metagenome capture was sufficient to partially reconstruct historical lake microbiomes. PCR gene amplification is likely the most practical approach for targeting the population dynamics of preserved groups identified in this study. Future works might also attempt to extract historical signals embedded in indigenous sediment microbiomes (62,63).
Conclusion and outlook. In conclusion, we put forth two ideas to meet some of the unrealized potential of molecular genetic analysis in paleolimnology. First, wholemetagenome shotgun sequencing can access a broad diversity of microorganisms preserved in lake sediment DNA archives. Second, microorganisms are underexploited paleoindicators whose ecotypic diversity can potentially retrace environmental flux. This study tested a proof of concept that DNA from historical lake microbiomes can be recovered from sediment metagenomes. Our strategy of baiting unassembled sediment metagenome reads with reference surface water metagenome assemblies was designed to disentangle historical microbial signals from the indigenous sediment background. Future works are encouraged to improve on this approach to optimize the recovery of historical signals, e.g., by querying population-level gene variation in microbial time series with PCR amplification. All told, our application of shotgun DNA sequencing has expanded the detectable microbial diversity available to paleolimno-logical reconstructions to include heterotrophic bacteria and viruses, which we posit are as important in historical contexts as they are in contemporary lake microbiomes.

MATERIALS AND METHODS
Lake water and sediment sampling. Surface water and sediment cores were sampled from three eastern Canadian lakes between 31 July and 25 August 2017 during the Canadian Lake Pulse Network field campaign (59) (Fig. 1; Table 1). Lakes were selected by meeting the criteria of a maximum depth of at least 1 m and accessibility within 1 km from a road (59). At each lake, surface water and sediment cores were collected at the site of maximum depth, located by depth sounding and using bathymetric maps as guides when available. All water sampling devices were acid washed and triple rinsed before sample processing. Water was collected from the euphotic zone over an integrated depth of up to 2 m below the surface with an integrated tube sampler. Carboys were stored in chilled coolers until same-day water filtration on shore. Up to 500 ml of water was prefiltered through 100-m synthetic nylon (Nitex) mesh and then vacuum filtered on 47-mm-diameter 0.22-m Durapore membranes (MilliporeSigma, Darmstadt, Germany) through a glass funnel apparatus at a maximum pressure of 8 in Hg. Sediment core target length was estimated to reach preindustrial depositional age based on regional sedimentation rates synthesized from the literature. Sediment cores were retrieved using a gravity corer fitted with a 68-mm-diameter plastic core tube. Sediment cores were extruded on shore and subsampled at top and bottom 1-cm intervals. Bottom sediment subsections collected at core depth below the water-sediment interface were relatively consistent: the 40-to 41-cm interval in Lac Paula and the 39-to 40-cm interval in Grand lac Touradi and Eightmile Lake. Samples for molecular genetic analysis were collected from innermost sediments, furthest from the walls of the core tube, using a sterile syringe or spoon. Water filters and sediment samples were stored at Ϫ80 and Ϫ20°C, respectively. To determine the depositional ages of bottom sediments, 210Pb and bismuth-214 (214Bi) radionuclide activities were measured using gamma spectrometry and a constant rate of supply model (64). When overlap was observed between 210Pb and 214Bi activities within 1 standard deviation, sediments were determined to have been deposited within the last 100 to 150 years (i.e., reaching preindustrial background).
DNA extraction. DNA was extracted from filters using the DNeasy PowerWater kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions with the addition of the following optional steps: after vortex bead beating and centrifugation (step 7 of manufacturer's detailed protocol), 1 l of RNase A was added to samples before a 30-min incubation at 37°C. DNA extractions from sediments were performed in a UV cabinet in a separate dedicated laboratory. DNA was extracted from 0.500 Ϯ 0.015 g of wet sediment using the NucleoSpin soil kit (Macherey-Nagel, Düren, Germany) according to the manufacturer's instructions with lysis buffer SL1 and enhancer SX options, without protocols for separating intra-and extracellular DNA. DNA was quantified using the Qubit dsDNA BR assay kit (Invitrogen, Carlsbad, CA, USA) in a Qubit 2.0 fluorometer.
Metagenome shotgun sequencing, assembly, and annotation. DNA samples were submitted to Genome Quebec for library barcoding and shotgun sequencing of 150-bp paired-end reads on an Illumina NovaSeq 6000 platform with flow cell type S2. The sequencing depth ranged from 25,717,497 to 73,554,802 reads per sample. Sequencing adapters, low-quality bases, and ultrashort reads were trimmed with Trimmomatic v. 0.38 (65). Metagenome assembly from paired reads was performed with MEGAHIT v. 1.0.6 using k-mer lengths 23, 43, 63, 83, 103, and 123 (66). Assemblies were submitted to the Integrated Microbial Genomes & Microbiomes (IMG) system for annotation in IMG Annotation Pipeline v. 4.16.6 (67). Taxonomy was predicted by phylodist (i.e., the taxonomic affiliation of the closest gene homolog) in the IMG pipeline.
Read recruitment. For each lake, read recruitment was conducted between surface water, top sediment, and bottom sediment metagenomes, always in the direction tracking the downward preservation of DNA (i.e., reads from the underlying metagenome were mapped to the overlying metagenome assembly; Fig. 2a). Global read recruitment was performed with BBMap v. 35.68 (68) using an initial, highly permissive nucleotide sequence identity threshold of ϳ70% to assess alignment trends between metagenomes. A subsequent, empirically derived sequence identity threshold of 90% was applied to align closely related sequences. This cutoff was selected after observing consistent shifts in recruitment trend across capture experiments toward 90% sequence identity, which appeared to delineate closely (Ն90%) and distantly (Ͻ90%) related populations (Fig. 2b). To reject false-positive results overestimating DNA preservation, scaffolds containing ribosomal or transfer RNA genes were removed from captured metagenomes because these conserved sequences are shared with relatively high similarity among distantly related organisms and can produce biased alignment. The number of mapped reads was normalized by the number of unassembled reads in the recruitment effort. Scaffold lengths and gene counts on scaffolds in captured metagenomes were evaluated to determine whether the taxonomic annotations of protein-coding sequences could be extended to taxonomic assignments of whole scaffolds (see Fig. S2 in the supplemental material).
To increase the taxonomic resolution for groups with well-described freshwater microdiversity, surface water and sediment metagenome reads were recruited to Polynucleobacter and Limnohabitans reference genomes available from IMG (listed in Table S4 in the supplemental material). Ribosomal and tRNA genes were masked in reference genomes with BEDTools maskfasta v. 2.26.0 (69) to avoid read recruitment from distantly related organisms to conserved sequences. Read recruitment from surface water and sediment metagenomes to concatenated reference genomes was performed in BBMap with the default ϳ70% sequence identity threshold in order to visualize recruitment success as a function of genetic similarity.

Statistical analyses.
Principal-component analysis (PCA) was performed on metagenome taxonomic composition data to evaluate variation among captured and free metagenomes from all study lakes. Relative coverage-weighted metagenome composition data of order-rank taxonomy were chordtransformed with the function decostand(method ϭ "norm") in the R package vegan (70). The PCA was computed with the vegan function rda(). Dendrograms of prevalent taxa were based on Ward's hierarchical clustering of the Bray-Curtis dissimilarities between metagenome taxonomic compositions across lakes. Data wrangling and visualization were programmed in R (71) using the tidyverse package suite (72).
Phylogenetic analyses. To resolve the host identities of captured bacteriophages, a Caudovirales phylogeny was constructed based on TerL, a common taxonomic marker for this group (73). Caudovirales TerL amino acid sequences detected in metagenomes were aligned with publicly available reference sequences in MAFFT online service using default settings (74). A maximum likelihood tree was inferred using the Jones-Taylor-Thornton substitution model (75) with 5 discrete gamma categories and 100 bootstrap replicates in MEGA-X (76).
Data availability. The annotated metagenome assemblies in this study are available from the Integrated Microbial Genomes & Microbiomes system (U.S. Department of Energy Joint Genome Institute, Berkeley, CA, USA) at https://img.jgi.doe.gov under GOLD Study Gs0136026. Links to the GOLD Analysis Projects are listed in Table S5. All scripts are available at https://github.com/rebeccagarner/paleocapture.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.

ACKNOWLEDGMENTS
We are grateful to the NSERC Canadian Lake Pulse Network's staff and field crews for their enormous efforts in collecting the lake water and sediment samples and for compiling the lake metadata. Thank you to Susanne Kraemer for performing the filter DNA extractions and for preparing the DNA samples for sequencing. Thank you to Adam Jeziorski for performing the sediment dating. This work was supported by the Natural Sciences and Engineering Research Council of Canada and the Canada Research Chairs Program.