Bwindi Impenetrable National Park is a 331 km² large forest (0°53′–1°08′ S; 29°35′–29°50′E) in southwest Uganda (Figure 1), which supports an exceptional diversity of at least 135 mammal species (10 primate species), 381 bird species, 34 reptile species, 29 amphibian species and 393 tree species (Plumptre et al., 2007). Fifty-six tetrapod species and 74 plant species of Bwindi are Albertine Rift endemics (Plumptre et al., 2007).

Bwindi is located in the Kigezi Highlands (Figure 1), which are part of the eastern shoulder of the Albertine Rift – the western arm of the East African Rift. The bedrock geology of Bwindi consists of Precambrian metamorphic rocks such as schist, quartzite, shale and granite and the soils are ferrallitic (Butynski, 1984). The elevational gradient of this relatively small reserve ranges from 1190 m in its northern most part to over 2600 m in the southeastern corner based on the digital elevation model of Figure 1. The land surface is strongly dissected, resulting in substantial relief differences and steep slopes.

At Ruhija (2350 m, Figure 1A) mean monthly temperatures are between 13.4 and 19.1°C (September 2001 to August 2002: monthly median 15.1°C; Nkurunungi et al., 2004) and mean annual precipitation is 1378 mm recorded over the period 1987–2006 (Kasangaki et al., 2012). Two rainy seasons occur between March to May and September to November associated with the biannual passage of the tropical rainbelt. Bwindi is covered with mid-altitude and montane rainforest and a small area in the southeast with bamboo (Arundinaria alpina) forest. The dominating tree species are Chrysophyllum gorungosanum, Entandrophragma excelsum, Neoboutonia macrocalyx, Newtonia buchanani, and Parinari excelsa (Butynski, 1984; Olupot and Plumptre, 2010).

In the southcentral part of Bwindi lies Mubwindi Swamp – a 1 km² flat, slightly inclined peatland fed by several streams from a 12 km² large, forested catchment (Figures 1, 2A). We cored Mubwindi Swamp due to its long sediment record and because it serves as habitat to various terrestrial and (semi) aquatic species, which enhances the possibility of DNA deposition. The swamp is mostly covered with sedge communities dominated by Cyperus latifolius and Cyperus denudatus, which are often associated with Thelypteris cf. confluens, Alchemilla johnstonii, Crassocephallum paludum, Lobellia mildbraedi, and other species.

Several large mammals use Mubwindi Swamp as a foraging or breeding site like Sitatunga (Tragelaphus spekei) and bushbuck (Tragelaphus scriptus) (Mugerwa et al., 2013; pers. obs.). Mountain gorillas visit the swamp’s edge to feed on the thistle Carduus kikuyorua (Ganas et al., 2009; Rothman et al., 2014; pers. obs.). The swamp is also home to breeding populations of several Albertine Rift endemics, including the globally endangered Grauer’s swamp warbler (Bradypterus graueri; Kahindo et al., 2017), Delany’s swamp mouse (Delanymys brooksi; Kasangaki et al., 2003), and several frog species such as Bururi long-fingered frog (Cardioglossa cyaneospila; Blackburn et al., 2016) and Ahl’s reed frog (Hyperolius castaneus; Drewes and Vindum, 1994). Bwindi forest apparently lies within the hybrid zone of forest and savanna elephants (Loxodonta cyclotis, L. africana; Mondol et al., 2015) and currently supports a population of less than 40 elephants (Kasangaki et al., 2012), which regularly visit the forests around Mubwindi and the swamp itself for foraging and drinking, particularly during the dry season (Babaasa, 2000, pers. obs.). In the catchment of Mubwindi Swamp occur other rare vertebrates such as African golden cat (Caracal aurata), yellow-backed duiker (Cephalophus silvicultor) and one of Africa’s rarest birds, the African green broadbill (Pseudocalyptomena graueri) (Olupot and Plumptre, 2010; Mugerwa et al., 2013; pers. obs.).

Between August 4 and 7, 2017 we collected seven sediment cores from Mubwindi Swamp (1°4′ S, 29°45′ E, ca. 2090 m a.s.l.) with a modified Livingstone-type corer equipped with a square rod and steel barrel, which was 5 cm in diameter (Wright et al., 1984). Prior to collecting cores, the piston and core barrel were washed with boiling water to reduce DNA contamination. In the swamp center, we collected two long cores from two sites (site 1 and 2, Figure 1) that both reached a basal resistant gravel layer (cores designated MUB17-1A: 620 cm, MUB17-2C: 656 cm). To this end, we successively recovered one-meter-long core sections from the same bore hole and extruded the cores in the field. Site 1 was located in a mixed sedge-fern community dominated by Cyperus denudatus and Thelypteris cf. confluens (Figure 2B) and site 2 in a tall sedge community covered by dense stands of Cyperus latifolius and within 5 m of a stream (Figure 2C). At these sites we collected additional surface cores within 1 meter from the main cores for assessing DNA degradation and modern metagenomic composition (site 1: core 1E: 87 cm, site 2: cores 2A: 71 cm, 2D: 93 cm). For site 2, we combined core sections 2D-1 and sections 2C-2 through 2C-6 into a composite core (here referred to as “master core”) for studying long-term patterns (Figure 3), whereas for site 1 we only present data from surface samples. Furthermore, we collected two short cores (4A, 5A) at the eastern edge of the swamp where elephants had just been active prior to our visit. Core MUB17-4A (83 cm) was recovered from a site covered with elephant dung (here referred to as “elephant dung site”; Figure 2D) and core MUB17-5A (79 cm) retrieved from a flooded elephant wallow (here referred to as “elephant wallow site”; Figure 2E).

All core sections were wrapped in plastic wrap and stored in ABS (acrylonitrile butadiene styrene) tubes for transport and permanent storage. Cores were shipped to the LacCore facility (University of Minnesota, MN, United States) for non-invasive analyses, sub-sampling and permanent cold storage. At LacCore, cores were first scanned non-invasively for wet bulk-density using gamma-ray attenuation with a Geotek multi-sensor core logger at 0.5 cm intervals (two runs per core section). Limited high-energy irradiation using Gamma or X-rays is unlikely to significantly impact sedimentary DNA preservation (M. Muschick, pers. com.; Wanek et al., 2013; Wanek and Rühli, 2016). Each core was split longitudinally with a bleached diamond-bladed band saw into a working split section used for destructive sub-sampling and an archive split section used for non-invasive analyses and permanent storage. Each fresh archive section was photographed with a GeoScan IV digital linescan camera at 300 dpi.

Under sterile conditions, we collected samples for sedimentary ancient DNA analyses from the freshly split working sections of cores 1A, 1E, 2A, 2C, 2D, 4A, 5A (Figure 3 and Supporting Data). Two-cm-thick samples were taken from the core’s center with bleached and acetone cleaned tools after removing one centimeter of sediment from the core surface to avoid any contamination (Figure 3). All samples were immediately frozen and shipped to the Smithsonian Conservation Biology Institute (Washington, DC) for metagenomic analyses. From the archive core sections we obtained point magnetic susceptibility profiles (0.5 cm resolution) with a Geotek XYZ core logger and X-ray fluorescence (XRF) elemental profiles (0.5 cm sampling resolution; 15 s dwell time) with an ITRAX X-ray core scanner at Large Lakes Observatory (Duluth, MN, United States).

The lithostratigraphy of the cores was determined by visual inspection of the core surfaces and with a dissecting scope. In addition, mineralogy was analyzed microscopically from smear slides and with a Hitachi TM 1000 tabletop scanning electron microscope. For measuring water and organic matter content we collected two-cm-thick samples contiguously from the master core and two-cm-thick samples parallel to every DNA sample location from the other analyzed cores. Water content was determined by drying sediment at 100°C for 12 h in an oven and organic matter (OM) content was determined by loss-on-ignition of the dried samples at 550°C for 4 h in a muffle furnace (Dean, 1974). In addition, 1-cm-thick samples were taken for elemental analyses (C, N, S, P) immediately adjacent to every sedDNA sample location. These samples were freeze-dried and ground to a powder. Total elemental contents of carbon (C), nitrogen (N), and sulfur (S) were determined by high-temperature combustion elemental analysis (infrared detection, Elementar Vario EL, IGB Berlin). Total phosphorus content was determined by molybdenum blue spectrometry (Murphy and Riley, 1962) after thermal combustion (550°C) and hot potassium peroxodisulfate hydrolysis (Andersen, 1976; Ebina et al., 1983).

From cores 2C and 4A we collected plant macrofossils of non-aquatic plants and wood for AMS radiocarbon dating (Table 1). Site 5 (wallow) was not dated because of obvious bioturbation. Radiocarbon activity for 11 samples was determined by accelerator mass spectrometry at Lawrence Livermore National Laboratory (Table 1). We developed a Bayesian age-depth model for the master core with the BACON software package (Blaauw and Christen, 2011). All ¹⁴C dates were also calibrated into calendar years with the IntCal13 calibration curve (Reimer et al., 2013) in CALIB 7.1 (Stuiver and Reimer, 1993). Ages are presented as calendar years before present (i.e., present is 1950; “cal BP”) and on the common era (“CE”) notation.

To assess the influence of the local environment on DNA preservation, we conducted field surveys on water chemistry and physical and chemical soil properties of Mubwindi Swamp in June and September 2018 and July 2019. At all our four coring locations we measured soil pH and temperature changes with depth. To this end we successively retrieved 50-cm-long sediment core sections with a D-section core sampler and immediately upon core retrieval measured soil temperature in intervals of between 1 and 10 cm with an Extech digital thermometer, followed by pH with a Lutron soil pH meter at the same intervals. Close to our four coring locations, we also collected surface water samples from open pools or dug pits for the analysis of major dissolved elements, anions, and nutrients. Water samples were filtered (0.45 μm) and preserved by acidification (2M HCl) for subsequent analysis in the laboratory at IGB Berlin. Water temperature, pH and EC were measured at each sampling location with a handheld multiparameter probe (WTW Multi 3530). Dissolved aluminum (Al), calcium (Ca), iron (Fe), potassium (K), magnesium (Mg), manganese (Mn), and sodium (Na) were determined by inductively coupled-optical emission spectroscopy (ICP-OES, Thermo Scientific iCAP 6300). Dissolved chloride (Cl^–) and sulfate (SO₄^2–) were determined in samples without acidification by ion chromatography (conductivity detection after chemical suppression, Metrohm CompactIC). Dissolved ammonia (NH₄⁺) and nitrate (NO₃^–) were determined by flow-segmented analysis (FSA, SEAL AutoAnalyzer 3). Dissolved organic carbon (DOC) was determined by thermocatalytic conversion infrared spectroscopy (Shimadzu TOC-L). Soluble reactive phosphorus (SRP) was determined by molybdenum blue spectroscopy (Murphy and Riley, 1962).

We extracted DNA from 44 sediment samples (∼0.25 g per extraction), 30 from our master core and the remainder from the short cores (Figure 3 and Supporting Data), using DNeasy PowerSoil kits (Qiagen Inc., Germantown, MD, United States). To monitor DNA contamination during the coring and DNA extraction procedures, we also extracted DNA from two ∼0.25 g samples of Lux bar soap (Unilever) used to grease the corer’s piston and processed two additional sham extractions that contained only extraction reagents. DNA concentrations were measured using a Qubit® 2.0 fluorometer with the dsDNA HS assay kit (Thermo Fisher Scientific, Waltham, MA, United States).

We built double-indexed, double-stranded DNA libraries using KAPA Library Preparation kits – Illumina (Roche Sequencing and Life Science, Kapa Biosystems, Wilmington, MA, United States) with iNext adapters (Glenn et al., 2019). To render the iNext stub compatible with the KAPA kit’s A-tailing step, the stub sequence was modified by the addition of a thymine residue to the 3′-terminus (revised sequence: 5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGT-3′) and the stub complement sequence was extended by a guanine at the 5′-terminus (revised sequence: 5′-[phos]GACAGAGAATATGTGTAGAGGCTCGGGTGCTCTG-3′). In addition to the previously described DNA samples and controls, we added four control reactions (“Library Controls”) containing only water to monitor reagent contamination during library preparation. Libraries were amplified by 18 cycles of indexing PCR using KAPA HiFi Uracil + polymerase (Roche Sequencing and Life Science, Kapa Biosystems, Wilmington, MA,

United States) according to the manufacturer’s instructions. All purification steps during library preparation and amplification were performed using carboxyl paramagnetic beads (Rohland and Reich, 2012). Amplified libraries were visualized on 2% agarose gels stained with GelRed (Biotium Inc., Fremont, CA, United States) and quantified using a Qubit® 2.0 fluorometer with the dsDNA HS assay kit and quantitative PCR with the KAPA Library Quantification Kit (Roche Sequencing and Life Science, Kapa Biosystems, Wilmington, MA, United States). Libraries were pooled and submitted to Admera Health (South Plainfield, NJ, United States) where residual adapter-multimers were removed using AMPure XP beads (Beckman Coulter Life Sciences, Indianapolis, IN, United States) and final library pool quality was confirmed via visualization on a TapeStation (Agilent Technologies, Santa Clara, CA, United States). The quality-controlled pool was then 2 × 151 bp paired-end sequenced on a single lane of a Nextseq 500 (Illumina, Inc., San Diego, CA, United States).

Library sequence qualities were inspected using FastQC 0.11.5 (Andrews, 2016). We trimmed residual adapter contaminants and low-quality bases from the sequences using Trimmomatic 0.39 (parameters LEADING:3, TRAILING:3, SLIDINGWINDOW:4:15, MINLEN:30, HEADCROP:1, CROP:149, ILLUMINACLIP:NexteraPE-A-tail.fa:2:30:10 where Nextera-PE-A-tail.fa includes the A-tailing Nextera adapter sequences; Bolger et al., 2014). We merged paired reads using FLASH 1.2.11 (parameter -M 149; Magoč and Salzberg, 2011) and removed PCR duplicates from merged reads using CD-HIT-EST 4.6 (parameter -c 1; Li and Godzik, 2006).

We estimated putative ancient DNA fragment length distributions from the deduplicated, merged reads using a previously developed custom Ruby script (Campana et al., 2014). While this estimate excludes a minority of library inserts longer than 288 bp, typical mean fragment lengths of ancient DNA are 150 bp or shorter (e.g., Green et al., 2009; Prüfer et al., 2010; Dabney et al., 2013a). We confirmed the accuracy of our fragment length estimates by inspection of library insert length distributions, both bioinformatically (Supporting Data) and by visualization on agarose gels.

We performed an ad hoc test to determine whether the Mubwindi Swamp sedimentary DNA demonstrated cytosine deamination patterns consistent with authentic ancient DNA preservation (Briggs et al., 2007). The sediment taxonomic profiles were dominated by Rhizobiales (see below). Many of these taxa, such as Bradyrhizobium japonicum, are plant symbionts providing functional benefits such as nitrogen fixation and nutrient provision (e.g., Kaneko et al., 2011; Erlacher et al., 2015). We observed intact plant tissues (roots, leaves and stems) primarily in the top layers of the sediment column. Therefore, we expect increased proportions of cytosine deamination in lower layers if “ancient” damaged Rhizobiales molecules are being preserved in situ (Briggs et al., 2007; Sawyer et al., 2012). Absence of an increasing cytosine deamination signal is consistent with these molecules originating from modern DNA, either from free-living taxa closely related to root-associated species or via leaching through the sediment column. To estimate the rates of terminal cytosine deamination, we aligned the deduplicated, merged reads against the Bradyrhizobium japonicum USDA6^T reference genome (GenBank accession NC_017249.1; Kaneko et al., 2011) using BWA-MEM 0.7.17-r1188 (Li, 2013). The alignment was converted to bam format and sorted using SAMtools 1.9 (Li et al., 2009). Deamination patterns were analyzed using mapDamage2 2.0.8-dirty (Jónsson et al., 2013).

We aligned the deduplicated, merged reads against the National Center for Biotechnology Information (NCBI) non-redundant nucleotide (hereafter “nt,” version dated 9 September 2019) and the Refseq Genomic (hereafter “Refseq,” version dated 3 February 2020) databases using megaBLAST 2.6.0+ (Camacho et al., 2009) under default settings. MegaBLAST results were analyzed using the naïve lowest common ancestor (LCA) algorithm in MEGAN Community Edition 6.17.0 (nt analysis) or 6.18.5 (Refseq analysis) under default settings (Huson et al., 2016) except that “MinSupportPercent” was set to 0.005. We compared the two databases’ read assignments using paired t tests in GraphPad QuickCalcs (GraphPad Software Inc.).¹ Relative metagenome compositions were compared using normalized counts to control for variation in sequencing depth, maintaining at least one read per identified taxon, and discarding unidentified sequences. Community compositions at the ordinal rank were compared by Principal Coordinates Analysis (PCoA) and neighbor-net analysis based on Bray-Curtis distance in MEGAN. We compared animal (Metazoa) and land plant (Embryophyta) taxa (ordinal level or higher) presence/absence in the sedimentary DNA record against regional species (Butynski, 1984; Kasangaki et al., 2003, 2008; Stanford and Nkurunungi, 2003; Olupot and Plumptre, 2010; Mugerwa et al., 2013; Decru et al., 2019) and pollen records (Marchant et al., 1997; Marchant and Taylor, 1998). Pollen data were obtained from the African pollen database hosted at ftp://ftp.ncdc.noaa.gov/pub/data/paleo/pollen/tiliafiles/apd/ (accessed 3 November 2019). We excluded samples with fewer than 50,000 unique reads from the animal and plant record comparisons as these produced too few eukaryotic sequences for accurate identification.

To better characterize the sediment microbial communities, we also analyzed the deduplicated, merged reads using MetaPhlan2 2.9.21 (Truong et al., 2015) and QIIME 2 2019.7 (Bolyen et al., 2019) following Ferrari et al. (2018). We performed MetaPhlAn2 analyses under default settings and generated heat maps clustering sediment samples and taxa using Euclidean distances. In the QIIME 2 analyses, we closed-reference clustered the sequences against a previously trained (using QIIME 2 2018.4) SILVA 16S database (build 132; Pruesse et al., 2007) at 99% identity using VSEARCH (Rognes et al., 2016). Clustered sequences were aligned with MAFFT (Katoh and Standley, 2013), and a phylogenetic tree was built using FastTree (Price et al., 2010). Both phylogenetic and non-phylogenetic diversity metrics were calculated with rarefaction to 50 and 500 sequences. Communities were compared using PCoA using Bray-Curtis and Jaccard distances in EMPeror (Vázquez-Baeza et al., 2013, 2017).

We examined the relation between DNA yield and environmental factors with exponential regression analysis and the linear dependence between multiple DNA quantity and degradation parameters by computing Pearson correlation coefficients in R 3.5.1 (R Core Team, 2018) using the PerformanceAnalytics 1.5.3 package (Peterson et al., 2019). Since we did not have an a priori model to explain DNA degradation in the Mubwindi Swamp sediments, we also assessed non-parametric relationships between environmental parameters and DNA yield, terminal deamination, and sequence length using Spearman’s ρ in R.

The master core (MUB17-2C/D) contains a stratigraphic sequence which grades from clastic to organic sediment from bottom to top and is divided into six stratigraphic units based on patterns of sediment composition, texture and geochemistry (Figure 4). The core bottom from 656 to 653 cm is an unsorted gravel layer (unit I), high in density (1.9 g cm^–3) and magnetic susceptibility, but low in water (<20%) and organic matter content of the dry mass fraction (<10%). High XRF counts were obtained for Ca, S, Al, and Fe (Figure 5). Unit II is a coarse sand layer from 653 to 642 cm containing wood and charcoal, but overall less than 10% organic matter. Its C (2.6%) and N (0.1%) content were the lowest of the entire core record, while XRF peaks occurred for Si, Al, Ti and Zn. The sediment type changes abruptly at 642 cm to a homogenous silty clay that continues until 563 cm (unit III). This sediment had, on average, a water content of ca. 50%, a dry mass organic matter content of ca. 20% and a C content of <10%, whereas average density is 1.3 g cm^–3. The major elements recorded by XRF are K, Ti, Fe and Al. Unit IV from 563 to 434 cm is a clay deposit with numerous horizontal layers of charcoal, buried leaf fragments and some smaller wood pieces. Its sediment had a density of ∼1 g cm^–3, an average sediment water content of 78%, a dry mass organic content of about 37%, a C content < 20% and a N content < 1%. The XRF elemental composition was similar to that of the previous unit, but showed higher counts for S, Co, and Ni. The sediment also contains pyrite and other Fe minerals. Unit V, which extends from 434 to 312 cm, is an organic-rich clay with nearly 60% average dry mass organic matter content and mean sediment water content of 86%. This sediment composition was generally reflected in slightly negative values for magnetic susceptibility, except for a high magnetic susceptibility interval between 372 and 340 cm corresponding to major peaks in XRF counts for Fe and Co and the occurrence of Fe oxides. In addition, relatively high XRF counts of S and a maximum S content of nearly 4% characterized this unit, consistent with the frequent occurrence of pyrite. The final unit VI from 312 to 0 cm is water-saturated, sedge peat (water content > 90%) with an average organic matter content of the dry mass of 84%, a wet bulk density of 0.9 g cm^–3 and consistently negative magnetic susceptibility values. This stratigraphic unit has the highest C (average 42%) and N (average 2.1%) content of all units. The P content rises toward the top of the unit and is between 1.1 and 1.7 mg g^–1 dry mass in the upper 25 cm. The peat consists mostly of sedge rootlets and contains few small wood pieces (Figure 3). Nearly all XRF elements show significantly lower counts in the XRF profiles (except Ni) and subdued variability in this unit. The stratigraphic units III–V (clay sediments) are interpreted to represent a shallow lake environment, which filled with sediment to form a peatland (unit VI). In contrast to the master core, the short cores from the swamp’s edge (4A, 5A, Figure 3) consist entirely of silty clay and the dry mass contained only between 15 and 50% organic matter. The elephant dung site (4A) had a water content of ca. 50–60% and about 20% dry mass organic matter. The elephant wallow core (5A) exhibited signs of bioturbation and consists of about 70–80% water and the dry mass had about 50% organic matter in the top 50 cm and about 30% below that depth. The most abundant element counts of the XRF analysis for these cores were Fe, Ti, and K.

The radiocarbon chronology of site 2 (master core) showed that this central part of Mubwindi Swamp was an accumulating basin since ∼2200 cal BP (Table 1). However, the difference of nearly 1000 years between the two deepest radiocarbon ages of the master core, less than 50 cm apart, and the abrupt change in sediment from coarse sand to clay at 642 cm (transition from unit II to III) suggests a possible erosional surface and an associated depositional hiatus. Marchant et al. (1997) also found episodic sedimentation and several hiatuses in their cores. We therefore implemented a hiatus at 642 cm in the BACON age model calculation (Supporting Data). The resulting age-depth model shows relatively constant and very fast deposition since ∼1320 cal BP (642–0 cm). The associated average long-term sedimentation rate is 4.9 mm/yr, which means that one centimeter of sediment was formed in only about two years. The master core therefore provides a temporal high-resolution record: excluding the oldest DNA sample, the average time between adjacent DNA samples is ∼50 years, decreasing to ∼20 years in the upper one meter (Figure 3). The elephant dung site at the edge of the swamp has a median age of ∼180 cal BP (∼1770 CE) at a depth of 30 cm, indicating a slower sedimentation rate of ca. 1 mm/yr at the swamp’s edge (Table 1).

Mubwindi Swamp is an acidic swamp in a relatively cool tropical climate. Daytime soil temperatures in the sediment profile of site 2 (master core) ranged from maximum 20.6°C (0 cm) to 14.6°C (150 cm) (Figure 4). The average soil temperature of the entire 630 cm profile was 15.5°C – very similar to mean monthly air temperatures. Within the upper 10 cm of the sediment column the soil temperature declined rapidly by over 5°C, then continued to drop slightly until 140 cm, but below this depth increased slightly to 15.4°C (Figure 4). At the elephant dung site (core 4A) soil temperatures showed subdued changes, varying between 17.5 and 16.5°C and lacking a maximum at the surface, which instead was recorded at 20–30 cm. In contrast, the elephant wallow site (core 5A) had a temperature maximum at the surface (18.4°C) whereas below 10 cm temperature remained relatively constant at ca. 17°C.

Soil acidity along a 630 cm deep sediment column of the master core site ranged from pH 5.1 to 6.4 (Figure 4). The peat in the upper 20 cm is acidic (average pH 5.4) and showed the greatest variation in pH values. Between 25 and 210 cm, constantly moderately acidic conditions prevailed (pH of ∼5.6) and below that depth pH continued to rise; the sediment was only slightly acidic below 330 cm (≥ 6.1). The surface of the elephant dung site (core 4A) had a pH of 4.4 and became only weakly acidic below 10 cm depth (≥6.1). The elephant wallow site (core 5A) exhibited very similar acidity of between pH 6.3 and 6.7 (slightly acidic to neutral) (Figure 4).

The low concentration of Ca and the moderately acidic pH values of the surface water classify Mubwindi Swamp as a transitional poor to intermediate fen (i.e., groundwater-fed peatland; Table 2). The water was generally depleted in nutrients, however the master core site (2) was slightly enriched in dissolved reactive phosphorus (SRP), ammonium (NH₄⁺), and potassium (K). Both swamp center sites (1 and 2) were very rich in dissolved organic carbon (DOC). At the swamp edge (sites 4, 5), the surface water had elevated sulfate concentrations.

All sediment samples contained Qubit-quantifiable DNA (range: 0.9-86.1 ng DNA/g wet sediment; Mean ± SD: 22.2 ± 21.1 ng/g). Qubit-quantifiable DNA was not detected in negative controls verifying that this DNA derives from the sediments and not contaminants. Between 193 and 24,239,573 read pairs (Mean ± SD: 2,840,581 ± 5,273,296 read pairs) were sequenced for each sediment library, yielding between 126 and 15,458,842 unique merged reads per library (Mean ± SD: 1,859,731 ± 3,425,701 unique reads). Between 1 and 46,643 read pairs (Mean ± SD: 14,440 ± 15,484 read pairs) were sequenced for each negative control, yielding between 0 and 76 unique merged reads per library (Mean ± SD: 19 ± 25.8 unique reads). These low numbers of unique reads in the blanks indicated that our results were not strongly biased by reagent contamination.

We observed a clear trend of decreasing quantities of DNA with greater depths (older age) of the cores (Figures 4, 6, 7), consistent with the preservation of ancient DNA in situ. The surface samples (n = 7) of all our cores contained between 28.6 and 86.1 ng DNA/g sediment (Mean ± SD: 53 ± 17 ng/g, Figure 6). In the master core the amount of sedimentary DNA exponentially declined with depth (r² = 0.755, p < 0.001) from a maximum amount of 54.7 ng/g at the surface (0–2 cm) to 0.9 ng/g at 645 cm in the basal sand (Figure 7). The largest decline in DNA mass occurred in the upper 100 cm with reductions of more than 10 ng/g between adjacent samples from 2 to 3 cm and 20 to 30 cm. Below 100 cm, the amount of DNA was consistently less than 10 ng/g sediment and, below 250 cm (sediment older than 500 years), it was generally less than 5 ng/g (≤10% of surface sample). Below this depth the mass of DNA remained relatively constant at an average fraction of 3 ng/g despite substantial changes in sediment type, water- and organic matter content and geochemistry in this part of the core (Figure 4).

The trend of rapidly declining DNA fraction in the upper meter was persistent across Mubwindi Swamp despite the obvious differences in sediment type and composition between swamp center and edge, i.e., peat vs. silty clay (Figure 6). In core 4A (elephant dung site) DNA declined exponentially (r² = 0.683, p = 0.5322) from 52.9 at 0–2 cm to 7.0 ng/g at 70 cm depth (Figures 4, 6). In contrast, the surface sample of core 5A (wallow site) contained only half as much DNA as the samples from 5 to 7 and 10 to 12 cm which have 55 ng DNA/g sediment. Only below this depth did the DNA concentration decline, but the amount was still larger than in samples of similar depths in the other cores (Figure 6). Additional core samples are required to determine whether this variation is a result of local environmental conditions or simply variability in DNA taphonomy, which has been shown to differ even within a single bone (Green et al., 2010).

Fragment lengths and cytosine deamination patterns also supported the persistence of authentic ancient DNA in the sediment. Mean fragment lengths ranged from 127 to 175 bp (Mean ± SD: 146 ± 13 bp) and molecular fragment lengths were highest in the upper 30 cm of the cores (Figure 6). In the master core, fragment lengths declined with depth (exponential decline r² = 0.543, p < 0.001), but not as strongly as the DNA fraction. Conversely, both 5′ C→T and 3′ G→A increased exponentially with sediment depth and age in this core (5′ C→T: r² = 0.519, p < 0.001; 3′ G→A: r² = 0.601, p < 0.001), strongly suggesting that a portion of the sediment sequences derived from preserved aDNA. The quantity of DNA and fragment length both decreased down core and were significantly correlated (r² = 0.701, p < 0.001), whereas both parameters were significantly negatively correlated with both 5′ C→T and 3′ G→A as shown in Figure 8.

The vast majority (nt database: 87.1–97.3%; Refseq database: 80.8–95.5%) of sediment sequences were not identifiable via megaBLAST and MEGAN analysis (nt: Mean ± SD: 94.1 ± 2.6%; Refseq: Mean ± SD: 90.9 ± 3.7%; Figure 7; Supporting Data). MEGAN-identified sequences were dominated by Bacteria (nt: range 79.0–96.9%, Mean ± SD: 88.2 ± 4.4%; Refseq: range 79.5–99.0%, Mean ± SD: 90.7 ± 4.4%), particularly Rhizobiales (nt: range 7.1–31.3%, Mean ± SD: 16.5 ± 4.8%: Refseq range 9.0–33.3%, Mean ± SD: 15.5 ± 4.8%) and Burkholderiales (nt: range 0.0–15.0%, Mean ± SD: 7.2 ± 3.0%; Refseq: range 1.3–14.1%, Mean ± SD: 6.2 ± 2.8%). The microbial profiling conducted in this project was undertaken to further our understanding of DNA preservation in this environment. So, while a full description of microbial profiles is beyond the scope of this paper, detailed microbial profiles and MEGAN results are available in the Supporting Data. The bacterial profiles were complex, representing 125 (nt database) to 156 (Refseq database) taxonomically described orders across all samples (Supporting Data: Taxonomic Profiles). Archaea, representing 15 (Refseq database) to 16 orders (nt database), comprised 0.0–7.5% of the nt assigned reads (0.0–6.9% of Refseq assigned reads) and increased with frequency with depth (nt: Mean ± SD: 4.2 ± 2.1%, r² = 0.69; Refseq: Mean ± SD: 3.4 ± 1.8%, r² = 0.28). The most frequent Archaea were Methanobacteriales (nt: range 0.0–4.3%, Mean ± SD: 1.1 ± 1.0%; Refseq: range 0.0–4.8%, Mean ± SD: 0.5 ± 1.0%), Methanomicrobiales (nt: range 0.0–7.4%, Mean ± SD: 1.0 ± 1.2%; Refseq: range 0.0–6.3%, Mean ± SD: 0.9 ± 1.0%), and Methanosarcinales (nt: range 0.0–5.7%, Mean ± SD: 0.9 ± 1.1%; Refseq: range 0.0–3.4%, Mean ± SD: 0.7 ± 0.8%). Eukaryotes were under-represented, comprising only 0.0–13.8% of the identified reads in total (nt: range 0.0–11.5%, Mean ± SD: 4.5 ± 2.3%; Refseq: range 0.0–13.8%, Mean ± SD: 4.2 ± 2.8%) with 0.0–5.8% (nt: range 0.0–3.1%, Mean ± SD: 1.6 ± 0.8%; Refseq: range 0.0–5.8%, Mean ± SD: 2.5 ± 1.6%) deriving from Metazoa and 0.0–6.9% (nt: range 0.0–5.8%, Mean ± SD: 1.4 ± 1.1%; Refseq: range: 0.0–6.9%, Mean ± SD: 1.2 ± 1.3%) from Viridiplantae (Figure 7). Other than the relative decrease in Bacteria and corresponding increase in Archaea with depth, we observed little evidence of temporal structure at an ordinal scale (Supporting Data: Taxonomic Profiles). The relative increase in methanogenic Archaea between strata may not necessarily indicate changes in the past community structure, but rather current structure due to differing microbiota in deeper Mubwindi Swamp sediments (Vuillemin et al., 2017). PCoA and neighbor-net analysis separated some of the surface samples from the short cores (samples 57, 59, 70, and 100) from the primary core samples. To a lesser extent, the short core samples 54, 56, 58, 60, 62, and 63 were also distinct. While these results were consistent with local community structure variation due to microenvironmental differences, the distinct samples were also the shallowest sequenced (120–37,261 unique sequences per sample), indicating that this pattern was probably a sampling artifact (Supporting Data).

The Refseq-aligned dataset assigned taxa to more sequences than the nt-aligned dataset (nt: Mean ± SD: 90,911 ± 150,070 sequences; Refseq: Mean ± SD: 142,931 ± 238,131 sequences; two-tailed p = 0.0003). The Refseq-aligned dataset assigned significantly more sequences to Eukaryota (nt: Mean ± SD: 4,042 ± 6,928 sequences; Refseq: Mean ± SD: 5,919 ± 10,897 sequences; two-tailed p = 0.0036), Bacteria (nt: Mean ± SD: 80,718 ± 132,524 sequences; Refseq: Mean ± SD: 129,720 ± 215,085 sequences; two-tailed p = 0.0003), and Archaea (nt: Mean ± SD: 3,598 ± 6,906 sequences; Refseq: Mean ± SD: 4,578 ± 8,665 sequences; two-tailed p = 0.0013). While the Refseq-aligned dataset assigned more reads to both Metazoa (nt: Mean ± SD: 1,569 ± 2,843 sequences; Refseq: Mean ± SD: 3,810 ± 7,390 sequences; two-tailed p = 0.0022) and Viridiplantae (nt: Mean ± SD: 1,038 ± 1,669 sequences; Refseq: Mean ± SD: 1,187 ± 1,911 sequences; two-tailed p = 0.0041) than the nt-aligned dataset, the effect was larger for Metazoa (mean 2.43 × more assigned sequences) than Viridiplantae (mean 1.14 ×). In fact, the increase in Metazoa was greater than the increase in total Eukaryota (mean 1.46 ×), likely reflecting the disproportionate number of animal genomes in the Refseq database (e.g., Brandies et al., 2019).

The taxonomic compositions of our MEGAN results were not significantly impacted by reagent contamination. The low contamination level (151 total unique reads in the negative controls) were consistent with rare cross-contaminations between sediment samples and negative controls and sporadic reagent contamination. For instance, the Schistosoma, Mustela and Percomorphaceae reads (see below) were probably the result of rare cross-contaminations from the sediment samples (e.g., via index switching: Kircher et al., 2012) as none are found frequently in reagents or were processed previously in our ancient DNA laboratory. Since our analyses were limited to high taxonomic ranks unlikely to be significantly biased by these rare contaminations, we list identified contaminants, rather than bias our dataset by excluding these taxa. Using the nt database, the majority of MEGAN-identified sequences (n = 32 of 37 assigned sequences; 86%) in the negative controls derived from common laboratory contaminants including Hominoids (n = 14), of which 12 were identified as human (Homo sapiens), Canids (n = 1), and Bacteria (n = 17). Identified bacterial species and strains in the blanks included Acinetobacter johnsonii XBB1 (n = 1), Alcaligenes faecalis (n = 1), Aquaspirillium sp. LM1 (n = 1), Brevundimonas sp. (n = 1), Citrobacter freundii complex (n = 1), Dietza sp. oral taxon 368 (n = 1), Fusobacterium sp. (n = 1), Pseudomonas putida (n = 1), Salincola tamaricis (n = 1), Sphingomonas hengshuiensis (n = 1), Staphylococcus saprophyticus subsp. saprophyticus (n = 1), and Stenotrophomonas sp. (n = 1). One contaminant sequence was identified as Schistosoma japonicum, and another three were identified as Percomorphaceae [including Dicentrarchus labrax (n = 1), Mastacembelus armatus (n = 1), and Scophthalmus maximus (n = 1)]. Using the Refseq database, MEGAN assigned taxa to 62 contaminant sequences, of which 33 derived from Bacteria. These included Acinetobacter sp. (n = 1), Alcaligenes faecalis subsp. faecalis NBRC 13111 (n = 1), Bradyrhizobium sp. LSPM299 (n = 1), Brevundimonas bullata (n = 1), Dietzia sp. (n = 1), Egibacter rhizosphaerae (n = 1), Fusobacterium sp. (n = 1), Henriciella litoralis (n = 1), Methylobacterium pseudosasicola (n = 1), Microbacterium sp. CGR2 (n = 1), Pedosphaera pravula Ellin514 (n = 1), Pseudomonas sp. (n = 2), Sphingobium phenoxybenzoativprans (n = 1), Sphingomonas sp. (n = 1), Staphylococcus sp. (n = 2), Streptomyces griseus subsp. griseus (n = 1), Terracidiphilus gabretensis (n = 1), and Williamsia sp. (n = 1). Eukaryotic contaminants identified using the Refseq database included Percomorphaceae (n = 7, including 1 assigned to Lates calcarifer), Primates (n = 14, including 8 assigned to Homininae), Canidae (n = 4, including 2 assigned to Canis lupus), and Mustela putorius furo (n = 1).

For land plants (Embryophyta) our MEGAN analysis against the nt database resulted in 18 taxonomic assignments, with 11 at ordinal rank. All 18 taxa were recorded in the master core, in which the richest samples occurred below 340 cm depth and single samples contained between 1 and 7 orders (Figure 9). The most commonly detected orders were in descending frequency Poales (100% presence), Fabales, Solanales, Rosales, and Brassicales. In comparison, the Refseq database yielded 21 plant taxonomic assignments with 14 orders. All these taxa were recorded in the master core, in which ordinal richness ranged from 0 to 10 orders per sample and was also highest below 340 cm depth (Figure 9). Fabales, Solanales, Poales, and Rosales were the most frequently detected plant orders with Refseq. Refseq identified all the taxa that were identified with the nt database, but in addition also Asparagales (1×), Gentianales (1×), and Caryophyllales (3×) (Figure 9).

Using the nt database, MEGAN assigned sedDNA sequences to 26 different animal taxa (Metazoa) of which 9 are at the level of order (Figure 10A). The most frequently recorded orders are Diptera (in all samples), Primates, Hymenoptera, and Cetartiodactyla. All of the detected taxa occur at present in Africa and members of all recorded orders in Bwindi Impenetrable Forest. In the master core between 1 and 5 orders/sample were detected and the number of assigned orders increased slightly with depth (age) (Figure 10A). In contrast, the Refseq database assigned sequences to 41 animal taxa representing 15 orders. Diptera, Hymenoptera, and Rodentia were detected in all samples, followed in frequency by Primates, Carnivora and Cetartiodactyla. In the master core, the number of detected orders per sample ranged from 3 to 13, with the minimum recorded at 4 cm and the maxima between 300 and 500 cm depth (Figure 10B). Four assigned taxa (Protacanthopterygii, Metatheria (= marsupials), Cetacea, Octopoda) do not occur in (tropical) Africa and we consequently considered them misidentifications. Cetacea and Octopoda were only detected once but Metatheria was detected regularly in the master core (Figure 10B).

The taxonomic assignments for Metazoa from both reference databases had 22 taxa with the following seven orders in common: Cypriniformes, Rodentia, Primates, Cetartiodactyla, Hymenoptera, Lepidoptera, and Diptera. Excluding misidentifications, Refseq uniquely identified the orders Squamata, Passeriformes, Eulipotyphla, Carnivora, Chiroptera, Hemiptera, and Coleoptera whereas the nt database uniquely yielded Cyprindontiformes and Rhabditida. Together both databases detected a total of 45 assigned animal taxa of which 17 are at ordinal rank.

MetaPhlAn2 only identified a total of 11 bacterial (n = 7) and archaeal (n = 4) species/strains representing 5 bacterial and 1 archaeal orders. No taxa were identified for 30 of the sediment samples or for any of the negative controls (Supporting Data). We did not analyze these data further given that they were uninformative.

QIIME 2 identified between 0 and 1823 unique 16S sequences per sediment sample (Mean ± SD: 238 ± 412 sequences; Supporting Data). No 16S sequences were identified in the negative controls. Excluding samples with no retained reads, 27.3–100.0% were assigned to Bacteria (Mean ± SD: 74.6 ± 20.2%) and 0.0–72.7% were assigned to Archaea (Mean ± SD: 25.4 ± 20.2%). As in the MEGAN analyses, the percentage of reads assigned to Archaea increased with depth (r² = 0.72). Community compositions were exceptionally complex, with no taxa dominating the assignments (Supporting Data). In the dataset rarified to 50 sequences, PCoA separated the samples from the first meter of sediment from the remaining sediment samples (Supporting Data). PCoA did not reveal informative patterns in dataset rarified to 500 sequences due to the low sequencing depth eliminating most samples (n = 8 after rarefaction; Supporting Data).

The deposition of DNA in a sedimentary basin and its representation of the locally present biota are influenced by the abundance of the locally present species, their biomass, their genome length, transport processes and by chance (but see Andersen et al., 2012; Yoccoz et al., 2012; Giguet-Covex et al., 2019). Four possible sources provide DNA to the sediments of Mubwindi Swamp including (1) DNA from micro- and macro-organisms living in the sediment, (2) DNA from organisms living on or utilizing the surface of the swamp, (3) DNA derived via streams and runoff from the catchment of the swamp (see Giguet-Covex et al., 2019), and (4) DNA derived by deposition from the air (e.g., pollen grains: Parducci et al., 2005). Given these sources, DNA extracted from the sediments will be a mixture of modern and ancient DNA.

Mubwindi Swamp was apparently an open water environment (shallow lake) from about about 1320 to ∼650 cal BP and then served as habitat for (semi)aquatic biota (e.g., Cypriniformes; Figure 10) and very likely as a water source to terrestrial fauna (e.g., Cetartiodactyla; Figure 10). At this stage the sediments received local authochtonous DNA and allochotonous DNA influx from the catchment that could be transferred to the central part of the basin. In the master core the taxonomic diversity is highest in the clayey sediment below ∼330 cm depth (older than ∼720 cal BP) corresponding to the shallow lake phase. Possibly the Mubwindi basin received more diverse DNA via erosion and runoff from its catchment during this phase than the peatland that formed later. Slightly higher sedimentation rates prior to 800 cal BP together with higher XRF counts of Al and Ti (Figure 5), which are indicators of erosion, support this interpretation. At about 650 cal BP the peatland had formed (Figure 3) and terrestrial habitat conditions became established that changed DNA transport pathways as a consequence.

Swamps and peatlands provide habitat to a diverse group of organisms that live in the organic deposits. Microbes will contribute modern DNA over the entire sediment column and thus disturb ancient DNA signals. They are expected to be most abundant in the aerobic zone of the peat surface of Mubwindi Swamp (water table: 0–14 cm measured during field work), where also high soil temperatures and in the center of the swamp also higher nutrient concentrations likely favor microbial activity (Dickinson, 1983; Dabney et al., 2013b; Table 2). Generally, the aerobic zone of peat deposits in swamps is habitat for various protozoa (e.g., amoeba) and soil fauna (e.g., mites, collembola, nematodes; Mason and Standen, 1983; Speight and Blackith, 1983) and also the rooting zone for most plants (Crawford, 1983). These organisms will add their DNA to the sediments as indicated by the detection of nematode DNA (Figure 10A). Deeper anaerobic conditions are likely to be associated with reduced microbial life but provide habitat for Archaea (Figure 7). Plants with anatomical transport mechanisms for oxygen will be able to survive in this waterlogged environment (Crawford, 1983). Radiocarbon dating of roots has shown that living herbaceous plants can insert their roots into waterlogged sediments of subtropical swamps to a depth of 60 cm (Glaser et al., 2012). Therefore, living plants may also be contributing DNA directly into the upper sediment column. We expect that the roots of the sedges Cyperus latifolius and Cyperus denudatus, which dominate most areas of Mubwindi Swamp and likely contribute a large proportion to the peat, represent a major source of sedDNA as indicated by the frequent detection of Poales (Figure 9).

Given these conditions most biotic activity should be concentrated in approximately the upper 25 cm of the sediment column. Indeed, from this zone most DNA was extracted from the cores and the DNA shows the lowest proportion of deamination and fragmentation (Figure 7). The extracted DNA pool from the upper sediment column thus contains a large contribution of modern DNA from living biota. Hence, the decline in the DNA yield with depth is not simply an indication of DNA degradation, but also of decreasing biotic activity in the sediment column. This high proportion of in situ derived DNA will also overwhelm signals of DNA derived from passing animals and influx from the swamp’s catchment. Moreover, the biotically active zone will contribute a certain signal of time-averaging by all those species that can exist across a wider zone (>20 cm) of sediment. Time-averaging of sedaDNA may therefore be a specific problem in slowly accumulating systems. Permanent and deep lakes with anoxic bottom waters may be better sources for catchment studies of past biota. Yet modern microbial DNA will always be a significant part of any sedaDNA profile.

The collected sediment cores differ substantially in terms of their composition and the sediments show a large range in water content (15–94%), organic matter content (6–92%), nutrients (N, P, K), pH (4.4–6.7), and temperature (14.6–20.6°C; Figure 4). DNA is preserved in all sediment types and persists under various environmental conditions with few clear explanatory trends (Figure 11 and Supporting Information). DNA is thus degrading at similar rates in different sedimentary environments with varying nutrient, pH and temperature conditions. General trends in our data are (1) the exponential decline in the amount of DNA with sediment depth and age, (2) the increase in DNA fragmentation with depth and age, and (3) the increase in 5′ C→T and 3′ G→A substitutions with depth and age (Figure 7). Nevertheless, the preserved DNA’s read lengths, deamination rates, and quantities show significant monotonic relationships with almost all environmental factors within the master core, including those unlikely to have a direct causative relationship (e.g., sediment magnetic susceptibility: Figure 11 and Supporting Information). This finding suggests that there is not a single environmental driver or simple set of factors that influence the degradation of DNA in swamp sediments (Figure 11; Dabney et al., 2013b) and complicates the generation of a comprehensive statistical model for DNA preservation in the Mubwindi Swamp due to environmental parameter non-independence. The roles of local sedimentation patterns and sediment types in DNA preservation demand further study.

At the master core site, the DNA in the basal sand layer was apparently deposited under drier conditions with neutral pH and lower nutrient levels than those above (Figure 4), which likely limited hydrolysis immediately after deposition (Lindahl, 1993; Pääbo et al., 2004). Although such conditions have been associated with comparably “good” sedaDNA preservation in caves (e.g., Hofreiter et al., 2003; Slon et al., 2017), at Bwindi the amount of remaining DNA and its fragmentation in the basal sediments are very similar to that of the overlaying wetter acidic organic peat deposits (Figure 7). Although higher taxonomic diversity was detected in the deeper minerogenic sediments than in the peat, we assume that catchment-scale transport processes rather than differential lithologic control on preservation account for these findings. Organic matter content also appears to have limited effects on the quantity of preserved DNA and its fragmentation (Figure 4). In the master core, DNA yield and fragmentation reach background levels below about 200 cm, where organic matter is present in a wide range of values (6–88%). Thus, the deposition in clastic sediment (sandy, silty, or clay deposits) or in organic peat deposits seems not to determine how well DNA is preserved in the long-term. This conclusion is supported by similar initial amounts of DNA yield in the surface of all cores and the similar decline in DNA yield in the first meter of the short cores (4A, 5A) (Figure 6A), which entirely consist of clay in contrast to the master core’s upper peat layer (Figure 3).

In addition, the variation in the elemental composition of the master core is also not reflected in variations of DNA yield or preservation stage. Despite large changes in the quantity of cations such as K, Al, or Fe we find no indication for preferential preservation through DNA adsorption in zone of higher concentrations of metals (Figure 5). The DNA patterns of the short cores with their high Fe XRF counts and clay content corroborate this inference.

As predicted by current models of DNA preservation (e.g., Dabney et al., 2013b; Kistler et al., 2017), the vast majority of DNA content loss occurs rapidly after deposition in the Mubwindi Swamp (Figures 6, 7). Not only was there a sharp decline in DNA content below the top meter of bioactive sediment, we observed very few eukaryotic sequences, even in sites of recent elephant activity, suggesting that deposited DNA degrades rapidly in the acidic and warm (17–21°C) swamp environment (Giguet-Covex et al., 2019). Based on laboratory and field experiments, acidic, warm conditions are known to promote DNA hydrolysis and are non-conducive to long-term DNA preservation (Lindahl and Nyberg, 1972; Lindahl, 1993; Strickler et al., 2015; Kistler et al., 2017; Seymour et al., 2018). However, inference of the influence of pH and temperature on DNA degradation by direct comparison of DNA yield with these parameters is misleading. Most DNA is deposited at the sediment surface (where DNA yield, acidity, and temperature are highest), whereas DNA yield declines with depth as acidity and temperature decrease. This finding likely indicates the concentration of DNA from living biota in the sediment surface where an aerobic, warm, and more nutrient-rich environment facilitates abundant plant growth and microbial and soil faunal activity. DNA degradation under surface conditions only becomes apparent when deposition time is considered. In the swamp’s center, DNA is deposited at the sediment surface at > 20°C and then exposed to the maximum burial temperatures for over 50 years (= 25 cm depth) until temperature stabilizes at about 15.5°C (Figure 4). This time of exposure to high temperatures (and low pH) will likely drive rapid DNA degradation, while at the same time the in situ microbial contribution to the DNA pool should also decrease as the sediment changes from aerobic to anaerobic as inferred from the presence of Archaea and the measured minimum position of the water table at -14 cm. Furthermore, the higher concentrations of sedimentary nitrogen and phosphorus near the surface likely contribute to DNA degradation by stimulating microbial decay (Dabney et al., 2013b; Figure 4) and are thus not reflecting good preservation conditions as suggested in Figure 11.

Although the reduced DNA content in earlier (deeper) strata likely correlates with decreased biotic activity at these depths, statistically, age and depth are the best predictors for the remaining fraction of DNA in the sediment (age/depth: r² = 0.755, p < 0.001). Furthermore, the decreasing fragment lengths and increased cytosine deamination of DNA sequences deriving from deeper, earlier sediments suggest that at least a portion of these molecules are preserved ancient DNA. Moreover, the pattern of increasing cytosine deamination with depth is not consistent with the DNA from deeper strata deriving solely from water-driven leaching of DNA from extant, living and recently deceased organisms through the sediment column despite up to > 90% sediment water content. In the case of the leaching scenario, we would expect all sediment DNA samples to exhibit roughly the same level of deamination since they would all be nearly of the same age. Furthermore, in the leaching scenario, we would expect a continuous gradient in DNA content down the core. Instead, we find that the DNA content stabilizes below the first meter of sediments. We conclude that, despite the acidic condition, genuine ancient DNA has survived in Mubwindi Swamp for over 2200 years and sedaDNA is therefore retrievable from tropical swamps.

Our data suggest a tropical sediment model in which most DNA degrades beyond recovery rapidly after deposition in the swamp (<200 years) (Bremond et al., 2017). The majority of DNA molecules in the most recent sediment layers likely derive from ongoing biotic activity rather than preserved ancient biomolecules. The small portion of DNA molecules that survive this initial stage of elimination become stabilized in the less acidic and cooler deeper sediment levels, which are more conducive to long-term DNA survival. These deeper sediments still preserve a sedaDNA record of diverse taxonomic assemblages. In contrast to other studies of African sedaDNA (Boessenkool et al., 2014; Bremond et al., 2017) we, in fact, observe no decline in the number of taxa with increasing sediment age. Instead the master core contains higher taxonomic diversity in sediment older than ca. 600 cal BP. The absence of a declining trend in taxonomic richness with age suggests that sedaDNA can provide information on past tropical biodiversity for several thousand years.