High resolution ancient sedimentary DNA shows that alpine plant biodiversity is a result of human land use

Alpine areas are well known biodiversity hotspots, but their future may be threatened by expanding forest and changing human land use. Here, we reconstructed past vegetation, climate, and livestock over the past ~ 12,000 years from Lake Sulsseewli (European Alps), based on sedimentary ancient DNA, pollen, spores, chironomids, and microcharcoal. We assembled a highly-complete local DNA reference library (PhyloAlps, 3,923 plant species), and used this to obtain an exceptionally rich sedaDNA record of 366 plant taxa. The vegetation mainly responded to temperature during the rst half of the Holocene, while human activity drove changes from 6 ka onwards. Land-use shifted from episodic grazing (Neolithic, Bronze Age) to agropastoral intensication (Medieval Age). This prompted a coexistence of species typically found at different elevational belts, thereby increasing plant richness to levels that characterise present-day alpine diversity. Our results indicate that traditional agropastoral activities should be maintained to prevent reforestation and preserve alpine plant biodiversity.


Introduction
Changing environmental conditions are displacing organisms out of their ranges, causing severe threats to biodiversity 1 . In mountains, vegetation is strongly determined by temperature and forms elevational belts, in the European Alps ranging from the warm lowland Colline belt to the cold highland Nival belt 2 .
Climate-based projections indicate an expected upward displacement of vegetation that will reduce habitat for present-day alpine species and especially cold adapted taxa that inhabit the highest elevations 3 . This process is already being observed on European summits, with an increase in total plant species richness due to the arrival of lowland vegetation over the last century 4,5 . Plant remains in lake sediments allow us to explore vegetation responses to past climate changes and human activity, particularly at the decadal and longer time scales relevant for understanding future vegetation response to global warming. Therefore, detailed palaeoecological records representing the full range of plant types and functional groups that compose alpine and subalpine vegetation are needed to understand how longterm interactions of climate and humans affect overall biodiversity and survival of high-altitude plants. However, some ecologically relevant groups such as grasses and sedges are poorly represented in conventional palaeoecological records due to limited taxonomic resolution 6 . Recent advances in sedimentary ancient DNA (sedaDNA) have greatly improved our ability to get detailed insight into past diversity processes [7][8][9] .
The European Alps are an important plant biodiversity hotspot 10 , with ~4,000 native plant species 11 . This diversity results from a complex interplay of both "natural" and "human" factors over both geological and more recent timescales. Natural climatic and environmental conditions in alpine regions as well as palaeoclimatological and palaeogeographic changes resulted in suitable conditions for plant immigration, speciation, and endemism 12,13 . However, traditional human activities over millennia have modi ed and in many cases helped to maintain a part of this diversity 14 . Humans have modi ed the alpine landscapes since the Mesolithic, ca. 10 ka (1 ka = 1,000 yr ago), by clearing small areas of forest to attract prey for hunting 15,16 while the introduction of agropastoral activities during the Neolithic (ca. 7 ka) drove a downward shift of alpine treelines [15][16][17][18] . Human-environment interactions in forested and open vegetation types such as the Subalpine zone led to a mosaic of different habitats that include species-rich meadows 2,9,16 . As a result, future changes in land use might imply a reduction in biodiversity of subalpine and alpine landscapes 19 . For example, the abandonment of high-mountain traditional practices during the last half-century has reduced the composition of alpine pastures in many mountain ranges such as the Alps 14,20 , the Pyrenees 21 , and the Himalayas 22 .
Here, we reconstruct the response of the vegetation around Lake Sulsseewli, located in the northern Swiss Alps (Figure 1), to climate and human activities over the past 12,000 years. We used a multiproxy approach consisting of plant sedaDNA, pollen, fossil chironomids for summer temperature reconstruction, geochemical proxies, and multiple independent indicators of human activity, that included microscopic charcoal (re ecting re activity) and grazing indicators (coprophilous fungi spores and mammalian sedaDNA). Also, we assembled trnL P6 loop locus data from a new comprehensive taxonomic DNA reference database consisting of 3,923 plant species collected in the Alps and 417 from the Carpathians (the PhyloAlps database; http://phyloalps.org/). We demonstrate that PhyloAlps provides superior identi cation accuracy for plant sedaDNA as compared to three other non-local databases. The exceptionally high taxonomic resolution of the plant sedaDNA data allowed us to reconstruct both longterm changes in plant diversity and changes in plants that are particularly temperature-sensitive (i.e. with restricted elevational distributions), or are considered pastoral and arable indicators 23 . Our results show that the vegetation was mainly driven by temperature during the rst half of the Holocene (11 to 6 ka), while human activity drove changes from 6 ka. We nd that millennia of intensive human pressure prompted the rise of diversity that characterizes the present alpine diversity by favouring the coexistence of taxa normally found in different vegetation belts. These ndings suggest that in order to maintain the current high plant diversity of subalpine and alpine ecosystems in the face of ongoing climate warming, governments should aim to maintain moderate land use practices.

Improved identi cation of plants and ecological indicators from sedaDNA
After nal data ltering, we obtained sequences from 366 unique plant taxa from 73 sediment samples of Sulsseewli (see Methods, Supplementary dataset S1, Supplementary Figure SF4, Supplementary Tables ST8, ST9). We rst identi ed plant taxa in our sedaDNA data set using four reference databases (PhyloAlps, PhyloNorway, ArctBorBryo, and EMBL). PhyloAlps provided the highest number of sequences assigned to vascular plants overall and increased the number of taxa assigned at genus or species level by 30% compared to EMBLat genus or lower. After consolidating the identi cations from the four reference databases, PhyloAlps identi ed 87% of the total sequences, followed by EMBL which assigned 12% (Supplementary Tables ST3, ST4.1, ST4.2).
A total of 91 of our 366 identi ed plant taxa were informative of land use or vegetation belts: 87 were identi ed as indicators of speci c elevational vegetation belts (Material and methods, Supplementary  Table ST2), one indicated arable land (Myosotis arvensis), eight taxa were used for pastoral inference  (Supplementary Table ST5, Figure 2, B) and ve taxa are indicators for both vegetation belts and pastoral inference. For data analysis, all sedaDNA results are expressed as the relative abundance index (RAI), which integrates information from the relative proportion of reads and replicability of metabarcoding PCR (see Material and methods).
A local Holocene temperature reconstruction Fossil chironomids are excellent indicators of past summer temperature change, as many chironomid taxa are indicative of prevailing temperature conditions 24 . The remains of chironomid larvae are well preserved and remain identi able in lake sediment records. Chironomids from Sulsseewli indicate several major community composition shifts, which are indicative of changing temperatures (Supplementary Figure SF3). The compositions of chironomid communities were translated to estimates of past summer temperature change using a Swiss-Norwegian chironomid -July air temperature transfer function 24 that has been extensively used and tested in the Alps 25 . Reconstructed mean July temperatures range between 7 and 11 °C (Figure 2, D). The earliest Holocene presents a phase of relatively cool inferred temperatures (ca. 7.9 °C, before 11 ka), which was followed by an extended phase of relatively warm temperatures in the mid-Holocene (ca. 10.8 °C, 9.2 to 5.5 ka) and again cooler temperatures in the late Holocene (ca. 9.7°C, from 4 ka). This temperature development is in agreement with other climate records from the Alps 25,26 .

Determining drivers of vegetation changes
We conducted a redundancy analysis (RDA) of the plant sedaDNA data across all samples and added 21 explanatory variables to assess how the relative abundance index (RAI) of plants changed in relation to potential drivers, such as early human activities (represented by plant sedaDNA of pastoral and arable indicators, coprophilous fungal spores, sedaDNA of key mammals groups, and microcharcoal) and climatic changes (represented by chironomid-inferred temperatures, and organic matter content) ( Figure  4). The results of this analysis show that the rst axis (RDA 1, 53.35% of variance explained) is mainly related with multiple independent variables related to human in uence (plant sedaDNA of pastoral and arable indicators, the coprophilous fungi spores), and also separated samples older or younger than ~6 ka. Axis 1 was also related with the RAI of species representing the Colline-Montane and Montane-Subalpine vegetation belts, a group that also includes some taxa used as pastoral/arable indicators. This suggests that the distribution of these taxa may be in uenced by human activity rather than climate (estimated summer temperatures). SedaDNA of domestic mammals such as domestic goat and sheep strongly correlate with these indicators, providing independent evidence of this ecological inference. While sheep are more related to Colline-Montane taxa, goats correlate with our arable indicator and Montane-Subalpine taxa. The second axis (RDA 2, 17.06%) is related to variables re ecting climate change, such as the temperature inferred by chironomids and organic matter content of the sediments, as well as the relative proportion of Alpine-Subalpine communities. This second axis also broadly separates the Early and Mid Holocene samples. Given the results of the RDA analysis, we split our description of the proxy results into the two periods highlighted by RDA 1.

Temperature-driven vegetation changes during the Early to Mid-Holocene
The oldest analysed sediments of the Sulsseewli record are challenging to date precisely based on the available chronological and sedimentological data (Supplementary Figure SF2) but likely originate from the Younger Dryas-Holocene transition at 11.7 ka or the period just preceding this transition. Highly variable and low chironomid-derived July temperatures are inferred for these lowest sediment sections, which is consistent with the transition from the Younger Dryas into the Early Holocene. Based on the plant sedaDNA, this period was associated with a high abundance of cold-adapted Alpine forbs such as Achillea atrata, Crepis rhaetica and Silene acaulis with some Dryas octopetala dwarf shrubs. This period had the lowest values of organic matter (LOI), suggesting limited vegetation cover and organic production for this interval (Figure 2, F, G, Zone 1, 12-11.35 ka, T range = 7-9 °C). An increase in Subalpine/Alpine plant taxa (Anthyllis vulneraria, Athamanta cretensis, Chaerophyllum and Carex frigida), suggests an upward migration of the vegetation that could be associated with an initial plateau of inferred temperature of ~8.5 °C at 11.35-10.55 ka (Figure 2, B, Zone 2). This resulted in the largest forb introduction in the catchment (~50 taxa) and the highest diversity of sedaDNA plant species for the Early Holocene (~120 taxa per sample)(Supplementary Figure SF7). This climate-related establishment of new species apparently triggered a series of biotic interactions that, together with further increasing temperatures, changed the community composition and displaced alpine species and caused a decrease in richness (Figure 2 Holocene interval, we observe only sporadic mammalian sedaDNA sequences of two wildlife taxa, ibex (Capra ibex) and chamois (Rupicapra rupicapra), suggesting a low mammalian biomass and natural landscape at this time ( Figure 3, B, Supplementary Figure SF9).
Temperatures then increased and reached a second plateau by ~9.5 ka that were as much as ca. 3 °C higher than in the inferred Younger Dryas interval. This represents the most dramatic temperature increase in our record and, based on plant sedaDNA, prompted the upward migration of Subalpine taxa like Crepis bocconi whereas Subalpine-Alpine taxa almost disappear, and richness decreases in the vicinity of the lake (Figure 2, B, Zone 3, 10.55 -9.2 ka). The warmer and relatively stable temperatures of the Holocene Climatic Optimum (9-5.5 ka, ~11 °C) also allowed the upward expansion of Montane-Subalpine taxa (Abies alba, Chaerophyllum aureum, Lonicera alpigena), along with a dominance of woody taxa (Supplementary Figure SF4, SF5, Figure 2, B, C, Zone 4, 9.2 to 6 ka). The rise of shrubs (Maleae, Rosoideae), which began at ~10.5 ka, might have transformed ecosystem productivity and produced the highest values of sedimentary organic matter (LOI) (Figure 2, A, E, F). As shrubs expanded, the relative abundance index of herbs decreased and diversity dropped to only a third of the taxa that dominated in the earliest Holocene (~50 taxa) ( Figure 2, C, Supplementary Figure SF7). We recovered a single occurrence of domestic cow (Bos taurus) at ~9 ka, which we attribute to stochastic contamination (see Methods) as there is no archaeological evidence to support this occurrence 27 . A rise in coprophilous fungi spores between 7.9-7.15 ka suggests a greater abundance of mammalian herbivores and therefore higher grazing pressure in the vicinity of Sulsseewli (Figure 3, A, B). The lack of arable and pastoral indicators and domesticated livestock sedaDNA suggest that this may have resulted from wild animals. The decrease in trees and shrubs in the plant sedaDNA record beginning at 7 ka led to the introduction of new herb species and the rise of total plant diversity (~90 taxa) ( Figure 2, A, D, Supplementary Figure  SF8). During this interval, red deer (Cervus elaphus) are rst detected in the record.

Human-driven vegetation changes during the Mid-Late Holocene
The rst major evidence of human disturbance through re and grazing appears at 6.35 ka during the latter half of the Neolithic (7.5-4.  (Figure 2, B). However, a drop in all human indicators and the reappearance of red deer around 2.2-2 ka preceding the Roman period suggests the temporary abandonment of agropastoral activities in the zone, which apparently favoured Subalpine plant taxa and conifer reforestation (Pinus cembra, Picea, Supplementary Figure SF4, SF5), leading to another decrease in meadows (Figure 2, A). By the Late Roman period (ca.1.6 ka), a marked decrease of trees re ected by plant sedaDNA functional groups and increased microcharcoal concentrations suggest a high re incidence to clear the forest. The high proportions of pastoral plant sedaDNA and co-occurrence of domestic goat (Capra hircus), sheep and cow sedaDNA also support that forest was cleared for pastures of diverse livestock (Figure 3, A

Discussion
The diverse multiproxy and high resolution approach of this study makes it the most detailed palaeoecological reconstruction of Holocene Alpine vegetation to date. With 366 identi ed plant taxa, Sulsseewli yields the richest single record studied to date with sedaDNA using an exact match. This is 2-3 times as many taxa as found in any previous study of ancient plant sedaDNA 7,8,28,29 , and highlights the advantage of using a region-wide comprehensive, rather than global and incomplete reference databases 9,30,31 . Our sedaDNA results uncover a hidden diversity of taxa that are unrepresented by pollen analysis [32][33][34] , including many insect-pollinated plants normally underrepresented by pollen, and highlights the importance of forbs in Subalpine communities 35 . The pollen of Sulsseewli (166 vascular taxa) present a similar diversity as nearby lakes I gsee (143 taxa) 17 and Lac de Bretaye (137 taxa) 36 . These results suggest that the high values of richness found in Sulsseewli are representative and not only related to good DNA preservation and recovery, thereby indicating that the study region has been a biodiversity hotspot over the entire Holocene.
The high number and resolution of detected taxa allowed us to identify those that are informative about elevational vegetation belts and therefore reconstruct how Alpine, Subalpine and Montane plant communities changed throughout the Holocene at Sulsseewli, which is coherent with the regional trends of pollen and macrofossil studies [37][38][39] . Pollen studies reconstructed the shifts of the Montane-Subalpine ecotone by grouping characteristic pollen taxa, with the limitation that pollen taxa can often include several species and autecologies 17,40 . Some sedaDNA studies used single genera such as Plantago to infer human activities 9,30 . However, this genus has ~16 species that are found in the Alps of which only four are favoured by grazing (P. alpina, P. lanceolata, P. media, P. major 23,41 ). On the other hand, Liu et al. 7 used single species such as Sanguisorba o cinalis to infer human activities. However, this species can also be found in undisturbed sites such as wetlands or meadows. Our sedaDNA approach provides a more robust inference since its ecological interpretations are based on groups of well-characterised taxa.
We have also overcome the taxonomic limitation of some species sharing the same p6 loop sequence 42 by retaining the sequences as indicators only if their autecological category is shared among all species assigned to that sequence (see Methods). The information obtained from these stricter indicator sequences opens up a range of possibilities to robustly reconstruct unexplored parameters in palaeoecology.
The rise of DNA richness in Sulsseewli during the Early Holocene contrasts with the minor increase in palynological richness in Sulsseewli and the Alps 43 (Figure 2, A). In this study, the higher representation of forbs by sedaDNA lls a gap of the herbaceous taxa that could not have been resolved by pollen or might have been masked by high pollen-producing taxa (Supplementary Figure SF7). Our sedaDNA results paint a more detailed picture than the pollen results of how higher temperatures led to increased plant diversity during the Early Holocene and suggest that temperature was the main driver of vegetation changes during the rst part of the Holocene. This nding diverges from the weak relationship between sedaDNA-inferred plant taxa richness and independent temperature reconstructions that Liu et al. 7 found in Lake Naleng (4,200 m a.s.l, subalpine meadows, Tibet) during the Holocene. However, the different thresholds of the identi cation match for the sequences used in both studies limits the comparability of these results.
Our multiproxy palaeoecological approach also reveals that from the Neolithic onwards human activities promoted the co-occurence of taxa typical for different elevation belts. This resulted in a rise of diversity in the vicinity of Sulsseewli, a change also supported by a clear increase of the regional pollen diversity 44,45 . Considering the strong evidence of intensi ed grazing, it seems likely that grazing activities during the Neolithic modi ed the structure and composition of the vegetation by patchy disturbance, dung deposition and grazing preference. This is in line with social and human genetic changes that occurred in the Late Neolithic population 46  Our study showcases how human activity can promote a co-occurrence of taxa typical for different elevational vegetation belts. Based on the intermediate disturbance hypothesis 57,58 , intermediate levels of agropastoral disturbance during the last millennia may have maximized the spatial heterogeneity and plant species diversity. This is corroborated by contemporary studies that relate land abandonment during the past decades to a signi cant reduction in plant biodiversity 59 as well as by observations of peak plant richness with moderate levels of grazing 58 . Therefore, our long term reconstruction of plant diversity based on sedaDNA and other palaeoecological indicators con rms earlier studies that suggested that maintaining high plant diversity in the Alps requires conserving or promoting moderate land use practices 60 . This contrasts with palaeoecological studies elsewhere that suggest that grazing effects might be too weak to compensate for climate warming impacts 7,61 . This discrepancy might be explained by a lesser human impact in their study site, located at a much higher altitude (4,200 m a.s.l). However, our ndings stand with palaeoecological and contemporary studies that registered the potential of grazing to mitigate the elevational advance of climate-driven shifts in plant communities through increasing total stress 45,62 . Furthermore, recent studies suggest that shifts from intensive to extensive agricultural land use can favour native species while controlling the oversimpli cation of the ora due to plant invasion 61 .
Considering the observed general decrease of species' richness with elevation 2 , we conclude that a rise in diversity in alpine systems might be a sign that land-use allows the coexistence of species adapted to different habitats. This is in line with many studies that point out that regional richness may increase with warmer temperatures 2 , prompting an acceleration of changes in mountain summits during the past century due to the upward movement of lower vegetation belts 4 . In the view of our results, we question the use of total diversity as a measure of ecosystem health in alpine systems and consider the introduction of other diversity metrics. In contrast with Liu et al. 7 , we propose metrics that are based on particular elevational vegetation belts, as demonstrated here, instead of the within dominant alpine families since they can contain taxa belonging to different elevational habitats and communities. Therefore, if preserving the high biodiversity of alpine landscapes is a conservation goal, then policies should support the maintenance of traditional land-use practices in high-elevation areas 19,20,44,59,61 that will ensure the continuity of the ecological niches for alpine species and limit the spread of forest species that threaten them 45,63 . These measures are also important for other organisms that inhabit such habitats, such as the mountain hare or ring ouzel, which are especially vulnerable to global warming 64,65 .

Study Site
Sulsseewli is a small lake (2 ha) in the Bernese Alps (northern Swiss Alps), located in the subalpine vegetation zone below the present treeline (Figure 1 The organic content of the sediments was measured by mass loss-on-ignition (LOI) following the Lamb method 68 . Samples were taken at the same intervals as DNA, dried overnight at 105 °C and burned at 550 for 2 h to oxidize the organic matter. Total LOI was calculated as the percentage loss of dry weight after ignition.
Twenty-three plant macrofossil remains were sampled and radiocarbon-dated using accelerator mass spectrometry (AMS) at either the Laboratory for the Analysis of Radiocarbon at the University of Bern ('BE' accessions) or the Poznan Radiocarbon Laboratory ('Poz' accessions, Poland) (Supplementary Table   ST1). AMS radiocarbon dates were calibrated using the terrestrial IntCal20 curve 69

Chironomid-based temperature reconstruction
For chironomid analysis, 75 sediment subsamples of 0.125 to 9 cm 3 were sieved (100 µm) from a correlated composite section of cores SUL A and B without chemical pretreatment following Brooks 71 .
Chironomid head capsules were recovered from the sieve residue with a ne forceps under a stereomicroscope at 30-50x magni cation and mounted on slides in Euparal mounting medium.
Identi cation was performed by examining the mounted specimens under a compound microscope (100-400x magni cation) and using keys for Chironomidae Larvae 71,72 . Pollen, spores and non-pollen palynomorphs (NPPs) were identi ed at 400× magni cation according to Moore et al. 74 and Reille 76 and the reference collection at the Institute of Plant Sciences (University of Bern). A minimum of 500 terrestrial pollen grains per sample was counted. Up to 166 pollen types of terrestrial and aquatic vascular plants were identi ed. NPP and coprophilous fungal spores were identi ed according to van Geel et al. 77 . Diagrams were plotted with the program Tilia. Statistically signi cant pollen zones were determined with partitioning using optimal sum of squares and the broken stick method 78 . To study the regional re activity, microscopic charcoal was analysed 79,80 . Particles between 10-500 µm were counted on the pollen slides following Tinner et al. 81 . Coprophilous fungi spores were analyzed to infer grazing pressure (Cercophora, Delitschia, Podospora, Sordaria, Sporormiella and Trichodelitschia). The palynological results are presented as percentages of the terrestrial pollen sum excluding aquatic plants whereas charcoals and fungal spores were presented as in ux (particles cm -2 year -1 , Figure SF5).

Sedimentary ancient DNA data generation
Core SUL C was subsampled at 10-cm resolution in the ancient DNA lab at TMU. DNA extraction was conducted in a clean ancient DNA laboratory and followed the protocols of Rijal et al. 8 . DNA was extracted from 80 sediment samples and 13 extraction/sampling negative controls using a modi ed DNeasy PowerSoil kit (Qiagen, Germany) protocol in the ancient DNA laboratory at TMU. DNA extracts and negative extraction/sampling controls, along with two positive PCR controls, were ampli ed using a uniquely dual-tagged generic primer sets (Supplementary dataset S5) that amplify either the trnL P6 loop region of the chloroplast genome, a locus that has proven most successful for studies of plant sedaDNA 82,83 , for plants (gh primers; 84 or a section of the mammalian mitochondrial 16S locus (MamP007 primers; 9 ). Reaction and cycling conditions for all PCRs followed Voldstad et al. 85 , with the following exceptions for the 16S PCRs: (1) forward and reverse primer concentration was reduced to 0.1 μM each, and (2) forward and reverse blocking primers were added at 1 μM each. We used a forward blocking primer modi ed from Giguet-Covex et al. 9 with a sequence of GGAGCTTTAATTTATTAATGCAAACAGTAGG-C3 and a reverse blocking primer with a sequence of CCCAACCGAAATTTTTAATGCAGGTTTGGTGA-C3. Eight PCR replicates were carried out for each sample or control for trnL, whereas four replicates were performed for 16S. The PCR replicates were pooled and cleaned, and the pools were converted into 4 DNA libraries using a modi ed TruSeq PCR-free library kit Each library was sequenced on ~10% of a ow cell on an Illumina NextSeq 500 platform (2x150 bp, midoutput mode) at the Genomics Support Centre Tromsø (UiT).

Database construction and sedimentary ancient DNA data analysis
The OBITools software package 86 was used for the bioinformatics pipeline, following the protocol and criteria de ned by Rijal et al. 8 . Brie y, paired-end reads were aligned using SeqPrep (https://github.com/jstjohn/SeqPrep/releases, v1.2). Merged reads were demultiplexed according to the 8 bp unique primer tags and identical sequences were collapsed. Singleton sequences and those shorter than 10 bp were removed and putative artifactual sequences were identi ed and removed from the dataset. 29 Previous metabarcoding studies of the Alps that targeted the trnL P6 loop have relied on global databases with a sparse representation of the Alpine ora resulting in limited, and potentially inaccurate, identi cations 9,30,31 . Here we generated a local DNA reference library. We used the PhyloAlps genome skim database that consist of 4604 specimens of 4437 taxa collected in the Alps (n=4280) and the Carpates (n=324) (https://data.phyloalps.org/browse/). A full description of this database including the standard barcodes ITS2, matK and rbcL are available in Alsos et al. 83 . Here, we compiled a P6loop database by running the ecoPCR software 87 on their corresponding raw databases with the gh primers.
The same approach was also used for the PhyloNorway genome skim database 29,83 , and the global EMBL rl143 database.
The assignment of the reads were then done to the four different reference libraries: (1) PhyloAlps; (2) ArctBorBryo (regional arctic/boreal reference library compiled from Willerslev et al. 28 , Sønstebø et al. 42 and Soininen et al. 88 ); (3) PhyloNorway 83 ; and (4) the global reference library based on the EMBL rl143 database. For mammals, we generated a reference library from the EMBL rl143 database as described above except using the 16S primers. The identi ed sequences were ltered in R using a custom script (available at https://github.com/Y-Lammers/MergeAndFilter). Only sequences with a 100% match to a reference sequence, represented by three reads per replicate, and with a minimum of 10 total reads and three replicates across the entire data set were retained. Furthermore, arti cial sequences that were the result of sequencing errors, such as homopolymer length variation, were merged with the source sequence (following Rijal et al. 8 ). Additionally, sequences that displayed higher average read counts in negative extraction or PCR controls than lake sediment samples were removed, as well as common laboratory contaminants (Supplementary Table ST6 Table ST6, Supplementary dataset S5). In the 16S data set, we determined nal taxonomic assignments based on scrutiny of NCBI BLAST 89 hits and comparison to known temporal biogeographic distributions (full justi cations are given in Supplementary dataset S5). Data for 16S sequences assigned to the same taxon were collapsed. We discarded all nonmammalian sequences and those assigned as human (Homo sapiens) or wild boar (Sus scrofa), as the former is a common contaminant (also found in the negative controls) and the latter cannot be differentiated from sympatric domestic pig at this locus (Supplementary dataset S5). Although absent from the negative controls, we note that domesticated taxa (pig, cow, sheep, goat) can be sources of sporadic PCR contamination. We therefore interpret sporadic occurrences of these taxa, de ned as single, isolated PCR replicate detections, as likely deriving from contamination. We note that these occur in otherwise low diversity samples that are most at risk of sporadic contamination detections (e.g. sheep at 6.1 ka; cow at 7.2, 8.9, 12.5 ka).
We present a novel relative abundance index (RAI) to integrate the information of the relative abundance of reads and PCR replicability. It is a proportional index multiplying the proportion of obtained reads by the proportion of weighted PCR replicates (weighted PCR replicates are de ned in Rijal et al. 8 ). This study follows the line of a growing number of environmental DNA modern studies that suggest that the number of DNA copies contain information about the species' relative abundance 90 . DNA sequences to identify plants associated with distinct elevational belts, arable and pastoral activities We used all the obtained sequences to get the maximum autecological information, even if not assigned at species level, for reconstructing distributional changes of plants typical for particular vegetation belts, arable and pastoral activities. From all the obtained DNA sequences, only taxa with narrow elevational, climatic and/or habitat requirements are good environmental indicators that can provide insights into past environmental change, while other sequences that do not originate from speci c indicator taxa might smooth the results, and by consequence, our palaeoecological interpretation. First, we matched the obtained sequences from Sulsseewli against the PhyloAlps database to obtain a list of the species that share the same sequence (haplotype-sharing species from now on). Then, we checked these haplotypesharing species with the ecological parameters of Flora Indicativa for temperature, which is the main parameter that characterizes the elevational vegetation belts. The obtained results were graded from optimal to not suitable indicators. e.g.: if one sequence has 4 haplotype-sharing species and all of them have the same ecological value, it is considered an optimal indicator for temperature, while con icting values indicate a non-suitable indicator. Similarly as for temperature, indicators for pastoral and arable activities were obtained from the sequence data based on the review of palaeoecological literature and Flora indicativa. The regional coherence of all sequences was checked. Optimal indicator sequences for elevational vegetation belts, and arable and pastoral activities as identi ed based on this procedure were plotted versus other palaeoecological indicators in Figures 2 and 3. These comparisons formed the basis for interpretations regarding changes in vegetation caused by immigrations or expansions of plant species typical for various altitudinal belts or promoted by human activities. If climate was the main driver of past vegetation change, then changes in taxa representative for different vegetation belts are expected to track the chironomid temperature record. If human impact was the main driver, then these changes would be expected to follow the patterns of other palaeoecological indicators of human activities. Constrained incremental sum of squares (Coniss) zonation for all samples was used to infer statistically signi cant changes of plant DNA sequences along the core. Finally, we performed a Redundancy analysis (RDA) of the relative abundance index of all samples with 21 added explanatory variables to explore the in uence of climate (represented by chironomid-inferred temperatures, organic matter (LOI)) and human impacts (represented by plant sedaDNA of pastoral and arable indicators, coprophilous fungi spores, sedaDNA from domesticated mammals, and microcharcoal) on the plant communities. Plots were made with R v3.4.2 using the Vegan, Rioja and Ggplot2 packages, RDA and CONISS were calculated in R using the vegan and rioja packages, respectively [91][92][93] .

Data and Code Accessibility
We will deposit raw DNA sequence data on the European Nucleotide Archive (ENA) with accessions xxxxxx-xxxxxx. The un ltered Obitools output tsv les will be on gshare (doi: xxxxxxx). All other raw and processed data and links to code will be within the manuscript or supplementary information. Chironomid count data and chironomid-inferred temperatures will be uploaded to the Dryad database (datadryad.org, doi: xxxxxxx) and pollen data to the Neotoma database (https://www.neotomadb.org). Author contributions IGA and SL designed the research, raised the funding, and provided resources; SL, PAC and EC generated and curated the PhyloAlps database; IGA, CS, and FR did the eldwork; SGP did the ancient DNA laboratory work with input from IGA and PDH; TG performed radiocarbon dating; CS built composite cores and SGP performed age-depth modeling with input from CS and OH; FR performed pollen, charcoal and non-pollen palynomorphs analysis; MH and AR performed chironomid analysis supervised by OH, who also built the temperature reconstruction; SGP veri ed and curated the plant barcode sequence taxonomic assignments with input from IGA and JPT, who also veri ed their botanic origin; PDH designed the blocking primers and veri ed the mammal barcode sequence taxonomic assignments; OW and SGP designed the bioinformatic pipeline to obtain the indicator sequences, which were veri ed by JPT; SGP and OW developed the Relative Abundance Index; SGP and YL performed bioinformatics and the quality control checks; SGP did the statistical analyses with input from OW; SGP curated the data; WT, IGA, SL, CS, JPT, AGB, FB and OH contributed with the palaeoecological and botanical interpretation; PDH and KW provided interpretation of the mammalian data; KW provided archaeological interpretation of the region.