Habitat-specific patterns of bacterial communities in a glacier-fed lake on the Tibetan Plateau

Abstract Different types of inlet water are expected to affect microbial communities of lake ecosystems due to changing environmental conditions and the dispersal of species. However, knowledge of the effects of changes in environmental conditions and export of microbial assemblages on lake ecosystems is limited, especially for glacier-fed lakes. Here, we collected water samples from the surface water of a glacier-fed lake and its two fed streams on the Tibetan Plateau to investigate the importance of glacial and non-glacial streams as sources of diversity for lake bacterial communities. Results showed that the glacial stream was an important source of microorganisms in the studied lake, contributing 45.53% to the total bacterial community in the lake water, while only 19.14% of bacterial community in the lake water was seeded by the non-glacial stream. Bacterial communities were significantly different between the glacier-fed lake and its two fed streams. pH, conductivity, total dissolved solids, water temperature and total nitrogen had a significant effect on bacterial spatial turnover, and together explained 36.2% of the variation of bacterial distribution among habitats. Moreover, bacterial co-occurrence associations tended to be stronger in the lake water than in stream habitats. Collectively, this study may provide an important reference for assessing the contributions of different inlet water sources to glacier-fed lakes.


Introduction
Glacier-fed lakes, as an important part of the cryospher e, ar e aquatic ecosystems dir ectl y affected by glaciers and are sensitive indicators of climate change (Hotaling et al. 2017, Zhang et al. 2020 ).Global warming has been shown to increase both the rate and extent of glacier melting in high-latitude and high-altitude regions (Lee et al. 2017 , Cauvy-Fraunie andDangles 2019 ).Rapid melting of glaciers is expected to result in an expansion of glacierfed lake areas due to increased meltwater (Slemmons et al. 2013, Zhang et al. 2015 ).Massive glacial meltwater inflows could affect micr obial assembla ges by exerting significant contr ols on abiotic and biotic features of glacier-fed lakes (Slemmons andSaros 2012 , Tiberti et al. 2020 ).For instance, glacier meltwater can transport high concentrations of mineral particles into downstream ecosystems, leading to high turbidity in glacier-fed ecosystems (Sommaruga 2015 ).The high turbidity limits light penetration into water, causing unfavorable conditions for primary producers and further indir ectl y affecting the micr obial comm unity assembla ges (Peter and Sommaruga 2016, Peter et al. 2018, Tiberti et al. 2020 ).Furthermore , glacial meltwater ma y also dir ectl y affect the micr obial diversity of downstream lakes by exporting microorganisms (Cameron et al. 2017, Kohler et al. 2020 ).
In addition to the recruitment of microbes from glaciers, glacier-fed lakes also can r eceiv e species fr om surr ounding envir onments, suc h as soil (Crump et al. 2012 ), permafrost melt-water (Bomberg et al. 2019 ) and groundwater (Echeverria-Vega et al. 2018 ).Given that glaciers are retreating rapidly and many will disappear within decades under global warming, the effects of glacier meltwater on microbial diversity in downstream lakes will diminish (Peter andSommaruga 2016 , Liu et al. 2019 ).Subsequently, the water and microbial sources of glacial lakes will shift to be dominated by non-glacial str eams (e.g.gr oundwaterfed streams) (Freimann et al. 2013 ).Such shifts in water and microbial sources will have major impacts on the physicochemical c har acteristics and micr obial comm unities in glacial lakes (Peter and Sommaruga 2016 ).In general, glacial lake systems are influenced by high disc har ge of glacial meltwater during summer ablation and an increasing influence of groundw ater to w ar ds winter (Freimann et al. 2013 ).Liu et al. ( 2019 ) demonstrated that bacterial abundances and alpha diversity increased as the amount of meltwater increased during the glacier melting seasons.In contrast to glacial meltwater supply, groundwater-fed systems pr ovide mor e spatiotempor al stability and a more homogeneous landsca pe (Br own et al. 2003, Battin et al. 2004 ), and the smaller fluctuations in physicochemical characteristics in such habitats cause reduced variability in bacterial communities.Taken together, it is important to e v aluate the impact of glacial and nonglacial stream sources on microbial diversity in glacial lakes.
SourceTr ac ker based on a Bayesian mixing model has been widely used to identify the sources of community assembly (Peter and Sommaruga 2016, Comte et al. 2017, Cameron et al. 2020 ).It is commonly recognized that microbial communities ar e highl y unbalanced: a small number of species ar e highl y abundant (r eferr ed to as "abundant biospher e"), while a lar ge number of other species have a low abundance (r eferr ed to as the "r ar e biospher e") (Lync h andNeufeld 2015 , Li et al. 2023 ).The abundant micr oor ganisms contribute most of the micr obial biomass, wher eas the r ar e micr oor ganisms may act as a "seed bank" for maintaining microbial diversity.Previous studies have found that abundant and r ar e comm unities sho w ed the opposite source patterns in "sink microbial communities".For instance, a pr e vious study on sediment bacteria in the Yarlung Tsangpo River sho w ed that the potential contributions of abundant and r ar e comm unities fr om upstr eam and tributaries to downstr eam wer e differ ent (Liu et al. 2022 ).By contr ast, se v er al studies hav e also shown that the total, abundant and r ar e comm unities hav e similar proportions of contributions for sink microbial communities (Wang et al. 2021, Xiong et al. 2021 ).Ho w e v er, it r emains unclear whether the source patterns of abundant and r ar e taxa in the microbial communities of the glacier-fed lake are similar.
To e v aluate the r ole of glacial and non-glacial sources for diversity of the glacier-fed lake, we sim ultaneousl y collected water samples from a glacier-fed lake named Amuco and its two external inlet streams on the Tibetan Plateau in June 2018, that is, a glacial stream and a non-glacial stream ( Fig. S1 ).Bacterial comm unities wer e obtained using the Illumina MiSeq sequencing method with 16S rRNA gene amplicons.Considering the lake is supplied mainly by glacial meltwater, we hypothesized that the species pool of the glacial stream contributes more to the microbial communities of the lake than the non-glacial str eam.Giv en that the glacier-fed lake acts as recipients of micr oor ganisms fr om div ersel y alloc hthonous r esour ces, w e further hypothesized that the bacterial diversity in the glacier-fed lake was higher than that in the two external inlet streams.

Study area and sampling
Lake Amuco (33.45 • N, 88.72 • E, 4960 m above sea level) is situated in the eastern Qiangtang Plateau of the central Tibetan Plateau.The area of the lake is a ppr oximatel y 34.8 km 2 and the maximum depth is 19 m.Lake Amuco receives meltwater from two streams.One is a glacial stream formed by glacial meltwater from the Qiangtang No.1 glacier (33.29 • N, 88.70 • E) in the southern part of the lake with a total length of 12 km (Li et al. 2017 ).The other stream is a non-glacial stream in the northwestern part of the lake, which is formed by precipitation and permafrost meltwater ( Fig. S1 ).
In June 2018, a total of 37 water samples were collected from the surface of the lake and its two inlet str eams.Specificall y, the 37 samples were collected from the surface (0.5 m) of the glacierfed lake (13), the glacial stream (15) and the non-glacial stream (9), r espectiv el y.Appr oximatel y 5 L of water was collected from each site with a Schindler sampler.Each water sample was divided into two subsamples: one for DNA extraction and the other for measuring physiochemical properties.First, 500 ml of water was filter ed thr ough a 20-μm mesh (Millipor e, USA) to r emov e lar ge particles and was then filtered through a 0.22-μm polycarbonate membr ane (47-mm diameter, Millipor e, USA) for DNA extraction.The filters were stored at −80 • C until molecular analysis.Also, 100 ml of the water sample was filtered through a 0.45-μm hydr ophilic pol yethersulfone (PES) syringe filter (25 mm, Anpel) to measure the dissolved organic carbon (DOC) and total nitrogen (TN) concentrations and was frozen at −20 • C for physicochemical analysis in the laboratory.
Sampling site coordinates were recorded using a Global Positioning System.The pH, conductivity (Cond), total dissolved solids (TDS) and water temper atur e (Temp) were monitored in situ with a YSI m ulti-pr obe Water Quality Sonde (Y SI EXO2, Yellow Springs , OH, USA).The DOC and TN concentr ations in the filter ed water wer e measur ed by a Shimadzu Total Organic Carbon Analyzer (TOC-VCPH, Shimadzu Corporation, Japan) with a TN measuring unit (TNM-1, Shimadzu, Ja pan) thr ough high-temper atur e catal ytic oxidation (680 • C).Befor e combustion, eac h sample was first acidified with 1 M HCl and sparged with carrier gas to r emov e all the inorganic carbon (Guo et al. 2022 ).

DN A extr action, 16S rRN A gene amplicon sequencing and Illumina MiSeq sequencing
Environmental DN A w as extr acted fr om the filters using the Fast DNA ® Spin kit (MP Biomedicals, Santa Ana, CA, USA) according to the manufacturer's instructions .T he ra w DN A w as quantified with a NanoDrop 1000 Spectrophotometer (T hermo-Scientific).T he V4 region of the bacterial 16S rRNA genes was amplified with a uniquely tagged primer pair 515F (5 GTGCCAGCMGCCGCGGTAA-3 ) and 806R (5 GGA CTA CHV GGGTWTCTAAT -3 ).The 50-μL pol ymer ase c hain reaction (PCR) systems were performed in triplicate, with each containing 10 ng of DNA template, 5 μL of 2x Premix Taq DNA pol ymer ase (Takar a Biotec hnology, Dalian, China), 1 μL of each primer (10 μM) and 20 μL of nuclease-free w ater.PCR w as performed under the following conditions: 94 • C for 3 min, follo w ed by 30 cycles of 94 • C for 30 s, 60 • C for 30 s and 72 • C for 1 min, after which we performed a final cycle of 5 min at 72 • C.After amplication, we pooled multiple samples together in equal volumes .P ooled samples were purified using Agencourt AMpure XP beads.PCR pr oducts wer e sequenced using the Illumina MiSeq platform 2 × 250 bp paired-ends (Illumina, San Diego, CA, USA).T he ra w sequencing data gener ated in this study wer e submitted to the NCBI short r eads arc hiv e (SRA) database under BioProject number PRJNA884020.

Sequence processing
The paired-end reads were assembled with FLASH (v. 1.2.11) using default settings (Magoc and Salzberg 2011 ).We processed the sequences mainly using the Quantitative Insights Into Microbial Ecology (QIIME) pipeline (v.1.8) (Ca por aso et al. 2010 ).After quality filtering, denoising and c himer a r emov al, high-quality r eads wer e clustered into operational taxonomic units (OTUs) at the cutoff of 97% with the UPARSE algorithm (Edgar 2013 ).Re presentati ve sequences fr om eac h OTU wer e determined using the r efer ence SILVA database (version 132 NR) at a confidence cutoff of 80% (Quast et al. 2013 ).After taxonomies had been assigned, OTUs that were affiliated with chloroplast, archaeal and unclassified sequences were removed from the following analysis.To avoid artifacts from sequencing depth, we used a randomly selected subset of 20 600 sequences based on the sample with the smallest sequencing for further analysis.

Sta tistical anal ysis
The Alpha-diversity indices (Shannon div ersity, Ric hness, Pielou's e v enness) wer e calculated with the pac ka ge "v egan" in R (v ersion 4.0.3).Non-metric multidimensional scaling analysis (NMDS) based on the Bray-Curtis dissimilarity was applied to c har acter-ize differences in bacterial community composition between samples (Clarke 1993 ).Analysis of similarities (ANOSIM) analyses of Bray-Curtis dissimilarity was used to e v aluate the differ ences between samples grouped by habitats.Similarity percentage (SIM-PER) analysis identified the principal OTUs responsible for the differences between sample groups, which can be performed in the PAST software (Hammer et al. 2001 ).The significantly discriminant taxa in each major cluster (habitats groups) were determined using the Linear discriminant analysis (LDA) effect size (LEfSe) method (Segata et al. 2011 ).The Venn dia gr am was drawn using the "VennDia gr am" pac ka ge (Chen and Boutr os 2011 ).
To identify the significance of environmental factors that may be influencing changes in community composition, a distancebased m ultiv ariate linear model (DistLM) analysis on Bray-Curtis distances was performed using the DISTLM_forw ar d3 pr ogr am (McArdle and Anderson 2001 ).And we assessed the impact of environmental factors on the indicator species by Randomforest anal ysis, whic h was performed with the pac ka ge "r andomFor est" in R (version 4.0.3).
SourceTr ac ker based on a Bayesian mixing model (Knights et al. 2011 ) was used to identify the different sources and estimate their contribution to the bacterial community composition of the lakes.It could estimate the r elativ e contributions of micr obes fr om m ultiple sources to an envir onment.In SourceTr ac ker anal yses, the r elativ e contributions fr om differ ent sources to a sink environment are modeled as a pr obabilistic mixtur e of the composition of sources (Baral et al. 2018 ).Independent SourceTr ac ker anal yses were carried out for different taxonomic community levels (total community, abundant community and rare community) and dominant bacterial phyla/classes a total of 14 times.We defined "abundant OTUs" as those having r elativ e abundances abov e 0.1% of total sequences, and "r ar e OTUs" as those having r elativ e abundances below 0.01% (Jiao et al. 2017 , Jiao andLu 2020 ).The surface water of the lake was set as the sink, and the contributions of the two str eam comm unities to the sink comm unities wer e quantified.
A co-occurrence netw ork w as constructed using the "igr a ph" and "Hmisc" pac ka ges in R. T he O TUs were those that occupied at least 20% of the samples that had been used to construct the co-occurrence network.The pairwise Spearman's correlations between OTUs were calculated.Each shown connection has a correlation coefficient > | 0.8 | and a P value < 0.01.The pairwise comparisons based on OTUs and the false discov ery r ate (FDR)-adjusted P v alue wer e performed using the "rcorr" function in the "Hmisc" pac ka ge .T he co-occurrence network was visualized using Gephi platform (v.0.9.2) ( https://gephi.org ) (Bastian et al. 2009 ).In addition, we performed indicator species analysis of whole OTUs based on the OTU r elativ e abundance (indv al v alue > 0.6 and P < 0.05 are strong indicators for a habitat) using the indval function in R (version 4.0.3).And using the results about habitats of indicator operational taxonomic, we colored the network.Statistical differences in node-level attributes between the matrices were determined using non-parametric Kruskal-Wallis tests.

Physioc hemical c haracteristics
The physioc hemical par ameters of the water samples fr om the lake and streams are summarized in Table S1 .As shown in Table S1 , the surface water of the lake is c har acterized b y lo w temper atur e and low pH values, but high conductivity.The water temper atur e (9.91 ± 2.39 • C), pH (8.35 ± 0.04), DOC (1.06 ± 0.75 mg/L) and TN (0.19 ± 0.07 mg/L) sho w ed the lo w est values in the surface water of the lake.Ho w e v er, Cond and TDS sho w ed the highest mean values in the lake surface water (mean 9.64 ± 14.68 ms/cm and 9585.77± 14802.76 mg/L, r espectiv el y).The lo w est values of Cond (0.42 ± 0.09 ms/cm) and TDS (208.96 ± 44.39 mg/L) were found in glacial stream water.

Bacterial taxonomic composition and di v ersity
A total of 762 200 reads were obtained after quality control and r ar efication, and these wer e cluster ed into 44 049 OTUs at 97% similarity le v el.Taxonomic anal yses of our sequence data r e v ealed that Pseudomonadota (original name: Proteobacteria) (mean r elativ e abundance, 50%) and Bacteroidota (original name: Bacter oidetes) (28%) wer e the two dominant phyla in all stream and lake samples (Fig. 1 ).Ninety-five per cent of the Pseudomonadota phylum from the three habitats in total was related to the classes Beta pr oteobacteria, Alpha pr oteobacteria and Gamma pr oteobacteria, with Beta pr oteobacteria (27.4% of all sequences) and Gamma pr oteobacteria (13.2% of all sequences) enriched in the non-glacial stream.The Pseudomonadota populations were dominated by Alpha pr oteobacteria in the lake (22.6% of all sequences), and Desulfobacterota (original name: Deltaproteobacteria) were more abundant in the lake water samples than in stream samples.Actinomycetota (original name: Actinobacteria), Bacillota (original name: Firmicutes) and OD1 were present in low abundance in all samples (Fig. 1 ).Furthermore, we used LEfSe analysis to identify specialized bacterial communities of different habitats and search for biomarkers from the phylum to the genus le v el.A total of 37 microbial taxa were found to be significantly different among three habitats ( Fig. S2 ).There were greater numbers of species enriched at a significant level (LDA > 4) in GS (15) and Lake (18) compared with those in NGS (4).Lefse analysis revealed that within the phyla Bacteroidota, the genera Flavobacterium , Acinetobacter , Leadbetterella , Emticicia and Mycoplana were enriched in the glacial str eam samples, wher eas the phylum OD1 and the genus Pseudomonas wer e enric hed in the non-glacial str eam samples .T he dominant bacteria in the lake water samples were Actinomycetota, Bacillota and Verrucomicrobiota (original name: Verrucomicrobia).On the genus level, the Planomicrobium and Luteolibacter wer e enric hed in the lake water samples.
The Shannon diversity index ranged from 6.3 to 9.5 across all samples, with the highest values in the surface water of the lake and the lo w est in the glacial stream.Pielou's evenness estimated for all habitats ranged from 0.5 to 0.8, with the highest values in the surface water of the lake and the lowest in the glacial stream.The Richness for three habitats varied from 1533 to 4928, with mean values being highest in the lake surface water and lo w est in the non-glacial stream.The alpha diversity indices of lake water samples were significantly higher than those in the two external inlet streams (Kruskal-Wallis test, all P < 0.05, Fig. 2 ).
NMDS anal ysis r e v ealed the samples fr om the lake water and stream water were clearly separated (Fig. 3 A).Ho w ever, tw o stream samples partially overlapped to a certain extent.This observation was further confirmed by ANOSIM, showing that the micr obial comm unity structur es wer e significantl y differ ent among habitats ( Table S2 ).Lake water samples were found to be significantly dissimilar to glacial stream (R = 0.752, P < 0.001) and nonglacial stream samples (R = 0.742, P < 0.001).Ho w e v er, glacial str eam samples wer e calculated to hav e low dissimilarities to non-glacial stream assemblages (R = 0.288, P < 0.01).
OTUs that were shared between sample groups were used as an indicator of the potential transfer of assemblages among habitats .T he Venn dia gr ams indicated a total of 4059 OTUs (9.2% of the detected OTUs) with 84.1% (640 871 sequences) of total sequences that were shared among all three habitats (Fig. 3 B and  C).The proportion of unique OTUs was highest in the lake (35.5%, 15 644 OTUs), follo w ed b y the glacial stream (24.6%, 10 837 OTUs) and non-glacial stream (10.0%, 4398 O TUs).T he number of OTUs shared between the glacial stream and the lake was 8043 (18.2%), which was higher than the OTUs between other pairwise habitats (Fig. 3 B and C).
The SIMPER analysis identified the top 11 OTUs that cumulativ el y contributed 25.21% to the differences in bacterial community composition among three habitats (Fig. 4 ).The top 11 OTUs w ere dominated b y species from the phyla Pseudomonadota and Bacter oidota, whic h together contributed 22.23% to the ov er all differences .T he O TU32 belonging to the genus Flavobacterium was abundant in three habitats, explaining 7.14% of the ov er all community dissimilarity, follo w ed b y Comamonadaceae O TU37, O TU237 and OTU65 (5.56%) and Acinetobacter OTU33 (2.83%) (Fig. 4 and Table S3 ).The r andomfor est model r e v ealed that pH, Cond and TDS wer e str ong pr edictors for differ ences in r elativ e abundances of most specific OTUs ( Fig. S4 ).For example, pH was most correlated with the r elativ e abundance of bacteria including

Linkage of environmental variables and bacterial biodi v ersity
Bacterial alpha diversity indices, including Shannon diversity, Richness and Pielou's evenness, exhibited strongly negative  correlations with pH, DOC and TN, r espectiv el y (Spearman's rank correlations, P < 0.05 in all cases, Fig. 2 D).Ho w e v er, Cond and TDS sho w ed significantl y positiv e r elationships with thr ee bacterial alpha diversity indices (Fig. 2 D).For community composition, DistLM analysis sho w ed that pH, Cond, TDS, Temp and TN were significantl y corr elated with bacterial comm unity (Table 1 ).These significant v ariables totall y explained 36.2% of the variation of the bacterial community composition.Of all the measur ed v ariables, pH (14.1%) was primaril y r esponsible for the variation of the bacterial communities across the three habitats, follo w ed b y Cond (7.7%), TDS (5.3%), Temp (4.6%) and TN (4.6%).

Source contributions
Using the Bayesian classifier SourceTr ac ker, we e v aluated the importance of two inlet stream sources of diversity for downstream lake comm unities (Fig. 5 ).SourceTr ac ker r esults r e v ealed that 64.67% of the total community in the lake water was seeded by the glacial streams (45.53%) and the non-glacial streams (19.14%) micr obial comm unities (Fig. 5 A).At the population le v el, str eam sources contributed 86.35% and 76.34% of the total sequences of Beta pr oteobacteria and Bacillota in lake surface water communities, r espectiv el y.Ho w e v er, the potential contributing sources for Desulfobacterota (88.6%) and Planctomycetota (original name:  Planctomycetes) (80.9%) in lake water comm unities wer e lar gel y unknown (Fig. 5 B).Furthermor e, onl y 6.96% of the abundant community (Fig. 5 C) and 11.72% of the rare community (Fig. 5 D) in the lake water were contributed by glacial stream and non-glacial stream bacterial communities , respectively.T he potential contributing sour ces w er e lar gel y unknown (93.04% and 88.28%, r espectiv el y).As for the abundant communities, lake water bacterial comm unities deriv ed fr om the glacial str eam (3.13%) wer e lo w er than those from the non-glacial stream (3.83%) (Fig. 5 C).Howe v er, the r ar e comm unities sho w ed the opposite pattern, with the glacial stream (7.21%) being a relatively significant source for lake water bacterial communities (Fig. 5 D).

Co-occurrence patterns of bacterial communities in different habitats
Co-occurrence patterns of bacterial taxa were estimated by constructing correlation networks for water samples from different habitats .T he correlation-based network consisted of 878 nodes (OTUs) and 16 880 edges (correlations) for the whole bacterial communities.A module is defined as a group of OTUs that are linked more tightly together.Here, bacterial netw orks w ere clearly parsed into se v en major modules, of whic h modules I, II and III accounted for 19.36%, 18.34% and 18.00% of the whole bac-terial network, r espectiv el y (Fig. 6 A).The nodes in the network were assigned to 24 bacterial phyla, among whic h thr ee phyla (Pseudomonadota, Bacteroidota, Actinomycetota) were widely distributed, accounting for more than 71.52% of all nodes (Fig. 6 B).
The network was colored in accordance with the indicator species of three habitats; we found these nodes sho w ed different pr efer ences to habitats (Fig. 6 C).For example, the majority of nodes in module 1, module 2 and module 3 were the most abundant in lake water samples, the majority of nodes in module 4 and module 5 were the most abundant in glacial streams samples, while nodes in module 7 were the most abundant in nonglacial streams samples .T his finding suggests that habitat difference plays a k e y role in determining the network modular structur e. Bacterial comm unities within each habitat could have more interactions instead of outside it.
To further e v aluate the inter actions of OTUs within a network of three habitats, their corresponding node-level topological properties were calculated ( Fig. S5 ).In general, the higher the values, the stronger the interactions of species (Banerjee et al. 2018 ).Values of the topological features including node degree (connectivity), betweenness centrality and eigenvector centrality were highest in lake surface water samples, indicating that bacterial communities in the glacial-fed lake had the strongest co-occurrence associations (Kruskal-Wallis test: P < 0.05).F igure 6. Co-occurrence netw orks of the bacterial comm unity based on pairwise Spearman's corr elations between O TUs .A connection denotes a strong (Spearman's ρ > 0.8) and significant ( P < 0.01) correlation.The nodes in the networks are colored according to modularity class (A), taxonomy (B) and habitats (C).Node size is proportional to the number of connections (i.e.degree).

Habita t-specific pa tterns of bacterial communities
Our finding indicated that the bacterial alpha diversity was significantly higher in the lake water than glacial and non-glacial str eams (Fig. 2 ).This r esult was inconsistent with se v er al studies that reported the bacterial alpha diversity was higher in inlet streams as opposed to surface water samples (Crump et al. 2012, Comte et al. 2017, Cavaco et al. 2019, Gu et al. 2021 ).A possible explanation for this might be that the inlet streams carry alloc hthonous micr oor ganisms into the lake, thus we could consider the surface water of the lake as a "sink" of glacial and terrestrial ecosystems .T hese str eams ar e likel y passiv e conduits for micr oor ganisms sourced fr om the upstr eam ecosystems (Cav aco et al. 2019, Zhang et al. 2021 ) and can be selectiv el y seeded in the downstream lake ecosystems (Sheik et al. 2015, Hauptmann et al. 2016 ).
The bacterial communities sho w ed clear separation betw een the three habitats (Fig. 3 A).Pseudomonadota and Bacteroidota were dominant bacterial phyla in the lake and two streams of water (Fig. 1 ).The high r elativ e abundances of Pseudomonadota and Bacter oidota observ ed in the glacial-fed lake w ater w ere not surprising as they have also been observed in many studies of glacierfed lake ecosystems (Gu et al. 2021, Liu et al. 2021, Zhang et al. 2021 ).Inter estingl y, we found that the phylum OD1 presented a high r elativ e abundance in the non-glacial stream water.The phylum Parcubacteria (OD1) belongs to the super phyla Patescibacteria, which has been found to be pr e v alent in water environments and streamlined many functions to adapt to the special environment, such as low and less nutrients, darkness and low oxygen (Tian et al. 2020 ).
Different habitat conditions probably lead to the percentage of shared OTUs among the three habitats being lo w er than those unique OTUs occurring in each habitat (Fig. 3 B).Ho w ever, the few shared OTUs (9.2%) accounted for a considerable proportion of the total sequences (84.1%), further substantiating the idea that the most ubiquitous taxa are often the most abundant (Fig. 3 B  and C) (Salazar et al. 2016, Li et al. 2023 ).Taxonomic composition analyses at the order le v el r e v ealed that species affiliated with Burkholderiales (Beta pr oteobacteria) and Flavobacteriales (Bacteroidota) dominated the shared OTUs ( Fig. S3a ).These bacteria are well known for their organic matter degradation abilities and wide distribution in high-altitude lakes and/or stream ecosystems (Newton et al. 2011, Hotaling et al. 2019, Liu et al. 2021 ).The most abundant unique taxa identified in each habitat (e.g.Bacteroidota in glacial stream and lake) were similar to those dominant shared taxa across the three habitats .T his finding is consistent with previous studies on some taxa displaying very different spatial distributions and environmental preferences and tolerances (Ruiz-Gonzalez et al. 2019 ).
SIMPER analysis identified 11 OTUs primarily responsible for the differences in bacterial community compositions observed across the three habitats (Fig. 4 ).These OTUs are strongly associated with environmental variables ( Fig. S4 ).For instance, the discriminant taxa belonging to the genus Flavobacterium were negativ el y corr elated with Cond and TDS, but positiv el y corr elated with pH and TN.Members of the Flavobacterium were highly abundant in freshwater and marine ecosystems, and were known for their ability to r a pidl y exploit bioav ailable or ganic matter (Wakie wicz and Irzykowska 2014 ).
Our results also illustrated that distribution of bacterial communities was driven by local en vironmental variables .We found that pH was the critical environmental factor in shaping the dis-tribution of bacterial communities across the three habitats (Table 1 ).The importance of pH in sha ping micr obial comm unities has been shown in pr e vious studies conducted in lakes (Lindstrom et al. 2005, Ren et al. 2015 ).The pH could affect microbial growth and metabolism by altering the balance of H + and OH -ions on the cell wall/membrane (Yang et al. 2019 ).Our results further demonstrated that conductivity was another driving factor that differed between bacterial communities among the three habitats.Previous studies have shown that conductivity is a major parameter driving the community patterns of streams (Wilhelm et al. 2013 ) and lakes (Liu et al. 2019, Gu et al. 2021 ).The conductivity is a measure of water conduction current, which is related to the total dissolved salt content of the water.Aquatic micr obes r equir e a r elativ el y stable concentration of the major dissolved ions in the water.Le v els too high or too low may limit microbial survival (Liu et al. 2019 ).

Bacterial communities exhibit distinctive co-occurrence patterns across three habitats
We further examined the associations between bacterial communities of the Lake Amuco and its two inlet streams.Network topological properties can reflect interactions between microbial species .For instance , the degr ee v alue describes the connectivity between OTUs in a network (Deng et al. 2012 ), and nodes with high betweenness centr ality r epr esent or ganisms that ar e important for maintaining the network (Zhu et al. 2019, Zhang et al. 2021 ).In addition, the closeness centrality value reflects how quickly information spr eads fr om a giv en node to other r eac hable nodes, and eigenvector centrality is used to describe the degree of a central node that is connected to other central nodes (Deng et al. 2012 ).Our results indicated that lake water bacterial OTUs have the highest node degree, betweenness centrality and eigenvector centr ality v alues ( Fig. S5 ).This suggests that bacterial OTUs in the lake water exhibited closer interconnections than those in the glacial and non-glacial streams.A possible explanation for this is that stoc hastic r atios of bacterial comm unities wer e gener all y lo w er in Amuco lake than in stream habitats (Liu et al. 2021 ), indicating that the r elativ e influence of deterministic processes on bacterial comm unities incr eased fr om inlet str eams to lake water.Deterministic pr ocesses involv e envir onmental filtering and biotic interactions (Liu et al. 2021, Yang et al. 2022 ).In our study, bacterial communities of lake water wer e mor e pr ominentl y affected by environmental filtering (e.g.lo w er w ater temper atur e), and had stronger biotic interactions, that is species co-occur more fr equentl y in the lake water.Another possible explanation is that high micr obial div ersity in the lake water may result in strong micr obial co-occurr ence associations.

Contribution of the two streams to bacterial communities in the lake water
Our r esults r e v ealed that the glacial str eam, r ather than the nonglacial stream, was an important source of diversity for bacterial community in the glacial-fed lake water, supporting our first hypothesis that the glacial stream contributes more species to the micr obial comm unity of the downstr eam lake than the nonglacial str eam.Se v er al phyla wer e pr edominatel y calculated to be sourced from the glacial stream, such as Actinomycetota, Bacillota, Alpha pr oteobacteria, Verrucomicr obiota and Beta pr oteobacteria (Fig. 5 B).These phyla are regularly identified in glacial environments (Wilhelm et al. 2013, Sharma et al. 2020, Zhang et al. 2021 ), suggesting that they may comprise cryophilic taxa, which is supported by culti vation-de pendent approaches (Cheng andFoght 2007 , Sherpa et al. 2018 ).A possible explanation for this re-sult is the presence of resident communities in the lake .T his finding was supported by pr e vious observ ations, whic h r eported that the most important source of bacteria in lakes was the resident community due to the importance of priority effects (Hanson et al. 2012, Comte et al. 2017 ).Consistent with this finding, higher relative abundances of Planctomycetota and Desulfobacterota were observed in the lake water than in the streams (Fig. 1 ).Another possible explanation is that other possible sources were not considered in our study, such as soil (Crump et al. 2012 ) and other strata of the lakes (Comte et al. 2017 ).
Our results also sho w ed that the r ar e comm unities in lake water contributed by the glacial stream are higher than those from the non-glacial stream, while abundant communities show the opposite proportions .T hese findings suggest that glacial streams tr ansport mor e r ar e taxa to lake water bacterial comm unities, while non-glacial streams may transport more abundant taxa to lake water.In addition, the lo w er Shannon index and higher Richness of the glacial stream compared with the non-glacial stream supports this deduction (Fig. 2 ).A pr e vious study reported that some taxa exhibited a high r elativ e abundance in glacial streams and their predominance may have resulted in their extremely low e v enness and lo w est Shannon index (Freimann et al. 2013, Liu et al. 2016 ).T his ma y be inter pr eted as a r esult of a dominance of a fe w r ar e taxa ada pted to the curr ent set of environmental conditions (Freimann et al. 2013 ).Once the glacial stream has decreased and shifted to w ar ds the non-glacial str eam, r ar e taxa transported to lake water will subsequently be diminished.In addition, the water flow rate changes of the glacial stream across the season will dir ectl y and indir ectl y c hange the bacterial comm unity composition.Bacterial community composition of the lake can be related to w ater flo w and the import of bacterial cells from the dr aina ge area.The w ater flo w rate, to some extent, likel y also mirr ors water residence time in the stream, although the latter was not directly measured in this study.The importance of water residence time in shaping microbial communities has been shown in previous studies (Lindstrom et al. 2005, Ruiz-Gonzalez et al. 2015 ), as it can regulate the balance between the transportation of bacteria from adjacent ecosystems, and the sorting of species by local environmental conditions .T hese findings confirm that the glacial stream is important and r epr esents a div erse source community for the glacier-fed lake (Wilhelm et al. 2013 , Peter andSommaruga 2016 ).

Conclusions
In summary, this study provides insights into the distribution patterns and drivers of bacterial communities in a glacier-fed lake in the central Tibetan Plateau.Our results demonstrated that the bacterial communities exhibited high heterogeneity of the glacierfed lake and its two inlet streams that was closely related to envir onmental factors, suc h as pH, conductivity and total dissolved solids.Our results also indicated that the glacial stream was an important source of diversity for the lake surface bacterial communities and that bacterial alpha diversity was highest in the lake surface water.Furthermore, species co-occurrences were more frequent in the lake water.Collectiv el y, the findings expand our knowledge of microbes in a glacier-fed aquatic ecosystem and offer valuable insights into the sources of diversity for glacier-fed lakes.

Figure 1 .
Figure 1.Relative abundance of dominant bacterial taxa at the phylum/class le v el.GS: the water samples of glacial stream; NGS: the water samples of non-glacial stream.

Figur e 2 .
Figur e 2. T he bacterial alpha diversity among three habitats.Indices of alpha diversity are shown as Shannon diversity (A), Pielou's evenness (B) and Ric hness (C).Differ ent lo w er case letters indicate significant differences among habitats .T he top and bottom boundaries of each box indicate the 75th and 25th quartile v alues, r espectiv el y, and lines within each box r epr esent the median values .T he mean values are represented by the white point.(D) Spearman correlations between the alpha diversity and environmental variables.Asterisks indicate the statistical significance ( * * * P < 0.001; * * P < 0.01; and * P < 0.05).GS refers to communities collected from water samples of the glacial stream.NGS refers to communities collected from water samples of the non-glacial stream.Temp: temperature; Cond: conductivity; TDS: total dissolved solids; DOC: dissolved organic carbon; TN: total nitrogen.

Figure 3 .
Figure 3. (A) Non-metric multidimensional scaling (NMDS) ordination visualization of bacterial community compositions (Bray-Curtis distance) among four habitats .T he 95% confidence ellipses are shown for each habitat.Venn diagrams showing (B) the number and proportion of shared and unique OTUs and (C) the number and proportion of shared and unique sequences across the GS, NGS and Lake.GS: the water samples of the glacial stream; NGS: the water samples of the non-glacial stream.

Figur e 4 .
Figur e 4. T he SIMPER analysis showing the principal operational taxonomic units (OTUs) responsible for the differences between habitats.Mean sequences indicate the number of sequences belonging to each OTU.GS: the water samples of the glacial stream; NGS: the water samples of the non-glacial stream.

Figure 5 .
Figure 5. Results from SourceTr ac ker anal ysis showing the contribution of the differ ent source comm unities to the lake water.(A) Based on the total O TUs .(B) Based on the 11 most abundant phyla/classes.(C) Based on the abundant O TUs .(D) Based on the r ar e O TUs .GS: the water samples of the glacial stream; NGS: the water samples of the non-glacial stream.

Table 1 .
Distance-based m ultiv ariate linear model of bacterial community composition showing percentage of variation explained by envir onmental v ariables.(Br ay-Curtis distances, 999 permutations.)

P Percentage v aria tion explained Cum ulati v e variation explained
Data in bold r epr esent significant correlations ( P < 0.05) TDS: total dissolved solids; Cond: conductivity; TN: total nitrogen; Temp: temperature; DOC: dissolved organic carbon.