Geographical variation in soil bacterial community structure in tropical forests in Southeast Asia and temperate forests in Japan based on pyrosequencing analysis of 16S rRNA

Natsumi Ito, Hiroko Iwanaga, Suliana Charles, Bibian Diway, John Sabang, Lucy Chong, Satoshi Nanami, Koichi Kamiya, Shawn Lum, Ulfah J. Siregar, Ko Harada and Naohiko T. Miyashita* Laboratory of Plant Genetics, Graduate School of Agriculture, Kyoto University, Kitashirakawa-Oiwake-cho, Sakyoku, Kyoto, Kyoto Prefecture 606-8502, Japan Sarawak Forestry Corporation, Lot 218, KCLD, Jalan Tapang, Kota Sentosa, 93250 Kuching, Sarawak, Malaysia Laboratory of Plant Ecology, Graduate School of Science, Osaka City University, 3-3-138 Sugimoto, Sumiyoshi-ku, Osaka, Osaka Prefecture 558-8585, Japan Department of Forest Resources, Faculty of Agriculture, Ehime University, Tarumi, Matsuyama, Ehime Prefecture 790-8566, Japan Department of Natural Sciences and Science Education, National Institute of Education, Nanyang Technological University, Nanyang Walk, Singapore 637616 Faculty of Forestry, Bogor Agricultural University and Southeast Asian Regional Centre for Tropical Biology, Jl. Raya Tajur Km. 6, Bogor 16134, Indonesia


INTRODUCTION
Soil microbes are among the most diverse and abundant organisms on Earth (Whitman et al., 1998;Torsvik et al., 2002;Venter et al., 2004), and they play important roles N. ITO et al. in ecological systems, promoting decomposition of organic matter (i.e., plant litter and dead animals), supplying nutrients and stimulating growth and health of aboveground plant life, and facilitating the carbon and nitrogen cycles (Chapin, 1980;Högberg et al., 2001;Kowalchuk and Stephen, 2001;Hättenschwiler et al., 2005;van der Heijden et al., 2008).Soil microbes can exist in or on the surface of plant roots in a mutualistic or commensal manner, or they can remain physically apart from but near plant roots; in the latter arrangement, the bacteria modulate the supply or availability of plant nutrients in the soil, and the fractional rate of resources (van der Heijden et al., 2008).On the other hand, plant litter is also a growth substrate and a source of nutrients for soil microbes (Wu et al., 2011), and substances secreted from the roots of living plants also support the growth and survival of microbes (Bais et al., 2006).This suggests a mutual reciprocal relationship between the growth and productivity of the aboveground vegetation and those of soil microbes (Wardle et al., 2004;Harris, 2009).
Molecular studies have recently examined the relationship between the above-and belowground biotas in forest environments.A 16S rRNA pyrosequencing analysis showed that the composition of the soil bacterial communities in natural forests in Australia varied with variation in the dominant aboveground vegetation (Hackl et al., 2004); for example, high-GC content Grampositive bacteria (Firmicutes and Actinomycetes) were most dominant (49% of all data) in pine forests, followed by Alphaproteobacteria.In contrast, in oak-hornbeam and spruce-fir-beech forests, Holophagae of Acidobacteria was most dominant, followed by Verrucomicrobia and Alphaproteobacteria, respectively.However, these forests also differed in soil chemical properties, microbe biomass and metabolic rate of nutrients, making it unclear which factor has the greatest influence on soil bacteria composition.A study using automated ribosomal internal transcribed spacer (ITS) analysis (ARISA) reported that the structure of bacterial and fungal communities in fynbos soils in South Africa is stable over time, but varies with the nature of the aboveground vegetation at different sampling points, suggesting that the growth of some bacteria and fungi may depend on the presence of particular plant species (Slabbert et al., 2010).In a study of forest soils in Oregon, USA, in which terminal restriction fragment length polymorphism (T-RFLP) analysis for eubacterial and archaeal 16S rRNA and length heterogeneity (LH)-PCR analysis for fungal ITS sequences were employed, it was reported that the fungal and archaeal community structures were different in forests dominated aboveground by Douglas-fir or red alder vegetation, but that the structure of the bacterial community did not vary (Yarwood et al., 2010).As noted above, these studies also suggest that the presence of certain plant species aboveground influences the growth and survival of spe-cific microbes belowground.
Nevertheless, many complex interacting factors, both biotic and abiotic, influence the distribution of biota in a single soil ecosystem, such that the exact factors that influence the structure of the soil microbial community are not completely understood (Wardle et al., 2004).As an example of biotic factors, both the developmental stage and the species of crops (alfalfa, clover and beans) were reported to influence the structure of the bacterial community in soil and in association with plant roots in cultivated land in Germany, based on the temperature gradient gel electrophoresis (TGGE) separation of 16S rRNA (Wieland et al., 2001).A change in land use management from forest to pasture also correlated with change in the structure of the soil bacterial community in Hawaii and Brazil, using GC content abundance analysis of SSU rRNA and T-RFLP of 16S rRNA, respectively (Nüsslein and Tiedje, 1999;Jesus et al., 2009).Additional physicochemical characteristics of soil, such as temperature (Fiere et al., 2005), salinity (Lozupone and Knight, 2007), pH (Fierer and Jackson, 2006;Tripathi et al., 2012), nutrient abundance (Fiere et al., 2012) and redox potential (DeAngelis et al., 2010), also affect growth, survival and function of soil microbes.A study of 98 forest soils in North and South America revealed a strong association between soil pH and the microbial community structure, based on T-RFLP and pyrosequencing analyses of 16S rRNA (Fierer and Jackson, 2006;Lauber et al., 2009).Even when two or more environmental conditions (e.g., temperature, moisture content, C/N ratio, texture, pH) were considered simultaneously, soil pH appeared to be most strongly correlated with bacterial diversity.
Considering the relationship between above-and belowground biotas, in areas with high biodiversity for both animal and plant species, such as tropical forests, the soil microbial ecosystem is expected to be more complex than in areas with less biodiversity (Myers et al., 2000).One study on pasture and forest soils in tropical Brazil, using a culture-independent metagenomic approach based on 16S rRNA sequences, identified previously unreported sequences, and suggested that "immense" microbial diversity existed in the studied regions (Borneman and Triplett, 1997).On the other hand, subtropical soils in Brazil and Florida and temperate soils in Illinois and Canada were comparable with respect to the level of soil bacterial diversity (10 4 species per sample), based on pyrosequencing analyses of the 16S rRNA region (Roesch et al., 2007).Recently, by studying 25 temperate grassland sites from four continents (Africa, Australia, Europe and North America) by high-throughput sequencing of 16S rRNA (bacteria and archaea) and an ITS region (fungi), it was shown that plant α-diversity was poorly related to that for microbial groups (Prober et al., 2015).In our previous work on five tropical forests in Malaysia and one temperate forest in Japan, using pyrosequencing analy-Soil bacterial community structure in the tropics ses of bacterial 16S rRNA (Miyashita et al., 2013), we also observed a similar composition of higher taxonomic taxa and similar bacterial species diversity in the six forests.
The present study investigates bacterial diversity and composition in 28 forest soils based on pyrosequencing analyses of the 16S rRNA region.This extends our previous study (Miyashita et al., 2013) by including an additional 21 tropical forests in Malaysia, Indonesia and Singapore, and one temperate forest in Japan.These 28 forests had different aboveground vegetation, including dipterocarp forest, acacia plantation, oil palm plantation, limestone forest and mangrove.Thus, it should be possible to determine whether the conclusion of our previous study would hold for forest soil bacterial communities having more diverse geographic origins and aboveground vegetation.The aim of this study was to elucidate the environmental factors and mechanisms that influence the spatial biogeography of a soil bacterial community.For this purpose, we conducted the following analyses: 1) comparison of higher taxonomic composition and species diversity among soil bacterial communities, geographic regions and aboveground vegetation (forest type); 2) comparison of operational taxonomic unit (OTU) occurrence within and between forests; 3) examination of the distance effect by testing the correlation between β-diversity and direct distance between forests; 4) examination of the effect of geographical region and aboveground vegetation (forest type) on the soil bacterial community (composition); and 5) evaluation of the association between α-and β-diversities and physicochemical properties of soil.Primary conclusions were as follows: 1) among the investigated forests, bacterial phyla were represented in similar proportions, despite diverse environmental differences, but species diversity differed considerably; 2) OTU distribution was skewed, and forest-specific OTUs were predominant; 3) a significant and positive correlation was detected between β-diversity and direct distance between forests, suggesting local adaptation and bacterial dispersal among forests; 4) both geographical region and aboveground vegetation (forest type) influenced soil bacterial composition; 5) α-and β-diversities were correlated with soil pH.

MATERIALS AND METHODS
Soil sampling and DNA extraction Soil samples were collected from 26 tropical forests in Malaysia (Sarawak), Indonesia (Java, Sumatra) and Singapore and from two temperate forests in Japan in 2010-2011 (Supplementary Fig. S1).A detailed description of the environmental condition and forest type at each sampling location is given in Table 1.In each forest, ~10 g of soil located ~10 cm below the surface was sampled at 10 sampling sites 1-10 m apart from each other.The soil samples were mixed and transferred into a tube containing extraction beads and buffer immediately after removal, and then frozen.Frozen soil samples were shipped to laboratories.Microbial DNA was extracted using a PowerSoil DNA Isolation Kit or a PowerMax Soil DNA Isolation Kit (MO BIO Laboratories, Carlsbad, CA, USA) according to the manufacturer's instructions.Extracted DNA was purified using a PowerClean DNA Clean-Up Kit (MO BIO Laboratories).For Indonesian forests, 1 kg of soil was collected at two randomly chosen sampling sites to determine soil physicochemical properties (soil type, moisture content, organic C content, N content, and pH).The average value of the two sampling sites for the Indonesian samples was used for analyses (Supplementary Table S1).For the other forests, 1 kg of soil pooled from 10 sampling sites was used to determine physicochemical properties.However, some samples were not examined for one or more soil properties (Supplementary Table S1).All necessary permits were obtained for the described field studies.Research permits for the studied forests and export permits for the extracted DNA samples were issued by the Forest Department Sarawak, the National Parks Board (Singapore Botanical Gardens), the Ministry of Environment and Forestry and Indonesian Academy of Sciences through SEAMEO BIOTROP (Bogor, Indonesia), the Ashiu Experimental Forest (the Field Science Education Center, Kyoto University), and the Ibuki Property District Manager (Maibara, Shiga Prefecture).

PCR amplification of bacterial 16S rRNA region
Extracted microbial DNA was used as template for PCR amplification of part of the 16S rRNA region as described (Miyashita et al., 2013;Miyashita, 2015).Briefly, two PCR primers were used for reactions in this study: 9F 5'-GAGTTTGATCCTGGCTCAG-3' and 533R 5'-TIACC-GIIICTICTGGCAC-3'.These primers amplify a 510-bp fragment that includes three variable regions (V1, V2 and V3) of the 16S rRNA gene in bacteria, archaea and eukarya (Baker et al., 2003).To distinguish the 28 samples from each other, a unique tag sequence was attached to the 5' end of the forward and reverse primers, as described (Miyashita et al., 2013).The PCR solution (50 μl) contained 0.2 mM dNTPs, 2.5 units of TaKaRa ExTaq (Takara Bio, Otsu, Japan), 10 × Taq buffer, 2 mM MgCl 2 , 0.2 μM each of the forward and reverse primers, and ~10 ng of template DNA.The PCR conditions used were, sequentially: 94 °C for 2 min; 30 cycles of 94 °C for 30 sec denaturation, 55 °C for 30 sec annealing and 72 °C for 1 min extension; and 72 °C for 5 min extension.For each sample, three replicate PCR products were combined and quantified by agarose gel electrophoresis, and ~200 ng of PCR amplicons from each of the 28 samples was prepared for pyrosequencing.The PCR amplicons were cleaned using the QIAquick PCR Purification Kit (Qiagen, Venlo, Netherlands) following the manufacturer's instructions.(Quince et al., 2009;Caporaso et al., 2010a).In this process, reads shorter than 300 bp and/or having a quality score lower than 50 were eliminated from the data set.A total of 267,073 sequences were used for further analyses.All of the sequences were clustered into OTUs based on their sequence identity using the uclust method (Caporaso et al., 2010b), and a cluster at 3% distance was defined as a "species" (OTU) following the conventional definitions (Bond et al., 1995;Schloss and Handelsman, 2005;Huber et al., 2007).For each OTU a representative sequence that was the most abundant sequence in an OTU cluster was selected, and a set of the representative sequences were aligned using the PyNAST method (Caporaso et al., 2010b) and taxonomically assigned based on the 16S rRNA database in the Ribosomal Database Project (RDP) at Michigan State University (Cole et al., 2009) using the uclust consensus taxonomy classifier.Chimeric sequences were identified using ChimeraSlayer (http://microbiomeutil.sourceforge.net/#A_CS), and OTUs represented by these chimeric sequences were excluded from all downstream analyses.The above-mentioned cluster analyses, alignment, taxonomic assignment and chimera identification were performed using QIIME tools.A total of 256,693 nonchimerical sequences were used for further analyses.To compare the composition of bacterial taxa among the forests, Spearman's rank correlation was calculated for all the pairwise comparisons between forests for 29 phyla detected and for the top 30 ranked taxa of class and order, using Excel and a web tool (http://www.gen-info.osaka-u.ac.jp/testdocs/tomocom/spea.html).Heterogeneity in the taxon frequency among the forests was tested by chisquare tests, using marginal frequencies of taxon and location to obtain the expectation with Excel.Based on the results of cluster analyses, soil bacterial species diversity (α-diversity) in each forest was estimated; α-diversity was represented by two indices, the Chao1 index (Chao, 1984), which estimates the minimum number of OTUs based on the observed number of OTUs, and the Shannon H' index (and equitability: Shannon index divided by log e (the number of OTUs)) (Shannon, 1948), which estimates the evenness of OTU distribution.Rarefaction curves (Sanders, 1968;Gotelli and Colwell, 2001) were created by a tool in QIIME.To normalize read number variation, 2,951 sequences (the minimum number of reads in all forest samples) were randomly resampled 100 times from each sample.For the resampled sequence sets, the average values for the 100 resampled sequence sets were estimated for the Chao1 index and Shannon H' index (and equitability).Pearson's correlation coefficients were used to examine the relationship between α-diversity and environmental factors.To estimate β-diversity (differentiation) between samples, weighted UniFrac (Lozupone et al., 2007) and the 1 -abundant Jaccard index (Chao et al., 2005), which measures similarity in OTU composition between forests, were calculated for all pairwise comparisons.Based on the β-diversity distance matrices, Principal Coordinates Analysis (PCoA) plots (Anderson and Willis, 2003) and Unweighted Pair Group Method with Arithmetic Mean (UPGMA) dendrograms were constructed.Calculation of β-diversity indices and construction of dendrograms and PCoA plots were performed using QIIME tools.Pearson's correlation coefficients were used to examine the relationship between coordinate values of PCoA and environmental factors.The latitude and longitude data of sampling locations (Table 1) were used to estimate the direct distance between all the pairs of sampling locations with a web tool (http://vldb.gsi.go.jp/ sokuchi/surveycalc/surveycalc/bl2stf.html) of the Geospatial Information Authority of Japan, Ministry of Land, Infrastructure, Transportation and Tourism.Since latitude and longitude data are missing for the mixed dipterocarp forest in Bako National Park (BKOmdf), the total number of pairs was 351 ( = 27 × 26 / 2).Statistical significance of the correlation coefficient between the pairwise direct distance and β-diversity (1 -abundant Jaccard index and weighted UniFrac) was examined by the Mantel test (Sokal and Rohlf, 2000).Mantel correlogram analysis (Sokal, 1986;Legendre and Legendre, 1998) was conducted to examine autocorrelation in the soil bacterial community structure.The number of distance classes was determined to be nine based on Sturge's rule (the number of classes = 1 + log 2 (m), where m = the number of distances in the β-diversity triangular matrix; in this study, m = 351).Two different distance classes were used (Legendre and Legendre, 1998): equal distance and equal count (Supplementary Table S2).The Mantel test and Mantel correlogram analysis were performed using PASSaGE v2 (Pattern Analysis, Spatial Statistics and Geographic Exegesis) software (http://www.passagesoftware.net/,Rosenberg and Anderson, 2011).Statistical significance of the correlation coefficient for the Mantel test and Mantel statistics for the correlogram analysis was tested by 10,000 permutations.For the Mantel correlogram analysis, the progressive Bonferroni correction was used.Permutational ANOVA (PERMANOVA, Anderson, 2001) was applied to the estimated triangular β-diversity matrices to test the effect of soil sampling geographical region and aboveground forest type on β-diversity in terms of community structure ("location" and "dispersion") within and between groups (region and forest type).One-way design for each N. ITO et al.
of the groups, in which factors were fixed, was set up.To test the statistical significance of the pseudo-F value, the distribution of pseudo-F statistics obtained by permutation of raw data for 9,999 times by the unrestricted method was used.PERMANOVA was performed using the PRIMER v7 and PERMANOVA + add-on computer programs (http://www.primer-e.com/).The 16S rRNA sequences have been deposited under DDBJ accession nos.DRR014854 to DRR014881.

RESULTS
Bacterial composition at higher taxonomic levels in investigated forest soils A total of 29 of 55 phyla listed in the RDP database were detected, representing 89.2% of all sequence reads (Fig. 1 and Supplementary Fig. S2).The average number of phyla per forest was 16.4.A mangrove forest in the Bako National Park (BKOman) had the highest number of phyla ( 26) and a kerangas forest in the Bako National Park (BKOker) had the lowest number of phyla ( 13).The two temperate forests in Japan (Ashiu and IBUlim) had a moderate number of phyla (16 and 18, respectively).Nine phyla (Acidobacteria, Proteobacteria, Verrucomicrobia, Planctomycetes, Bacteroidetes, TM7, Actinobacteria, Firmicutes and OP10) were detected in all 28 forests, and three phyla (Bacteria_incertae_sedis, Gemmatimonadetes and OP11) were found in 27 forests, the exceptional forests being BKOman, a peat swamp forest in Bako (BKOpts) and BKOpts, respectively.Several phyla were found uniquely in one forest, including Deferribacteres, Fusobacteria, Deinococcus-Thermus, and Thermotogae in the BKOman forest, Synergistetes in Bakam secondary forest (BER2nd), and Chlamydiae in Lambir Hills National Park Valley area (LHNval).A total of 10.8% of the 16S rRNA sequence reads were assigned to unclassified bacteria, a group belonging to the domain Bacteria but not assigned to any existing phylum, ranging from 4.8% in a mixed dipterocarp forest in Bako (BKOmdf) to 26.2% in a secondary forest in Bau (BAU2nd).
As a whole, Acidobacteria were the most dominant phylum (41%) among the 29 phyla detected, and Proteobacteria followed (23%) (Fig. 1).The predominant five phyla (Acidobacteria, Proteobacteria, Verrucomicrobia, Planctomycetes and Bacteroidetes) accounted for more than 80% of the phyla detected.However, there was considerable variation in frequencies of bacterial phyla among forests.Exceptional composition of phyla was observed in the BKOman mangrove forest (Fig. 1), where Acidobacteria were represented at a lower frequency (3%) and Proteobacteria, Bacteroidetes, and Cyanobacteria were represented at higher frequencies (37%, 11% and 10%, respectively) than in other forests.To examine the similarity in phylum composition in the forests, Spearman's Fig. 1.Composition of phyla in the 28 soil samples.Soil bacterial community structure in the tropics rank correlation was calculated for all pairwise comparisons (378 comparisons) of 28 forests (Table 2).Despite the variation in phylum frequency among the forests and above-mentioned environmental conditions, statistically significant rank correlations were detected at least at the 1% level in all comparisons, indicating similar taxonomic composition at the phylum level.The statistical significance of rank correlations probably reflects the dominance of the top five phyla in all forests studied.However, when heterogeneity in frequency of individual phyla was evaluated by chi-square tests, statistically significant heterogeneity was detected at least at the 1% level for all phyla detected.These results indicated that, despite the overall similarity in phylum composition, the representation of individual phyla in these forests was influenced by some forest-specific factor(s).
At the class and order levels, about 87% and 77% of the total sequence reads were assigned to 61 classes and 67 orders, respectively (Supplementary Fig. S2).Compared with the phylum composition, the class and order compositions were more variable among forests (Supplementary Fig. S3).Some dominant taxa were common, and the top five or six dominant taxa accounted for more than 50% of assigned classes (Supplementary Fig. S3A).Acidobacteria subgroup Gp1 was the most dominant class (20%) of all classes detected, and Acidobacteria Gp2 (14%) and Alphaproteobacteria (10%) followed.In most forests, Acidobacteria Gp1 and Gp2 were the major classes of the phylum Acidobacteria, accounting for 80-90% of Acidobacteria (Supplementary Fig. S4A).However, instead of Acidobacteria Gp1 and Gp2, Gp4 and Gp6 were predominant in an oil palm forest in Jasinga (JASoil), a secondary forest in Bau (BAU2nd), and a limestone forest in Mt.Ibuki (IBUlim).Similarly, Acidobacteria Gp4 and Gp10 were predominant in BKOman, and Gp5 and Gp6 were predominant in limestone forests in Bau (BAUlim) and Niah (NIAlim).The class composition within Proteobacteria also differed among forests.In most forests, Alphaproteobacteria was the most dominant of five classes, but in some forests (BAU2nd, BAUlim, BKOman, NIAlim and IBUlim) mentioned above, Betaproteobacteria was the most dominant class (Supplementary Fig. S4B).At the class and order levels, Spearman's rank correlations were lower than at the phylum level, and were not sig-nificant for some pairs, but the percentage was low (11.6 and 6.6% for class and order, respectively, Table 2), still consistent with the generally similar taxonomic composition in the forests studied at the class and order levels.The non-significant rank correlations involved BAUlim and BKOman forests, ndicating that these two forests differed in class and order composition from the other forests.However, statistically significant heterogeneity was detected at the 1% level for all classes and orders as well as phyla, implying that forest-specific factors influence the frequency of taxa at the level of class and order.
Soil bacterial community structure at the species level To examine bacterial composition and diversity at the species level, cluster analysis based on 97% sequence identity was carried out for the entire data set.Rarefaction curves at the 3% divergence level did not plateau for most of the forests, so that the observed number of OTUs was insufficiently represented for actual bacterial diversity (Supplementary Fig. S5).The total number of OTUs obtained from the cluster analysis was 21,223 for the 28 forests (Table 3).The average number of OTUs per forest was 1,414, ranging from 855 (BKOmdf, a mixed dipterocarp forest in Bako) to 2,992 (BKOman).The estimated Chao1 index for all 28 forests was 45,130.4(Table 3).The average Chao1 index per forest was 2,569, ranging from 1,245.7 (SEMmac, a Shorea macrophylla plantation in Semengoh) to 6,861.8 (BKOman).These results imply a large variation in bacterial species diversity among investigated forests.The two Japanese forests had intermediate levels of species diversity, indicating that temperate forests did not necessarily have lower bacterial species diversity than tropical forests.All the Shannon indices estimated were high, indicating highly skewed OTU distribution in all the forests (Table 3).In fact, Pearson's correlation between the Shannon index and Chao1 index or percentage of singleton OTUs was highly significant (r = 0.770, df = 26, P < 0.01 for Chao1 index and r = 0.689, df = 26, P < 0.01 for percentage of singleton OTUs), indicating that the bias of OTU distribution was associated with the abundance of overall and rare OTUs in the forest soils.The BKOman forest had the highest Shannon index and percentage of singleton OTUs, while forests with low Shannon indices (for example, TIMrid, a ridge area of mixed dipterocarp forest in the Bukit Timah National Park, BER2nd and SEMmac) had lower percentages of singleton OTUs.Although Pearson's correlation between the equitability and percentage of singleton OTUs was also highly significant (r = 0.527, df = 26, P < 0.01), relatively high equitability values indicated the absence of highly predominant OTUs, consistent with the low maximum number of reads per OTU in all forests, except for a secondary forest in the Bakam Experimental Reserve (BER2nd, Table 3).
Although Pearson's correlation coefficients between  was highly significant (r = 0.783, df = 26, P < 0.01 for the number of OTUs, and r = 0.854, df = 26, P < 0.01 for Chao1 index).This indicates that the overall conclusion with respect to soil bacterial species diversity was not influenced by sequence read variation, and that tropical forests did not necessarily have higher soil bacterial species diversity than temperate forests in Japan (Tables 3  and 4a).Soil bacterial species diversity also varied according to the geographical area of the soil sample.Forest soils in Indonesia, especially Java, had the highest bacterial spe-cies diversity, while forest soils in Singapore had the lowest bacterial species diversity (Table 4b).As for the forest type, as noted above, the mangrove forest (BKOman) had exceptionally high bacterial species diversity and mixed dipteroparp forests, which may be considered natural forests with minimal human activity, had lower diversity than acacia and oil palm plantations, which are artificial forests with a single tree species (Table 4c), indicating that aboveground tree composition was not strongly related to forest soil bacterial diversity.It should be noted that there should be some caution when N. ITO et al.
inferring these relationships among soil bacterial communities from these data, since almost all samples did not actually plateau in rarefaction curves (Supplementary Fig. S5).
Occurrence of OTUs within and between investigated forests Based on the concerted evolution and conserved copy number variation of bacterial 16S rRNA cistrons, the number of 16S rRNA sequence reads per OTU may be regarded as the representative of the number of individual bacteria per species (OTU) in the forest soils investigated, although it is difficult to convert the number of sequence reads to the exact number of individual bacteria (Miyashita et al., 2013;Miyashita, 2015).For the total data, the most abundant OTU, which belonged to Acidobacteria subgroup Gp1, represented approximately 3% (7,934) of all reads (256,693), and the second most abundant OTU, which belonged to Acidobacteria subgroup Gp2, represented approximately 2% of reads (5,384) (Table 3).For individual forests, the average maximum number of reads per OTU was 4.2%; however, BER2nd had an exceptionally predominant OTU representing 13.8% of sequence reads.Besides these dominant OTUs with many sequence reads, the average number of reads per OTU was only 12.1 (0.00047%) for all 21,223 OTUs, ranging from 2.9 (0.06% in BAU2nd) to 13.0 (0.07% in TIM2nd) over the 28 forests.In addition, on average, 43.8% of OTUs in a forest were singletons, ranging from 27.5% (SEMpin, Shorea pinanga plantation in Semengo) to 59.4% (BKOman), and 50.0% of OTUs were singletons for all data.These results indicate that most OTUs (species) exist at very low density in each forest, and only a small number of OTUs reach high frequency.
No OTUs were represented in all 28 forests.Five OTUs were represented in 27 forests, and four OTUs were represented in all forests except BKOman.On the other hand, forest-specific OTUs accounted for over 70% (14,936) of all OTUs in the total data (Table 3), of which 71.0% (10,607) were singletons.The forest-specific OTUs accounted for 32.7% of OTUs in each forest on average, and, surprisingly, 94.5% (2,826) of OTUs (2,992) in BKOman were uniquely detected in BKOman.This indicated that bacterial composition at the species (OTU) level was largely differentiated by forest, especially for BKOman.

Relative divergence of soil bacterial communities in investigated forests
To investigate the differentiation of soil bacterial communities, distance matrices calculated for all pairwise comparisons of 28 forests (378 pairwise comparisons), based on the abundant Jaccard index (Chao et al., 2005) and weighted UniFrac (Lozupone et al., 2007), were used to construct UPGMA trees and PCoA plots.In the UPGMA dendrogram based on the abundant Jaccard index (Fig. 2A), the BKOman forest, which had a distinct composition at the phylum level Fig. 2. UPGMA dendrogram for the 28 forests constructing using (A) abundant Jaccard index and (B) weighted UniFrac distance.and an extremely high proportion of forest-specific OTUs, was located at the periphery of the tree.Although their external branches were the longest, two Japanese forests, Ashiu and IBUlim, formed a cluster and were included in a large cluster of all Southeast Asian forests, except BKOman.Forests in Singapore and Indonesia (two different islands, except for JASoil) also formed a distinct cluster.In general, forests adjacent to one another in Sarawak formed their own cluster, with the exception of BKOman, NIAlim and NIAmdf.These results indicate that geographic origin (i.e., physical distance) influenced the similarity of soil bacterial composition, rather than aboveground tree vegetation.However, in addition to the singularity of BKOman, the separation of NIAmdf and NIAlim, and the proximity of NIAlim to BAUlim and IBUlim, all of which had limestone soil, suggest that soil properties are important per se, in addition to geographical origin.
In the UPGMA dendrogram based on weighted UniFrac (Fig. 2B), the 28 forests were separated into three large groups.BKOman formed the outer branch, representing a solo cluster.Three limestone forests (IBUlim, BAUlim and NIAlim), the oil palm plantation in Jasinga (JASoil) and the secondary forest in Bau (BAU2nd) formed another cluster separated from the third cluster of 22 forests.The five forests in the second cluster were also in close proximity in the tree based on the abundant Jaccard index (Fig. 2A), although IBUlim and JASoil were in different clusters.The two Japanese forests (Ashiu and IBUlim) were distantly located in different clusters based on weighted UniFrac, although they were close to each other in the tree based on the abundant Jaccard index (Fig. 2A).Furthermore, BTNmdf in Indonesia was separated from the other three Indonesian forests, but was still in the third cluster.Despite these discrepancies, relative relationships were similar in the two trees based on different distance measures.In other words, geographic origin, rather than aboveground tree vegetation, also influenced the similarity, when based on weighted UniFrac, which considers DNA sequence variation (Lozupone et al., 2007).
Two PCoA plots were constructed based on the two different distance measures, and both indicated differentiation of soil bacterial communities in the investigated forests (Fig. 3).In both plots, despite large variation in the aboveground environment, forests in Malaysia (except for limestone forests) and Singapore formed a distinct cluster, and limestone forests and the mangrove forest (BKOman) were relatively isolated.The two Japanese forests were well separated.However, there were also slight differences between the plots based on the different distance measures.In the plot based on the abundant Jaccard index (Fig. 3A), five Indonesian forests were separated from the rest and located on the negative side of the 2 nd axis.On the other hand, in the plot based on weighted UniFrac (Fig. 3B), four Indonesian forests clustered with the forests in Malaysia and Singapore.In addition, in this plot BKOman was distantly located in the 2 nd quadrant.Since the PCoA based on weighted UniFrac explained a larger fraction of the total variation (PC1: 43.2%) than that based on the abundant Jaccard index (19.1%), the former plot may be a more realistic representation of relationships among soil bacterial communities in these 28 forests.Factors that influence soil bacterial composition (β-diversity) We also investigated the relationship between β-diversity (1 -abundant Jaccard index and weighted UniFrac) and direct distance between forests (Fig. 4).Pearson's correlation coefficient was positive for both β-diversity estimates (r = 0.45, P < 0.0001, for 1abundant Jaccard index and r = 0.11, P < 0.204 (NS), for weighted UniFrac), according to the Mantel test.This result indicated that a distance effect existed not only for adjacent forests, as noted in the UPGMA trees and PCoA plots above, but also for forests that are widely distributed over Southeast Asia and Japan.These positive correlations may reflect relatively lower β-diversities for pairs of forests less than 1 km apart and higher β-diversities for pairs of forests close to 10 4 km apart (between Southeast Asia and Japanese forests), especially for 1 -abundant Jaccard index.Pairs of forests separated by 1 km to 1000 km had large variations in β-diversity, which may be due to large variation in above-and belowground environmental conditions.The Mantel correlogram analysis revealed that significant autocorrelation was detected for both β-diversities in the first distance classes for the equal-distance distance class data (Fig. 5); however, for the equal-count distance class data, none of the Mantel statistics was significant (Supplementary Fig. S6).The detected significant autocorrelation is consistent with the lower β-diversities between forests located relatively close to each other (less than 1 km), indicating that the autocorrelation is limited within a short distance.However, for the equal-distance distance class data, the number of counts (pairwise distances) was very low (3 -6) in the distance classes with significant value (Supplementary Table S2), so that the significance in the permutation test might not be reliable due to low power of the test.Nevertheless, the positive correlation (Fig. 4) and autocorrelation (Fig. 5) indicated that biogeography of soil bacteria was not random and was influenced by factors, such as local adaptation and bacterial dispersal, which could reflect a distance effect.However, the exact factor(s) responsible for the distance effect and autocorrelation cannot be identified using the available data.
Table 5 summarizes the average β-diversity within and between geographic regions and forest types.The average β-diversity within a geographic region was generally smaller than between regions (Table 5a), except for two Japanese forests, in which soil properties, especially pH, are distinct.This overall pattern was confirmed by the  6a).This result was consistent with the distance effect detected above.On the other hand, it was not clear that average β-diversity within a forest type was smaller than between forest types, except for dipterocarp plantations (Table 5b).These dipterocarp plantations were located close to one another in Semengo, so the lower value might not reflect forest type.However, for both β-diversity estimates, the Relationship between α-and β-diversities and physicochemical properties in forest soil Although physicochemical properties were not measured for all the investigated forests (Supplementary Table S1), available data were used to examine the influence of the soil properties on both diversity estimates.Since Chao1 and Shannon indices were correlated as shown above, both measures of α-diversity were significantly and positively correlated with soil pH for the original and normalized data (Fig. 6, Table 7, and Supplementary Fig. S7).The forests with neutral and alkaline soil had higher α-diversity than forests with acidic soils.The other environmental factors (latitude, longitude, altitude, moisture content %, organic C content %, N content % and C/N ratio) did not show statistically significant correlation with α-diversity (Table 6).
The principal coordinate analysis (PCoA) based on  PERMANOVA test showed significant influence of forest type on soil bacterial community structure (Table 6b), even after eliminating singleton forest type from the test (Supplementary Table S3).These results indicate that forests, geographically adjacent and/or with the same forest type, generally tend to have similar belowground soil bacterial composition.Soil bacterial community structure in the tropics weighted UniFrac and the 1 -abundant Jaccard index revealed a statistically significant association between soil pH and PCoA 1 st axis value (Table 8), which explained 43.2% and 19.1%, respectively, of the variation in the original data (Fig. 3).These results indicated that soil pH strongly influenced not only the α-diversity of soil bacterial communities in individual forest soil but also the differentiation of soil bacterial communities among the investigated forest soils.A significant association was also detected between C/N ratio and soil texture (sand %) and PCoA 2 nd axis value based on weighted UniFrac and the 1 -abundant Jaccard index (Table 8), which explained 10.8% and 10.2%, respectively, of the variation in the original data (Fig. 3), although the C/N ratio was not measured for forests in Singapore and many in Sarawak.No statistically significant relationships between the other environmental factors and axis 2 value were detected.

DISCUSSION
Taxonomic composition of soil bacterial communities in 28 forests in Southeast Asia and Japan Despite differences in the aboveground vegetation and environmental conditions, soil bacterial composition at the phylum level was largely similar in the 28 forests in Southeast Asia and Japan investigated in this study (Fig. 1).The top five phyla in all forests were Acidobacteria, Proteobacteria, Verrucomicrobia, Planctomycetes and Bacteroidetes, and these phyla have been reported to be major phyla in forest soil in Australia, North America and South America (Chow et al., 2002;Hackl et al., 2004;Roesch et al., 2007).Two phyla, Acidobacteria and Proteobacteria, were dominant in all forests except BKOman.These two phyla were reported to be dominant in tropical forests in the Malay Peninsula (Tripathi et al., 2012) and Sarawak (Miyashita et al., 2013) in Malaysia.The present study shows that Acidobacteria and Proteobacte-ria are also predominant in the soil of tropical forests in Indonesia (Java and Sumatra) and Singapore in Southeast Asia.
Acidobacteria and Proteobacteria are Gram-negative bacteria represented by many species that grow and survive under many environmental conditions (Stackebrandt et al., 1988;Quaiser et al., 2003); however, Acidobacteria are found primarily in soil (Sogins et al., 2006), while Proteobacteria are found in soil (Barns et al., 1999;Zinger et al., 2011;Wang et al., 2012), fresh water (Hugenholtz et al., 1998) and seawater (Hugenholtz et al., 1998;Venter et al., 2004;Zinger et al., 2011).In BKOman, an exceptional phylum composition was observed, characterized by high representation of Proteobacteria (≈40%) and low representation of Acidobacteria (4%).This may be related to the ability of Proteobacteria to thrive in seawater, which was detected in the mangrove forest BKOman but not in the other 27 forests studied.
In addition to BKOman, in some forests, such as JASoil, BAU2nd, BAUlim, NIAlim and IBUlim, class compositions in Acidobacteria and Proteobacteria were different from those in the majority of the forests (Supplementary Figs.S3 and S4, A and B).In these six forests, the frequencies of Acidobacteria classes Gp1 and Gp2 were lower and other subgroups (Gp4, Gp5 and Gp6) were higher than in the other forests (Supplementary Fig. S4A).In BKOman, the frequency of Acidobacteria class Gp10 was distinctively high.The three limestone forest soils (BAUlim, NIAlim and IBUlim) had a relatively high frequency of Acidobacteria classes Gp5 and Gp6.On the other hand, in these six forests, except for JASoil, the frequency of Alphaproteobacteria was lower and that of Betaproteobacteria was higher than in the other forests (Supplementary Fig. S4B).Thus, the changes in class composition in Acidobacteria and Proteobacteria seemed to be linked, suggesting some kind of interaction between these dominant phyla, although there are many other possible explanations.This possibility needs to be further examined in future studies.
Besides these six forests of unusual class composition of Acidobacteria and Proteobacteria, in the other forest soils in Southeast Asia and Japan, Acidobacteria Gp1 and Gp2 accounted for approximately 90% of Acidobacteria classes, and Alphaproteobacteria was most frequent among the Proteobacteria classes.These class compositions of Acidobacteria and Proteobacteria in Asia differ from those in other regions of the world.For example, in a study of 16S rRNA sequences in 32 soil samples from diverse environments in Europe and Australia in which Acidobacteria and Proteobacteria were the dominant phyla, class composition depended on the geographic origin of the soil sample, and Gp1, Gp4 and Gp6 of Acidobacteria and Alpha-and Betaproteobacteria of Proteobacteria were the dominant classes (Newton et al., 2011).A study on Acidobacteria by T-RFLP and DGGE methods on the 16S rRNA gene for 27 grassland and 30 forest plots in Germany showed that class composition of Acidobacteria differed among plots within an aboveground vegetation and between aboveground vegetations, and that Gp6 and Gp1 were dominant in grassland and forest soils, respectively (Janssen, 2006).Furthermore, pyrosequencing and clone library analyses of Acidobacteria 16S rRNA sequences from 87 soil samples from North and South America showed that Gp1 was dominant (Naether et al., 2012).These results indicate that Gp1 and Gp6 of Acidobacteria and Alpha-and Betaproteobacteria of Proteobacteria are dominant in soils from Europe and Australia.Although Alpha-and Betaproteobacteria of Proteobacteria were also dominant in Asian forest soils in this study, the predominance of Acidobacteria Gp2 and Gp5 may be a characteristic of Asian forest soils.
Diversity of soil bacterial communities in investigated forests A cluster analysis showed that the number of OTUs and the number of species estimated (Chao1 index) were of the order of 10 3 in all forests (Tables 3  and 4), in agreement with the number of species estimated by PCR amplification of the 16S rRNA gene region (Curtis and Sloan, 2005;Gans et al., 2005).However, the actual diversity of the soil bacterial community in these samples is likely to be higher than 10 3 , because, in all cases, the rarefaction curves for these data sets did not plateau.Many OTUs were singletons (43.8% on average) with few sequence reads per OTU per forest; this is consistent with the "rare biosphere" concept described in previous analyses of bacterial communities in deep-sea water (Sogins et al., 2006), agricultural soil (Ashby et al., 2007), prairie soil (Elshahed et al., 2008) and forest soil in Southeast Asia (Miyashita et al., 2013).Soil bacterial species diversity (the number of OTUs and Chao1 index) varied among forests.However, the relationship among forests with respect to the level of bacterial species diversity was not influenced by 16S rRNA sequence read variation, as indicated by a highly significant Pearson's correlation between diversity estimates for raw and normalized data.These results suggested that factor(s) inherent to each forest affected the level of bacterial species diversity.The bacterial species diversity also varied by geographic area: Java samples had high diversity, and Singapore samples had low diversity.However, bacterial species diversity in Japanese forests was similar to that of the tropical forests.In fact, bacterial species diversity in Mt.Ibuki (IBUlim) was the third-highest value observed in this study (Table 4).Furthermore, forest type was not associated with bacterial species diversity.Thus, it can be concluded that aboveground condition did not directly influence soil bacterial species diversity in these forests.
Bacterial diversity was reported to correlate with soil pH in American forests (Fierer and Jackson, 2006;Lauber et al., 2009), such that diversity was lowest in acid soil, highest at nearly neutral pH, and then decreased with increasing pH.This suggests that extreme pH is stressful for unicellular bacteria.Although soil with pH > 8 was not detected in this study, Chao1 and Shannon indices correlated positively with soil pH (Fig. 6); the diversity increased as pH became alkaline (near pH 8) from acid (near pH 3), as reported for tropical forests in Malaysia (Tripathi et al., 2012;Miyashita et al., 2013).However, the influence of pH on soil bacterial diversity was not straightforward, since there was considerable variation in the Chao1 index among five forest soils with high pH (around 7), but slight variation in soil with lower pH (Fig. 6A).Although MARaca5 and BAUlim forest soils had almost the same pH (6.60 and 6.61, respectively, Supplementary Table S1), their Chao1 indices were 2404.5 and 6051.8,respectively.Furthermore, BAU2nd soil had a high pH (7.78), but its Chao1 index was not high (3702.4).As indicated by the positive correlation between Chao1 and Shannon indices, as pH increased, the Shannon index increased.This result suggests that as the number of species increases in a soil environment with pH, the density of species (the number of individuals per species) decreases.The effect of pH on the Shannon index was not the same as on the Chao1 index; there was little variation in the Shannon index among soils with higher pH, while considerable variation existed among soils with lower pH (Fig. 6B).This would indicate that the effect of pH on species density varied among forest soils at low pH.Taken together, these results suggest that the influence of pH differs between the two measures of α-diversity and that factor(s) other than pH influence the α-diversity of soil bacteria, although the effect of pH is the strongest.
Relationship among soil bacterial community structures in investigated forests In the UPGMA dendrogram based on the abundant Jaccard index, which took account of the similarity of OTU composition, geographically adjacent forests formed clusters, except for BKOman, NIAlim, NIAmdf and JASoil (Fig. 2A).This indicated that soil bacterial composition at the species (OTU) level was related to geographical distance, but not to the type of aboveground vegetation.In contrast, the topology of the dendrogram based on weighted Uni-Frac distance (Fig. 2B) differed, in that the 28 forests were divided into three large clusters, not necessarily in agreement with the geographical relatedness among forests.Because weighted UniFrac considers phylogenetic relationships among OTUs, the difference in the topology between the two dendrograms indicated that OTUs in geographically adjacent forests did not necessarily share similar evolutionary history in terms of sequence evolution of 16S rRNA.The PCoA analysis based on the abundant Jaccard index divided the forests roughly into four quadrants (Fig. 3A), while that based on the weighted UniFrac distance showed three distantly separated groups (Fig. 3B), corresponding with the topology of the weighted UniFrac-based UPGMA dendrogram.The difference in the scattering of both PCoA plots also reflects the fact that they are based on different distance measures.
It is worth noting that in the dendrograms based on the abundant Jaccard index or weighted UniFrac, and in the PCoA plot based on weighted UniFrac, the BKOman forest was clearly separated from the other forests.This result suggests that the structure of the bacterial community in BKOman is unique at the level of species.In particular, 94.5% of the OTUs in BKOman were unique to BKOman, in the context of this study.A mixture of fresh water and seawater irrigates the soil in BKOman, a mangrove forest, and it was reported that salinity influences the structure of bacterial communities (Lozupone and Knight, 2007).This could promote growth of salttolerant and anaerobic bacteria in BKOman soil, contributing to the unique structure of its bacterial community at the levels of phylum and species.Consistent with this idea, of the OTUs detected only in BKOman, the two most abundant were Cyanobacteria, which is also a major bacterial phylum of seawater (Gotelli and Colwell, 2001;Gans et al., 2005).However, many characteristics of BKOman other than the salinity of its soil, such as water temperature and oxygen level associated with tidal action, could contribute to the unique structure of its soil bacterial community.To examine the uniqueness of soil bacterial community structure in BKOman and the effect of salinity on the soil bacterial community structure, it would be necessary not only to study other mangrove forest soils, but also to measure salinity in other forest soils in future studies.
Environmental properties and mechanisms that influence soil bacterial species composition To investigate the association between soil bacterial community structure, especially species composition, and soil environmental properties, Pearson's correlation coefficient was examined for the 1 st and 2 nd ordination scores of PCoA (Table 8).A highly significant correlation was detected between the 1 st ordinate values and soil pH, with a less strong correlation observed for eight other physicochemical properties of soil.We conclude that soil pH strongly influences α-diversity of soil bacteria as well as composition of the soil bacterial community at the level of species, most likely because the pH of the environment has a large effect on bacterial growth and survival.However, in the two PCoA plots based on different β-diversity, three limestone forests, IBUlim, BAUlim and NIAlim, in addition to Bau2nd, lie close together in a single cluster.Although limestone usually makes soil alkaline, NIAlim had acidic soil (Supplementary Table S1), indi-cating that pH as well as other properties of soil influence soil bacterial composition.In addition to pH, a highly significant Pearson's correlation was detected between the 2 nd ordination values and C/N ratio and soil texture (sand %), indicating that these two variables also affect the composition of the soil bacterial community.Indeed, it has been reported that bacterial diversity and composition differed between Bornean rainforest soils with different C/N ratios and soil textures, suggesting the influence of the composition of carbon and nitrogen, and soil pH, in addition to oxygen availability (Russo et al., 2012).However, in this study, C/N ratio and soil texture were not associated with α-diversity, but instead were correlated with the 2 nd ordinate values of PCoA, which have secondary importance.Since C/N ratio was not measured for all the forests studied, this discussion of the C/N ratio should be taken with caution.
Another important observation of the present study was the statistically significant correlation between measures of β-diversity (1 -abundant Jaccard index) and physical distance between forests (Fig. 4), confirmed by the PERMANOVA test for the effect of geographical region on soil bacterial community structure (Table 6).This implies that the biogeography of soil bacterial composition is non-random, such that bacterial composition is influenced by local adaptation and limited dispersal (Hughes Martiny et al., 2006).In our previous studies, multiple samples were taken ca. 10 meters apart in each of six forests in Sarawak and Japan, with the result that bacterial species composition was more similar within a forest than between forests (Miyashita et al., 2013;Miyashita, 2015).These previous results are consistent with a very small spatial scale distance effect, probably reflecting local adaptation and bacterial movement within a forest.In this study, Mantel correlogram analysis revealed a significant autocorrelation in β-diversity between forests further separated (up to 28.8 km for the 1 -abundant Jaccard index and 0.55 km for weighted UniFrac, Fig. 5 and Supplementary Table S2), suggesting a relatively longer scale of distance effect.However, as mentioned above, the significance of this autocorrelation should be taken with some caution, because of the low power of the permutation test.On the other hand, the distance effect in this study is over a much longer spatial scale (up to 10,000 km), such that bacterial dispersal was probably limited, and bacteria or bacterial spores could be subject to long-distance passive dispersal by wind, rain or contact with birds (Hughes Martiny et al., 2006).In fact, some OTUs were shared between forests in Southeast Asia and Japan, and β-diversity estimates were less than the maximum value of one.Actually, in the Japanese forests Ashiu and IBUlim, 1-the percentage of forestspecific OTUs, which is the percentage of OTUs in a forest shared with other forests, was 0.575 and 0.3 (Table 3), respectively, indicating that a high percentage of OTUs in N. ITO et al.
Japanese forests were shared with the Southeast Asian forests, given that the β-diversity estimates between the two Japanese forests were relatively high (Table 5).
Although it was not clear from the average values of β-diversity (Table 5), PERMANOVA tests also indicated that aboveground forest type has a significant influence on soil bacterial species composition (β-diversity, Table 6), indicating that forests with the same forest type are likely to have similar soil bacterial composition.On the other hand, it was suggested in this study that soil bacterial species diversity (α-diversity) was not strongly influenced by aboveground forest type (Tables 3 and 4c).These results may be relevant to the finding from a study on temperate grassland sites that aboveground plant α-diversity can predict patterns of belowground soil bacterial composition (β-diversity) but not the pattern of α-diversity (Prober et al., 2015).However, since a diverse range of aboveground forest types was investigated in this study, it is reasonable to assume that many different factors other than plant α-diversity might have contributed to our results.Therefore, it would be very informative to take into account aboveground plant α-diversity for studies of soil bacterial community structure in future.
Taken together, the results of this study suggest that soil bacterial community structure (species composition in particular) was influenced significantly by soil physicochemical properties, mainly pH (and salinity), which influence bacterial growth; by distance effects, which might be related to local adaptation, evolutionary history from common ancestors and migration of bacterial species among forests; and finally by aboveground forest type.However, one caveat is that the range of soil pH examined in this study was not wide, because high alkaline soil was not included.Also, the influence of salinity was inferred from the bacterial community structure in BKOman, from which the effect of mangrove itself was not eliminated.Moreover, salinity itself was not measured in this study, and should be quantified in future studies.In addition, the underlying causes of the distance effect observed here are not clear.In future studies on the environmental, ecological and historical factors that influence soil bacterial community structure in tropical and temperate forests, it will be important to analyze forests that are more ecologically diverse and more geographically distant than those in the present study.

Table 1 .
List of sampling locations

Table 2 .
Rank correlation of taxa at higher taxonomic levels (378 pairwise comparisons) eliminate the effect of read number variation, the read number was normalized into the minimum number of reads per sample among the 28 forests (2,951), and then the number of OTUs, Chao1 index and Shannon index were recalculated for each sample (Table4).There was still a large variation in species diversity among the for-2).Even after adjusting the read number, corresponding with the positive correlation between Shannon and Chao1 indices detected for raw data, BKOman had the highest Shannon index (6.71),and TIMrid had the lowest (5.19).In fact, Pearson's correlation coefficient between diversity measures for raw and normalized data

Table 3 .
Summary of bacterial diversity and abundance of the total data

Table 4 .
Summary of α-diversity normalized to the minimum number of sequence reads (2951 reads)

Table 5 .
Average β-diversity within and between regions (a) and forest types (b)

Table 6 .
PERMANOVA results of testing the effect of geographical region and forest type on β-diversity (1 -abundant Jaccard index and weighted UniFrac) Classification of geographical region and forest type is the same as in Table5.

Table 8 .
Pearson correlation coefficients between the ordination score of PCoA and soil and environmental factors