Extensive Phylogenetic Analysis of a Soil Bacterial Community Illustrates Extreme Taxon Evenness and the Effects of Amplicon Length,Degree of Coverage,and DNA Fractionation on Classification and Ecological Parameters

To thoroughly investigate the bacterial community diversity present in a single composite sample from an agricultural soil and to examine potential biases resulting from data acquisition and analytical approaches, we examined the effects of percent G+C DNA fractionation, sequence length, and degree of coverage of bacterial diversity on several commonly used ecological parameters (species estimation, diversity indices, and evenness). We also examined variation in phylogenetic placement based on multiple commonly used approaches (ARB alignments and multiple RDP tools). The results demonstrate that this soil bacterial community is highly diverse, with 1,714 operational taxonomic units demonstrated and 3,555 estimated (based on the Chao1 richness estimation) at 97% sequence similarity using the 16S rRNA gene. The results also demonstrate a fundamental lack of dominance (i.e., a high degree of evenness), with 82% of phylotypes being encountered three times or less. The data also indicate that generally accepted cutoff values for phylum-level taxonomic classification might not be as applicable or as general as previously assumed and that such values likely vary between prokaryotic phyla or groups.Efforts to describe bacterial species richness and diversity have long been hampered by the inability to cultivate the vast majority of bacteria from natural environments. New methods to study bacterial diversity have been developed in the last two decades (32), many of which rely on PCR-based procedures and phylogenetic comparison of 16S rRNA gene sequences. However, PCR using complex mixtures of templates (as in the case of total microbial community DNA) is presumed to preferentially amplify certain templates in the mixture (23) based on their primary sequence, percent G+C (hereafter GC) content, or other factors, resulting in so-called PCR bias. Moreover, the amplification of template sequences depends on their initial concentration and tends to skew detection toward the most abundant members of the community (23). To further complicate matters, subsequent random cloning steps on amplicon mixtures are destined to result in the detection of numerically dominant sequences, especially where relative abundance can vary over orders of magnitude. Indeed, any analysis based on random encounter is destined to primarily detect numerically dominant populations. This is especially of concern where limited sampling is performed on highly complex microbial communities exhibiting mostly even distribution of populations with only a few showing any degree of dominance, as typically perceived for soils (17). These artifacts and sampling limitations represent major hurdles in bacterial community diversity analysis, since the vast majority of bacterial diversity probably lies in “underrepresented minority” populations (24, 30). This is important because taxa that are present only in low abundance may still perform important ecosystem functions (e.g., ammonia-oxidizing bacteria). Of special concern is that biases in detection might invalidate hypothesis testing on complex communities where limited sampling is performed (5).Recently, there has been a concerted effort toward addressing problems impeding comprehensive bacterial diversity studies (7, 13, 24, 26, 28). In recent years, studies have increased sequencing efforts, with targeted 16S rRNA gene sequence libraries approaching 2,000 clones (11) and high-throughput DNA-sequencing efforts (e.g., via 454 pyrosequencing and newer-generation high-throughput approaches) of up to 149,000 templates from one or a few samples (25, 30). These technological advances have come as researchers recognize that massive sequencing efforts are required to accurately assess the diversity of populations that comprise complex microbial communities (29, 30). Alternatively, where fully aligned sequence comparisons need to be made, novel experimental strategies that allow more-comprehensive detection of underrepresented bacterial taxa can be applied. One such approach involves the application of prefractionation of total bacterial community genomic DNA based on its GC content (hereafter GC fractionation) prior to subsequent molecular manipulations of total community DNA (14). This strategy has been successfully applied in combination with denaturing gradient gel electrophoresis (13) and 16S rRNA gene cloning (2, 21) to study microbial communities. This approach separates community genomic DNA, prior to any PCR, into fractions of similar percent GC content, effectively reducing the overall complexity of the total community DNA mixture by physical separation into multiple fractions. This facilitates PCR amplification, cloning, and detection of sequences in fractions with relatively low abundance in the community, thereby enhancing the detection of minority populations (13). Collectively, this strategy reduces the biases introduced by PCR amplification and random cloning of the extremely complex mixtures of templates of different GC content, primary sequence, and relative abundance present in total environmental genomic DNA.Any large molecular survey that relies on sequencing further requires the analysis of large amounts of data that must be catalogued into phylogenetically relevant groups. This is usually done using high-throughput methods like RDP Classifier or Sequence Match (6) or a tree-based method like Greengenes (8) or ARB (18). Two major pitfalls that are encountered using these former approaches are the presence of huge numbers of unclassified sequences in databases and the lack of representative sequences from all phyla. This leads to most surveys having large portions of their phylotypes designated as unclassified. The latter tree-based approaches, although better suited for classification schemes, are also dependent on having a comprehensive database with well-classified sequences for reproducible results. This reproducibility becomes especially important when trying to compare data across different studies, especially those that utilize different approaches and study systems.In the current study, we analyzed an extensive (∼5,000 clones) partial 16S rRNA gene library from a single soil sample that was generated using very general primers and GC-fractionated DNA. Total DNA was extracted from soil at a cultivated treatment plot at the National Science Foundation Long Term Ecological Research (NSF-LTER) site at the Kellogg Biological Station (KBS) in mid-Michigan (http://www.kbs.msu.edu/lter). To test the effect of GC fractionation on recovery of 16S rRNA gene sequences, we conducted a direct comparison with a nonfractionated library generated from the same soil sample. Using the GC-fractionated library, we also calculated several measures of bacterial diversity and examined the effects of sampling size and sequence length on Shannon-Weaver diversity index, Simpson''s reciprocal index (1/D, where D is the probability that two randomly selected individuals from a sample belong to the same species), evenness, and Chao1 richness estimation. The results show that GC fractionation is a powerful tool to help mitigate limitations of random PCR- and cloning-based analyses of total microbial community diversity, resulting in the recovery of underrepresented taxa and, in turn, reducing the sampling size needed for accurate estimations of bacterial richness. The results also provided evidence for the need to expand the typical scale of sequence-based survey efforts, particularly in environments where evenness abounds or where minority bacterial populations may have important effects on community function and processes. We suggest that there is a need for the establishment of standardized approaches for the analysis of sequence data from community diversity studies in order to maximize data comparisons across independent studies and show examples of software programs developed to facilitate comparative analysis of large sequence datasets.