A census-based estimate of Earth's bacterial and archaeal diversity

Abstract

The global diversity of Bacteria and Archaea, the most ancient and most widespread forms of life on Earth, is a subject of intense controversy. This controversy stems largely from the fact that existing estimates are entirely based on theoretical models or extrapolations from small and biased data sets. Here, in an attempt to census the bulk of Earth's bacterial and archaeal ("prokaryotic") clades and to estimate their overall global richness, we analyzed over 1.7 billion 16S ribosomal RNA amplicon sequences in the V4 hypervariable region obtained from 492 studies worldwide, covering a multitude of environments and using multiple alternative primers. From this data set, we recovered 739,880 prokaryotic operational taxonomic units (OTUs, 16S-V4 gene clusters at 97% similarity), a commonly used measure of microbial richness. Using several statistical approaches, we estimate that there exist globally about 0.8-1.6 million prokaryotic OTUs, of which we recovered somewhere between 47%-96%, representing >99.98% of prokaryotic cells. Consistent with this conclusion, our data set independently "recaptured" 91%-93% of 16S sequences from multiple previous global surveys, including PCR-independent metagenomic surveys. The distribution of relative OTU abundances is consistent with a log-normal model commonly observed in larger organisms; the total number of OTUs predicted by this model is also consistent with our global richness estimates. By combining our estimates with the ratio of full-length versus partial-length (V4) sequence diversity in the SILVA sequence database, we further estimate that there exist about 2.2-4.3 million full-length OTUs worldwide. When restricting our analysis to the Americas, while controlling for the number of studies, we obtain similar richness estimates as for the global data set, suggesting that most OTUs are globally distributed. Qualitatively similar results are also obtained for other 16S similarity thresholds (90%, 95%, and 99%). Our estimates constrain the extent of a poorly quantified rare microbial biosphere and refute recent predictions that there exist trillions of prokaryotic OTUs.

Fig 1. Estimating global prokaryotic OTU richness.

(A, B) Accumulation curves showing the number of bacterial (A) and archaeal (B) OTUs discovered, depending on the number of distinct studies included. Curves are averaged over 100 random subsamplings, and whiskers show corresponding standard deviations. Continuous curves were calculated using all studies (worldwide), while blue dashed curves were calculated using solely studies performed in the Americas or near American coasts. (C, D) Global OTU richness of Bacteria (C) and Archaea (D), estimated using the iChao2, iChao2split, ICE, CatchAll, breakaway, and tWLRM estimators. The number of OTUs discovered by the GPC is included for comparison (last bar). Whiskers indicate standard errors, estimated from the underlying models; most standard errors are likely underestimated by the models, so the variability between models is probably a more honest assessment of uncertainty. (E, F) Illustration of two methods used to estimate global bacterial OTU richness (dashed circle). (E) The iChao2split richness estimator is based on the numbers of OTUs discovered once, twice, thrice, or four times when studies are randomly split into four complementary "sampling units" (shaded circles). Average estimates were obtained by repeating the random split multiple times. (F) Based on the fraction of bacterial nonredundant (NR99) sequences in SILVA (right shaded circle) that could be matched to the GPC (left shaded circle), we estimated the fraction of global bacterial OTU richness represented in the GPC and, given the total number of bacterial OTUs in the GPC, the total number of extant bacterial OTUs. For analogous results at 99% clustering similarity, see S4 Fig. GPC, Global Prokaryotic Census; ICE, incidence coverage-based estimator; NR, nonredundant; OTU, operational taxonomic unit; tWLRM, transformed weighted linear regression model.

Fig 2. Mean relative OTU abundances and OTU distances to SILVA.

(A) Frequency histogram of MRAs of prokaryotic OTUs discovered by the GPC (grey continuous line), of OTUs discovered by the rGPC (grey dashed lines), and of all extant OTUs as estimated using a probabilistic model of OTU discovery (blue continuous curve). The probabilistic model was fitted separately for each MRA interval by comparing the discovery rates of the GPC and the rGPC. The blue dashed curve shows a log-normal distribution model [11] fitted to the estimated MRA distribution of extant OTUs. For analogous results at other phylogenetic resolutions (99%, 95%, and 90% similarity), see S3 Fig. (B) Frequency histogram of the phylogenetic distances of GPC's prokaryotic OTUs (97% similarity in the 16S-V4 region) to SILVA (release 132, set NR99). The distance of an OTU to SILVA was defined as the minimum difference (fraction of nucleotide mismatches over the aligned 16S region) of the OTU to any entry in SILVA. Observe how almost all OTUs have a distance to SILVA below 20%. GPC, Global Prokaryotic Census; MRA, mean relative abundance; NR, nonredundant; OTU, operational taxonomic unit; rGPC, rarefied variant of the GPC.

Fig 3. Richness of phyla and coverage by SILVA and the GPC.

(A) Estimated number of OTUs (97% similarity in the 16S-V4 region) globally, within various prokaryotic phyla. Estimated based on the coverage of SILVA by the GPC (subfigure C) and the number of OTUs in the GPC. Only phyla including at least 10 entries in SILVA (release 132, set NR99) [14] and estimated to contain at least 10 extant OTUs are considered. Phyla are sorted in decreasing estimated OTU richness; only the 25 richest phyla are shown. (B) Fraction of GPC OTUs that could be mapped to SILVA NR99 at similarity ≥97%, as a proxy for global OTU richness covered by SILVA, within the same phyla as in A. (C) Fraction of SILVA NR99 sequences that could be mapped to the GPC at similarity ≥97%, as a proxy for global OTU richness covered by the GPC, within the same phyla as in A. For additional phyla not shown here, see S6 Fig. For analogous results at other phylogenetic resolutions, see S7 Fig (99%), S8 Fig (95%), and S9 Fig (90%). For analogous results at the class level, see S10 Fig (99%), S11 Fig (97%), and S12 Fig (95%). GPC, Global Prokaryotic Census; NR, nonredundant; OTU, operational taxonomic unit.