OUP user menu

Cultivation gives context to the microbial ecologist

Dominica Nichols
DOI: http://dx.doi.org/10.1111/j.1574-6941.2007.00332.x 351-357 First published online: 1 June 2007
  • uncultivable bacteria
  • cultivation strategies
  • metagenomics
  • genome assembly
  • ‘great plate count anomaly’


The posed debate asks whether cultivation has a place in contemporary environmental microbial ecology. Approximately two decades ago, cultivation-independent methods experienced an advance (Olsen et al., 1986) concurrent with the establishment in the literature of the following sentence: ‘It is well known that 99.9% of all organisms are unculturable’ (Rothschild, 2006). While technical advancements have been made in sequence-based and cultivation-based microbial ecology, the ‘uncultivability’ of environmental microorganisms has stood as a beacon to the molecular community of the futility of cultivation. The same phenomenon has served as a welcome challenge to others, leading to the development of novel cultivation approaches and technologies. It appears now that microbial cultivation is an emerging frontier in environmental microbiology, in part because it provides information about communities that cannot be obtained directly from sequencing efforts alone (Ellis et al., 2003). As the barriers to cultivation are reduced, the complications of undertaking microbial ecology in its absence are becoming more apparent.

Cultivation-independent ecology

Sequence collection efforts may appear, at first glance, far superior to cultivation because they bypass the discrepancy between the number of microbial cells present in a sample and the number of colonies produced in vitro (Schloss & Handelsman, 2005). The promise of genomes without cultivation has made metagenomics a popular cultivation-independent approach to microbial community assessment (Handelsman, 2004; DeLong, 2005). Isolation and sequencing of microbial DNA from mixed communities en masse, rather than focusing on select phylogenetic markers, potentially provides access to the full complement of microbial genes and the opportunity to reassemble genomes without the requirement of environmental isolates (Handelsman et al., 1998). However, closing genomes from these efforts proves exceedingly difficult, apparently owing to genomic micro-heterogeneity (Tringe & Rubin, 2005), compounded by species diversity within samples (Venter et al., 2004). Even with closure, typically 40% or more of the proteins in any one genome cannot be assigned a function. Microbial cultivation, on the other hand, provides a context in which to corroborate the theoretical findings of metagenomics and much more direct access to genomes of environmental isolates.

Genome closure attempts from metagenomic sequencing efforts

Simple communities may hold the best promise to closing the genomes of environmental organisms because few genomes mean fewer assembly puzzles to resolve. However, even metagenomic projects conducted on simple communities have seen significant barriers to assembly.

A success story often provided by the metagenomics community is the work of Tyson in (2004) (Meyerdierks et al., 2005; Schloss & Handelsman, 2005; Tringe & Rubin, 2005; Tyson & Banfield, 2005; Abulencia et al., 2006; Green & Keller, 2006; Podar & Reysenbach, 2006; Wilmes & Bond, 2006). From a simple community of six bacterial and archaeal lineages found in an acid mine drainage system, they were able to obtain five genomes of various levels of completion (Tyson et al., 2004). The authors obtained two 10 × coverage genome categories by first binning scaffolds into high and low GC content categories, then separating them by coverage and comparing the total length to known genome sizes. The discovery of a full complement of tRNA synthases and rRNA genes in each of these categories further suggested that each could belong to one organism. The 3 × coverage genome categories were assigned to organism types using rRNA gene markers and gene order (Tyson et al., 2004). In total, this work reports the almost complete closure of one genome, and a somewhat less complete closure of another. The heterogeneity of scaffolds of this second genome suggests that they could be a mosaic of closely related organisms, rather than the genome of an individual. Venter et al. (2004) observed a similar phenomenon in the Prochlorococcus assemblies from the Sargasso Sea database, resulting in a comparable genus-level composite genome (Venter et al., 2004).

Although subsequent summaries of this work refer to five nearly complete genomes, three of these genomes have only 3 × coverage and all five were obtained by grouping sequences that may or not belong to single organisms. Although heavily weighted to one variant of the Leptospirillum group II based upon their 16S rRNA gene library, the acid mine drainage community appears to have further strain diversity given the 17 other variants observed (Tyson et al., 2004). Other taxonomic groups observed in the AMD biofilm showed similar micro-diversity, with variants observed with c. 1% variation from each other (Tyson et al., 2004). In other words, isolates with identical 16S rRNA genes have been observed with disparate phenotypes and even vastly disparate genomes (Welch et al., 2002; Hahn & Pockl, 2005; Davelos Baines et al., 2007). The problem with grouping strains within species, or species within higher order taxonomic categories, for the purposes of genome assembly is that they may not be similar enough to yield relevant composite sequences. Yet this ‘grouping’ approach had to be taken even in the case of a simple community (Tyson et al., 2004).

To date, the only completely closed genomes obtained by a community metagenomic effort are the Shewanella and Burkholderia strains found by Venter in the Sargasso Sea database (2004). It has been argued that these microorganisms were contaminants because their sequences hyper-dominated a few samples over the others and they appeared to be most closely related to terrestrial species, rather than marine organisms (DeLong, 2005). Irrespective of the nature of these strains, difficulties in assigning genes to organisms were experienced even by the most comprehensive sequencing effort to date (Rusch et al., 2007; Yooseph et al., 2007). In fact, only 9% of the base pairs sequenced from the reported Sorcerer II Global Ocean Sampling expedition database could be assembled into scaffolds greater than 10kbp in length (Rusch et al., 2007). This percentage drops to just 0.3% when scaffolds greater than 50kbp were considered (Rusch et al., 2007). Assembly of the genomes of natural community members thus appears a daunting task (Venter et al., 2004; Tringe & Rubin, 2005; Rusch et al., 2007).

Sequencing one cell at a time

Technologies allowing for the isolation of single microbial cells and the amplification of their DNA provide tools to circumvent some barriers to assembly intrinsic in metagenomics. Specifically, DNA extracted from a single cell reduces the complexity of the sample to one genome, thereby removing the possibility that genes from disparate genotypes be confused as belonging to one (Hutchison & Venter, 2006; Kvist et al., 2007; Zhang et al., 2006). Cells from two laboratory strains of the heterogeneous Prochlorococcus genus could be differentiated using a sequencing technology that creates polymerase clones from the DNA of a single cell. This approach is being used to sequence two wild Prochlorococcus cells from the Pacific Ocean (Zhang et al., 2006). Consequently, it is not only the traditional laboratory model organisms that can be differentiated using the single cell approach, but also a wild organism whose genome cannot be thus far assembled using metagenomics (Venter et al., 2004).

The downside of sequence analysis of one cell at a time is that it requires amplification. This complicates the process because the methodology must provide a high yield of precisely amplified sequence, genome-wide (Kvist et al., 2007). Polymerase cloning using isothermal multiple displacement amplification, as described by Zhang et al. (2006), is a robust methodology that appears to provide just that. However, the entire genome is not likely to be amplified (Raghunathan et al., 2005; Kvist et al., 2007) and the un- and under-represented regions appear to have a random distribution (Zhang et al., 2006). This concern is not unique to single cell sequencing. Genomic sequencing of isolates has similar complications, albeit to a lesser extent. Even without amplification, metagenomics is also likely to miss sections of its target organisms (Furrie, 2006). However, the reduced complexity intrinsic in single cell sequencing more than compensates for possible bias when multiple displacement amplification is used for primary sequencing and PCR-based methods are only used to complete unrepresented regions after the fact (Zhang et al., 2006).

Hypothetical microbial ecology

In the coming years, genome assembly algorithms will undoubtedly continue to advance, genome sequencing will become less expensive, and the number of assembled microbial genomes will grow significantly. This will likely facilitate the assembly of individual genomes in community sequencing projects. In the interim, metagenomic data can be mined extensively in the absence of complete genome assembly by comparing the data against pre-existing DNA and protein sequence databases (Abulencia et al., 2006).

Even for the most similar genes we often do not know if their functional role is the same as predicted by similar genes in the database. A prime example of this is the gene encoding proteorhodopsin in SAR11. Proteorhodopsins are light-driven proton pumps originally discovered in the surface waters of the ocean by cloning and sequencing large DNA fragments (Beja et al., 2001; DeLong & Karl, 2005). It has been estimated that each bacterial cell maintains c. 25000 copies of this protein (Beja et al., 2001). The gene is present and expressed in the SAR11 isolate Pelagibacter ubique obtained by Giovannoni et al. (2005). Similar proteorhodopsin proteins were isolated directly from the ocean, suggesting that natural populations of this organism also express these genes. Finally, the proteorhodopsin gene in P. ubique also functions as a light-driven proton pump in Escherichia coli. Surprisingly, it does not confer enhanced growth to P. ubique in the presence of light or retinal under laboratory conditions (Giovannoni et al., 2005). The bacterium maintains a large number of these molecules. This suggests that they must play a role; one that is not apparent from the sequence information alone. Possible alternative functions for the gene product are as a signal receptor (Sudo & Spudich, 2006) or to provide a proton motive force under ultra oligotrophic conditions (Giovannoni et al., 2005). Clearly, the precise function of proteorhodopsin in P. ubique remains unknown.

Empirical microbial ecology

As exemplified by SAR11, cultivation of environmental bacteria, when successful, gives microbial ecologists a context in which to investigate theoretical molecular findings empirically. One elegant example is the work of Tyson et al. (2005) done to follow-up their previous metagenomic sequencing of an acid mine drainage (AMD) biofilm (Tyson et al., 2004). They found members of Leptospirillum group II and III in biofilms; a genus with a previously cultivated representative capable of nitrogen fixation. No other relatives of known nitrogen-fixing acidophiles were identified in the AMD metagenome that they sequenced (Tyson et al., 2005). Other AMD sequencing efforts have not found any acidophiles known to fix nitrogen in communities at similar pH (Tyson et al., 2005). It is noteworthy that they identified a nif operon within a genomic fragment of Leptospirillum group III. Since this was the only nif operon found in the AMD metagenome, and it is not found in the near-complete genome of Leptospirillum group II, group III representatives may be responsible for nitrogen fixation. Selective media designed with this in mind allowed the Banfield group to cultivate and isolate the novel species Leptospirillum ferrodiazotrophum, a member of group III (Tyson et al., 2005). This approach has been termed reverse metagenomics, as it uses information from metagenomics to obtain cultures of ecologically relevant microorganisms (Podar & Reysenbach, 2006).

The work of Parro & Moreno-Paz (2003) further emphasizes how direct access to isolates can be used to accumulate accurate annotation of genes involved in a regulon of interest. They printed a random genomic library from the bacterium Leptospirillum ferrooxidans upon a micro-array which was then challenged with RNA extracted from the bacterium grown in the presence and absence of the nitrogen source ammonium. Using this strategy, most of the bacterial genes with known involvement in nitrogen fixation were identified by sequencing relatively few clones from the array because of the extent of fragment overlap from the shotgun library with which the microarray was made. Many genes with putative nitrogen fixation functionality were also identified. This appears robust for testing theoretical findings from a metagenome using environmental isolates (Parro & Moreno-Paz, 2003).

Microbial cultivation: problems and solutions

Many microorganisms have proved difficult to cultivate, let alone isolate. In most cases, we cannot search the genome for the answer to how to cultivate a fastidious organism of interest because the majority of genes are of unknown function. Two additional barriers to cultivation are the ‘great plate count anomaly’ and the fact that isolates might display unnatural behaviors in a synthetic environment. Both are surmountable obstacles.

The ‘great plate count anomaly’ and its challengers

The ‘great plate count anomaly’, a term coined in 1985 by Staley & Konopka, put a name to the frequently reported observation of a vast discrepancy between the number of cells in a sample and the number of colonies they produce on traditional media (Staley & Konopka, 1985). The need for in vitro access to the uncultivated microbial majority is overwhelming in both basic and applied microbiology and microbial ecology (Young, 1997; Fry, 2000).

Over the past decade, there have been significant efforts to remove the restrictions inherent in cultivation using the standard methodologies. Amongst others, the groups of Button, Epstein and Lewis, Giovannoni, Janssen, Keller, Overmann, Stevenson, and Hahn have each made contributions to these efforts (Button et al., 1993; Janssen et al., 1997; Bruns et al., 2002, 2003; Connon & Giovannoni, 2002; Kaeberlein et al., 2002; Sait et al., 2002; Zengler et al., 2002; Joseph et al., 2003; Hahn et al., 2004; Stevenson et al., 2004). Each approach attempts to match the in vivo and in vitro conditions but approach this goal in various ways. The more recent techniques combine high throughput cultivation with simulation of natural growth conditions including use of natural media (i.e. seawater) or synthetic media with low concentrations of nutrients, mirroring the oligotrophic conditions of their samples (Bruns et al., 2002; Connon & Giovannoni, 2002; Cho & Giovannoni, 2004). Other methods include extinction and dilution culturing (Button et al., 1993), the extension of incubation times (Stevenson et al., 2004; Davis et al., 2005), the addition of cell-to-cell signaling compounds to culture media (Bruns et al., 2002), and employing diffusion via devices such as the environmental growth chamber (Kaeberlein et al., 2002). In this latter methodology, microorganisms are incubated in chemical contact with their source environment. Other methods have followed suit, using dialysis membranes to deliver environmental growth factors to microorganisms during incubation (Ferrari et al., 2005). Collectively, these studies have yielded an increase in microbial in vitro recovery over standard cultivation approaches, in terms of both percent and diversity. Large isolate libraries have also been obtained, including organisms of high value to marine and terrestrial ecosystems, using these new cultivation methodologies (Connon & Giovannoni, 2002; Zengler et al., 2005).

The recent advancements in microbial cultivation technology have not completely eradicated the uncultivated majority. However, high throughput cultivation approaches hold immense promise at a relatively low cost for those organisms capable of in vitro growth. For example, Zengler et al. (2002) describe their high throughput encapsulation approach to cultivation as capable of extracting ‘more than 10000 bacterial and fungal isolates per sample’ from soil (2005). High throughput cultivation approaches developed for marine (Connon & Giovannoni, 2002) and aquatic (Bruns et al., 2002) water column bacteria have likewise contributed a large number of ecologically important isolates, including members of previously uncultivated clades (SAR11, SAR86 and others). High throughput in situ cultivation methodologies under development (Schering et al., unpublished data) will likely yield additional novelty at low cost and with comparable efficiency. Other recently developed cultivation approaches continue to inform the development of high throughput approaches and have likewise provided ecologically relevant isolates. Together, these methodologies support the contention that the barriers of the ‘great plate count anomaly’ are not insurmountable.

The issue with isolates

The secondary issue of the incongruence between laboratory and natural conditions often begs the question, ‘Am I observing an ecologically relevant phenomenon in my isolate?’ This concern is of greater long-term impact, given that microbial cultivability appears to be simply an issue of technology development with pending solutions. Laboratory conditions are often a poor environmental mimic. The physical, chemical and biological complexities of the environment likely all play a role in the resultant behavior of a microbe, including commitment to division and metabolite production. Evidence of this can be seen under laboratory conditions. Temperature, oxygen level, pH, salinity and growth matrices, as well as many other factors, are integral to the growth regimens employed for cultivable microorganisms (Kopke et al., 2005). Selective media can enhance the growth of microorganisms with specific metabolisms or antibiotic resistance. Enzyme or antibiotic production can be induced in one bacterium in the presence of another (Maldonado et al., 2004; Xavier & Bassler, 2005). Some otherwise ‘uncultivable’ bacteria form colonies in the presence of another bacterium (Kaeberlein et al., 2002). The concern of ecological relevance is unpalatable when these in vitro observations are considered.

Two simple solutions to this quandary come to mind: to either utilize the limits of cultivation to one's advantage, or to bypass in vitro cultivation conditions completely by growing and monitoring organisms in situ.

Ellis et al. (2003) suggested that cultivation might be a better environmental assessment tool than molecular signatures obtained directly from nature because the percent colony recovery may be a better measure of the physiological health of environmental microorganisms (Ellis et al., 2003). The 16S rRNA gene approach used in that study showed similar community profiles in both uncontaminated and the most contaminated samples. Denaturing gradient gel electrophoresis (DGGE) profiles of amplified environment DNA also showed little difference between samples. In contrast, when samples of various contamination levels were cultivated, the grown material yielded a diversity of DGGE profiles. The proportion of culturable bacteria was also significantly less in samples with the highest heavy metal content. Overall, that study described how the response of environmental microorganisms to growth in laboratory media could be used to differentiate between contaminated and uncontaminated soils. This highlights the usefulness of cultivation-based assessments and ‘suggest[s] that the proportion of microscopically counted bacteria that form colonies on laboratory media is an ecologically relevant parameter’ (Ellis et al., 2003).

Although a possible indicator of overall community health, the behavior of specific microorganisms differs under in vitro and in vivo conditions. In some cases, cultivation in situ could be the best solution. This technology is available (Kaeberlein et al., 2002), and has been adapted to soil, wastewater and both freshwater and marine water column and sediment microorganisms (Schering et al., 2006). One possible concern with in situ cultivation is the ease of monitoring during experiments. Improvements in microscopy and related probing techniques, combined with more accurate and adaptable micro-sensors, will be of great use to microbial ecologists working outside a synthetic laboratory setting. These advances will make it possible to perform ecological studies in nature and relate findings to observations made under in vitro conditions, a luxury that plant and animal ecologists often take for granted.

Concluding remarks

Both cultivation-independent and cultivation-dependent approaches to microbial community assessment are in need of improvement, and should be employed synergistically because their findings are complementary. The question is: ‘which approach is the better starting place?’ In my mind, the answer is cultivation. This is primarily because environmental genomics data rely heavily upon sequenced genomes of cultivated species. As a starting place, cultivation has a significant advantage over metagenomics because determining whether two genes came from one organism is a nonissue. Although single-cell sequencing technology shares this quality, it does not provide information on the physiological status of the organism in question. Furthermore, sequence data rarely inform a successful cultivation strategy. The reverse is not true – sequencing a cultured microbe is requiring less and less time and fewer resources (Shendure et al., 2005). Although the majority of microorganisms have still to be cultured, recent cultivation advancements suggest this is not an intractable problem. Novel cultivation-dependent approaches, if married to genomics, will likely provide the best ecological context with which to interpret the microbial world.


I would like to thank Slava Epstein, Kim Lewis and Annette Bollmann for helpful discussions and critical reading of drafts of this manuscript.


  • Editor: Jim Prosser


View Abstract