With readily applicable hybridization assays, mainly based on rRNA-targeted nucleic acid probes, and direct, cultivation-independent sequence retrieval, microbiologists can for the first time determine the true composition of microbial communities. Phylogenetic identification and exact spatiotemporal quantification of microorganisms will in the future become prerequisites for high quality studies in microbial ecology just as good taxonomy and solid quantification have always been for macroecology. This review is intended to give a short history of the development of rRNA-targeted nucleic acid probes and probe technologies, as well as of their application in microbial ecology. The current state of the art is described, and we will try to look into the future. Over the last decade, rRNA-targeted probes have become a handy tool for microbial ecologists. In order to speed up the transformation of microbial ecology from a mostly descriptive to a hypothesis-driven, experimental science more intense use must be made of the taxonomic precision and quantitativeness of rRNA-targeted probes.
Do we need to describe microbial community structure in a taxonomically precise and quantitative way?
Yes, of course! Ecology is the science that studies the interactions of organisms with their biotic and abiotic environment. Those interactions determine not only the occurrence and distribution of microbial species but also their activities. In macroecology, nobody would lump all vertebrates in one category. For studying a specific interaction such as the effects of predation on earthworms, the abundance of moles is the topic of interest and not the abundance of vertebrates. Through lack of suitable monitoring methods, lumping of discrete bacterial populations happened and still happens much too often in microbial ecology despite the tremendous functional differences among microorganisms. Many studies have worked with very broad categories like photoautotrophic microorganisms (=‘algae’, including the blue green ones) vs. heterotrophic prokaryotes (=‘bacteria’). If we want to get beyond descriptive knowledge such as ‘one milliliter of sea water contains about one million bacteria’, and reach an improved functional understanding such as ‘the summer bloom of the alga Phaeocystis globosa increases the availability of substrate X and, as a direct consequence, causes an almost contemporaneous bloom of bacterium Yz which is an important degrader of X’, we have to learn how to monitor microbial species at high spatiotemporal resolution. This requires reliable identification by objective methods so that results can be compared between laboratories. Furthermore, the methods need to be quantitative, not semiquantitative or qualitative. For example, protozoal predation will usually not eliminate a certain bacterial species, but rather reduce it to lower density, from which it will recover upon, e.g. a seasonal increase of a defined substrate.
Why should nucleic acid probes be used for describing microbial community composition?
Simply, because nucleic acid probes currently allow the taxonomically most precise and quantitative description of microbial community structures. Traditional cultivation-based methods are laborious and slow and, due to their inevitable selectivity, usually fail to achieve even a half complete picture of the true composition of complex communities. There are numerous examples. For example, in many oligotrophic ecosystems the number of colony forming units is 2 to 4 orders of magnitude below the microscopic cell number (for review see ). Often the important microbial populations seem to be difficult to cultivate. A nice example for the biased view that can result from cultivation even in more easily studied eutrophic systems has been reported for an activated sludge system treating municipal wastewater . Three different media were used on the identical sample and yielded three different results, of which none matched the composition determined by molecular methods.
Compared to cultivation-based methods for describing community structure, hybridization with nucleic acid probes is faster and more reliable. There are two basic techniques. In quantitative slot- or dot-blot hybridization, total nucleic acids, DNA or RNA or both, are isolated from the sample and subsequently hybridized with nucleic acid probes [3–5]. Alternatively, whole fixed cells are directly enumerated by in situ hybridization [6, 7]. Both assays have the potential to be fully quantitative and they also work for uncultured microorganisms. Nevertheless, like any other method they need some expertise to assure both specificity and quantitativeness.
In recent years, there have also been attempts to use PCR-based techniques for community composition analysis [8, 9]. In the simplest case, a certain gene is amplified from environmental DNA using conserved primers, the different amplification products are separated by cloning, and a certain number of clones are sequenced. The frequency of a certain clone type is then taken as a measure of the frequency of the respective organism. This assumes that all organisms have the same number of gene copies; that all the different genes are equally well amplified despite differences within the primer binding sites and the length, primary structure and GC content of the amplified region; and that the cloning step is not selective. Another source of erroneous conclusions might be (residual) rRNA in the nucleic acid preparations used for PCR in combination with the reverse transcriptase activity of (many) thermostable polymerases. Finally, even if none of these parameters is a problem, the number of genome copies per cell may vary depending on the growth phase of the organism.
Quantitative PCR assays, recently developed for certain enterobacteria, pseudomonads, and staphylococci , were used to evaluate the power and reliability of quantitative PCR [11–15] as well as to study the correlation of the amount of target DNA, colony forming units, and growth phase of the respective reference organisms. It was shown that the quantitative approach can be applied even in complex samples to obtain precise and reproducible estimates of the number of target genes. A clear correlation between the number of colony forming units and the amount of target nucleic acid was demonstrated. However, depending on growth phase and species, the colony numbers assigned to a certain amount of target gene varied by one order of magnitude. Therefore, even though they already allow quite accurate measurements of gene copy numbers, PCR-based techniques provide only semiquantitative estimates of cell numbers.
Considering the many experimental steps and potential biases, clone frequencies should not be used at all to speculate about cell frequencies. Although it is en vogue to perform several rounds of PCR and cloning followed by statistical analysis of the clone frequencies , and to conclude full coverage of sample biodiversity from the statistics, this does not necessarily reflect reality. However, there are methods available to reduce problems of discrimination by applying multiple primers for in vitro amplification of rDNA, specific immobilization of cells via polynucleotide probes  or rDNA by subtraction hybridization  before cloning.
What is so special about ribosomal RNA (rRNA)?
The composition of complex microbial communities is most often analyzed by rRNA-targeted nucleic acid probes. There are many good reasons for targeting probes to 16S/18S rRNA of the small subunit of the ribosome (SSU rRNA) or to the 23S/28S rRNA of the large subunit of the ribosome (LSU rRNA). Many of these have been outlined before : the large amount of rRNA in most cells; the apparent lack of lateral gene transfer; and a good length of about 1500 and 3000 nucleotides for 16S and 23S, respectively, with a range of very conserved to quite variable sites. The foremost advantage, however, is the availability of huge rRNA databases [19–21] for comparative sequence analysis. At the time we are writing this review, the ARB database contains 22 000 aligned SSU rRNAs and 500 LSU rRNAs. With the ARB software package, rRNA-targeted oligonucleotide probes can be designed in a straightforward fashion. Specific organisms (database entries) or (phylogenetic) groups can be selected by mouse click or search tools. Parameters such as probe length, G+C content, and target region can be defined and the ARB probe design tool will then search for potential target sites against the background of the full sequence data set. The output shows a ranking of these sites according to a set of optimality criteria. These concern the number, position, and quality of discriminatory positions  within the proposed probe target sites. Fig. 1 demonstrates the design of a probe in ARB from the definition of the target organism in the rRNA-based tree to the ordered output of potential probe target sites. The rapidly growing rRNA databases are also essential for in silico probe specificity checks. For practical reasons, experimental evaluation of probe specificity is usually only possible for a limited selection of non-target reference organisms or sequences. Therefore, in silico probe evaluation is of central importance. The ARB probe match function provides a listing of sites on both strands of the rDNA which share sequence similarity with the target site under evaluation. These sites can also be visualized in the ARB primary structure editor by highlighting. The reliability of such an evaluation depends on a data set with the broadest possible phylogenetic diversity. This is important because usually only one to a few diagnostic base changes make a stretch of 15 to 20 nucleotides a probe target site. Especially in the case of species- and genus-specific probes, these diagnostic positions are often among the most evolutionary variable in the database. Therefore, identical target sites may occur in phylogenetically distant organisms as a consequence of plesiomorphy (false identity resulting from multiple base changes during the course of evolution). Consequently, any probe sequence has to be routinely checked for specificity with the growing database. Ten years ago rRNA-targeted oligonucleotide probes were developed based on data sets of only several hundred rRNA sequences. Old probes should not be used without checking!
What has been done in microbial ecology with rRNA-targeted probes in the last decade?
In 1993, when the Deutsche Forschungsgemeinschaft (DFG) initiated the special research program entitled ‘Structure/function analysis of natural microbial communities’, rRNA-targeted nucleic acid probes were mostly still a specialized tool for molecular biologists. This has changed over the last six years. The hybridization techniques enabled many interesting findings and quickly became an important driving force for the special research program.
The applications of rRNA-targeted nucleic acid probes in microbial ecology have been too numerous to even briefly review all of them. We will instead look at some highlights in three categories: (i) method development, (ii) in situ identification of as yet uncultured bacteria, and (iii) composition of complex microbial communities.
Most rRNA-targeted nucleic acid probes used today are oligonucleotides. The question of whether poly- or oligonucleotide probes are more suitable was however unresolved in the early 1990s and still is a very basic one. It recently gained new attention when polynucleotide probes were used for the visualization and enumeration of marine planktonic archaea and bacteria by FISH . DeLong and co-workers acknowledged in their report that their almost full length 16S and 23S rRNA probes lack the specificity to discriminate much below the level of Crenarchaeota vs. Euryarchaeota vs. Bacteria. Polynucleotide probes are, however, not necessarily so broad. Already in the mid 1980s a polynucleotide DNA probe targeting an evolutionarily variable region of 23S rRNA was used to identify members of the Pseudomonas fluorescens group . This method was subsequently improved by designing a conserved primer set flanking the variable region in domain III of 23S rRNA, with which a 150–250 bp template for the in vitro transcription of a single-stranded RNA probe could be rapidly generated . Under stringent hybridization conditions, the specificity of such polynucleotide probes can be at or even below the genus level. These probes were adapted to FISH . The integration of multiple fluorescently labeled nucleotides during in vitro transcription results in probes of higher sensitivity. It could indeed be shown that the signals improve up to 26-fold as compared to mono-labeled oligonucleotides. However, this was true only under relatively relaxed hybridization conditions. At the high stringencies required for intragenus discrimination, the probe signals were much lower. Polynucleotide probes have their niche where the demands for specificity are not too high and detection is sensitivity limited. They have a special application in cell enrichment as mentioned above . Furthermore, for statistical reasons the significance of identification achieved with a polynucleotide probe is higher than that of an oligonucleotide probe.
There are, nevertheless, several good reasons why oligonucleotides became the common hybridization probes. Compared to the enzymatic production and simultaneous labeling of RNA transcripts the solid phase synthesis of a mono-labeled oligonucleotide is much easier and cheaper and results in a defined product. Furthermore, large 16S rRNA databases and software packages like ARB facilitate the rapid rational design and in silico specificity profiling of oligonucleotides. Much, but not all, of the specificity evaluation can be done at the computer, e.g. by identifying characteristic mismatches in non-target sequences that need to be discriminated. The specificity of a polynucleotide at a given stringency must be evaluated empirically which means that large, representative strain collections must be screened.
Within the DFG special research program there have been additional attempts to improve the sensitivity of FISH. For its brightness and relative photostability the fluorochrome Cy3 quickly became the label of choice for oligonucleotides and enabled much higher FISH detection rates in environmental samples . Horseradish peroxidase-labeled oligonucleotides were combined with fluorescent tyramide substrates  and yielded signals strong enough to identify even strongly autofluorescent cyanobacteria . The probe permeability of the cell periphery and the in situ accessibility of specific target sites on the rRNA were identified as additional problems for FISH. The large differences in the latter were addressed for the 16S rRNA of Escherichia coli with a set of more than 200 fluorescein-labeled oligonucleotides . Permeabilization of Gram-positives remains a problem, although several protocols have been described (e.g. [30, 31]).
In situ identification of as yet uncultured bacteria
rRNA sequences can be retrieved directly from the environment without prior cultivation of the organism of interest (e.g. [1, 32–34]). Over the last decade, FISH with rRNA-targeted oligonucleotide probes has often been used for the visualization and quantification of the bacterium behind an rRNA sequence. Initial applications focused on magnetotactic bacteria and bacterial symbionts of protozoa (for review see ), and more recent ones on two well-known bacteria with conspicuous morphologies, Nevskia ramosa and Achromatium oxaliferum. N. ramosa is a neuston bacterium that forms typical, dichotomically branching rosettes on the surface of shallow freshwater habitats. From an enrichment three different 16S rRNA sequences were retrieved of which one, affiliating deep in the gamma-subclass of Proteobacteria, could be assigned to the N. ramosa morphotype . A simultaneously obtained pure culture showed the almost identical 16S rRNA sequence . The same combination of direct rRNA sequence retrieval and subsequent identification and quantification with rRNA-targeted nucleic acid probes was also applied to A. oxaliferum. This huge bacterium (cell length up to >100 μm, diameter up to 50 μm) contains sulfur globules and massive calcite inclusions and inhabits the upper layers of freshwater sediments. It is visible to the naked eye and has by its resistance to cultivation puzzled generations of microbiologists. By the rRNA approach a distant, but significant relationship to the phototrophic Chromatiaceae could be demonstrated [37, 38]. Furthermore, 16S rRNA sequences differing up to 7% were found in clones retrieved from this morphotype. The presence of several different genotypes even in morphologically homogeneous physical enrichment obtained from one lake sediment was proven by FISH .
Composition of complex microbial communities
rRNA-targeted oligonucleotide probes are ideally suited to investigate the composition of complex microbial communities. Within our special research program, many FISH studies were done with domain- and group-specific oligonucleotide probes. By targeting more conserved sites of the rRNA, these probes discriminate between the three domains Archaea, Bacteria and Eucarya or identify members of larger phylogenetic groups such as the alpha-, beta-, and gamma-subclasses of Proteobacteria, the Cytophaga–Flavobacterium cluster , Gram-positive bacteria with a high  or low  DNA G+C content, or planctomycetes . The group-specific probing yields important data on the abundance of different phylogenetic groups in different environments. The physiological diversity in all of these groups is much too high to link one of them to a specific process. However, they give good hints as to which groups to screen for the environmentally relevant populations. Here, we would like to list just few of the most interesting findings: Archaea occur in oxic marine waters with densities above 105 ml−1 making up more than 10% of all cells . Whereas beta-proteobacteria dominate many freshwater habitats, they are hardly detected in the bacterioplankton of the oceans . Soils and marine sediments contain high numbers of planctomycetes and members of the Cytophaga–Flavobacterium cluster [45, 46]. Information we have from cultivated members of the two groups suggests that they might be responsible for an important part of the polymer degradation. Attempts to characterize environmentally relevant polymer degraders should therefore focus on the enrichment and isolation of members of these groups.
Hybridizations with rRNA-targeted probes have in the meanwhile yielded the first insights into the composition of many other complex microbial communities (e.g. [1–4, 30, 31, 47]). The technique is ideally suited to monitor the reactions of selected groups of bacteria to environmental changes. In response to the input of cyanobacterial biomass to anaerobic marine sediments members of the Cytophaga–Flavobacterium cluster rapidly and strongly increased in number, corroborating their involvement in polymer degradation .
Insights in processes, interactions and their regulation rely in their precision and significance on the homogeneity of the population which is investigated. Hypotheses on structure–function correlations, therefore, should to be derived and tested for genera, species or sometimes even strains rather than for broad phylogenetic groups. For defined populations also the accessory FISH data like localization (e.g. on particulate organic matter), cell size distribution, and cellular rRNA content become ecologically more significant. Localization contributes to the description of the ecological niche of a population. Cell size distributions are essential for the understanding of the interactions of bacteria with their predators [49–52]. Grazing by protozoa and other top-down effects are at certain times more important for the composition of microbial communities than substrate availability and other bottom-up effects. The meaning of the ribosomal rRNA content has recently been discussed in detail . The role of the ribosomes as protein factories is so central that changes in the cellular rRNA of specific populations have high significance even though it might not be possible to discern from it specific parameters such as growth rates or turnover of a given substrate at the time of sampling.
Can rRNA-targeted nucleic acid probes be used in all environments?
Yes; the only question is which hybridization format is the most reasonable in a given setting. Presence/absence of a specific organism may today be addressed by PCR assays (see above) or by reverse hybridizations in which multiple rRNA-targeted nucleic acid probes are immobilized on various supports ranging from membranes, microtiter plates to miniaturized glass chips [54–57]. These techniques have the potential to be quantitative. In the future, they might increasingly supplement and, for the quantitative slot-blot hybridization, ultimately replace the currently established methods. However, for the time being, microbial ecologists interested in monitoring population dynamics in a strictly quantitative way have two options, FISH and the quantitative dot- or slot-blot hybridization. Limitations remain for both methods. As mentioned above, hybridization intensity cannot be correlated directly with cell numbers, since cellular RNA and DNA contents depend on the physiological status of the cells. On the other hand, cell counts by FISH may be underestimates where rRNA contents are below the detection limits or cell permeability is limited. Although the applicability of FISH has been expanded in the last few years from eutrophic to oligotrophic environments and from relatively clean samples to sediments and soils, it still cannot be universally applied. There are, for example, problems with statistical significance in patchy environments. In contrast, essentially all types of samples can be used for quantitative slot-blot hybridizations, which makes it the method of choice in those systems which are difficult for FISH. The absolute and relative (as compared to total rRNA) amounts of a specific rRNA are no direct measure of cell counts since they are also influenced by the cellular rRNA content which might change over at least one order of magnitude. However, since an increase in activity of a certain population is usually linked to higher cellular ribosome contents and cell numbers the monitoring of a parameter that summarizes both effects should enable reasonable correlations between population dynamics and a defined function. We recommend, especially if a new environment is examined, to combine at least two techniques for analysis of community composition. Results do not need to be identical, but they should at least be consistent and not contradictory.
What are the limitations of rRNA-targeted probes?
Some basic, theoretical limitations have to do with the target molecule. The 16S rRNA may be too well-conserved to discriminate between closely related populations. Different species may have almost identical 16S rRNA sequences . In such cases, the 23S rRNA may be useful. It is approximately twice as long and contains several highly variable regions. The heterogeneity sometimes found among the rRNA operons of a single organism may also be a problem. 16S rRNA sequence dissimilarities of 5% were identified in the two rrn operons of the archaeon Haloarcula marismortui. Significant interoperon differences have also been found in bacterial species (e.g. [60, 61]).
Another limitation originates from the fact that the rRNA diversity has only been partially described [1, 33]. Even if a probe was designed to be specific on the most complete current data set and shown to be specific for a range of test organisms, it may also hybridize with as yet unknown organisms. Conversely, there may be unknown microorganisms which are phylogenetically members of a probe target group, but do not contain a perfectly matching target site. The latter problem often occurs when designing group-specific probes. For many phylogenetically well-defined groups no common diagnostic target sites can be found. Therefore, it is recommended to follow one population with more than one probe [62, 63]. According to this multiple probe concept  sets can consist of nested probes specific for the genus, species and sequence of interest, or of two or three probes that target the same population. Simultaneous FISH with two or three differently labeled probes offers an elegant way to check whether ‘cross-hybridizing’ populations are present in the sample . If there are indications of hybridization to non-target organisms, further rRNA sequences should be recovered for comparative sequence analysis. This brings us back to the point that probe design relies on the quality of the rRNA database. Our ability to quantify populations in complex microbial communities is directly correlated to the effort that goes into maintaining and enlarging high-quality rRNA databases and into the continuous development of tools for rational probe design and in silico specificity control.
The most important practical limitation for the wide application of rRNA-targeted nucleic acid probes in microbial ecology is the lack of automation. Currently, only a limited number of samples can be processed with a restricted set of probes. Just imagine how much information could be retrieved from a semiquantitative oligonucleotide chip containing two or three specific probes for every rRNA sequence known.
In the last decade many molecular methods have been developed and their potential for microbial ecology has been proven. What will the future bring? For certain, there will be automation and parallelization of hybridization assays. Currently, each hybridization is performed separately and, in the case of FISH, each microscopic field is counted manually. With microarray/DNA chip technology [54–57] environmental samples will soon be probed with hundreds to thousands of different probes in a single reverse hybridization. Flow cytometers may become the standard instruments for quantitative analysis and sorting of hybridized or otherwise labeled microorganisms [66–69]. It has been demonstrated that following sorting, cells can be further analyzed by PCR-based techniques [68, 69], thereby giving direct access to the genetic information of uncultured bacteria.
There is a clear need for further systematic method development. This goes hand-in-hand with automation; for example, consider the recent flow cytometric study of the in situ accessibility of the 16S rRNA of E. coli to more than 200 fluorescein-labeled oligonucleotides . Tremendous differences were found in the amount of fluorescence conferred by probes to different sites. By targeting sites yielding bright fluorescent signals, in situ identification in environmental samples will become even more reliable.
Another trend is the combination of microbial community composition analysis by rRNA probing with in situ measurements of function. Recent studies have combined FISH with microsensors (e.g. [70–72]) and microautoradiography [73, 74]. Thereby, important ecological information on the physico-chemical environment of individual identified microbial cells and their substrate spectrum is obtained. Furthermore, gene expression could be visualized by in situ probing of mRNA in some cases (e.g. [75–77]). Combined community structure/function studies applying these or other techniques will in the future significantly promote our understanding of the interactions of microbial species with their biotic and abiotic environment independent of whether they are already culturable.
Another category of future studies could address the problem of limited culturability. There is clear evidence that only a small part of the extant microbial diversity has so far been retrieved in the form of pure cultures [1, 34]. rRNA-targeted probes must in the future be used to find out whether the bacteria already cultured are just ‘laboratory weeds’ or really important in certain environments, and, more importantly, to direct new cultivation attempts to those groups that are abundant in nature.
After all, for detailed physiological studies we still need pure cultures. Those will also be needed to rapidly determine whole genome sequences which are nowadays becoming prerequisites for truly detailed analyses of gene expression and its regulation in response to environmental changes.
The work of the authors has been supported by Deutsche Forschungsgemeinschaft. Additional funding came from the Bundesministerium für Bildung und Forschung, the European Union, the Max-Planck-Society, the Fonds der chemischen Industrie and the Körber Foundation. We thank Barbara MacGregor for critically reading the manuscript and all co-workers that contributed over the past years to the rRNA databases and the development and application of rRNA-targeted nucleic acid probes.