1We examined the relative importance of resource composition (carbon : phosphorus molar ratios which varied between 9 and 933) and growth rate (0·5–1·5 h−1) to biomass carbon : nitrogen : phosphorus stoichiometry and nucleic acid content in Escherichia coli grown in chemostats, and in other heterotrophic prokaryotes using published literature.
2Escherichia coli RNA content and the contribution of RNA-P to total cellular P increased with increasing growth rate at all supply C : P ratios. Growth rate had a much stronger effect on biomass C : P than did supply C : P, and increased RNA content resulted in low biomass C : P and N : P ratios.
3However, we observed only twofold variations in biomass C : P and N : P ratios in the experiments, despite a difference of two orders of magnitude in C : P and N : P supply. The response of biomass C : P and N : P ratios to alteration of the supply C : P and N : P ratios revealed that E. coli was strongly homeostatic in its elemental composition.
4This result, and a literature survey, suggest that each heterotrophic bacterial strain regulates its elemental composition homeostatically within a relatively narrow range of characteristic biomass C : P and N : P ratios.
5Thus shifts in the dominance of different bacterial strains in the environment are probably responsible for the large variation in bacterial biomass C : P, as has been suggested for crustacean zooplankton. These findings indicate that bacteria are more like animals than plants in terms of biomass C : P and N : P homeostasis.
There can be an imbalance in the ratio of carbon (C) to other elements, especially phosphorus (P), among different trophic levels or functional groups (Sterner et al. 1998; Sterner & Elser 2002). At the base of the food web, phytoplankton have a wide range of biomass C : P mole ratios (C : P hereafter), from 100 up to ≈4000. This large range probably reflects interspecific and intraspecific variations among marine (Goldman, McCarthy & Peavey 1979) and freshwater species (Rhee 1978; Gächter & Bloesch 1985; Sterner et al. 1993). The C : P ratios of phytoplankton consumers such as crustacean zooplankton are typically much smaller and less variable; the total range of body C : P in taxa studied to date varies from 50 to 200, showing limited intraspecific variability (strong physiological homeostasis) with most variation associated with differences among species (Elser et al. 2000b). Such imbalances of biomass C : P between predator and prey have important ecological consequences, such as impacts on population dynamics and nutrient recycling (Elser & Urabe 1999; Sterner & Elser 2002).
One proposed cause of variable C : P or nitrogen (N) : P ratios in different organisms is interspecific differences in RNA content, reflecting differences in their life history (Elser et al. 1996; Elser et al. 2000a). For example, fast-growing cladocerans (especially Daphnia) contain as much as 1·5% P and 10% RNA dry weight, while slow-growing copepods contain only 0·6% P and 2% RNA. Thus differences in P content between Daphnia and adult copepods can be explained largely by the difference in RNA content (assuming RNA is ≈9% P; Sterner 1995; Elser et al. 1996). This mechanism relating C : N : P stoichiometry, RNA allocation and the life history of organisms is called the growth rate hypothesis (GRH, Elser et al. 1996; Elser et al. 2000a). While the GRH originally emerged from interspecific comparisons among biomass C : P, N : P and maximum growth rate for different species under ideal growth conditions (Elser et al. 1996; Main, Dobberfuhl & Elser 1997), it may also be relevant to intraspecific variations such as ontogenetic differences in biomass C : P and N : P ratios (Hessen 1990; Hessen & Andersen 1990; Villar-Argaiz, Medina-Sánchez & Carrillo 2002) because growth rates and RNA contents generally decline with age (Gorokhova & Kyle 2002).
Although many questions relating to the GRH have been examined in crustacean zooplankton, the dominant decomposers in lakes and oceans are heterotrophic bacteria (Cotner & Biddanda 2002), and therefore these organisms are likely to have large impacts on the stoichiometric composition of these waters. In heterotrophic bacteria collected from lakes and coastal waters and grown in the laboratory, biomass C : P varied from 8 to ≈500 (Tezuka 1990; Kirchman 2000), suggesting that the biomass C : P of these organisms is more variable than that of multicellular consumers such as zooplankton, and is perhaps comparable with that of phytoplankton (see Figure 1 of Sterner et al. 1998). However, little attention has been paid to factors causing large variations in bacterial C : P. Tezuka (1990) found positive relationships between the C : nutrient ratio of culture media and that of lake bacteria, implying that biomass elemental composition could closely reflect the elemental composition of supplied resource (weak homeostasis), as in phytoplankton (Rhee 1978; Sterner et al. 1993). Given that both phytoplankton and bacteria are osmotrophic, incorporating nutrients as individual rather than ‘prepackaged’ units, as herbivores and predators do, it is probable that bacteria would have a biomass stoichiometry that varied similarly to that of phytoplankton, although this question has not yet been examined.
On the other hand, many studies of enterobacteria such as Escherichia coli have shown that RNA (more specifically ribosomal RNA) content increases with growth rate (Maaloe & Kjeldgaard 1966; Neidhardt, Ingraham & Schaechter 1990; Hanegraaf & Muller 2001). There are also considerable variations in RNA content or RNA : DNA ratios in natural bacteria, with implications for growth and productivity (Kemp, Lee & LaRoche 1993; Lee & Kemp 1994; Jeffrey et al. 1996). Among these studies, Kemp et al. (1993) documented positive correlations between RNA content and growth rate for four marine bacterial isolates. Thus there is ample evidence that increased RNA could reduce bacterial biomass C : P as suggested by the GRH. Indeed, phytoplankton biomass C : P decreases from ≈1000 to 100 with increasing growth rate in continuous culture experiments (Goldman et al. 1979; Gächter & Bloesch 1985). However, simultaneous measurements of biomass C, N, P and nucleic acid content are still required for a direct evaluation of these correlations.
We now have alternatives (resource dependency and growth rate dependency) for biomass C : P and N : P stoichiometry of various organisms in the basal part of planktonic food webs. Although a contrast between phytoplankton (weak homeostasis) and zooplankton (strong homeostasis) is reasonably well established, information on the relative importance of and/or the interactions between these alternatives for microbes is incomplete. To improve understanding of stoichiometric homeostasis and its determinants in heterotrophic bacteria, we measured biomass C, N, P and nucleic acid content simultaneously in E. coli K-12 under a wide range of substrate C : P and growth rates. We also surveyed the literature on the homeostasis of biomass C : P and N : P for cultured and natural freshwater lake bacteria, and compared these with the results of our experiments. We propose a possible mechanism for highly variable biomass C : N : P stoichiometry in bacteria in nature.
Materials and methods
We chose to culture E. coli, admittedly not a very representative species for aquatic studies. But because RNA production is an essential outcome of growth in all organisms, we presumed that physiological responses are similar in bacteria growing in many different habitats. Our literature survey, which included several aquatic bacteria (see below), supported this view.
All culture media were defined as shown in Table 1 (Neidhardt, Bloch & Smith 1974; Wanner, Kodaira & Neidhardt 1977). We used glucose and ammonium chloride for C and N sources, respectively, and held C and N concentrations constant. The concentration of potassium phosphate was manipulated to create a gradient in supply C : P mole ratio from 9·3 to 934 and in supply N : P from 7·2 to 721. The pH of media after mixing all compounds was 7·3–7·4.
Table 1. The composition of defined media used in the present study
Concentrations in medium (mm)
Tris–HCl, pH 7·6
Tricine, pH 7·6
5 × 10−4
3 × 10−6
4 × 10−4
3 × 10−5
1 × 10−5
8 × 10−5
1 × 10−5
Panthothenate, hemicalcium salt
KH2PO4 (supply C : P & N : P = 9·3 & 7·2)
KH2PO4 (supply C : P & N : P = 93 & 72)
KH2PO4 (supply C : P & N : P = 233 & 180)
KH2PO4 (supply C : P & N : P = 467 & 361)
KH2PO4 (supply C : P & N : P = 933 & 721)
At the beginning of the experiment, three 33 ml chemostats were inoculated with E. coli grown in batch culture in a culture medium with a particular target C : P, and the same medium was fed to the chemostats at dilution rates of 0·5, 1·0 and 1·5 h−1 (except for the first run at a supply C : P of 9·3, in which dilution rates were 0·5, 1·1 and 1·7 h−1). The chemostats were mixed continuously with filtered, hydrated air, and maintained at 37 °C in the dark. We began collecting samples when the chemostat reached steady-state (≈36 h after inoculation). Outflowing medium containing E. coli was collected and filtered onto precombusted glass-fibre filters (Whatman GF/F) or polycarbonate filters (Osmotics, 0·2 µm pore size). The former were used to determine bacterial C and N contents; the latter were processed for bacterial P and nucleic acids. C and N contents were determined using a CHN analyser (model 2400, Perkin-Elmer, MA, USA), and P contents were measured by acid-persulfate digestion and subsequent soluble reactive phosphorus analysis (APHA 1992) using an Alpkem Flow Solution 3000 Analyser (ANTEC GmbH, Pinneberg, Germany). The pH of outflowing medium was 7·2–7·3.
Nucleic acids were determined by extracting the cellular contents via sonication followed by staining with the fluorochrome RiboGreen (Molecular Probes, OR, USA), which reacts with DNA and RNA (Jones et al. 1998; Gorokhova & Kyle 2002). A polycarbonate filter containing E. coli and 5 ml TE buffer (1×, Molecular Probes) with 0·167% w/v N-laurosarcosyl (Sigma, MO, USA) were transferred to a 50 ml snap-cap vial. Samples were sonicated on ice for 2 min using a 5 mm tip on a Microson Ultrasonic Cell Disruptor XL set at 10% maximum output, then incubated for 1–2 h on a shaker at room temperature. From each tube, three replicate 75 µl subsamples were pipetted into wells of a black microplate, and 75 µl RiboGreen working solution (750 nm in 1 × TE buffer) were added to each well. Negative control samples (containing all reagents but no E. coli) and standard DNA (calf thymus, Sigma) and RNA (Type III from baker's yeast, Sigma) solutions, diluted in the TE buffer with sarcosyl, were also distributed into the microplates and processed in the same way. The microplate was incubated in the dark at room temperature for 5 min. The samples were then measured in an FL600 Microplate Fluorescence Reader using KC4 software (Bio-Tek, VT, USA) with peak excitation at 480 nm and peak emission at 520 nm. After an initial scan, 7·5 µl RNase A (Promega, WI, USA), 10 µg ml−1 in 1 × TE buffer, were added and the plate was rescanned after 20–40 min. Differences in the sample scans relative to differences observed in RNA and DNA standards were used to determine RNA and DNA concentrations in the extracts.
Biomass N, P, RNA and DNA contents were converted to relative values (percentage of dry weight) by assuming that C content was 50% of dry weight (Neidhardt et al. 1990). The relative amounts of RNA-P and DNA-P to total biomass P were calculated by assuming that P occupies 9% of the mass of nucleic acids (calculated from the elemental composition of five nucleotides). The effects of dilution rate (= growth rate) and supply C : P on E. coli C : N : P stoichiometry and nucleic acid contents (Y) were examined by stepwise multiple regression analysis using the following model:
Y = a1(growth rate) + a2 (supply C : P) + intercept,
where a is the partial regression coefficient.
To compare the nutrient ratios of bacteria and culture media, we modelled stoichiometric homeostasis (Sterner & Elser 2002) by describing a linear relationship between the relative changes in these two parameters with the following equation:
where x is a supply nutrient ratio in the culture media, y is the same ratio in bacteria (measured with identical units), and Η is a regulatory coefficient measuring the degree of homeostasis. Equation 1 can be integrated to give:
where c is a constant of integration. Equation 2 can be expressed in linear form using logarithms:
Equation 3 indicates that if one plots the logarithm of a nutrient ratio in a consumer (bacteria) vs the logarithm of the ratio of the same nutrients in its resources, the slope (1/Η) relates directly to the strength of homeostatic maintenance of nutrient ratio within the consumer biomass. Η will vary between 1·0 (no homeostasis) when log(y) = log(c) + log(x) and ∞ (no variation in consumer nutrient ratios, or strict homeostasis) when log(y) ≈ log(c). Sterner & Elser (2002) present several examples of the use of this analysis to quantify different degrees of homeostasis in diverse consumer–resource pairs.
Results and discussion
escherichia colibiomass in chemostats
Adding more P to the culture media increased E. coli biomass in chemostats at low growth rates (Fig. 1a). When the growth rate was 0·5 h−1, biomass was ≈80 µg C ml−1 for supply C : P ratios of 9·3 and 93, but only 7 µg C ml−1 at a supply C : P of 933. However the differences in biomass diminished when growth rates increased. Theoretically, the biomass of organisms in a chemostat should increase with increasing concentration of a limiting resource at a constant growth rate (Fiechter, Käppeli & Meussdoerffer 1987; Neidhardt et al. 1990). With our various media P : C ratios, there was some evidence that E. coli biomass was P-limited at low growth rates (Fig. 1b). At the highest growth rate, however, there was little effect of increased P concentration (higher P : C ratio) in the medium.
biomass cnp stoichiometry, nucleic acid content and growth – evaluation of the growth rate hypothesis
Consistent with the GRH, the biomass C : P and N : P ratios of E. coli decreased with increasing growth rate in all media, while biomass C : N changed little (Fig. 2; Table 2). The 200% increase in growth rate from 0·5 to 1·5 h−1 led to a 20% reduction in biomass C : P from 65·1 to 52·4 (the average of the treatments), and a 21% reduction in biomass N : P from 16·5 to 13·0. These responses reflect the increase in P content (from 2 to 2·8%), with growth rate while the N content did not increase. N content (14–15%) and subsequently biomass C : N were affected by the supply C : P. However, supply C : P was not statistically significant in stepwise multiple regressions for biomass C : P and N : P ratios (Table 2), suggesting that variations in biomass C : P and N : P were independent of supply C : P.
Table 2. Results of stepwise regression analysis for E. coli C : N : P stoichiometry and biochemical composition
Partial regression coefficient
Standard regression coefficient
R2 is the coefficient of multiple determination; F is the variance ratio of multiple regression; t is the t-statistic. D, S and I denote dilution rate, supply C : P ratio and Y intercept of the regression lines, respectively.
7·7 × 10−4
Biomass C : P
Biomass N : P
Biomass C : N
2·1 × 10−4
3·7 × 10−3
RNA : DNA
Proportion of DNA-P to total P
1·4 × 10−3
Proportion of RNA-P to total P
1·6 × 10−2
Bacterial RNA content is strongly associated with growth rate (see Introduction). Consistent with this and with the GRH, we also observed that the RNA content of E. coli increased considerably with increasing growth rate, whereas DNA content was stable around 2·5% of E. coli dry weight (Fig. 3). Thus RNA : DNA ratios rose from 3·5–5·5 to 7–9 as growth rate increased. This increased RNA was probably ribosomal RNA, as it generally dominates the cellular RNA pool (65–81% cellular RNA; Neidhardt et al. 1990; Lehninger, Nelson & Cox 1993) and increases in proportion to other types of RNA with increasing growth rate (Neidhardt et al. 1990). Similar trends of growth-dependent variation in RNA content have also been observed in cyanobacteria (Binder & Liu 1998; Lepp & Schmidt 1998); yeast (Brown & Rose 1969; Aiking & Tempest 1976); and freshwater eukaryotic algae (Rhee 1978). Within the range of growth rates that we applied, the RNA and DNA contents in our experiments were similar to those of E. coli recorded by Jacobsen (1974; cited by Neidhardt et al. 1990) and the E. coli RNA : DNA ratios were almost identical to those reported by Skjold, Juarez & Hedgcoth (1973) and of freshwater and enterobacteria summarized by Dortch et al. (1983).
Were increased RNA amounts sufficient to affect the cellular P pool? The contributions of RNA-P to total cellular P were at least 40–50% at the lowest growth rate of 0·5 h−1 and 70–80% at 1·5 h−1 (Fig. 3). DNA-P, on the other hand, was ≈10% of total cellular P and decreased with increasing growth rate. Thus RNA was the dominant pool of biomass P, especially at high growth rate; this result is also consistent with the GRH. Although similar observations are limited, Vadstein (1998) also found that about 60% of cellular P was bound in DNA + RNA + lipid fraction at the early stationary phase in two bacterial strains isolated from a eutrophic lake. The lipids of those isolated strains could have been entirely phospholipids, as in E. coli (Neidhardt et al. 1990). However, cellular lipid content (9·1% dry weight for E. coli; Neidhardt et al. 1990) as well as phospholipid P content (≈4%, Elser et al. 1996) are generally less than those of RNA (see above). Furthermore, as growth rates increase, cell size also increases (Cotner, Ogdahl & Biddanda 2001), decreasing the surface-to-volume ratio and probably decreasing the percentage of cellular P that would be in phospholipids. These results indicate that RNA-P is the internal pool having the largest effect on cellular-P in bacteria, at least under balanced growth conditions. The P content of E. coli in our experiments (Pec) can be expressed as a linear function of RNA content (RNAec):
Stepwise multiple regression revealed that although supply C : P was marginally associated with three out of five variables in Fig. 3, the effects of dilution rate on nucleic acid components (especially RNA content) were much stronger (Table 2). We conclude that variations in E. coli RNA content were regulated mainly by changes in the growth rate. To explore specifically whether RNA content could be responsible for changes in E. coli stoichiometry, we plotted E. coli biomass elemental ratios against RNA content (Fig. 4). Increased RNA content with growth rate corresponded with low biomass C : P and N : P, strongly supporting the GRH within this organism.
There are substantial data to support the relationship between bacterial nucleic acid content and growth, that indirectly support the GRH. Herbert, Phipps & Strange (1971) pointed out ‘... and in the case of bacteria, (P) may account for up to 3% of the bacterial dry weight. For example, nucleic acids contain roughly 10% phosphorus and fast-growing bacteria may contain 25% of their dry weight as RNA and DNA.’ The importance of RNA for cellular C : P is also given by the literature survey of Kirchman (2000) who compared the amount of RNA, DNA and phospholipid among fast- and slow-growing bacteria. However, few studies have directly examined the importance of the increased amount of P due to growth (= RNA) in terms of biomass C : N : P stoichiometry. We examined this issue and found strong support for the GRH. It appears that the GRH is valid for intraspecific variations in biomass C : N : P stoichiometry for many microbes because the response of increased RNA content with increased growth rate is apparently ubiquitous, as discussed above.
implications of variable c : n : p stoichiometry in natural bacteria
Our data show that variations in E. coli C : P and N : P are strongly determined by growth rate. However, the observed variation, ≈20% reduction in the biomass C : P and N : P ratios caused by a 200% increase in growth rate, was small compared with the large variations observed in lake bacteria; biomass C : P ratios vary between 8 and 500 (Tezuka 1990). So we are left with the question of whether growth rate variation is sufficient to explain this large variation in bacterial biomass stoichiometry in nature.
One might suspect that our experiments did not sufficiently represent the natural variance in growth rate, and that we therefore observed low stoichiometric variability in the chemostat relative to nature. Growth rates (= dilution rates) of E. coli (0·5–1·7 h−1 in our experiments, corresponding to the specific growth rates of 8–28 day−1) were high compared with those of bacteria in natural waters [up to 7·7 day−1, but typically less in terms of specific growth rate calculated from Table 1 of Ducklow & Carlson (1992)]. However, our multiple regression analysis predicted that E. coli biomass C : P would be around 73 (SE = 4·0; Table 2) at zero growth, and Jacobsen (1974; cited by Neidhardt et al. 1990) demonstrated that E. coli RNA content was nearly constant at growth (= dilution) rates <0·5 h−1, supporting the view that our range of growth rates adequately represented stoichiometric variation in E. coli. In Jacobsen's study, RNA content was still ≈10% of dry weight (and that of DNA increased to 5% dry weight) even at growth (= dilution) rates of 0·03–0·05 h−1, corresponding to specific growth rates of 0·50–0·83 day−1 (within the range of in situ rates). Thus E. coli biomass C : P may never reach 500, even at very low growth rates. In addition, Chrzanowski & Kyle (1996) cultured Pseudomonas fluorescens at a growth (= dilution) rate of 0·03 h−1 and found that biomass C : P was 77–210, even though they used much higher supply C : P ratios (1200–2500) than ours (see below). From these observations we argue that the intraspecific growth-dependent variation in biomass C : N : P stoichiometry is not sufficient completely to explain the wide variation in bacterial biomass C : N : P stoichiometry.
Therefore we return to the prediction of a strong relationship between bacterial biomass C : N : P stoichiometry and supply C : N : P stoichiometry as shown by Tezuka (1990). In contrast to this prediction, E. coli biomass C : P and N : P ratios in the present study were quite stable, despite the growth-dependent variations, across a broad range of supply C : P and N : P ratios. We applied the model of Sterner & Elser (2002) to identify the degree of stoichiometric homeostasis (Fig. 5), and found that there was no significant correlation between log(biomass C : P or N : P) and log(supply C : P or N : P): Η in equation 3 was effectively ∞ in both cases (Table 3). These findings indicate that E. coli was strongly homeostatic in its C : P and N : P ratios. This conclusion was also supported by stepwise multiple regression, which did not indicate significant independent effects of supply C : P on biomass C : P and N : P ratios, respectively (Table 2). Thus, resource dependency is also insufficient completely to explain the wide variation in biomass C : N : P stoichiometry in natural bacteria.
Table 3. Summary of the regression analyses for data shown in Fig. 5 of log(biomass elemental ratio) vs log(supply elemental ratio)
H values in equation 3 (= 1/slope), treated as infinity in insignificant relationships (P > 0·05).
Culture conditions for each data set:  continuous culture (dilution rate, 0·5–1·5 h−1, present study) and batch culture (early stationary phase, Nakano 1994);  continuous culture (dilution rate, 0·03–0·09 h−1, Chrzanowski & Kyle 1996) and batch culture (early stationary phase, Nakano 1994)  batch culture (early stationary phase, Nakano 1994), and  batch culture (stationary phase, Tezuka 1990).
E. coli (N = 18) 
C : P
N : P
P. fluorescens (N = 23) 
C : P
N : P
Strain from Lake Biwa (N = 6) 
C : P
N : P
Mixed strains from Lake Biwa (N = 8) 
C : P
N : P
An important difference in the study of Tezuka (1990) relative to ours was that he worked with lake water (<1 µm fraction). That inoculum would have contained many bacterial strains, each with different characteristic C, N and P requirements. The slopes of a regression line for his data in Fig. 5 were close to 1 (0·91 and 0·71 for C : P and N : P, respectively; Table 3), indicating that this community exhibited almost no homeostasis within a range of supply C : P similar to ours (50–1200), and even within the smaller supply N : P of 7–60. A possible explanation as to why Tezuka (1990) observed that lake bacteria C : P and N : P ratios varied directly with the supply C : P and N : P ratios is that he may have selected different strains with characteristic biomass C : P and N : P values that best matched the C : P and N : P ratios he supplied. Variations in biomass C : P (and N : P) of in situ bacteria could have been generated by shifts in the dominance of different strains in the environment, as has been observed for crustacean zooplankton (Gulati, Siewertsen & Liere 1991; Hessen, Andersen & Faafeng 1992).
The study of Nakano (1994) supports this suggestion. He isolated a bacterium from Lake Biwa, where the work of Tezuka (1990) was performed. That single strain showed strong homeostasis, as indicated by the modest slopes of the log–log plots in Fig. 5 (cf. Table 3) even though supply C : P was 1·5–2000 (1·0–50 for supply N : P; Nakano 1994). At least one strain of bacteria in Lake Biwa shares the lack of responsiveness of biomass C : P and N : P ratios that we observed in E. coli. Furthermore, we found that P. fluorescens studied by Nakano (1994) and Chrzanowski & Kyle (1996) was also strongly homeostatic (Fig. 5; Table 3). Together with the previously discussed observations from zooplankton, these results suggest that two of the main groups of heterotrophs in lakes, zooplankton and bacteria, and not just phagotrophic heterotrophs (metazoans such as crustacean zooplankton), are likely to be more strongly homeostatic than autotrophs. Therefore heterotrophic bacteria are more like animals in terms of C : N : P homeostasis.
Finally, we propose one possible explanation of how the biomass C : N : P of in situ bacteria varies through shifts in the dominance of different strains. Generally, in situ bacteria grow slowly (Ducklow & Carlson 1992) and may have RNA : DNA ratios as low as 0·3–1 (by mass), as observed by Lee & Kemp (1994) for marine bacteria. We might speculate that such bacteria have biomass C : P as high as 500, simply because of their low RNA content. However, DNA content in terms of percentage dry weight increases at low growth rate. Kirchman (2000) points out that the DNA content of slow-growing bacteria in the ocean can reach up to 10% dry weight (also see Simon & Azam 1989; Posch et al. 2001). We calculated the biomass C : P of such bacteria, assuming that C is 50% dry weight, nucleic acid is 9% P, and RNA : DNA is 0·3 or 1 (Fig. 6). In contrast to speculation, the calculated biomass C : P is 110, nearly the Redfield ratio of 106 (Redfield 1958), and 72 when RNA : DNA is 0·3 and 1, respectively, and when DNA is 10% dry weight. Our calculation also suggests that bacteria would have a biomass C : P as high as 500 if they contained little DNA relative to their biomass (less than 2% dry weight). This might be possible, mechanistically at least, because the DNA content of bacteria in natural waters is variable. For example, cellular DNA content varied from 1% up to 25%, with the mode around 10% dry weight, in an Austrian lake bacterial community (Figure 2 of Posch et al. 2001; also see Button & Robertson 2001).
There are many bacterial strains in natural waters that grow slowly (low RNA : DNA ratio), and those with low DNA content (e.g. a few per cent) may have higher biomass C : P than their conspecifics. Biomass C : P at a community level may increase if bacteria with such low DNA contents dominate, and this may be caused by increases in supply C : P, as observed in the experiment of Tezuka (1990). This high biomass C : P would not be observed in E. coli as its DNA content would still be ≈5% at low growth rate (Jacobsen 1974; cited by Neidhardt et al. 1990). Rather, the growth of bacteria with intrinsically low DNA content close to that of eukaryotes such as yeasts (0·1–0·6% dry weight; Brown & Rose 1969; Fiechter et al. 1987) and phytoplankton (0·4–0·8% dry weight; Holm-Hansen, Sutcliffe & Sharp 1968; Dortch et al. 1983) may be responsible for high biomass C : P at a community level.
Addressing these issues will require more comprehensive studies of natural microbial organisms, to determine whether they behave homeostatically in response to changes in resource stoichiometry. Our work suggests that this should be the case, but these studies need to be performed on more strains isolated from natural waters and soils. If they are not homeostatic, and instead bacterial community C : N : P ratios routinely track supply C : N : P ratios, this would suggest little effect of heterotrophic bacteria on the relative availability of nutrients at any external supply ratio–nutrients would be regenerated in about the same ratio as they are supplied (Elser & Urabe 1999). However, the existence of certain supply ranges where bacterial communities behave homeostatically could have quantitative effects on the relative availability of different nutrients, as has been shown for crustacean zooplankton (Elser & Urabe 1999; Sterner & Elser 2002).
Anthony Dean, University of Minnesota provided the E. coli strain and suggestions on chemostat design. Elena Gorokhova and Marcia Kyle, Arizona State University established fundamental parts of the DNA and RNA quantification method in this study. Judith Olson and Andrea Plevan, University of Minnesota performed E. coli C, N and P analyses. The US-NSF provided financial support (DEB-9977047).