A sample of cells harvested from each chemostat was preserved in formalin (final concentration of 3.7%) and stored at 4°C. The fixed cells were diluted in 118 mM pyrophosphate to reduce clumping. Also, after a 30 min period of shaking at 200 rpm, the cells were sonicated on ice for 2 min in 30-s intervals, then stained with acridine orange, filtered onto 0.2 μm pore-size black polycarbonate filters and examined by epifluorescence microscopy (Hobbie et al., 1977). Cell abundance was estimated by counting a minimum of 10 fields. The bacteria were photographed with a SPOT digital camera (Diagnostics Instruments, Inc) for image analysis. Mean cell length and width were measured for a minimum of 75 randomly selected cells from each chemostat using Image Pro Plus software (Media Cybernetics, version 4.5.1). Cell volume and surface area of Agrobacterium sp. and Flavobacterium sp. were determined using equations assuming the cell shape was a cylinder capped by two hemispheres (Sun and Liu, 2003). Cell volume and cell surface area of Arthrobacter sp. was determined by using equations derived for a prolate spheroid (Sun and Liu, 2003).

Elemental analyses for carbon and phosphorus were performed on biomass samples from the chemostats to assess the elemental composition of each strain. Bacteria were harvested and filtered (<5 ^* 10⁻⁵ atm) onto pre-combusted Whatman GF/F filters for particulate carbon analysis and onto acid-rinsed GF/F filters for particulate phosphorus analysis. All samples were collected and analyzed in triplicate. The particulate organic carbon content was determined using a CHN analyzer (PerkinElmer, model 2400). The P content was determined spectrophotometrically following an acid-persulfate digestion, reacting with molybdenum blue and absorption measurements at 880 nm (APHA, 1992).

Cells from each chemostat were collected on 0.2 μm white polycarbonate filters in triplicate. Bacterial nucleic acids were determined after extraction via sonication and staining with the fluorochrome RiboGreen (Molecular Probes), which reacts with both DNA and RNA (Gorokhova and Kyle, 2002). Bacterial samples on the filters, negative control samples (containing all reagents but no bacteria), and standards of RNA (E. coli, Invitrogen) and DNA (calf thymus, Sigma) were prepared for analysis as described in Makino et al. (2003). Two identical 96-well black microplates were prepared with one plate receiving an addition of RNase (Promega; conc in 1X ^* TE buffer) before the 30-min digestion period. After the digestion reaction period, 75 μL of RiboGreen was added to both plates followed by a second reaction period of 10 min. The fluorescence of the samples was measured on both plates (excitation 480 nm and emission at 520 nm) on a Micromax 384 plate reader (Fluormax, Horbia Jobin Yvon) using DataMax software. The fluorescence of the digested sample was used to determine DNA concentration and differences between the undigested and digested plates were used to determine RNA concentrations.

The effect of varying C:P_R and temperature on the molar ratio of biomass C:P in the replicated chemostats was evaluated using analysis of variance (ANOVA) using the JMP® statistical software (version 11.2). ANOVAs were conducted on biomass parameters with temperature, resource C:P and strain as fixed effects. We did a three-way ANOVA initially and if there were statistically significant interactions, we conducted a two-way ANOVA for each strain (with temperature and P as the independent variables). If there was a significant interaction in the two-way ANOVA, then we did a one-way test for each variable at each level of the other variable. We used a p-value of less than or equal to 0.05 to determine statistical significance.

Three-way ANOVAs on cell volume and length-to-width ratios indicated that there were significant effects of both C:P_R ratio and different strains but not temperature (Table 1, Figure 2, Table S1) on morphology. There was a strain^*temperature interaction with length to width ratios but not biovolume, and there was a strain^*C:P_R interaction with both length to width ratios and biovolume. Two-way ANOVAs revealed significant effects of temperature only on the biovolume of Arthrobacter and the length to width ratio of both Agrobacterium and Arthrobacter (Table 1, Table S1). C:P_R ratio had significant effects on both morphometric variables in all three strains and there were no interactions between C:P_R ratio and temperature (Table 1, Table S1). As C:P_R increased, the biovolume and length to width ratios also increased. There was a significant effect of C:P_R on both biovolume and L:W ratios in the 3-way (p < 0.0001) and two-way ANOVAs (all p < 0.006). There was a great deal of variability in both the volume and the L:W ratios among the different strains, however (Figure 2). There were large differences in the mean length: width ratios of the three strains with Flavobacterium sp. and Agrobacterium sp., having higher ratios (3.68 and 3.20, respectively) than Arthrobacter sp. (1.50). Arthrobacter also had the lowest volume (mean = 0.46 μm³) followed by Agrobacterium (1.30 μm³) and Flavobacterium (2.03 μm³).

Temperature and P availability interactively affected biomass element content. Consistent with differences in the sizes of the three strains, the strains also differed in terms of C and N content per cell, but not P content per cell (Table 1, Figure 3, Table S2; C quota: p < 0.0001; N quota: p = 0.0003; P quota: p = 0.068). Temperature had no effect on C and N cell quotas but it did affect P quotas (p < 0.0001), with increasing quotas with temperature increases and at the lowest C:P_R (Table 1, Figure 3). C, N, and P cell quotas were strongly affected by C:P_R and there were significant interactions between temperature and C:P_R for C, N, and P quotas, especially C and N quotas for Agrobacterium and Flavobacterium (Table 1). C:P_R had effects on the C and N quotas of all three strains. P quotas were affected by different factors among the strains. C:P_R had a significant effect on P quotas for Agrobacterium (p = 0.006) while the other two strains were affected more by temperature (p < 0.0001 and p = 0.0069 in Arthrobacter and Flavobacterium, respectively). There was also a temperature^*C:P_R interaction in Flavobacterium.

The changes in cell quotas were not consistent among C, N, and P, leading to substantial variation in the effects of temperature and P on biomass stoichiometry (Table 1, Table S3). The biomass C:P (C:P_B) ratios were high at a C:P_R of 1000:1, with all three strains having ratios near 500:1 at the lowest temperatures and decreasing with increased temperature (Figure 4). The median biomass C:P for the three strains was similar, varying from 144 in Flavobacterium to 181 in Agrobacterium and 202 Arthrobacter. Although, biomass C:P decreased with temperature at all levels of C:P_R, the response of biomass C:P to temperature was strongest at the highest C:P_R, indicating that biomass C:P_B ratios were the most sensitive to temperature changes when P was least available. A similar pattern was observed for N:P but less so for C:N. The mean N:P biomass ratio for the three strains varied from 25 in Flavobacterium to 31 in Arthrobacter and 36 in Agrobacterium. The mean C:N_B for the three strains varied much less among strains, with a mean of 5.9 in Flavobacterium, to 6.3 in Arthrobacter and 5.1 in Agrobacterium.

Consistent with the cell quota observations, temperature did not affect biomass C:N ratios, rather only biomass N:P and C:P ratios (Figure 4, Table 1). There were significant effects of temperature on both C:P_B and N:P_B in all three strains, with the exception of Arthrobacter biomass N:P. C:P_R also had significant effects on all three biomass ratios in each of the strains, with the exception of the C:N of Arthrobacter.

RNA per cell varied by strain, temperature and C:P_R, while DNA per cell varied only among strains (Figure 5, Table 1, Table S4). There were no interactions for either of these factors. RNA per cell was highest when C:P_R was highest (lowest P supply) in two of the strains (Agrobacterium and Flavobacterium), perhaps due to increased cell size when P was least available. In addition, RNA per cell increased with increasing temperatures in Agrobacterium and Arthrobacter. Agrobacterium, Arthrobacter, and Flavobacterium had mean values of 4.8, 4.7, and 9.5 fg RNA cell⁻¹, respectively. Across all levels of temperature and C:P_R, Agrobacterium, Arthrobacter, and Flavobacterium had mean values of 9.9, 9.8, and 16.8 fg DNA cell⁻¹, respectively, giving RNA/DNA ratios of 0.48, 0.48, and 0.56, respectively.

The proportion of cellular P in nucleic acids (RNA and DNA) increased with increasing C:P_R particularly at 1,000:1 and varied among strains (Figure 6). There was an effect of temperature on the proportion of total P in RNA (p = 0.0139) but no effect on total nucleic acid content per cell. The percentage of P in nucleic acids varied from a low of 10% (Agrobacterium, high P supply and 10°C) to 92% (Flavobacterium, low P supply, 25°C) with mean percentages of 32% for both Agrobacterium and Arthrobacter and 48% for Flavobacterium.

As mentioned above, there were mixed effects of temperature on elemental quotas and stoichiometry. In previous studies, temperature has been shown to have important effects on cell size and stoichiometry in diverse organisms, from bacteria to fish and terrestrial, freshwater and marine systems (Daufresne et al., 2009; Forster et al., 2012; Borer et al., 2013; Morán et al., 2015). Specifically, past studies have shown that increasing temperature generally leads to increased C:P, N:P, and C:N ratios (Woods et al., 2003; Cotner et al., 2006; Martiny et al., 2013), but in the present study we found that increasing temperatures often led to decreased biomass C:P and N:P ratios (p < 0.05 for N:P in two strains and p < 0.05 for C:P in all strains). This contradiction is likely due to differences in growth and temperature manipulations. In a chemostat at a constant dilution rate, as temperature increases, the relative growth rate will decrease due to a uniform realized growth with increasing μ_max, in contrast to the uniform relative growth in the present experiments.

Of the previous studies that have examined the effects of temperature on biomass stoichiometry, Chrzanowski and Grover (2008) manipulated both the dilution rate and temperature in a chemostat study using Pseudomonas. They observed that growth rate had a larger effect on biomass stoichiometry than did temperature and this observation was one of the motivations for why we maintained uniform relative growth rates in the present study. However, unlike the present study, Chrzanowski and Grover (2008) observed that cell quotas for C, N, and P were highest at either low temperature and low dilution rate or high temperature and high dilution rate. At low dilution rates, C, N, and P content decreased with increasing temperatures, but in the present study there was no significant effect of temperature on C and N cell quotas, while P cell quotas increased with increasing temperatures, particularly when P availability was high. Also, in their study, at high dilution rates, C, N, and P quotas increased with temperature, similar to our observations with E. coli for C and P (Makino et al., 2003).

The lack of a consistent effect of temperature on cell size (Table S1) and quotas (Table S2) was also surprising in light of past work showing an inverse relationship between temperature and cell size. Among species, there is a general decline in maximal growth rates and metabolic rates with increasing size (Sommer, 1989; Niklas and Enquist, 2001) but variation within a species is influenced by the relative growth rate, due to the fact that growth and structural components scale differently with size (Clark et al., 2013). As relative growth rates increase, single cell organisms tend to become larger and more dilute internally (Kempes et al., 2012), whereas increasing temperatures tend to make cells smaller (Woods et al., 2003; Morán et al., 2015). We have recently observed that there is a great deal of variability in how different bacterial strains regulate their C:P and N:P ratios (Godwin and Cotner, 2015) and some of the capacity to do this may be coupled with the capacity of cells to change their morphology.

An obvious question that needs to be addressed is whether, in natural systems the effects of temperature on organism stoichiometry are mediated largely through the effects of temperature on cell size and thus, internal pools as we observed here, or through effects on relative growth rate, as observed in our recent work (Godwin et al., 2017). Biomass stoichiometric ratios (C:P, N:P, or C:N) generally decrease with latitude, a pattern that has been documented in marine plankton (Martiny et al., 2013), multiple ecosystem types (Borer et al., 2013), plant leaf tissue (Reich and Oleksyn, 2004; Kerkhoff et al., 2005), marcrophyte biomass (Xia et al., 2014), and freshwater plankton (Dobberfuhl and Elser, 2000), but many variables other than temperature change along a latitudinal gradient (Martiny et al., 2016), complicating interpretation of these data.

Although temperature had important effects on several internal pools and some stoichiometric ratios, C:P_R affected nearly all of these parameters. This result suggests that, across the range of temperatures and C:P_R, nutrient availability was the more important determinant of bacterial stoichiometry and physiology at uniform relative growth rates. While it is not surprising that C:P_R had a strong effect on the C:P and N:P ratios of the bacterial strains (and to a less extent C:N), the temperature^*C:P_R interaction suggests that temperature modified the response to C:P_R (Figure 3). Specifically, biomass C:P was more responsive to differences in C:P_R at the lowest temperatures.

Some of the effect of supply stoichiometry on biomass stoichiometry and elemental quotas was attributable to changes in RNA content. There was an increase in RNA content at the highest C:P supply ratio (Figure 5) which is similar to other studies looking at the GRH when P was not limiting (Elser et al., 2003). This suggests that there may be significant changes in RNA unrelated to the growth rate. One possibility is that changes in RNA were caused by changes in cell size. If the cellular ribosome concentration varied little with C:P supply ratios, but there were increases in cell size, it would have increased the RNA content per cell. Biovolumes increased with increasing C:P supply ratios from 56% for Agrobacterium, and 108% for Arthrobacter, to 255% for Flavobacterium.

Temperature had more significant effects on the P quotas of the three strains (p < 0.05 for Arthrobacter and Flavobacterium and p = 0.08 in Agrobacterium) than did C:P_R, while C:P_R had stronger effects on C and N cell quotas (Figure 3, Table 1). These differences were likely due in a large part to the strong effects that C:P_R had on morphology in all three strains (Table 1). The lack of a strong C:P_R effect on P quotas likely reflects the compensating effects of P availability on cell volume. Higher P availability leads to decreased cell volume but higher internal concentrations of some pools, while decreased P availability has the opposite effect, leading to a “buffered” P quota per cell. Similar observations have been made by others examining the effects of C and P limitation on cell morphology with P limitation leading to larger and more elongate cells (Løvdal et al., 2008; Godwin and Cotner, 2015). But Løvdal et al. also noted in their study of Vibrio splendidus that C, N, and P scaled differently, with larger cells having proportionately less P than C and N.

Temperature is a fundamental environmental variable that affects the behavior of everything from molecules to ecosystems. The metabolic theory of ecology (Brown et al., 2004) demonstrated that metabolic rates vary in predictable ways based on organismal size (mass) and temperature. In a literature survey of organisms from bacteria and algae to plants and animals, Woods et al. (2003) observed that most organisms were more N-, P-, and RNA-rich at colder temperatures. More recently, it was shown that bacterial size (Morán et al., 2010) and plankton stoichiometry (Martiny et al., 2013) in the oceans may both be influenced by temperature. Although temperature has effects on both metabolism and biomass composition, the present study shows that there are important effects of temperature on stoichiometry independent of growth rates. This mechanism is not widely recognized in biogeochemical modeling, but has important implications for predicting how the pools and fluxes will respond to warming temperatures and increased availability of nutrients.

One thing that was clear in examining the effects of temperature on biomass stoichiometry was that there was much greater change in stoichiometry at different temperatures when P was most limiting. In particular, C:P and N:P slopes relative to temperature were much steeper at the lowest P availability. These results imply that temperature effects on nutrient cycling are likely to be greatest in the least productive (oligotrophic) ecosystems such as large lakes and the open ocean. Furthermore, increased eutrophication in freshwater and marine ecosystems is likely to dampen feedbacks between climate change and biogeochemical processing, due to this decreased sensitivity at high nutrient levels.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer RM and handling Editor declared their shared affiliation.