• Robert P. Goldman,

    1. College of Osteopathic Medicine of the Pacific, Western University of Health Sciences, Pomona, California 91766
    2. E-mail: rgoldman@westernu.edu
    Search for more papers by this author
  • Michael Travisano

    1. College of Osteopathic Medicine of the Pacific, Western University of Health Sciences, Pomona, California 91766
    2. Department of Biology and Biochemistry, University of Houston, Houston, Texas 77204
    3. Department of Ecology, Evolution and Behavior, 1987 Upper Buford Circle, University of Minnesota, St. Paul, Minnesota 55108
    4. E-mail: travisan@umn.edu
    Search for more papers by this author


Ultraviolet (UV) light is a major cause of stress, mutation, and mortality in microorganisms, causing numerous forms of cellular damage. Nevertheless, there is tremendous variation within and among bacterial species in their sensitivity to UV light. We investigated direct and correlated responses to selection during exposure to UV. Replicate lines of Escherichia coli K12 were propagated for 600 generations, half with UV and half as a control without UV. All lines responded to selection, and we found strong positive and negative correlated responses to selection associated with increased UV resistance. Compared to Control populations, UV-selected populations increased in desiccation and starvation resistance approximately twofold but were 10 times more sensitive to hypersalinity. There was little evidence for a persistent large competitive fitness cost to UV resistance. These results suggest that natural variation in UV resistance may be maintained by trade-offs for resistance to other abiotic sources of mortality. We observed an average twofold increase in cell size by the UV-selected populations, consistent with a structural mode of adaptation to UV exposure having preadaptive and maladaptive consequences to other abiotic stresses.

Ultraviolet (UV) radiation is an important cause of selection and has been throughout the Earth's history (Rothschild and Cockell 1999; Cockell 2000; Karam 2003). UV exposure causes widespread damage in proteins, membranes, and DNA, acting as both an agent of mortality and as a mutagen. Direct negative consequences of ambient UV exposure are observed across a broad array of taxa including grasses (Day et al. 1999; Milchunas et al. 2004), humans (West 1999), salamanders (Belden et al. 2000), and microbes (Meador et al. 2009). Indirect consequences of ambient UV exposure also have been observed, including loss of diversity in coral reef communities (Baird et al. 2009) and altered plant–herbivore interactions (Sullivan 2005). Numerous studies have documented deleterious consequences of elevated UV exposure (e.g., Armstrong and Kricker 2001).

Microbes are particularly sensitive to the effects of UV because their small size creates a high surface area-to-volume ratio. They are without a UV shielded germline, a trait that affords multicellular organisms some protection to the deleterious effects of UV. Bacteria do have multiple, redundant repair pathways that respond to UV-induced damage, the redundancy suggesting that UV has long been an important factor in shaping microbial evolution (Hanawalt et al. 1979; Goosen and Molenaar 2007). Despite the impact of UV on microbial mortality and evolution, and early interest in the effects of UV on microbes (Demerec 1946; Witkin 1947), there have been relatively few studies on the evolutionary consequences of increased UV exposure (but see Ewing 1995; Alcántara-Díaz et al. 2004; Weigand and Sundin 2009).

UV resistance varies greatly among bacteria (Nasim and James 1978; Gascón et al. 1995; Joux et al. 1999), and we hypothesize that associated costs of UV resistance are a cause for variation in UV resistance. For many species, even brief exposure to direct sunlight is lethal, a surprise given that many UV-sensitive bacteria have repair pathways to reduce UV damage and are thought to frequently experience UV exposure. UV damage is either avoided, repaired enzymatically, or tolerated—but each may be costly, either in terms of direct costs due to avoidance or enzymatic repair, or indirect costs in the form of slower growth rate, higher mutational load or pleiotropic consequences of resistance (Friedberg et al. 1995; Synder and Champness 1997).

To investigate the costs and consequences involved in the evolution of UV resistance, Escherichia coli populations were subjected to daily acute UV exposure. The outcome of selection was assessed by measuring fitness in the selective UV environment, UV resistance, and resistance to other abiotic sources of mortality. We predicted that there would be correlated responses to selection, as adaptive responses can dramatically alter physiologically associated traits, often generating greater variation in these nonselected traits than in the selected ones themselves (Travisano and Lenski 1996). Moreover, there are numerous studies demonstrating correlated stress sensitivities of wild and genetically manipulated bacterial genotypes (e.g., Shukla et al. 2007). This extensive literature and the existence of global stress response pathways induced by a variety of stresses, including UV (Khil and Camerini-Otero 2002), lead us to focus on other abiotic stressors. We anticipated selection for UV resistance could result in preadaptation (positively correlated fitness responses, see Cullum et al. 2001) to different environmental conditions that cause similar types of cellular damage. In contrast, the evolution of increased UV resistance by stress-specific mechanisms could potentially engender enhanced sensitivity to other stresses (Ferenci and Spira 2007). Such environment-specific responses could cause shifts in niche breadth (Bennett and Lenski 2007) due to the specificity of adaptation (Travisano and Lenski 1996).

Twelve replicate populations (lines) of E. coli K12 were exposed to a daily intense UV dose for 60 days (approximately 600 generations under our experimental conditions). A set of 12 replicate Control populations was maintained simultaneously under the same conditions without UV exposure. We subsequently assayed UV survival, competitive fitness, and other growth and survival traits over the course of selection. A follow-up selection experiment was performed for an additional 20 days, without UV exposure, to determine if UV resistance declined in the absence of selection.

Our analysis of direct and correlated responses to selection is structured on two possible mechanisms for increases UV resistance: enzymatic DNA repair and increased cell size. Improved enzymatic repair could reduce the immediate deleterious consequences of mutations, and prior studies have observed such adaptation after similar UV selection (Alcantara-Diaz et al. 2004). If improved enzymatic DNA repair was the primary mode of adaptive evolution, we expected that UV resistance would be strongly positively correlated with each of the three unselected stress resistances, because the three abiotic stresses are known to cause DNA damage. Alternatively, increased cell size could be the primary resistance mechanism. Increases in cell size have been repeatedly observed in response to UV selection (Zelle and Ogg 1957) and cell size is thought to be a potential mechanism to either reduce the initial amount of UV damage (Karentz et al. 1991) or reduce its effect (Davies and Sinskey 1973). If increased cell size was the primary mechanism for resistance, we expected correlated responses to differ depending upon the abiotic stress. Increased cell size can maintain larger reserves of water and other resources, promoting desiccation, and starvation resistance. But increased cell size is likely to cause greater negative consequences of osmotic stress. Hypersaline osmotic stress causes water to move out of a cell, shrinking the cell volume (Koch 1984), deforming and damaging the cytoplasmic membrane (Houssin et al. 1990; Wood 1999), and potentially causing cell death (Mille et al. 2002). Larger cell sizes are likely to be especially susceptible to increased cytoplasmic membrane stress, as deformation increases with cell volume.

The UV-selected lines showed substantial adaptation to the UV selective conditions, and we observed a threefold increase in UV resistance over the ancestral value. We also observed that UV-selected lines are larger on average than Control lines. There was little evidence for a persistent direct fitness cost of resistance in the selected environment. The UV-selected lines have greater survival during desiccation and stationary phase, but substantially poorer survival during saline exposure, consistent with the cell size mediated mechanism for UV resistance.



Escherichia coli K12 (MG1665, Blattner et al. 1997) was used as the ancestral strain for the selection experiment. For competition assays, we used E. coli B REL 606 (Lenski et al. 1991), which lacks the ability to grow on arabinose, as the common competitor (K12 MG1665 can grow on arabinose). Escherichia coli K12 (K12) and E. coli B (B) strains can be distinguished on Tetrazolium Arabinose (TA) agar plates, appearing red and white, respectively. Competition plates were supplemented with 10 g of arabinose and 100 μL of tetrazolium dye per liter. The growth media for all experiments was Luria Broth (LB): per liter 10 g tryptone, 10 g NaCl, 5 g yeast extract. Agar plates were made by the addition of 16 g agar per liter to LB broth.

Experimental evolution

To start the experimental evolution study, E. coli K12 was streaked from a −80°C freezer stock onto an LB plate and incubated it at 37°C for 24 h. The following day, sterilized toothpicks were used to pick 24 single colonies, which were then used to inoculate 24 LB liquid cultures. Single colonies were used to ensure that genetic variation arising within populations over the course of selection occurred as a consequence of independent de novo mutations. Hence, observations of parallelism among replicates could not be attributed to shared standing genetic variation. Twelve of the cultures were designated as UV lines and 12 as Control lines. Each liquid culture contained 10 mL of LB and was maintained with shaking at 120 rpm and incubated at 37°C. After 24 h, 100 μL from each culture was plated and spread onto LB agar with a glass spreading rod and incubated for an additional 24 h. The following day plugs of the resulting bacterial lawn were removed using sterile 13-mm glass test tubes. Each plug was then tapped down to the bottom of the test tube and suspended in 1 mL of 0.8% saline and vortexed to suspend the bacteria. An aliquot of 100 μL was then spread as a lawn onto fresh LB agar.

The 24 lines were propagated on a daily basis for 60 days. For the UV treatment, each of the LB agar plates was exposed to UV for 25 sec in a darkroom. The 25-sec time point was chosen as it caused 75% mortality in the ancestral population, as determined by a killing curve. UV exposures were conducted with a UVP Inc. (LLC Upland, CA) Chromato-Vue TM-36 transilluminator using 15 W bulbs that produce UV light at 302 nm, which is in the UV-B spectrum. We exposed the bacteria by inverting the plate, placing it directly on the UV transilluminator surface. A timer was started as soon as the plate was placed over the UV source and each plate was immediately removed from the light source after 25 sec. The transilluminator was switched on 30 min prior to use, to allow for stabilization of the light output. After each of the 12 plates was exposed to UV, they were wrapped in aluminum foil and incubated at 37°C for 24 h. The plates were handled entirely in the darkroom until covered for incubation.

The Control plates were handled identically, except that they were not exposed to UV. The following day, plugs were removed (as above) from each of the 24 plates, (UV and Control lines) resuspended as before and replated. The difference in selection (UV vs. Control) resulted in a smaller population size immediately after UV selection in those replicates. Population size can have potentially profound affects on adaptation, as smaller populations typically maintain smaller amounts of genetic variation. In addition, the appearance of beneficial mutations is strongly affected by population size, which is particularly relevant in a clonally reproducing population (Desai et al. 2007). In this experiment, the minimum population size in each 13-mm area was approximately 5 × 107 in the lines exposed to UV and 20 × 107 in the Control lines not exposed to UV. Although this difference in these population sizes is large, we anticipated that the occurrence of beneficial mutations would be sufficient so that the impact of clonal interference would reduce the consequences of the population size differences (Gerrish and Lenski 1998; Sniegowski and Gerrish 2010). Moreover, we excluded comparison of the UV and Control replicates under the Control environment; conditions that might favor an impact of the different population sizes on adaptation. This is because it is under the Control conditions that the larger population sizes are maintained and selection for UV-resistance alleles is absent.

Three additional potential complications arose from the selection. The UV lines would go through up to two additional generations per day (making up the growth of the UV killed cells), for a total of 120 additional generations of selection. The UV lines would have access to the resources of the killed cells, although this was unlikely to be an issue, as LB growth on plates is not limited by the availability of nutrient resources. Finally, selection in a spatially structured environment can lead to increases in yield, rather than in growth rate, although this selection is present in both the UV and Control treatment environments.

Every 10th day, an additional plug was removed from each plate. This plug was resuspended in 1 mL of saline, vortexed for 1 min, and the cell suspension added to 1 mL of 30% glycerol in sterile two dram glass screw cap vial to make a −80°C freezer stock, with samples frozen on days 10, 20, 30, 40, 50, and 60. Representative single colonies were isolated from each time point, were themselves viably preserved at −80°C, and were used throughout the single colony assays described below.

A 20-day follow-up selection experiment was performed to determine if UV resistance declined during a period of relaxed selection. Our goal was to distinguish between fixed (persistent) costs of increased UV resistance with transient costs, such as those arising via linkage. For this portion of the experiment, each lineage was propagated as before, except all the UV lines were propagated without UV exposure and samples were frozen on days 70 and 80.


We used single colonies isolated from freezer stock populations for all experiments measuring phenotypic responses to selection. For each assay, single colonies were inoculated into 10 mL of LB and incubated overnight at 37°C with shaking. The liquid cultures were then used as the source for each phenotypic assay. Unless noted otherwise, all freezer stocks were from day 60 of the selection experiment.

Measuring UV survival

UV survivorship was measured for days 30, 40, and 60 in a manner similar to the selection regime, except bacteria were diluted to a low density prior to UV exposure. One hundred microliters from overnight freezer cultures were plated onto fresh LB agar medium and incubated for another 24 h. The following day, 13-mm plugs of agar were removed from the plates and resuspended in 0.8% saline. Cultures were diluted by pipetting 100 μL into 9.9 mL of sterile 0.8% saline and vortexing. Two additional 100-fold serial dilutions were done and two 100-μL aliquots were plated from the final test tube on to two separate LB agar plates. One plate was immediately incubated at 37°C overnight—this plate was used to determine initial density. A second plate was exposed to UV for 25 sec, wrapped in foil, and incubated at 37°C overnight. The following day colonies were counted on both the unexposed and exposed plates. The ratio of colonies appearing on the exposed versus the unexposed plates gives an estimate of UV survival. The assays were replicated four times for day 60 and 80. A separate, single assay was also conducted directly from day 39 cultures during a periodic check of the populations for contamination. The assay procedure was the same as described above.

Measuring competitive ability under UV selective conditions

Competitive ability with UV exposure was assessed against a common competitor, as described above. To perform a competition assay, single colonies of both K12 and B were inoculated into LB liquid and grown overnight. The following day, both cultures were plated onto LB agar as described above. After 24 h of incubation, plugs were removed from both K12 and B plates and resuspended. Equal volumes of both competitors (0.5 mL) were added to 1.5-mL microcentrifuge tubes and vortexed. The resulting mixed cell suspension was diluted 106 and a 100-μL aliquot was plated onto TA to determine initial density. Another 100-μL aliquot was plated onto LB agar and exposed to UV and incubated as described above. The following day, plugs were removed and resuspended from the agar and plated onto TA and incubated overnight, and colonies were counted after another 24 h. Three replicate assays were conducted for day 60 fitness. Competitive ability was quantified via the selection rate constant (Lenski et al. 1991; Travisano and Lenski 1996) as the difference in Malthusian parameters between the two competitors:


where F and I refer to the final and initial population estimates determined by plate counts. The difference in Malthusian parameters is less sensitive to sampling error than the ratio (relative fitness), which is particularly important when comparing strains having large differences in competitive ability (Travisano and Lenski 1996).

Measuring growth rates

A Thermo Labsystems BioScreen C microtiter plate reader was used to measure growth rates of UV and Control lines. Cultures grown to stationary phase overnight and used to establish subcultures of each line by transferring 100 μL into fresh LB. After 4 h, the cultures were subcultured again into the microwell plate, by aliquoting 4 μL into 400 μL of LB in each well. Each line was replicated in four wells, along with six replicates for the ancestor. The plates incubated with continuous shaking at 37°C and optical density (OD) measured every 10 min over 24 h. OD measurements were log-transformed and maximum growth rate was determined by performing a least-squares regression (log OD vs. time) over 120-min sliding windows (see Joseph and Hall 2004). For each replicate, the window with the largest slope was designated the maximum growth rate. Correlation coefficients associated with each maximum growth rate regression ranged from 0.988 to 0.999.

Measuring survival with longer UV exposures

Replicates stored at day 60 were revived as described above. The following day, 100-μL aliquots were plated on LB agar, both diluted and undiluted. Plates were exposed to UV for various times from 5 to 180 sec and incubated wrapped in aluminum foil as described previously. After 24 h, colonies were counted on the dilution/exposure series and the initial density plates, and the ratio of survivors to initial density was used to obtain a dose survival curve.


Measuring cell size

Each day 60 replicate was grown to stationary phase overnight, from which 1 mL was removed and added to 2 mL of PBS buffer. One microliter of a 100 μM solution of BacLight Green (Molecular Probes, Carlsbad, CA.) was then added to the cell suspensions and the cells were stained for 15 min. The cultures were then sampled in a Becton Dickinson FACSCalibur flow cytometer. Green fluorescence was measured using the FL1 channel (530 ± 15 nm) and the 488-nm blue laser. Relative cellular sizes were estimated by using the forward scatter (FSC) signal. Validation of FSC as a measure of cell size was done by differential interference contrast (DIC) microscopy.

Starvation during stationary phase

Three colonies from each day 60 line were used to inoculate separate 25-mm tubes with 10 mL of LB, for a total of 72 tubes. The tubes were incubated with shaking at 37°C for 24 h and then plated by dilution on LB agar. The tubes were placed back into the incubator for another 24 h and sampled again. This was repeated at 72 and 96 h. Replicates were plated daily and the surviving fraction was estimated by the ratio of CFUs from the later time point relative to that after the initial 24 h of growth. The bacteria showed exponential declines in viability, and the data were analyzed by analysis of covariance (ANCOVA) on log and arcsine transformed data.

Desiccation survival

Overnight cultures of day 60 lines were obtained as described above. On the following day, 20-μL droplets of each of the 24 cultures were deposited at the bottom of sterile 13-mm tubes, and another 100-μL aliquot was diluted by 106 to estimate the initial density of the culture. The 20-μL droplets were incubated for 24 h at 37°C, after which they had completely dried, as verified by visual inspection. Dried cells were resuspended in 1 mL of saline and vortexed. The suspension was then diluted by 104 and plated onto LB agar and incubated overnight. The percentage of survivors was used to estimate desiccation resistance. The assay was replicated three times.

The four replicates of each day 60 time point line were assessed for the surviving fraction over 1 day of desiccation at 37°C (with four missing values). The data were analyzed by analysis of variance (ANOVA) arcsine-transformed data.

Salinity survival

Day 60 cultures were grown to stationary phase overnight. From these cultures, 100 μL from each was inoculated into 9.9 mL of LB with 8% saline (LB broth with 80 g salt/L). One hundred microliters aliquots were diluted from the saline culture before incubation to measure initial density. After 24 h of incubation at 37°C and 120 rpm, 100-μL aliquots were diluted onto LB agar plates for final density counts. The difference between final and initial densities was used to calculate salinity survival. The assay was replicated three times. The data were analyzed by a Mann–Whitney U test due to the large difference in variance between the two groups. We log and arcsine transformed data, as above, to perform within-treatment group analyses.



UV-selected lines showed substantial adaptation to the UV selective conditions, with a greater than twofold increase in fitness. In contrast, after 60 days, there was no statistically significant change in fitness of the Control lines in the UV-selected environment, even though there were slight increases in fitness after 30 days (Fig. 1). Single colonies isolated from UV and Control lines were competed against the E. coli B common competitor under the UV selective conditions. The evolutionary responses differ between the two groups (P < 10−5) as compared by ANCOVA with a fixed intercept to the initial fitness. These differences in the UV environment occurred despite the smaller minimum population size of the UV-selected lines. The strongly divergent differences between the two treatment groups by 30 days, persisting through 60 days, indicate that differences in the number of generations likely had no more than a small effect on fitness in the UV environment. No differences were observed between replicate lineages (P > 0.5) nested with selective history as a random factor, while both time (P < 10−4) and time2 (P < 0.005) were statistically significant. The absence of differences within the UV treatment suggests that sufficient beneficial mutations were present to result in a consistent evolutionary response. The changes in fitness differ strongly between the two groups, but the temporal dynamic of the responses over the course of the experiment are similar. Both groups have curvilinear responses, as indicated by support for the time2 term above and by partial F analyses (Zar 1999, p. 453) for the quadratic time term for the UV-selected lines (F1,23= 4.25, P= 0.0507) and Control lines (F1,23= 13.98, P= 0.0011) when separate ANCOVAs are done for each group.

Figure 1.

Competitive fitness of selected lines relative to a common competitor in the UV selective environment. UV-selected lines (squares) increase in fitness, whereas Control lines (triangles) have no directional responses over 60 days of selection from the initial value (circle). Prior to selection, the unselected ancestral genotype had a fitness of 1.47 (± 0.19, SEM = 0.087, df = 11) relative to the E. coli B competitor. The evolutionary responses differ between the two groups (F1,22= 35.4, P < 10−5) as compared by ANCOVA with a fixed intercept to the initial fitness. Both time (F1,47= 23.75, P < 10−4) and time2 (F1,47= 10.27, P < 0.005) were statistically significant. No differences were observed between replicate lineages (F22,47= 0.82, P > 0.5) nested with selective history as a random factor. Estimates are means of the 12 respective lineages and confidence intervals are 95%, determined by a t distribution with n− 1 = 11 df for each point.


UV survival

After 60 days of selection, UV and Control lines were assayed to determine the extent of UV resistance beyond the UV selection regime conditions. The UV-selected lines have a mean survival of 25% after 40 sec of exposure, which declines to an average of 5% with an additional 10 sec of exposure (Fig. 2). The Control lines did not have any detectable survival at either 40 or 50 sec of exposure (not shown).

Figure 2.

Survival of ancestral E. coli K12 M6165 (circles) and means of 60 days UV-selected lines (squares) to UV-B. Regressions are back transformed quadratic fits to log arcsine transformed data (ancestor adj. r2= 0.92, selected adj. r2= 0.69). Error bars are 95% CI, although there is only a single value for the ancestral line with 45 sec exposure.

At the selected 30-sec exposure, UV survival increased in the UV-selected lines to three times the ancestral value, whereas the Control lines declined to two-thirds of the ancestral value (Fig. 3). The selective responses of the two groups differed (P < 10−13), as assessed by ANCOVA with fixed y-intercept to the ancestral value. There is no support for divergence among replicate lineages (P > 0.5) nested with selective history as a random factor. Statistically, significant effects were observed for time (P < 10−10) as a continuous factor and an interaction term between treatment (selective history) and time (P < 0.001). The relatively rapid decline in average UV survival of the Control lines (0.18% per day) suggests a trade-off between competitive fitness and UV resistance.

Figure 3.

Fraction of surviving cells after 25 sec of UV exposure. UV-selected lines (squares) increase in survival (P < 0.0001), whereas Control lines (triangles) have a steady decline (P < 0.0001) over 60 days of selection from the initial value (circle). Prior to selection, survival was 27.9% (± 2.76%, SEM = 0.0127, df = 14 − 1 = 13). The selective responses of the two groups differed (F1,24= 244.7, P < 10−13), as assessed by ANCOVA with fixed y-intercept to the ancestral value. Statistically significant effects for were observed for time (F1,35= 86.1, P < 10−10) as a continuous factor and an interaction term between selective history and time (F1,35= 13.1, P < 0.001). There is no support for divergence among replicate lineages (F22,35= 0.68, P > 0.5) nested with selective history as a random factor. Estimates of derived lines are means of the 12 respective lineages and confidence intervals are 95%, determined by a t distribution with n− 1 = 11 df for each point.


However, no decline in UV resistance is observed in the UV lines over the additional 20 days of selection in the absence of UV (F1,11= 0.014, P > 0.9). The absence of a detectable decline in UV resistance, from 60 days of UV selection through and additional 20 days of only plate selection, is an indicator of a lack of persistent strong trade-offs for fitness and UV resistance.


Maximal growth rate

We did not observe a statistically significant difference in maximal growth rate between the UV- and Control-selected lines (Table 1). Mean maximal growth is 0.000474 and 0.000488 log10OD/min for the UV and Control lines, respectively.

Table 1.  Effects of treatment and other factors on responses to selection.
 TreatmentaLineagebReplicatecTreatment x RepTimed
  1. aFixed factor, UV, and Control.

  2. bRandom factor, nested within treatment.

  3. cRandom factor.

  4. dContinuous factor.

 Cell size5818114.580.0009

Cell size

The UV-selected lines have, on average, larger stationary phase cell sizes than the Control lines. The geometric mean forward scatter for the UV-selected lines was 63.68 and mean FSC for the Control lines was 32.54. A single factor ANOVA on FSC (Table 1) shows that cell size is significantly different between the two groups (P < 0.001). FSC was validated as a proxy for cell size by direct cell size measurement of six lines that spanned the range of cell sizes, and cell length and FSC are highly correlated (correlation = 0.99).

Simultaneous assessment of fitness components

We simultaneously assessed the combined contribution of UV resistance, maximal growth rate and cell size on fitness, using path analysis regression (Sokal and Rohlf 1981; Maruyama 1998). We hypothesized that each of the three factors could directly affect fitness in the UV environment, and that there was also a potential cascade of effects on fitness: growth rate affecting cell size and cell size affecting UV resistance. A cascade of effects could in principle result in indirect effects on fitness, which is supported by the path analysis (Fig. 4). Both cell size and UV resistance have statistically significant effects, although the impact of cell size is mediated via UV resistance. No impact of maximal growth rate, directly or indirectly, is statistically supported. Statistical significance for the path coefficients was assessed by Jackknife of the z-transformed path correlation coefficients (Sokal and Rohlf 1981; Lenski and Service 1982).

Figure 4.

Path analysis regression of effects on fitness of the selected strains in the UV selection environment. Path analysis partitions the total effect of a trait into direct and indirect components. Here, direct effects are shown by arrows pointing to fitness and indirect effects are shown by arrows pointing to other traits. UV resistance and cell size have statistically significant direct and indirect, respectively, effects on fitness.



UV-selected lines have greater survival during desiccation (P= 0.008, Table 1). Mean desiccation resistance for all UV lines was 33.7% (± 6.6%, df = 11), whereas the Control lines have an average desiccation survival of 13.9% (± 1.8%, df = 11). Significant genetic variation in viability was detected within treatments, due to variation among UV lines (F11,32= 3.41, P= 0.003) and not Control lines (F11,30= 0.569, P= 0.839).


UV-selected lines have greater survival during starvation in stationary phase than Control lines (P= 0.009, Table 1). All lines declined in viability but UV lines died slower than Control lines (Table 1). Significant genetic variation in viability was detected within treatments, due to variation among UV lines (F11,94= 9.45, P < 0.001) and not Control lines (F11,94= 0.642, P= 0.786).


UV-selected lines have substantially poorer survival during saline exposure (Table 1). All lines decline dramatically in viability in the saline environment, but all the UV lines suffer greater mortality than any of the Control lines, indicating a much greater UV line susceptibility to killing by saline exposure (Mann–Whitney U12,12= 144, P < 0.001). Mean saline resistance for UV lines was 0.018% (median 0.014%), whereas the Control lines have a mean survival of 0.16% (median 0.17%). Significant genetic variation in viability is detected within treatments, largely due to variation among Control lines (F11,22= 2.45, P= 0.035) and not UV lines (F11,22= 1.57, P= 0.176).

Are the evolutionary responses correlated?

Six correlations are statistically significant prior to Bonferroni correction, and two, saline/UV resistance and starvation/UV resistance, remain statistically significant after correction for conducting multiple nonindependent simultaneous tests (Table 2). None of the expectations were fully supported, possibly indicating a mixed adaptive response involving multiple mechanisms. The negative correlation between growth and starvation resistance suggests adaptation to induction of growth after damage, which could be deleterious in starvation conditions.

Table 2.  Correlation coefficients of UV and Control lineages. We assessed the effects of the fitness components on three unselected abiotic resistances. Italicized values are statistically significant prior to sequential Bonferroni correction, and bolded values are significant after correction.
TraitCorrelation with
Cell sizeGrowth rateUV
 Desiccation  0.52−0.014  0.44
 Saline−0.49  0.13−0.75
 Starvation  0.26−0.51  0.62

How much of the phenotypic variation among selected lines is associated with treatment differences?

A principal components analysis on correlations, which included all phenotypic assays made on the day 60 isolates (Sokal and Rohlf 1981), clearly differentiates the two treatment groups along the first axis (Fig. 5). This indicates that the differences in selective histories between the two treatment groups had a significant effect on their overall phenotypes. A t-test on the first principal component values confirms the distinct clustering of the two groups (t22= 11.38, P < 10−8), but there is no statistical support for differences between the two groups along either the second (P > 0.8) or third (P > 0.3) principle component axes. The first axis includes 51.6% of the variation among lines, indicating that half of the phenotypic differences are associated with UV selection.

Figure 5.

Principal components presentation of the day 60 selection isolates. All phenotypic assays were included, and the two axis contributing the most variance among the isolates are shown (51.6% and 19.8%, respectively). Along PC1, the UV-selected lines (squares) form a clearly disjoint cluster relative to the Control lines (triangles) and the difference is supported by a t test (df = 22, P < 10−8).

How well does a causal model based upon the path analysis fit the direct and correlated responses to selection?

The path analysis regression of the a priori model shows statistical significance for an indirect effect of stationary phase cell size and a direct effect of UV resistance. We used a posteriori exploratory model building to expand understanding of the causes and consequences of UV adaptation. We performed exploratory model building on fitness and the three correlated abiotic resistance traits (desiccation, saline, and starvation resistance). We use the Akaike Information Criterion (AIC, Burnham and Anderson 2001) on a mixed stepwise regression of full factorial models of cell size, growth rate, selective treatment, and UV resistance. AIC model selection trades-off the number of parameters with model likelihood, thereby resulting in a simple model with a potentially high likelihood. After AIC evaluation, the reduced models were statistically assessed (Table 3). The model for fitness is consistent with our path analysis model by including cell size, growth rate, and UV resistance (Fig. 4), and also includes an interaction term for cell size and UV resistance (Table 3, Column 2): the fitness benefit of increased cell size declines with increased UV resistance. This suggests that there may have been multiple mechanisms for adaptation to the UV selective environment. It is notable that the model does not include the selection treatment (UV vs. Control) as a factor, indicating that the model sufficiently captures the diversity of responses with the factors already in the model and without reference to the selective histories of the lines. Addition of a "Treatment" factor to the AIC model does not substantially increase the adjusted R2 value, from 0.69 to 0.70, and the Treatment factor itself would not be statistically supported (P= 0.199).

Table 3.  Models for fitness and three correlated responses, identified by Akaike Information Criteria (AIC). Different models best satisfied AIC, each of which is largely supported by subsequent statistical analysis. Only factors included in at least one of the AIC models are listed.
Fitness P-valueDesiccation P-valueSaline P-valueStarvation P-value
Cell size0.00030.05210.00410.008
Growth rate0.0939 0.00050.0012
UV resistance0.3030.0055 0.31
Cell size×Growth0.0004
Cell size×Treatment  0.0035 
Cell size×UV0.0008
Growth Treatment  0.0006 
Treatment×UV   0.0363
Cell size×Growth×Treatment0.0004
Complete model2.08×10−58.37×10−43.27×10−53.53×10−6
Adjusted R20.690.490.760.80

In contrast, treatment is included in all three AIC models for the correlated abiotic resistance traits. The AIC models for both starvation and saline resistance have high adjusted R2 values, 0.80 and 0.76, respectively. Both are complex models with multiple interaction terms. Treatment occurs in these models due to interaction effects, having no support as a single factor. As suggested by the correlation analysis (Table 2), cell size has a significant impact on both resistances, but in opposing directions. Another contrast with the fitness model is the impact of growth rate, both as a main effect and as an interaction term. The model for desiccation resistance is the simplest, but is also the least informative. The adjusted R2 relatively low (0.49), the significant treatment effect indicates that unidentified factors associated with the treatment had large impacts on desiccation resistance.


Exposure to UV light is common to much of life and can cause cellular damage (and mortality) to both uni- and multicellular life. For most of Earth's history, life was predominantly microbial, especially before the development of significant atmospheric shielding. This early microbial life was frequently exposed to potentially lethal UV damage (Rothschild and Cockell 1999; Cockell 2000). Despite atmospheric shielding, UV exposure can still cause substantial damage and mortality to microbes. Moreover, it is possible that anthropogenic changes to stratospheric ozone and climate may lead to increased UV exposure and damage to some species (Boucher 2010). UV is also of particular interest in astrobiology; recent evidence of water on Mars (Smith et al. 2009; Whiteway et al. 2009) increases the possibility of finding microbial life on or near the Martian surface. Understanding how extant Earth microbes adapt to high UV irradiance is vital both for understanding the possibility of Earth-to-Mars contamination (Berry et al. 2010), as well as determining where else we might look for life in the solar system.

To investigate adaptation to severe environmental conditions, we exposed replicate populations of E. coli to acute daily UV exposure. All UV-selected populations rapidly developed resistance to UV in less than 600 generations and increased in fitness in an environment with daily UV exposure. Increased UV survival was a major component of adaptation to the UV environment and an indirect contribution of stationary phase cell size to increased fitness was also detected. We observe abundant correlated responses to selection. The UV-selected populations increased in starvation and desiccation resistance, but there was not a general trend for increased resistances. Relative to the ancestral genotype and Control populations, the UV populations are more sensitive to saline conditions. Resistances to other abiotic stresses were correlated with fitness components underlying the response to UV selection. The observation of parallel direct and correlated responses to selection suggests that the underlying physiological mechanisms of adaptation to the UV selective environment were largely common across lines. This degree of parallelism is typical only for traits tightly associated with fitness (Vasi et al. 1994; Travisano and Lenski 1996).


Most studies of microbial resistance to harsh abiotic conditions have emphasized active enzymatic mechanisms, such as increased expression of heat shock proteins (Guisbert et al. 2008), cold shock proteins (Phadtare 2004), efflux pumps (Li and Nikaido 2004), ion scavenging enzymes (Imlay 2002), and many other factors. We anticipated that increased UV resistance could occur via increased DNA or other cellular repair pathways. Indeed, Alcantara-Diaz et al. (2004) exposed five lines of E. coli to periodically increasing amounts of UV and observed that resistance was due to mutations in DNA repair and replication genes. Their experiment differed from our current work in that the cells were exposed in liquid culture during early stationary phase and the UV dose was increased after every 10 exposures. Moreover, they did not measure fitness, nor look for correlated responses. Also we limited the range of possible repair mechanisms, by excluding photoreactive repair, to reduce the scope of diverse selective responses.

We did observe increased UV resistance, but did not observe the anticipated pattern of correlated responses. Prior studies have observed a positive correlation between resistance mechanisms (e.g., Fontaine et al. 2008), due to common regulation of stress resistance pathways. Importantly, we did not observe an absence of correlations, but clearly opposing responses, even after statistically conservative sequential Bonferroni correction, indicating divergent correlated responses to differing stresses. This pattern of responses suggests that increased DNA repair was, at best, only one of multiple mechanisms underlying adaptation to the selective environment.

Resistance to damage can potentially occur via structural changes to the cell, in addition to active enzymatic mechanisms. Several previous experiments have observed cell size increases as a response to UV and ionizing radiation (Zelle and Ogg 1957; Idziak and Thatcher 1964; Licciardello et al. 1969). This relationship is not unique to laboratory-adapted organisms as Karentz et al. (1991) demonstrated a positive correlation between cell size and UV resistance in natural isolates of marine diatoms. They observed that smaller surface area to volume ratio in diatoms led to less damage per unit of DNA. Another possible benefit of large cell size could be an increase in cellular redundancy (e.g., extra chromosomal copies or repair enzymes). Davies and Sinskey (1973) subjected Salmonella typhimurium to repeated rounds of UV exposure and reported that the resultant larger UV resistant cells had approximately two times more RNA and protein than the ancestral cells, but no difference in DNA content.

Our evidence for a beneficial effect of stationary phase cell size in a UV selective environment is circumstantial, but strong. Unfortunately, manipulating cell size directly is technically challenging (Mongold and Lenski 1996). We partially overcame this difficulty by using a path analysis regression approach, to determine the impact of cell size within the factors thought to be potentially important for resistance. The observation of a significant impact of cell size is consistent with the standing hypotheses on a potential beneficial impact and lends support for a structural protective role. Moreover, the pattern of correlation of cell size with abiotic other stresses besides UV, positive with desiccation and starvation (albeit weakly) but negatively with saline, are also consistent with a structural mechanism. Finally, the consistency of the AIC model with the path analysis approach is also supportive, in that the AIC model identifies cell size and an interaction term of cell size with UV resistance as the two supported factors affecting fitness in the UV selective environment. The absence of a correlation between growth rate and cell size is not surprising, even though numerous studies have demonstrated that cell size increases with growth rate (Schaechter et al. 1958), as our measures of cell size were undertaken with stationary phase nongrowing cells, under conditions similar to those in which the UV selection was performed.


Natural variation in UV resistance among microbes could result from direct costs of UV resistance on competitive fitness and differences in the frequency of exposure. If UV resistance directly engenders fitness costs in the absence of UV exposure, then reductions in UV resistance could be a beneficial evolutionary response for microbial species that rarely experience intense UV exposure. Variation in UV resistance could also arise from its effect on other traits besides competitive fitness, and it is this possibility that is supported by our results. Broadly, we investigated the potential for trade-offs among abiotic resistances traits to lead to variation in levels of resistance for those traits. In particular, we observed that increased UV resistance, in E. coli, engenders substantially increased sensitivity to saline environments.

Previous studies have found fitness costs to biotic and abiotic resistances in microbes, but the costs themselves can be extremely small, evolutionarily labile, and highly environmentally dependent (Lenski 1988; Bennett and Lenski 2007; Quance and Travisano 2009). Most trade-offs in microbial niche breadth have focused on exploitative traits (Rainey and Travisano 1998; Brockhurst et al. 2007), and not resistance to abiotic stress. There is evidence that UV resistance in some microbes arises as byproduct of desiccation resistance (Mrazek 2002), although it is unclear if this is a general consequence.

There was little evidence of a strong persistent trade-off between UV resistance and competitive fitness in the selective environment. The decline in UV resistance of the Control lines was surprisingly rapid, 0.18% per generation, and suggests a direct fitness cost. In contrast, no statistically significant decline in UV resistance was observed in the 20-day follow-up selection of the UV lines in the absence of UV selection. This pattern of responses is consistent with a cost of UV resistance that is readily ameliorated during selection (Schrag et al. 1997). In addition, the rapid and essentially linear increase in UV resistance of the UV-selected lines does not support a fitness trade-off, because increases in resistance were readily achievable. This is especially true, given the soft-selection experimental regime. Although the UV exposure itself was for survival, hard selection, the surviving populations were then allowed to grow up to the maximal population size prior to transfer to fresh medium and repeated UV exposure. Hence, the fitness benefits of UV resistance should decline with increasing resistance, as the benefits are scaled to total growth rather than survival (see the fitness calculation in Methods). Taken together, these results suggest that trade-offs between UV resistance and fitness in the selective environment are evolutionary labile and readily ameliorated.

However, we observed that the evolution of increased UV resistance has negative fitness consequences in other abiotic environments. All the UV-selected lines were far more sensitive to saline conditions, with an average survival of one-tenth that of the Control-selected lines. We hypothesize that increased cell size is the mechanism by which increased saline sensitivity arises, as osmotic stress increases with cell size. The volume of a completely spherical cell (inline image) increases as the cube of cell radius (r), whereas cell surface area (r2) increases only as square of the cell radius, so that osmotic stresses (= pressure per unit area) on the cytoplasmic membrane increase with cell size. Although E. coli cells are not spherical, increases in osmotic pressure occur generally for increases in cell size. The role of cell size on the direct and correlated responses to selection is supported by the correlational and exploratory AIC analyses. Additional experiments are required to firmly demonstrate a trade-off between UV and saline resistance, but a cell size mediated mechanism suggests that the trade-off may be recalcitrant to evolutionary modification. Our results therefore suggest that trade-offs do limit the extent of abiotic resistances but are among resistance to different abiotic stresses.

The exploratory models for the basis of the direct and correlated responses are surprisingly informative. This is most evident between desiccation and starvation resistance, both of which increase in the UV-selected lines. The models are largely disjoint from one another, even though the phenotypic responses are similar, suggesting complex responses. We anticipated that if there were a common physiological basis for both correlated responses that it would be reflected in the exploratory models. An obvious limitation of this approach is that the factors included in the model may not be causally associated with the traits under examination. This concern is greatest for desiccation resistance, in which the resulting model accounts for roughly half of the trait variation. As such, some of the benefits of these models are in planning future research. However, the models do provide immediate insight on two additional issues. We observe support for some commonality in the physiological basis for increased starvation resistance and saline sensitively in the UV lines, but also divergence in the consequences potentially arising from different physiological interactions. The main effects are shared between the two, but the six total interaction terms included in the two models are completely nonoverlapping. We also observe that selection history, distinguished by the two treatment groups, was not a significant main effect in most of the models. The phenotypic consequences of UV selection, particularly for fitness in the UV environment, can largely be explained by the phenotypic factors included in the model, without reference to the selection regime.

The pattern of correlated responses may be surprising given the complex adaptive dynamics observed. Both UV and Control treatment groups show quadratic evolutionary responses, the Control group apparently increasing and then decreasing in fitness. The complexity of fitness responses is likely due to at least two factors. The experimental selection was carried out on agar plates to standardize UV exposure. Agar plates are spatially structured environments, and spatial structure can promote diversity (Rainey and Travisano 1998), particularly by antagonistic interference competition (Chao and Levin 1981; Czárán et al. 2002; Greig and Travisano 2008), which can lead to nontransitive fitness interactions (Kerr et al. 2002). We hypothesize that the absence of any significant differences between treatment groups for maximal growth rate, and the apparent importance of growth rate on starvation, results from adaptation to the structural environment. We also performed our competition experiments with a closely related, but nonancestral, bacteria: E. coli B. Use of E. coli B as a common competitor allowed us to perform competition experiments without introducing markers and their potential confounding effects to our selected genotypes. If E. coli B was particularly sensitive to evolutionary changes in the selected lines, the fitness responses could have been magnified or reduced. That the direct and correlated responses were readily observable, regardless of the ecological complexity of the selective environment, is an indication of the robustness of the responses.


Populations of an enteric bacterial species are readily able to evolve increased resistance to UV light, a commonly experienced abiotic source of mortality. Decreased UV mortality was detected; involving increased stationary phase cell size. There was no persistent fitness cost to UV resistance in the UV selective environment. Adaptation to the UV selective environment engendered strong correlated responses to selection to other abiotic stresses. UV selection preadapted lineages to starvation and desiccation, but simultaneously caused an increased sensitivity to saline-induced mortality. Variation to UV sensitivity may therefore arise as a trade-off for resistance to other abiotic stresses. Future work will investigate the mechanistic basis of increased UV resistance, the relationships in abiotic resistance suggested by the exploratory model analyses.

Associate Editor: C. Burch


We thank D. Hall, S. Joseph, M. Lunzer, W. Ratcliff, and M. Sanders for technical advice. A. Escalante and R.F. Denison provided valuable encouragement and criticism, as well as four helpful anonymous reviewers. This work was funded by the Texas Space Grant Consortium, the Environmental Institute of Houston, the Houston Coastal Center, and the U.S. National Science Foundation.