Antagonistic Pleiotropy Is Unexpectedly Rare In New Mutations

Pleiotropic effects of mutations may underlie diverse biological phenomena such as ageing and specialization. In particular, antagonistic pleiotropy (“AP”: when a mutation has opposite fitness effects in different environments) generates tradeoffs, which may constrain adaptation. Models of adaptation typically assume that AP is common – especially among large-effect mutations – and that pleiotropic effect sizes are positively correlated. The rare empirical tests of these assumptions have largely focused on beneficial mutations observed under strong selection, whereas most mutations are actually deleterious or neutral, and are removed by selection. We quantified the incidence, nature and effect size of pleiotropy for carbon utilization across 80 single mutations in Escherichia coli that arose under mutation accumulation (i.e. weak selection). Although ~46% of the mutations were pleiotropic, only 11% showed AP, which is lower than expected given the distributions of fitness effects for each resource. In some environments, AP was more common in large-effect mutations (but not synergistic pleiotropy, SP); whereas pleiotropic effect sizes were positively correlated for SP (but not AP). Thus, AP is generally rare, is not consistently enriched in large-effect mutations, and often involves weakly deleterious antagonistic effects. Our unbiased quantification of mutational effects therefore suggests that antagonistic pleiotropy is unlikely to cause maladaptive tradeoffs.


INTRODUCTION 31
Biologists have long observed that organisms maximize resource allocation to one trait while 32 compromising allocation to another trait (Goethe) (Lenoir 1984). Such tradeoffs manifest as negative 33 correlations between traits, and may constrain evolution by limiting the breadth of phenotypes 34 available to organisms (Rees 1993). The nature and strength of tradeoffs between traits can thus 35 dictate whether organisms evolve to be generalists or specialists (Ferenci 2016). Tradeoffs also 36 underlie diverse biological phenomena such as life-history strategies (Zera and Harshman 2001;Sgrò 37 and Hoffmann 2004), ageing (Kirkwood 2005), and assembly of microbial communities and host-38 microbe interactions (Litchman et al. 2015). Although tradeoffs in resource allocation are undeniable, 39 they remain relatively poorly understood at the mechanistic level. Tradeoffs can occur when multiple 40 neutral or deleterious mutations accumulate and degrade traits under weak selection, leading to a 41 negative correlation with other traits evolving under positive selection (Elena and Lenski 2003). For 42 instance, in Lenski's long term experimental evolution lines, bacteria evolving under strong selection 43 for one metabolic function (growth on glucose) lost multiple other metabolic functions because 44 selection on these traits was very weak, allowing deleterious mutations to accumulate (Cooper 2014;45 Leiby and Marx 2014). Alternatively, tradeoffs may occur when a single mutation increases fitness in 46 a specific environment (or trait), simultaneously reducing fitness in alternate environments (or a 47 second trait) (Cooper and Lenski 2000). Such mutations are antagonistically pleiotropic for the two 48 traits or environments, and the phenomenon is called antagonistic pleiotropy (henceforth "AP"). 49

50
The evolutionary impact of AP clearly depends on its incidence and magnitude. If AP is frequent or 51 involves large-effect mutations, the resulting tradeoffs are more likely to constrain adaptation. 52 Historically, models of adaptive evolution have assumed that AP is the predominant form of 53 pleiotropy (Otto 2004, Lande 1983, implying that synergistic pleiotropy (SP; when a mutation 54 simultaneously either increases or decreases fitness in two different environments) is relatively 55 uncommon. However, for single beneficial mutations in Escherichia coli, AP between fitness on 56 glucose and alternate carbon sources was rare compared to positive SP (Ostrowski et al. 2005). 57 Similarly, most of the first-step beneficial mutations isolated from laboratory-evolved E. coli 58 populations showed SP, while only a few were strongly antagonistically pleiotropic (Dillon et al. 59 2016). Thus, contrary to model assumptions, empirical data suggest that AP may not be the 60 predominant form of pleiotropy. A second assumption of theoretical models is that large effect 61 mutations are more predisposed to show AP (Lande 1983), potentially explaining the prevalence of 62 small effect mutations during adaptation in natural populations (Lande 1983;Orr 1992;Dillon et al. 63 2016). However, no empirical study has explicitly tested this assumption. Finally, the pleiotropic 64 effect size of mutations is assumed to be proportional to their fitness effect in the selective 65 environment where the mutation arose, i.e. its primary effect size (Orr and Coyne 1992). Contrary to 66 this assumption, previous studies found that the antagonistic effect size was not correlated with the 67 primary effect size (Ostrowski et al. 2005;Dillon et al. 2016). Taken together, empirical studies 68 indicate that SP is more common than AP, at least among beneficial mutations. Additionally, the 69 direct and pleiotropic effects of beneficial mutations appear to be positively correlated when 70 pleiotropy is synergistic, but not when pleiotropy is antagonistic. Thus, widely used models of 71 adaptive evolution make assumptions that are either empirically untested or are poorly supported. To obtain unbiased estimates of AP, we evolved replicate populations of E. coli under mutation 84 accumulation (henceforth "MA") for multiple generations on a rich medium (Fig. 1). This regime of 85 experimental evolution minimizes the strength of selection due to repeated bottlenecking of the 86 populations, allowing all but lethal mutations to accumulate. We sequenced several time points frozen 87 during experimental evolution to identify lines that had a single mutation relative to their immediate 88 ancestor. Across 38 MA lines, we identified 80 isolates carrying new single mutations (including 89 single nucleotide changes and small indels < 10 bp; henceforth "mutants") relative to their immediate 90 ancestor. To determine the incidence of AP (i.e. the proportion of mutants that showed increased 91 fitness on resource A and decreased fitness on resource B), we measured the growth rate of each of 92 these mutants and their respective mutational ancestors on 11 different carbon sources. For each pair 93 of resources, we compared the observed incidence of AP with null distributions generated by 94 randomly sampling from the independent DFEs for each resource (Fig 1). We find that while 95 pleiotropy is not rare among new mutations, AP is quite uncommon and variable across resources, 96 even when compared to the null distribution. Although the incidence of AP often increases with the 97 effect size of the mutation, the form of the relationship varies across resources. Finally, we find that 98 the fitness effect sizes of mutations showing AP are either uncorrelated or negatively correlated. 99 Taken together, our results suggest that AP is more rare than previously thought, indicating that AP-100 mediated tradeoffs are generally unlikely to constrain adaptation. 101

Bacterial strains and media 105
We obtained the wild-type strain of Escherichia coli K-12 MG1655 from the Coli Genetic Stock 106 Centre (CGSC, Yale University). We streaked this strain on an LB agar plate, picked one colony at 107 random, made glycerol stocks, and used this isolate as the wild-type (WT) in all subsequent 108 experiments. For fitness assays, we used liquid culture media: LB broth (Miller, Difco), or M9 109 minimal salts medium + 5mM of a carbon source (glucose, trehalose, fructose, maltose, lactose, 110 galactose, succinate, pyruvate, melibiose, malate, fumarate; Sigma-Aldrich). We inoculated 2µL of each frozen stock (and ancestors of each strain) in 2mL LB and allowed the 125 cells to grow overnight at 37ºC with shaking at 200rpm. We extracted genomic DNA using the 126 GenElute Bacterial Genomic DNA kit (Sigma-Aldrich), quantified genomic DNA using the Qubit HS 127 dsDNA assay (Invitrogen), and prepared genomic DNA libraries for each isolate using the Nextera 128 XT DNA Library Preparation Kit (Illumina), using the manufacturer's instructions in each case. We 129 sequenced libraries on the Illumina Hi-seq 2500 platform using either 2x100bp paired-end reaction 130 chemistry (giving an average coverage of 150x per genome; range 102x to 240x) or the 1x100 single-131 end reaction chemistry (giving an average coverage of 70x per genome; range 21x to 140x). We 132 discarded reads with quality scores less than Q30, retaining >95% reads per genome. We aligned 133 filtered reads to the NCBI reference Escherichia coli K-12 MG1655 genome (RefSeq accession ID 134 GCA_000005845.2) using the Burrows-Wheeler short-read alignment tool BWA (Li and Durbin We 138 discarded all mutations that occurred at <80% frequency or were supported by <10 reads on both 139 strands. Compared to the NCBI reference genome, our WT ancestor contained one SNP (single 140 nucleotide change) and one indel that were present in all evolved MA lines, and were hence discarded 141 from further analysis. We did not find any long indels in our sequenced isolates. The mutations 142 observed in the MA lines will be described in detail in a separate publication. 143

144
For this study, we identified all isolates carrying a single mutation with respect to their immediate 145 ancestor. For instance, if a line had one mutation on day 39 and an additional mutation at day 200, we 146 retained both these isolates for further analysis, but discarded intermediate time points (days 104 and 147 140) since they did not represent single mutational steps. In this case, we obtained two distinct single-148 mutation steps from the same line: for the evolved isolate at day 39, we considered the WT as the 149 ancestor; and for the evolved isolate at day 200, we considered the evolved isolate at day 39 as the 150 ancestor. If a line already had two mutations on day 39 (the first frozen time point) and four mutations 151 in the subsequent time point (day 104), we discarded all these isolates from further analysis since we 152 could not find single-mutation steps for this line. Details of the 80 isolates representing single 153 mutational steps ("mutants") are given in Table S1. 154 155

Measuring growth rate as a fitness measure 156
We measured the fitness of all mutants, along with their respective ancestors (see summary in Fig.  157 1B). We inoculated each mutant from its freezer stock into M9 minimal salts medium with 0.4% 158 glucose, kept at 37ºC with shaking at 200 rpm for 16 hours. We then inoculated 6µL of this culture 159 into 594µL growth media (LB broth or M9 minimal salts medium + 5mM carbon source) in 48-well 160 plates (Costar), incubated in a shaking tower (Liconic) at 37ºC. Plates were read by an automated 161 growth measurement system (Tecan, Austria) every 40 minutes for 18 hours. We tested the growth 162 rate of three technical replicates per line per carbon source; meaningful biological replicates could not 163 be obtained since we had a single colony at the end of each MA line transfer. In each 48 well plate, 164 we included the WT and ancestor of the MA lines being tested as controls, and we used these to 165 estimate variation in growth rates across plates and to calculate relative fitness of evolved isolates. We 166 estimated maximum absolute growth rates using the Curve Fitter software (Delaney et al. 2013). For a 167 subset of 40 mutants, we repeated fitness assays in glucose, galactose, and pyruvate to ensure that 168 growth rates were consistent across independent runs (Fig. S1). 169 170

Estimating fitness effect sizes and calculating observed and null estimates of AP and SP 171
For each mutant,we calculated relative fitness as: (Growth rate of mutant -Growth rate of 172 ancestor)/Growth rate of ancestor (see Fig. 1B), using the average for three technical replicates. A 173 negative value indicated that growth rate had decreased compared to the ancestor, while a positive 174 value indicated increased growth rate compared to the ancestor. Growth rates for WT measured in 175 different plates run on different days varied by less than 5%. Similarly, the error in measurement of 176 growth rates across technical replicates (run on the same day) was also less than 5%. To account for 177 this variability, we considered mutants with <5% change in fitness from the ancestor as showing no 178 change. For each pair of carbon sources, we calculated the proportion of mutants showing evidence of 179 AP (relative growth rate <-0.05 in carbon source A but relative growth rate >0.05 in carbon source B) 180 or SP (relative growth rate <-0.05 in both carbon source A and carbon source B as synergistic 181 decreases in fitness; relative growth rate >0.05 in carbon source A and B as synergistic increases in 182 fitness) (Fig 1C). To determine the proportion of comparisons showing AP or SP for each focal 183 resource (Fig. 2), we calculated the total number of mutants showing AP or SP across all pairwise 184 combinations with the focal resource. Because there were 10 possible resource pairs for each focal 185 resource and 80 mutants for which AP or SP was measured, there were a total of 800 comparisons (10 186 resource pairs x 80 lines) for each isolate. Thus, we calculated the "observed" proportion of 187 showing AP or SP. We performed 1000 iterations of this process to generate a null distribution of the 195 incidence of AP or SP for each resource pair. When generating null distributions of proportions of 196 pleiotropy for beneficial mutations, for each resource pair (A and B), it was possible for a beneficial 197 mutations to occur in either resource A or resource B. We accordingly generated two separate null 198 distributions for each resource pair, leading to a total of 110 null distributions (Fig. 1D). For each null 199 distribution, we estimated the average proportion of AP (or SP) as the "expected" incidence of AP (or 200 SP), for comparison with the observed incidence of AP (or SP) for the specific resource pair (Fig. 1E). Similarly, to calculate the null expectation for the relationship between fitness effect size and 215 proportion of pleiotropy, we binned, as described above, fitness values randomly drawn from the 216 DFEs for individual resources. We measured the proportion of pleiotropy (AP or SP) within the null 217 distribution, and asked if it was correlated with the fitness effects for each of the 55 resource pairs. 218 219 Testing for a correlation between primary and pleiotropic fitness effect sizes 220 For each resource pair, we computed the Spearman's rank correlation between the magnitudes of 221 effect sizes (absolute values of relative fitness, as above) in the two resources, for all mutants that 222 showed pleiotropy (AP or SP, as appropriate). We included fitness data for LB in this analysis, as LB 223 is the environment in which our MA lines evolved. Thus, for this analysis, the number of resources is 224 12, and the number of resource pairs is 66. We excluded those resource pairs from analysis in which 225 less than 5 mutants showed the specific type of pleiotropy. Since AP is rare, we could therefore 226 compute effect-size correlations for 50 of 66 resource pairs. For SP, we computed effect-size 227 correlations for all 66 resource pairs. 228

Antagonistic pleiotropy is rare, and varies across environments 231
To estimate the incidence of pleiotropy, we measured the fitness effect (growth rate) of single 232 mutations obtained during an MA experiment, on 11 different carbon sources (Fig. 1). As expected, 233 the distribution of fitness effects (DFEs) observed for each resource showed that on average, ~49% of 234 all sampled mutations were deleterious, and would have been missed if we focused only on beneficial 235 mutations (Fig. S2). Combining data across all mutants and resource pairs (80 mutants x 55 resource 236 pairs = 4400 data points), we observed pleiotropy in ~46% of the cases (Fig. 2). However, most 237 pleiotropic mutations were synergistic (SP, ~35% of total) rather than antagonistic (AP, ~11%). 238 Importantly, resource identity had a significant impact on the incidence of both AP and SP ( Fig. 2; Tables S2 and S3; also see Tables S4 and S5 for 240 all pairwise resource comparisons). Malate had the highest incidence of AP (~23%), while melibiose 241 showed the highest incidence of SP (50%). Overall, AP was relatively rare compared to SP. 242 243 Another way to quantify the incidence of pleiotropy is to ask whether a given mutation shows 244 pleiotropy across multiple resource pairs. Most mutations (72 of 80) showed AP for at least one pair 245 of resources, with a median of 6 and a range of 0-24 resource pairs (out of 55 total resource pairs; Fig  246   S3). In contrast, all mutants showed SP for at least one resource pair, with a median of 16 resource 247 pairs (Fig. S3). These results again highlight the relative rarity of AP compared to SP. The relatively 248 high frequency of SP suggests that the paucity of AP cannot be explained by a general inability to 249 simultaneously detect small, pleiotropic fitness effects in multiple environments. 250 251 Finally, we compared the observed incidence of AP and SP with the null expectation derived from 252 DFEs for each resource in a given resource pair combination (Fig. 1C-E). Using random, repeated 253 sampling from observed DFEs for each resource pair, we estimated that the expected incidence of AP 254 was ~16 % (average across all resource pairs; Fig. S6); this is greater than the observed incidence of 255 ~11% described above. For each resource pair, we tested whether the observed proportion of mutants 256 showing AP was significantly greater or lower than expected from the null distribution for the specific 257 resource pair. We found that for most resource pairs (39 of 55), significantly fewer mutations showed 258 AP than expected by chance (Table 1; Fig S6). In contrast, in most cases SP was observed 259 significantly more often than expected (46 of 55 resource pairs;  S9). Together, these results reinforce our conclusion that AP is very rare in new 266 mutations. In contrast, SP is more common than expected, except when considering only beneficial 267 mutations. Overall, our results may explain why AP-mediated tradeoffs have been difficult to uncover 268 in empirical studies: AP is not only rare, but also depends on the environment. Theoretical models of adaptation assume that large-effect mutations are more commonly associated 272 with pleiotropic effects, and that these pleiotropic effects are mostly deleterious (Lande 1983). To test 273 this hypothesis, for each focal resource we grouped fitness effect sizes into four arbitrary classes: very 274 low (relative fitness 0.05 -0.1), low (relative fitness 0.1 -0.2), medium (relative fitness 0.2 -0.3), 275 and high (relative fitness 0.3 -0.4). Across all resources, ~37%, 45%, 14% and 4% of fitness effects 276 were classified in the respective classes. We then tested the relationship between the incidence of AP 277 and fitness effect size in two ways. Considering each focal resource in turn, we observed distinct relationships between the proportion of 281 AP and the mutational effect size. Four resources showed the predicted, monotonic positive 282 correlation (Kendall's rank correlation, p < 0.05; first row in Fig. 3A; Table S6); three resources 283 showed a concave positive relationship (second row in Fig. 3A); lactose showed a significant negative 284 correlation; and the remaining three resources did not show a significant correlation between the 285 incidence of AP and the fitness effect size. The correlation patterns for 7 of 11 resources supported 286 the prediction that large-effect mutations are more likely to show AP; but the form of this relationship 287 was not consistent across resources. Since a large fraction of mutations (37%) fall within the smallest 288 effect size class, the relatively low incidence of AP in this bin is consistent with the conclusion that 289 AP is generally rare. For SP, we observed more consistent relationships: the incidence of SP was 290 positively correlated with effect size class for 10 of 11 focal resources (Fig. S5A). 291 292 Next, we asked: conditional on the occurrence of AP, do antagonistically pleiotropic mutations occur 293 more frequently in large effect size classes? We again found variable patterns across resources: three 294 resources showed a monotonic or saturating increase (first row, Fig 3B); four resources showed a 295 convex relationship with highest AP incidence at intermediate fitness effect sizes (second row, Fig.  296 3B); and the remaining four resources showed no correlation (Table S8). In contrast, for datasets 297 generated from randomly sampling DFEs for each resource, we found that effect sizes were 298 consistently negatively correlated with the proportion of AP (Fig S10; Table S9). Thus, the observed 299 positive relationship between proportion of AP and effect size cannot be explained by a greater 300 chance of detecting AP in large-effect mutations. A similar analysis for SP showed that 4 of 11 301 resources showed a positive correlation between effect size and incidence of SP ( Fig. S5B; Table  302 S10), compared to the null expectation of a consistently negative correlation (Fig. S11, Table S11). 303 Thus, while the incidence of AP in observed mutations is often positively correlated with the fitness 304 effect size of those mutations, this pattern is not generally true for SP. 305 306 Together, these results offer partial support for the prediction that large-effect mutations may be more 307 like to show AP, with the caveat that the results vary dramatically across environments. For AP 308 involving glucose, we observed a consistent, strong positive correlation in both analyses (compare 309 Figs 3A and 3B), indicating that AP-mediated tradeoffs for glucose are more likely to occur for large-310 effect mutations. However, for other resources, the relationship between effect size and AP incidence 311 is either inconsistent, insignificant, or more complex with intermediate maxima or minima. Hence, 312 with respect to model assumption, this relationship is not robust and requires more careful attention. 313 314

Primary fitness effects sizes are correlated with synergistic, but not antagonistic effect sizes 315
We tested the relationship between primary and pleiotropic effect sizes for our set of random 316 mutations, measuring primary effect sizes in LB, the growth medium in which our MA lines evolved. 317 We measured secondary effect sizes in M9 minimal medium + 5 mM single carbon sources as above. 318 Contrary to expectation, we found that for AP, in most cases the primary fitness effect sizes (in LB) 319 were uncorrelated with the secondary effect sizes in specific carbon sources (bottom row, Fig. 4A; 320 Table S12). Thus, the magnitude of fitness change in LB is unrelated to fitness change in other 321 resources. For pairwise comparisons across single carbon sources, all significant correlations (25 of 322 39 possible comparisons; ~64%) were negative (Fig. 4A). Thus, a large benefit in one carbon source 323 was often associated with a small deleterious effect in another carbon source, or vice versa. Overall, 324 antagonistic pleiotropic mutations either do not exhibit correlated fitness effects, or show negatively 325 correlated fitness effects in different environments. Synergistic pleiotropic effect sizes were also 326 uncorrelated with primary effect sizes in LB ( Fig. 4B; Table S13), suggesting that changes in fitness 327 in a rich medium such as LB may generally not be related to fitness on individual carbon sources. 328 However, all other pairwise resource combinations were strongly positive (Fig. 4B), indicating that 329 large-effect beneficial (or deleterious) mutations in one carbon source also had a large benefit (or 330 disadvantage) in another carbon source. Thus, the predicted positive effect size correlations hold for 331 synergistic, but not antagonistic pleiotropic effects. Our results provide three clear lines of evidence suggesting that AP due to single mutations is unlikely 351 to be an important mechanism generating tradeoffs that hinder adaptation. First, we find that AP is 352 generally rare in new mutations. In fact, among beneficial mutations, AP is much rarer than expected, 353 indicating that beneficial mutations fixed during adaptation are unlikely to reduce fitness in other 354 environments. Previous studies also found that only 10-14% of ~20 beneficial mutations showed AP 355 (Ostrowski et al. 2005;Dillon et al. 2016). Second, we find that large-effect mutations are more likely 356 to show AP in some (but not all) environments. Hence, AP may impose a major constraint only in 357 specific environments and when adaptation involves large-effect mutations. Finally, we find that 358 antagonistically pleiotropic mutations often have negatively correlated fitness effects, such that a 359 highly beneficial mutation in one environment is only weakly deleterious in an alternate environment. 360 Thus, such mutations are unlikely to impose a significant fitness disadvantage in new habitats. 361 Together, our results contradict the prevalent idea that tradeoffs generated by AP may often constrain 362 adaptation. 363 364 Our analysis of 80 randomly sampled single mutational steps has several advantages over previous 365 studies. First, we determined the expected distribution of the proportion of AP given the underlying 366 distributions of fitness effects in different carbon sources, providing a general framework to determine 367 the occurrence of AP by chance alone. This null distribution allowed us to determine that the observed 368 proportion of AP is significantly lower than the expected proportion of AP for ~71% of all resource 369 pairs. A second advantage of our experiment is that we measured fitness effects in 11 distinct carbon 370 sources (55 resource pairs), which is a much larger set of environments than previous analyses. This 371 allowed us to detect many more instances of pleiotropy: all but eight of our mutants showed AP for at 372 least one pair of resources, and each mutant showed AP for a median of 6 resource pairs (out of 55). 373 Finally, since our lines evolved under very weak selection, we were able to explore not only highly 374 beneficial mutations, but the entire DFE for the occurrence of pleiotropy. This in turn allowed us to 375 measure pleiotropic effects of a large set of mutations, making it possible to empirically test the 376 relationship between fitness effect size and AP incidence. 377

378
We also note some important limitations of our work. First, to minimize false positive cases of 379 pleiotropy due to error in measuring growth rates, we considered that all mutations showing <5% 380 change from the ancestor were neutral. Effectively, we may have thus ignored mutations with effect 381 sizes <5%, potentially underestimating the incidence and effect sizes of antagonistically pleiotropic 382 mutations. However, this seems unlikely because we found that for many resources, small-effect 383 mutations are depleted in AP. Second, we measured the incidence and nature of pleiotropy only for 384 metabolic traits; specifically, for carbon utilization. Although we measured many more traits than 385 previous studies, this is still a small fraction of traits that are probably relevant for ecological and 386 evolutionary processes in bacteria. It is possible that antagonistic pleiotropy may be more frequent 387 across diverse traits, such as those related to metabolism vs. stress response. Despite these limitations, 388 our work represents the largest systematic analysis of single step mutational effects, and thus 389 represents an important test of long-held assumptions in evolutionary biology. 390

391
In summary, we provide new insights into the incidence, nature and effect sizes of pleiotropic 392 mutations. Although phenotypic tradeoffs clearly influence many biological processes, we suggest 393 that at the genetic level, tradeoffs are rarer than expected. Antagonistic pleiotropy is thought to 394 underlie the evolution and maintenance of generalists: AP may impose a cost of specialization on 395 resource specialists, such that in heterogeneous environments, generalists that do not pay this cost are 396 favoured (Cooper and Lenski 2000; Gompert and Messina 2016). Our results suggest that this broadly 397 intuitive explanation needs to be more nuanced, because the incidence of AP varies significantly 398 across environments. Thus, a generic "cost of specialization" cannot always explain the occurrence of 399 generalists, but may have explanatory power in specific heterogeneous environments that include 400 resource pairs showing high incidence of AP. We hope that empirical quantification of the incidence 401 and magnitude of AP across various organisms, environments, age classes, and genetic backgrounds 402 will provide further insights into these issues. Ultimately, we need to integrate across mechanistic and 403 phenotypic effects to better understand the role of tradeoffs in evolution.  Exp) when p < 0.05 for a Student's t-test comparing the observed proportion of AP or SP with the 504 mean of the null distribution for each resource pair (see Fig. 1C-E). Null distributions of the incidence 505 of AP and SP for each resource pair, and observed incidences, are shown in Fig. S7 and Fig. S8. resource. Colored blocks indicate the Spearman's rank correlation coefficient between a given focal 538 resource (x-axis) vs. all other resources (y-axis), for mutants that showed (A) AP or (B) SP between a 539 given pair of resources. In panel A, black blocks represent cases where correlations could not be 540 computed because very few isolates showed AP (<5). Asterisks indicate a significant correlation 541 between the magnitude of primary and pleiotropic effects (p < 0.05). Sample size, Spearman's rho, 542 and p values for each pairwise resource combination are given in Tables S12 and S13. 543 544 545