Evolution of age-specific decline in stress phenotypes is driven by both antagonistic pleiotropy and mutation accumulation

Efforts to more fully understand and test evolutionary theories of aging have produced distinct predictions for mutation accumulation (MA) and antagonistic pleiotropy (AP) mechanisms. We build on these predictions through the use of association mapping and investigation of the change in additive effects of polymorphisms across age and among traits for multiple stress response phenotypes. We found that cold stress survival with acclimation, cold stress survival without acclimation, and starvation resistance declined with age and that changes in the genetic architecture of each phenotype were consistent with MA predictions. We used a novel test for MA and AP by calculating the additive effect of polymorphisms across ages and found support for both MA and AP mechanisms in the age-related decline in stress tolerance. These patterns suggest both MA and AP contribute to age-related change in stress response and highlight the utility of association mapping to identify genetic shifts across age.


Introduction 23
The intensity of natural selection changes over an organism's lifespan, having greatest 24 effect early in life, as individuals reach reproductive maturity, and smaller effect as organisms 25 age (Charlesworth, 2001;Charlesworth and Hughes, 1996;Fisher, 1930;Haldane, 1941;26 Hamilton, 1966;Medawar, 1952;Williams, 1957). Decreased effectiveness of natural selection 27 at old age results in the accumulation of deleterious polymorphisms in populations and leads to 28 decline in age-specific fitness, characteristic of senescence (Hamilton, 1966;Medawar, 1952;29 Williams, 1957). Senescence is expected to negatively impact phenotypes related to fitness and 30 is thought to have evolved through two non-mutually exclusive genetic mechanisms (Bowler and 31 Terblanche, 2008;Charlesworth, 1994;Ricklefs and Finch, 1995). Under mutation accumulation 32 (MA; Medawar, 1952), decreased effectiveness of natural selection over lifespan allows the 33 retention of deleterious polymorphisms that are only expressed later in life (Ricklefs and Finch, 34 1995). Under antagonistic pleiotropy (AP; Williams, 1957), genes that are expressed over a wide 35 window of an individual's lifespan have positive effects on fitness at young age and negative 36 effects on fitness at old age (Charlesworth, 2001;Maklakov et al., 2015;Ricklefs and Finch, 37 1995; Williams, 1957). Both mechanisms rely on the relaxation of natural selection later in an 38 organism's life but have unique predictions for how age-dependent genetic control of phenotypes 39 changes. Promislow, 2003; Tatar et al., 1996), but far less is known about how stress response phenotypes 45 change with age (Bowler and Terblanche, 2008). Stress response over an organism's lifespan is a 46 critical component of fitness, and is an important modulator of lifespan (Colinet et al., 2015). 47 Variation in stress response can influence the persistence and evolution of populations over short 48 time scales, especially in variable environments (Bergland et al., 2014). In species that 49 experience seasonal change in thermal regime, changes in the demographic structure of 50 populations can also drastically influence the ability of individuals to tolerate stressful 51 average (Behrman et al., 2015). Thus, measures of thermal tolerance at one point in the season 55 therefore do not reflect the influence of seasonal variation in age on thermal tolerance. Such 56 shifts in the age structure of populations coupled with age related changes in the genetic control 57 of fitness phenotypes have the potential to dramatically influence short-and long-term responses 58 to environmental variation. 59 The MA and AP aging mechanisms make predictions about age-specific changes in 60 multiple quantitative genetic parameters. Under MA, genetic variance is expected to increase 61 with age because of the expression of age-restricted polymorphisms (Charlesworth, 2001; 62 Charlesworth and Hughes, 1996; Hughes et al., 2002;Leips et al., 2006). These late acting 63 polymorphisms are retained in the population because the individuals that possess them have 64 successfully reproduced, allowing such alleles to evade natural selection (Charlesworth, 2001;65 Haldane, 1941;Maklakov et al., 2015). Additionally, because the genetic control of the 66 phenotype across ages is independent, the genetic correlation of the phenotype between young 67 and old individuals is expected to be non-negative (Charlesworth, 2001;Maklakov et al., 2015; 68 evaluation of age-related shifts in the additive effects of associated polymorphisms, thus 91 facilitating the detection of weak antagonistic effects across age or phenotypes. 92 As an example, consider a hypothetical phenotype measured in young and old individuals 93 that is associated with non-overlapping sets of polymorphisms at each age. Two different 94 polymorphisms are associated with the hypothetical phenotype at young age and have positive 95 additive effects on the young phenotype. The polymorphism that is consistent with MA will shift 96 from a significant positive additive effect at young age to an effect that is near zero or of the 97 same sign at old age. In contrast, the polymorphism that is consistent with AP will shift from a 98 positive additive effect at young age to a negative additive effect at old age. Thus, even though 99 association mapping may not detect the antagonistic polymorphism with small effect in old 100 individuals, calculation of additive effects of polymorphisms across age can be used to detect 101 signals of AP ( Fig. 1; Maklakov et al., 2015). 102 In the current study, we used a combination of association mapping and quantitative 103 genetic analysis to dissect the variation in age-related changes in four environmental stress 104 response phenotypes and tested the influence of MA and AP. To do this, we measured age-105 related survival after cold-stress with acclimation and without acclimation, thermal phenotypic 106 plasticity, and starvation resistance in a genetically diverse D. melanogaster mapping population 107 related change in cold tolerance has been done in a single genotype (Czajka and Lee, 1990), and 113 little research is available to inform how starvation resistance will change with age (but see 114 (Colinet et al., 2015). We predicted that starvation resistance and cold tolerance (measured as 115 acclimation and non-acclimation survivorship) would decline with age. 116 Genetic variation has also been documented for various forms of thermal phenotypic 117 plasticity (Fallis et al., 2014;Gerken et al., 2015). Short-term acclimation through rapid cold-118 hardening (RCH score; Lee et al., 1987) is one form of plasticity that occurs when organisms are 119 exposed to a mild thermal stress before experiencing more stressful conditions (Coulson and 120 Bale, 1990; Czajka and Lee, 1990;Gerken et al., 2015;Lee et al., 1987;Powell and Bale, 2005). 121 In flies and other ectothermic species, this pre-treatment usually results in increased cold 122 survivorship and provides a simple model of the physiological response of ectotherms as they 123 respond to episodic fluctuations in temperature (Bozinovic et

Cold-stress responses in physiologically aged individuals 177
We conducted an additional experiment to test for variability in the rate of senescence 178 among DGRP lines. For this experiment, 10 lines were randomly selected from the 101 included 179 in the cold stress experiment (Table S1), and experimental flies were obtained as described

Genome-wide association analysis 201
We used association mapping to identify regions of the genome that were significantly 202 associated with variation in acclimation survivorship, non-acclimation survivorship, acclimation 203 score, and starvation resistance. Association mapping was performed on each age and phenotype 204 separately, and significance was assigned at -log 10 (5) (Durham et al., 2014;Gerken et al., 2015;205 Mackay et al., 2012). Shifts in genetic architecture across age and phenotype were assessed by 206 comparing the significant polymorphisms associated with each age-specific phenotype. We 207 performed gene ontology (GO) enrichment analysis using FlyMine (Lyne et al., 2007) to 208 determine whether specific classes of genes or pathways were overrepresented in the loci 209 associated with each phenotype and age. 210 211

Quantitative genetic analyses 212
Heritability, variance components, and genetic correlations were estimated using the 213 program H2boot, which applies bootstrap resampling to quantitative genetic data (Phillips, 214 1998). Acclimation survivorship, non-acclimation survivorship, acclimation score, and starvation 215 resistance for one-and four-week-old flies were treated as eight phenotypes. Data were analyzed 216 using a one-way ANOVA, resampling lines 10,000 times with replacement. Because DGRP lines 217 are inbred homozygous lines, reported heritability estimates are broad sense, and were estimated 218 as: 219 where ! ! is the among line homozygous genetic variance component, and ! ! is the 223 environmental variance component. The coefficient of homozygous genetic variance was used to 224 assess the effect of age on changes in homozygous genetic variance, and was estimated as: 225 where ! is the phenotype mean. Genetic correlations across ages were estimated as: 229 to population-level decline in a phenotype, late acting alleles will inflate CV G in four-week-old 253 flies. Second, if MA is responsible for age-specific decline in phenotypes, unique regions of the 254 genome should be associated with the phenotype at young and old age. Under AP, regions of the 255 genome that are associated with the phenotype in young and old individuals should overlap. 256 Third, under MA, we expect the genetic correlation between ages for each phenotype to be non-257 negative, due to the expectation that the additive effects at different ages are independent 258 individuals that are smaller but of the same sign ( Fig. 1), while under AP, polymorphisms 264 associated with a phenotype in young individuals are expected to have additive effects on the 265 phenotype in old individuals that are of the opposite sign ( Fig. 1). 266 In addition to testing these predictions of MA and AP within each phenotype across age, 267 we also tested the role of MA and AP in age-related change between phenotypes. As above for 268 each phenotype, we assessed the level of overlap of polymorphisms and genetic correlations 269 between phenotypes and calculated the additive effects of polymorphisms associated with each 270 phenotype on every other phenotype and age in our study. For example, the additive effects of 271 polymorphisms associated with acclimation survival at one week were calculated for both the 272 one and four-week non-acclimation survival and starvation resistance responses. 273

Tests for selection 275
We used the QTL sign test (QTLST) to the test the direction of the additive effects of 276 associated polymorphisms identified through association mapping. The QTLST was developed 277 by (Orr, 1998) to determine whether the signs of QTL effects were indicative of directional 278 selection acting on a phenotype. The probability for rejecting the null hypothesis that selection 279 does not influence the phenotype was calculated as in Orr (1998): In this study, we treated n as the number of associated polymorphisms detected for each 284 phenotype, n +obs as the number of these polymorphisms that had a positive additive effect, and G 285 as a vector of all additive effects. Because association mapping does not involve generation of a 286 mapping population from distinct lines, R was simply the standard deviation of the phenotype of 287 the population. An exponential distribution of the polymorphism effects was assumed in our 288 adaptation of this model as in applications of QTLST to QTL data. Because QTLST is sensitive 289 to high variance among additive effects, we also performed the QTLST-EE, which assumes that

Phenotypic responses 296
All phenotypes measured in this study were variable across ages, lines, and sexes ( Fig. 2; 297 Table S2). Two-way interactions among these effects were also significant, except for age by sex 298 in non-acclimation survivorship (Table S2). The three-way interaction between age, line, and sex 299 explained a significant amount of variation for acclimation survivorship and starvation resistance 300 as well (Table S2). On average, acclimation and non-acclimation survivorship and starvation 301 resistance decreased with age as expected ( Fig. 2A, D, J; Table S2). However, age-related 302 decline was stronger in non-acclimation survivorship compared to acclimation survivorship, 303 resulting in an average acclimation score that increased significantly with age ( Fig. 2G; Table  304 S2). When the sex-specific cold tolerances were analyzed, male flies maintained their capacity to 305 survive the acclimation treatment across age ( Fig. 2B; adj. P = 0.64). However, this maintenance 306 across age was not observed in the non-acclimation treatment ( Fig. 1E; adj. P < 0.001). In 307 females, flies tended to lose the survival capacity at an equal rate for both acclimation and non-308 acclimation treatments (Fig. 2B, E). As a result of these sex-specific age-related responses, 309 female acclimation score did not change with age (adj. P = 0.46; Fig. 2H), but male acclimation 310 score increased (adj. P < 0.001). Thus, the population-level increase in acclimation score was 311 likely driven by retention of cold tolerance in the acclimated male flies. Post hoc comparisons of 312 sex-specific starvation resistance revealed that the age-related average decrease in starvation 313 resistance was primarily driven by a significant decrease in resistance in females (adj. p < 0.001; 314  Genotype-specific responses for each phenotype were highly variable (Fig. 2, right 316 column; Table S3); an age-related increase in stress resistance was observed for some lines, 317 while responses in other lines remained constant or decreased (Fig. 2, right column; Table S2). 318 Negative acclimation scores were obtained for several lines screened at one week of age, 319 suggesting that the acclimation treatment had a detrimental effect on survivorship; however, the 320 vast majority of lines responded positively to this treatment (Fig. 2I). When screened at four 321 weeks, the negative acclimation effect largely disappeared as only two lines had acclimation 322 scores below 0. This change in the pattern of cold tolerance with age may have important 323 implications for the role of plasticity in maintaining stress response with age. 324 325

Variation in senescence 326
To assess the relationship between chronological and physiological age, we measured 327 acclimation survival, non-acclimation survival, and acclimation score on ten randomly selected 328 lines from the DGRP (Table S1) Table S4). While 338 longevity does vary among DGRP lines, variation in lifespan did not significantly alter the rank 339 order of acclimation survival among the lines. This suggests that variation in longevity among 340 the DGRP lines does not influence the age-related change in detected in phenotypes between 341 young and old flies. 342

Genetic architecture 344
In one-week-old flies, association mapping identified 24 polymorphisms and 23 345 genes associated with acclimation survival, 22 polymorphisms and 14 genes associated with non-346 acclimation survival, 45 polymorphisms and 23 genes associated with acclimation score, and 20 347 polymorphisms and 9 genes associated with starvation resistance (Table 1). In four-week-old 348 flies, association mapping identified 31 polymorphisms and 28 genes associated with acclimation 349 survival, 69 polymorphisms and 48 genes associated with non-acclimation survival, 26 350 polymorphisms and 6 genes associated with acclimation score, and 27 polymorphisms and 22 351 genes associated with starvation resistance (Table 1). Surprisingly, no polymorphisms or genes 352 were shared within phenotype across age or between phenotypes ( Fig. S2; Table S5). 353 Several polymorphisms were associated with genes that have been previously associated 354 with cold-, starvation-, or age-related phenotypes, and were distributed across the phenotypes 355 measured in this study (Table S5 and references therein). Out of all genes identified in our study 356 (Table 1, Table S5), 54 have been previously associated with cold acclimation or with a cold-357 sensitive phenotype in Drosophila, 18 have been previously associated with starvation response 358 or stress, and 59 have been previously associated with aging or lifespan. For example, Cht2, 359 involved in chitin binding, has been previously associated with cold acclimation response 360 (MacMillan et al., 2016) and was associated with four-week starvation resistance in this study. 361 Meltrin, associated with one-week starvation resistance in our study, has been previously 362 associated with cold acclimation response and age-specific fitness ( (28) were also associated with oxidative stress resistance, which has been associated with aging 367 and senescence (Schwarze et al., 1998). For example, decay, rg, and Pde1c have been previously 368 associated with oxidative stress and were associated with four-week acclimation survival or one-369 week non-acclimation survival in our study (Table S5). Additional details describing the function 370 of each gene and associated references are listed in Table S5. Despite the previous reporting of 371 genes that are associated with aging or stress phenotypes, no gene ontology (GO) categories 372 were overrepresented following enrichment analysis,. 373 374

Evolutionary theories of aging and the decline in stress response 375
Shifting genetic architecture within phenotypes across age 376 Each phenotype was associated with a unique set of polymorphisms across age and 377 among phenotypes (Table 1, Fig. S2, Table S5). Thus, the lack of overlap of associated 378 polymorphisms across age within phenotypes suggests the genetic architecture shifted and that 379 genetic control of the phenotypes was age-specific. MA not only predicts that associated 380 polymorphisms at each age are unique, but also that associated polymorphisms have positive 381 additive effects in young individuals that remain positive or approach 0 in older individuals. 382 Conversely, AP predicts associated polymorphisms with positive additive effects in young 383 individuals will have negative additive effects in old individuals. We calculated the additive 384 effects of all significantly associated polymorphisms for the stress response phenotypes that 385 declined with age (acclimation survivorship, non-acclimation survivorship, and starvation 386 resistance; Fig. 4) for the one-and four-week response. When additive effects of the associated 387 polymorphisms in one-week-old flies were calculated for the same phenotype in four-week-old 388 flies, the additive effects were either closer to 0 or of the same sign (i.e. not antagonistic; Fig. 4; 389 Table S6). The reverse comparison resulted in the same pattern; for example, additive effects of 390 four-week acclimation survival polymorphisms on one-week acclimation survival were smaller 391 and closer to 0 than the additive effects of four-week acclimation survival polymorphisms on 392 four-week acclimation survival ( Fig. 4; Table S6). This pattern of unique associated 393 polymorphisms and additive effects that decrease with age was observed for each of the 394 phenotypes that declined with age (starvation resistance, acclimation survival, and non-395 acclimation survival) and is consistent with MA. 396 MA also predicts that, for phenotypes that decline with age, the coefficient of genetic 397 variance (CV G ) will increase with age and that the genetic correlation between the phenotype in 398  Table S8). The lack of negative genetic correlations 408 for each phenotype between young and old individuals combined with the increase in the 409 coefficient of genetic variance with age provides additional evidence that supports the MA 410

mechanism. 411
Shifting genetic architecture between phenotypes with age 413 Comparisons of associated polymorphisms for each phenotype measured in this study 414 demonstrated unique genetic control for each phenotype across age. However, the lack of 415 overlap in associated polymorphisms does not necessarily mean that the polymorphisms detected 416 for each phenotype do not influence variation in other phenotypes at other ages, but rather that 417 the additive effect was too small to be detected by association mapping. Polymorphisms 418 associated with one phenotype (e.g. one-week acclimation survival) may have small but 419 important additive effects on other phenotypes (e.g. four-week starvation resistance), and if this 420 is the case, the interpretation is similar to the comparison of additive effects within phenotypes 421 across age reported above (e.g. Fig. 1). If the additive effect is close to 0 and of the same sign, 422 this supports the MA mechanism; if the additive effect is different from 0 and of the opposite 423 sign, this supports the AP mechanism (Fig. 1). 424 To test for the presence of pleiotropic effects of associated polymorphisms on other 425 phenotypes measured in this study, we calculated the average standardized additive effect of 426 associated polymorphisms for each phenotype on every other phenotype and age ( Fig. 4; Table  427 S6). For example, the additive effects of the set of associated polymorphisms with acclimation 428 survival in one-week-old flies were calculated for one-and four-week non-acclimation survival 429 and one-and four-week starvation resistance (Fig. 4A). Confidence intervals were used to 430 determine if the calculated average additive effects were different from 0 (Table S6). 431 Approximately half of the calculated average effects were not different from 0 (55.6%), 432 and 27.8% of the comparisons resulted in average effects with a sign opposite that of the average 433 effect of the polymorphisms in the phenotype with which they were significantly associated 434 (suggesting an antagonistic relationship; Table S6). The antagonistic effects of polymorphisms 435 between phenotypes were often, but not always, reciprocal (Fig. 4). For example, polymorphisms 436 associated with one-week acclimation survival had an average antagonistic additive effect on 437 starvation resistance at both ages (Fig. 4A), while polymorphisms associated with four-week 438 starvation resistance had an average antagonistic additive effect on only one-week acclimation 439 survival (Fig. 4F). In an aging context, evidence from additive effect comparisons support both 440 AP and MA across age, depending on the phenotypes being compared; however, MA was more 441 common based on apparent independence of additive effects (average additive effects were not 442 different from 0). 443 Phenotypes that appeared to be antagonistically pleiotropic such that polymorphisms 444 increased the phenotype in young flies but decreased the phenotype in old flies include 445 acclimation survival and starvation resistance (Fig. 4A, B, F) and non-acclimation survival and 446 starvation resistance (Fig. 4C, D, F). Polymorphisms associated with one-week starvation 447 resistance did not have an antagonistic effect on average on any other phenotype, although 448 several individual polymorphisms did have antagonistic effects on other phenotypes ( Fig. 4E; 449 Table S6). The antagonistic relationship between one-week acclimation survival and four-week 450 starvation resistance was further supported by a significant negative genetic correlation between 451 traits across ages (R G = -0.47 ± 0.2 S.E.; Table S8). However, all other combinations of 452 phenotypes involved effects and genetic correlations that were not different from 0 ( Fig. 4; Table  453 S6, S8), and were thus more consistent with the predictions of MA. 454 455

Evidence of selection and phenotypic trade-offs 456
With the exception of acclimation score, the additive effects of the majority of 457 polymorphisms significantly associated with the phenotypes measured in our study were of the 458 same sign (i.e. most additive effects were positive or negative; Fig. 4 and 5, Table S5). To 459 determine if more additive effects of positive sign were associated with the phenotype than 460 expected by chance, we used the QTLST to test for evidence of selection. More positive additive 461 effects than expected by chance were observed for one-week acclimation survival, non-462 acclimation survival, and acclimation score suggesting selection increased these phenotypes in 463 the founding population of the DGRP (Fig. 4 and Fig. 5A, C, E). The QTLST was also 464 significant for four-week acclimation survival, non-acclimation survival, and both one-and four-465 week starvation resistance, suggesting selection has acted to decrease these phenotypes (Fig. 4  466 and Fig. 5B, D, G, H). However, because the effectiveness of natural selection is expected to 467 decline with age, this significant result is likely the result of a correlated response resulting from 468 selection on young phenotypes. The signs of the additive effects of polymorphisms associated 469 with acclimation score in four-week-old flies were more mixed than any other phenotype, 470 leading to a non-significant QTLST for this phenotype (Fig. 4 and Fig. 5F). 471 472 Discussion 473

Genetic variation in age-specific decline in stress tolerance 474
As expected, the average phenotypic responses for most phenotypes declined with age, 475 with the exception of plasticity measured as acclimation score (Fig. 2). For each phenotype, we 476 observed significant genetic variation across ages, with some genotypes exhibiting increased 477 stress resistance with age. We investigated variation in longevity among DGRP lines by 478 comparing cold tolerance responses measured at four weeks to those in physiologically aged flies 479 at the point when the population reached Td50. We know that lifespan for virgin female flies 480 varies from approximately 20 days to approximately 80 days in the DGRP (Ivanov et al., 2015). The only phenotype in our study that behaved unexpectedly and did not decline with age 493 was plasticity measured as acclimation score. Acclimation score was significantly higher in four-494 week-old flies (Fig. 2G), where we expected this phenotype to remain constant across age. The 495 age-related response in acclimation score was driven by the male response. Four-week-old male 496 flies had a stronger age-related decline in non-acclimation survival compared to acclimation 497 survival. Therefore, the observed increase in acclimation score for our population has at least two 498 interpretations. First, plasticity at the population level may increase with age, potentially as a 499 compensatory mechanism to overcome the overall loss of basal cold tolerance (Fig. 2F). 500 Throughout the season, natural populations of D. melanogaster are expected to be composed of 501 increasingly old individuals such that by the time temperatures begin to cool at the beginning of 502 the fall season, a greater proportion of populations is composed of older individuals (Behrman et 503 al., 2015). If older individuals are less cold tolerant, they may still be able to tolerate cold 504 temperature exposures through increased capacity for adaptive plasticity through acclimation. 505 Second, acclimation pretreatment appears to have had a less detrimental effect on survival in 506 four-week-old flies compared to one-week-old flies (Fig. 2I). In one-week-old flies, acclimation 507 improved survival in the majority of lines; however, several lines (14%) had negative 508 acclimation scores, indicating that exposure to 4°C prior to the -6°C exposure was more 509 damaging than the -6°C exposure alone ( Fig. 2I; Gerken et al., 2015). Only 2% of lines tested 510 had negative acclimation scores at four weeks suggesting that acclimation may be more likely to

MA describes age-related change within individual phenotypes 518
When each phenotype was considered separately across age, we found support for MA, 519 satisfying predictions based on analysis of quantitative genetic parameters (Charlesworth, 2001; 520 First, the coefficient of genetic variance increased with age for each phenotype (Table 1). The 522 increase in CV G in starvation resistance was less drastic than other phenotypes, but when sexes 523 were analyzed separately, the increase was more dramatic (Table S7). An increase in CV G 524 indicates that a greater proportion of the phenotypic variance can be explained genetically at four 525 weeks of age (Charlesworth, 2001;Charlesworth and Hughes, 1996;Houle et al., 1994) and this 526 increase is consistent with the hypothesis that the age-related decline in the phenotype is the 527 result of the accumulation of deleterious age-specific polymorphisms that influence variation in 528 the phenotype (Charlesworth, 2001;Charlesworth and Hughes, 1996;Engström et al., 1989). 529 Second, MA predicts that a phenotype is controlled by unique sets of genes across age 530 (Charlesworth, 2001(Charlesworth, , 1994Charlesworth and Hughes, 1996;Maklakov et al., 2015;Medawar, 531 1952;Partridge and Barton, 1993;Rose, 1991). We detected unique sets of polymorphisms that 532 were associated with each phenotype across age. This is consistent with age-specific association 533 patterns presented by (Durham et al., 2014) who determined age-related change in fecundity was 534 also influenced by MA. The genetic independence of phenotypes across age was further 535 supported by the non-significant or significantly positive genetic correlations for each phenotype 536 across age (Table S8) window of ages but with lower additive effects (Maklakov et al., 2015). Thus, the positive 543 genetic correlations across age are the result of the associated polymorphisms having a slightly 544 wider window of age-specific effects that ultimately influence the phenotype at other ages. This 545 pattern was observed for acclimation and non-acclimation survival and starvation resistance 546 across age; for all phenotypes, the four-week associated polymorphisms had negative additive 547 effects on the one-week phenotype that were smaller than the effect of the polymorphisms on the 548 four-week phenotype. This suggests that four-week polymorphisms that led to decline in each 549 phenotype do have small pleiotropic effects (in the same direction) at one week of age. The Our third piece of evidence to support MA comes from our novel approach of calculating 555 the additive effects of polymorphisms across age. We calculated the additive effects of 556 polymorphisms associated with the one-week phenotype in the four-week phenotype data (and 557 vice versa). Under MA, we expected the additive effect of the one-week polymorphisms to be 558 smaller and in the same direction (i.e. they have the same sign) when the additive effects were 559 calculated for the four-week response data (Maklakov et al., 2015). If the sign of a particular 560 polymorphism had flipped (a polymorphism with a positive effect in one age had a negative 561 effect in the other age), this would have suggested that an antagonistic relationship existed and 562 would have supported AP (Maklakov et al., 2015). For all phenotypes, the calculated additive 563 effects of age-specific polymorphisms across age were either closer to 0 and/or in the same 564 direction, providing definitive support for the role of MA in the age-related decline in the stress 565 responses measured (Fig. 4). 566 567

MA and AP describe age-related variation between phenotypes 568
Our novel extension of association mapping through the calculation of additive effects of 569 polymorphisms across phenotypes allowed us to investigate the role MA and AP on age-specific 570 responses between phenotypes as well. Though each phenotype and age was associated with a 571 unique set of polymorphisms (consistent with predictions for MA), we found support for AP 572 between several phenotypes (Fig. 4). We observed a significantly negative phenotypic 573 correlation between one-week acclimation survival and both one-and four-week starvation 574 resistance, corroborating a pattern reported by (Hoffmann et al., 2005) (Table S8). We also 575 observed a significantly negative genetic correlation between one-week acclimation survival and 576 four-week starvation resistance (Table S8), suggesting that AP (between one-week acclimation 577 survival and four-week starvation resistance) influenced age-related change in these phenotypes. 578 When the additive effects of polymorphisms associated with four-week starvation were 579 calculated for both one-week cold tolerance phenotypes (Table S6), all four-week starvation 580 resistance associated polymorphisms, which had negative additive effects on four-week 581 starvation resistance, had positive additive effects on one-week acclimation survival and one-582 week non-acclimation survival (Fig. 4F). The change in sign of the additive effects of four-week 583 starvation resistance associated polymorphisms on both of the one-week cold tolerance 584 phenotypes is strong evidence to support the role of AP in age-related decline in starvation 585

resistance. 586
Additional examples of AP existed between phenotypes in our data as well and were 587 identified through confidence interval analysis of additive effects ( Fig. 4; Table S6). Specifically, 588 one-week acclimation and non-acclimation survival polymorphisms had positive effects on their 589 respective phenotypes across age but negative additive effects on four-week starvation resistance 590 ( Fig. 4A -D; Table S6). This pattern is consistent with that discussed above and again suggests 591 that age-related change in starvation resistance is influenced by AP with cold tolerance. 592 Interestingly, four-week acclimation and non-acclimation survival polymorphisms had largely 593 positive additive effects on one-week starvation resistance (Fig. 4B, D; Table S6). This pattern 594 suggests that AP may also be contributing to age-related decline in acclimation and non-595 acclimation survival. With evidence from our examination of acclimation and non-acclimation 596 survival across age (discussed above), and the apparent role of MA and positive pleiotropy for 597 age-related change within these phenotypes, it is evident that it may not be possible to fully 598 disentangle the roles of MA and AP on the age-related decline in phenotypes. In essence, 599 polymorphisms that increase one phenotype in young individuals and decrease another 600 phenotype in old individuals through AP may also contribute to age-related change within the 601 phenotype through positive pleiotropy under MA. These results demonstrate the need for caution 602 in interpreting the lack of overlap in significant associated polymorphisms as support for MA in 603 isolation of other evidence because AP may still be playing an important role in the age-related 604 change in phenotypes. 605 We also found support for the role of MA between phenotypes across age. Age-related 606 change in acclimation survival and non-acclimation survival appears to be evolving largely 607 independently of the other phenotype under MA as we found either non-significant or 608 significantly positive phenotypic and genetic correlations between all age combinations of these 609 phenotypes (Table S6, Table S6). 611 One-week acclimation survival polymorphisms, which had positive effects on one-week 612 acclimation survival, all had positive additive effects when calculated for both one-and four-613 week non-acclimation survival (Fig. 4A). On average, additive effects of polymorphisms 614 associated with four-week acclimation survival had additive effects that were not different from 615 zero, although some individual polymorphisms did have antagonistic effects on one-and four-616 week non-acclimation survival. All but two polymorphisms associated with one-week non-617 acclimation survival had additive effects of the same sign on one-and four-week acclimation 618 survival, and all polymorphisms associated with four-week non-acclimation survival had 619 additive effects of the same sign on one-and four-week acclimation survival. While the small 620 number of individual polymorphisms with antagonistic additive effects across phenotype may 621 impact age-related change in these cold tolerance phenotypes, it likely that this impact is small in 622 comparison to the role of MA and positive pleiotropy. 623 624

Natural selection shapes phenotypic variation across age 625
Our data recapitulate previously reported relationships between different measures of 626 cold tolerance and starvation resistance (Table S8;  acclimation survival, this suggests that natural selection favored polymorphisms that increase 640 cold tolerance phenotypes in the population from which the DGRP was established (Fig. 4A, C  641 and 5A, C; Table S5). This finding is consistent with evidence of selection for cold tolerance in 642 natural populations of D. melanogaster (Bergland et al., 2014), as well as previous reports of 643 majority positive additive effects of polymorphisms associated with chill coma recovery 644 (Mackay et al., 2012). 645 Conversely, most of the polymorphisms associated with cold tolerance at four weeks of 646 age were negative (Fig. 4B, D), indicating that the major alleles decreased survival following 647 cold stress. While the QTLST was significant for these late-acting polymorphisms, it is very 648 unlikely that natural selection directly led to this pattern. Instead, polymorphisms associated with 649 acclimation and non-acclimation survival in old individuals likely arose through mutation and 650 were maintained in the population because their negative effect on survival in young individuals 651 was small relative to the four-week additive effects (Charlesworth, 2001;Houle et al., 1994;652 Maklakov et al., 2015) Fig. 4B, D). Similarly, in both young and old individuals, most of the 653 additive effects of starvation resistance associated polymorphisms were negative (Fig. 4E, F). 654 This pattern suggests that age-related change in starvation resistance is influenced by positive 655 selection on acclimation and non-acclimation survival in young individuals. Alternatively, the 656 effectiveness of natural selection on starvation resistance may be constrained by pleiotropy 657 between one-week acclimation survival and one-week starvation resistance (many one-week 658 acclimation and non-acclimation polymorphisms had negative effects on one-week starvation 659 resistance; Fig. 4A, C). Some positive selection on starvation resistance in young individuals 660 may provide a mechanism for the retention of four-week acclimation and non-acclimation 661 survival associated polymorphisms that have increasingly negative effects with age. 662

Implications for evolutionary theories of aging 664
Efforts to more fully understand and test evolutionary theories of aging have encouraged 665 expansion and clarification of predictions of both MA and AP mechanisms (Charlesworth, 2001;666 Houle et al., 1994;Maklakov et al., 2015;Reynolds et al., 2007;Wachter et al., 2014Wachter et al., , 2013. Our 667 novel extension of existing methods not only revealed the relative importance of MA and AP for 668 age-related change in stress response, but also verified recent hypotheses that present expansions 669 on the theory of MA. When originally formulated, the MA mechanism predicted that fitness was 670 controlled by polymorphisms that had very narrow windows of effect (Charlesworth and 671 Hughes, 1996;Medawar, 1952;Rose, 1991), but evidence from several studies has indicated that 672 it is more likely that polymorphisms which contribute to late-life decline in fitness and age-673 related change in phenotypes have wide windows and increasingly large effects across age 674 However, by comparing the effects of polymorphisms calculated for each age and phenotype 688 pair, we are able to overcome this bias against polymorphisms that have small effects. 689 Very few studies present convincing evidence of the influence of both MA and AP on 690 age-related change within and among phenotypes (but see Leips et al., 2006), but we have 691 demonstrated that both mechanisms contribute to age-related change in stress response. It is clear 692 from our results that individual polymorphisms that are significantly associated with phenotypes 693 at different ages can contribute to age-related decline within and among phenotypes in patterns 694 that are consistent with both MA and AP. Thus, the evolution of senescence and associated 695 decline in fitness is influenced by a combination of natural selection acting on correlated 696 phenotypes that have non-independent antagonistic genetic architectures, as well as the 697 accumulation of polymorphisms with negative effects that strengthen with age. It is likely that 698 similar patterns will be observed for other phenotypes related to fitness as well, adding to our 699 understanding of how evolution of aging and multivariate evolution are tightly intertwined. Acclimation score significantly increased in males, but remained consistent across age for 927 females (age by sex: F 1,1212 = 8.68, P < 0.01). I. Acclimation score significantly varied among 928 the 101 DGRP lines (age by line: F 100,1212 = 3.18, P < 0.001). J. Average starvation resistance 929 decreased significantly with age (F 1,1623 = 893.0, P < 0.001). K. Starvation resistance 930 significantly decreased in both sexes, but to a larger degree in females (age by sex: F 1,1623 = 931 567.0, P < 0.001). L. Starvation resistance significantly varied among the 164 DGRP lines (age 932 by line: F 163,1623 = 6.0, P < 0.001). 933  Table 1. Quantitative genetic estimates (± S.E.) for all phenotypes as they vary with age and the 975 number of polymorphisms (generalized as SNPs) and genes significantly associated with each 976 phenotype identified by GWAS with a threshold of -log10(5). All heritabilities reported are 977 broad-sense and are greater than 0. 978 Figure S1. Graphical representation of acclimation (A) and non-acclimation treatment (B). Flies 982 were maintained at 25°C during rearing and recovery, and lights on occurred at 07:00hrs. A. 983 Flies were transferred from 25°C to 4°C for two hours for the acclimation (AC) treatment and 984 then were transferred immediately to -6°C for one hour for the cold shock treatment (CS). Flies 985 were placed on fresh media and allowed to recover for 24 hours at 25°C. B. Flies were 986 transferred to -6°C for one hour for the cold shock treatment (CS) and were allowed to recover at 987 25°C for 24 hours on fresh media.