Analysis of lifespan across Diversity Outbred mouse studies identifies multiple longevity-associated loci

Lifespan is an integrative phenotype whose genetic architecture is likely to highlight multiple processes with high impact on health and aging. Here, we conduct a genetic meta-analysis of longevity in Diversity Outbred (DO) mice that includes 2,444 animals from three independently conducted lifespan studies. We identify six loci that contribute significantly to lifespan independently of diet and drug treatment, one of which also influences lifespan in a sex-dependent manner, as well as an additional locus with a diet-specific effect on lifespan. Collectively, these loci explain over half of the estimated heritable variation in lifespan across these studies and provide insight into the genetic architecture of lifespan in DO mice. Methods Study Designs Dietary Restriction (DRiDO): This study is extensively described elsewhere (Di Francesco et al. 2023). Briefly, female DO mice were received at ~4 weeks of age in 12 waves from March 2016 through November 2017. Mice were housed in groups of 8 in single large-format ventilated pens with nestlets, biotubes, and gnawing blocks. Mice were randomized to one of five dietary interventions which were initiated for the surviving mice at 6 months of age: ad libitum (AL; n = 188), 1 day per week fasting (1D; n = 188), 2 days per week fasting (2D; n = 190), 20% caloric restriction (20; 2.75g/mouse/day; n = 189), and 40% caloric restriction (40; 2.06g/mouse/day; n = 182). Mice were extensively phenotyped as described (Di Francesco et al. 2023) and maintained until they died naturally. The mouse room was on a 12/12 hour light/dark schedule from 6:00 am to 6:00 pm and kept at 73o +/- 2o F. Harrison: Founder DO mice (167 retired breeder pairs) were obtained from the Jackson Laboratory and female offspring were accumulated for the lifespan study over 5 months. All mice were microchipped at 4 weeks of age. Mice were housed 22 per large-format double pens connected by a tunnel on an open-air rack. All 22 mice per pen were from different breeder pairs. Pens had pine shaving bedding with acidified water. Every week, one pen of the connected pair would be changed. Mice would be herded into one pen and the tunnel blocked off. The dirty pen would be removed, and a new clean pen attached with fresh water and grain. The ad libitum control mice (n = 349) were on a non-irradiated diet (5LG6, or “5S84”, TestDiet, Purina) from weaning. The diet restricted (n = 335) mice received 2.2 g/day/mouse of ground non-irradiated diet via modified fish feeders that were programmed to dump the ground diet onto the floor of the cage between 6-7 pm after lights were off. Modified feeders were restocked every 7 days. Any grain left in the feeders after 7 days was dumped on the cage floor. Proper feeder performance was indicated by a weighted string that was wound around a screw when the feeders dumped food. Diet restriction began at 4 weeks of age after being microchipped. An additional 339 mice received non-irradiated diet until they were 16 months of age, whereupon they started on 5LG6 diet with 142 ppm encapsulated rapamycin (Rapamycin Holdings, actual concentration of rapamycin in diet is 14 ppm, TestDiet, Purina). Mice were maintained until they died naturally. The mouse room was on a 12/12 hour light/dark schedule from 6:00 am to 6:00 pm and kept at 73o +/- 2o F. Svenson: We obtained female DO mice from the Jackson Laboratory breeding colony at ~4 weeks of age. Mice were obtained in eight waves over the course of 1 year and enrolled by randomization to dietary intervention protocols. Mice were housed 8 per group in single large format pens. Interventions were implemented as described for the Harrison study with a few differences indicated here. The ad libitum fed control mice (n = 319) were on a 4% irradiated diet (5K52, aka “5KOG”, TestDiet, Purina). The diet restricted mice (n = 316) received 2.2 g/day/mouse of ground 4% irradiated diet. DR mice were fed at ~7am daily and food was placed directly onto the bottom of the pen by a technician. On Friday the DR mice received a triple feeding (6.6 g/mouse) and were fed again on Monday morning. A third group of mice (n = 317) were maintained on the ad libitum protocol until 16 months of age and were then switched to rapamycin diet as described above. Mice experienced minimal handling (monthly body weights and weekly pen changes) and were maintained until they died naturally. The mouse room was on a 12/12 hour light/dark schedule from 6:00 am to 6:00 pm and kept at 70o +/- 2o F. For this study, DNA samples were collected for genotyping on the MUGA array, as described for other studies below. However, irregularities with sample labeling/handling made us question the integrity of our ID matches between mice and samples. We performed quality-control assessment by comparing genotype-predicted versus recorded coat colors across the mice (Silvers 2012) and confirmed extensive sample mismatches (data not shown). We therefore excluded these data from our genetic analyses. Shock: We obtained female (n = 244) and male (n = 240) DO mice from the Jackson Laboratory breeding colony at ~4 weeks of age. Mice were obtained in five waves from June 2011 through August 2012. Mice were housed in single-sex groups of 5 in standard ventilated duplex pens. All mice were fed ad libitum on 6% sterilized gain (5K52, aka “5KOG”, TestDiet, Purina). Mice experienced minimal handling (body weights and other non-invasive procedures). At 6, 12, and 18 months we obtained 3x100ul retroorbital blood draws, with 2 weeks recovery time between each. Mice were maintained until they died naturally. The mouse room was on a 12/12 hour light/dark schedule from 6:00 am to 6:00 pm and kept at 70o +/- 2o F. All procedures used in these studies were reviewed and approved by the Jackson Laboratory Animal Care and Use Committee. Genotyping Genotypes for all studies reported here were obtained using the mouse universal genotyping array (MUGA) (Morgan et al. 2016). DNA was isolated from tail tips using standard methods and shipped to Neogen Genomics (Lincoln, NE, USA) for analysis. Samples were genotyped using the MUGA (Harrison), MegaMUGA (Shock), or GigaMUGA (DRiDO) genotyping arrays. Founder haplotypes were reconstructed using the R/qtl2 software and samples with call rates at or above 90% were retained for analysis. Genome coordinates were from mouse genome GCRm39 and gene locations were taken from the Mouse Genome Informatics databases (Blake et al. 2021). Data analysis Survival analysis: We compared survival among each cohort of animals, as well as between experimental groups within studies. This was done by plotting Kaplan-Meier curves and by testing the equivalence of survival distributions among each cohort or experimental group using log-rank tests using overall tests (across cohorts and within each cohort) as well as pairwise comparisons between each experimental group and its respective within-study control group (ex: comparing rapamycin treatment to ad libitum within the Harrison study). p-values are reported with no correction for multiple comparisons, and are considered significant at p < 0.05. Median lifespan was estimated in each cohort as well as within each experimental group. The effects of dietary interventions and/or sex were estimated via Cox proportional hazards regression analysis, and are reported as hazard ratios with 95% confidence intervals. p-values are reported without correction for multiple comparisons and are considered significant at p < 0.05. Survival analysis was conducted using the “survival” (Therneau and Grambsch 2000; Therneau 2024) package in R and plotted via the “ggsurvfit” package (Sjoberg et al. 2024). Mortality doubling times and baseline hazards were estimated, beginning at the time of intervention, from a Gompertz log-linear hazard model with a 95% confidence interval and percentage change relative to female mice on an ad libitum diet via the “flexsurv” package in R (Jackson 2016). Additive whole-genome scans: All genetic analysis was conducted using the “rqtl2” package in R (R Core Team 2024; Broman et al. 2019). Whole-genome scans for lifespan QTL were carried out via the scan1() function using a mixed effects model in which lifespan was regressed on 8-state allele probabilities for each individual in a dataset. Within the Dietary Restriction and Harrison studies, dietary intervention and DO generation were included as additive covariates. In the Shock study, Sex and DO generation were included as additive covariates. In the meta-analysis, Study, Diet, Sex, and DO generation were included as additive covariates. For each genome-wide scan, 1000 permutations of the data were performed in which phenotypes were randomized and a whole-genome scan was run. The maximum LOD score observed in each permuted scan was recorded, and the 95th percentile of the distribution of 1000 maximum LOD scores was used as the significance threshold (ɑ = 0.05). QTL with LOD scores greater than this threshold were considered significant at our permutation-based threshold. In addition to this significance level, a significance level of LOD ≥ 6 was also used to identify loci contributing to variance in lifespan. While less conservative, this threshold is more stringent than a previously reported method (Wright et al. 2022). We report 2 LOD support intervals (‘2LOD SI’), corresponding to a 2 LOD drop around each peak position, about each peak marker identified in whole-genome scans. Forward regression analysis: In the meta-analysis, forward regression analysis was performed to account for the effects of genome-wide significant QTL when searching for additional loci influencing lifespan. This was done via the scan1() function using a mixed effects model including dietary intervention, sex, and DO generation as additive effects and kinship as a random effect. In addition to these covariates, previous QTL identified at a genome-wide significance level were included in the model as additive effects. QTL were encoded as numeric variables representing the genotype state at the marker with the highest LOD score as reported by association mapping. Only QTL reaching permutation-based significance thresholds were included in the model as additive covariates. Effect size estimation and percent variance explained: Best linear unbiased predictors (BLUPs) and corresponding 95% were computed for all QTL using the scan1blups() function in “rqtl2” using the additive covariates listed above. Phenotypic variance explained by each QTL was calculated using the following formula: 1 - 10-(2/n)*LOD Where n is the number of samples in a particular dataset, and LOD corresponds to the LOD score of the peak marker at each QTL. Variant association / fine mapping: Fine mapping was conducted within a 2LOD drop of the peak position associated with each QTL. Fine mapping was performed using the scan1snps() function in “rqtl2” using the same additive covariates listed in the Additive whole-genome scans section above. Variant and gene SQLite datasets used in this analysis are available at the “rqtl2” user guide website: https://kbroman.org/qtl2/assets/vignettes/user_guide.html. Single-QTL models for diet- and sex-specific loci: Within individual studies, QTL were tested for interaction with experimental factors unique to those studies. In the Shock cohort, QTL were tested for interaction with sex, while in other cohorts QTL were tested for interactions with one or more of the dietary interventions. Interaction tests were not conducted genome-wide using “rqtl2”, since this package does not fit interactions as random effects. For each QTL, tests were run using the 8-state allele probabilities at peak positions identified in whole-genome scans using the fit1() function in “rqtl2”. To assess significance, two models were run: an additive model in which experimental factors (sex, dietary intervention) and DO generation were included as additive covariates and an interaction model that included the same additive covariates with an additional interaction term corresponding to the experimental factor being tested. The reported LOD is the LOD of the interaction model minus the LOD of the additive model. To establish significance, 1000 permutations were run at each QTL in which phenotypes were randomized before running the additive and interaction models. After ordering the 1000 resulting LOD scores, the 95th percentile was chosen as the significance threshold (ɑ = 0.05). Interactions between QTL and experimental conditions were considered significant if their LOD was greater than the ɑ = 0.05 threshold. Interaction effects were plotted as residual lifespan values as a function of sex after correcting for additive covariates. Genome-wide scans for diet- and sex-specific loci: Gene by environment mixed effects models (GxEMM) were run using the ‘do-qtl’ software package in Python version 3.8.16 (Wright et al. 2022). Study, sex, and diet were included in the model as fixed effects, while diet and kinship were supplied as random effects. P values were computed based on 1,000 permutations of the data. A significance threshold of p <= 1*10-4 was used to define loci as statistically significant (Wright et al. 2022).寿命是一种整合表型，其遗传结构很可能揭示多种对健康和衰老具有高影响的生物学过程。本研究对多样性远交（Diversity Outbred, DO）小鼠的长寿性状开展遗传元分析，纳入了来自三项独立寿命研究的2444只实验动物。我们鉴定出6个可独立于饮食与药物处理显著影响寿命的位点，其中1个位点还以性别依赖的方式影响寿命，此外还发现1个具有饮食特异性寿命调控效应的位点。综上，这些位点可解释本研究中约半数以上的寿命可遗传变异，为解析DO小鼠的寿命遗传结构提供了新见解。 ## 研究方法 ### 研究设计 #### 饮食限制（Dietary Restriction, DRiDO） 该研究的详细信息已在其他文献中报道（Di Francesco等, 2023）。简言之，我们于2016年3月至2017年11月分12批次获取了约4周龄的雌性DO小鼠。小鼠以8只为一组饲养于大型通风笼盒中，笼内配备巢垫料、生物管与啃咬玩具。待小鼠至6月龄时，将存活个体随机分配至5种饮食干预组：自由进食组（ad libitum, AL; n=188）、每周禁食1天组（1D; n=188）、每周禁食2天组（2D; n=190）、20%热量限制组（20; 2.75g/只/天; n=189）以及40%热量限制组（40; 2.06g/只/天; n=182）。如先前报道（Di Francesco等, 2023），我们对小鼠进行了全面的表型分型，并维持饲养至其自然死亡。动物房采用12小时光照/12小时黑暗的光周期（光照时间为早6:00至晚6:00），环境温度控制在73℉±2℉。 #### Harrison研究 我们从杰克逊实验室（The Jackson Laboratory）获取了167对退休种鼠作为DO小鼠创始种群，在5个月内收集了其雌性后代用于寿命研究。所有小鼠于4周龄时植入电子芯片。小鼠以22只为一组饲养于通过隧道连接的大型双笼盒中，置于开放式笼架上，每组笼盒内的22只小鼠均来自不同的创始种鼠对。笼盒内铺设刨花垫料，提供酸化饮水。每周更换一次笼盒：将小鼠赶至其中一个笼盒并封堵隧道，移除脏污笼盒并接入装有新鲜饮水与谷物饲料的清洁笼盒。自由进食对照组小鼠（n=349）自断奶起喂食非辐照饲料（5LG6，亦称“5S84”，TestDiet, Purina）。饮食限制组小鼠（n=335）每日每只喂食2.2g碾碎的非辐照饲料，使用改装式喂食器在关灯后的18:00-19:00将饲料倾倒至笼盒底部，每7天补充一次饲料，若7天后喂食器内仍残留饲料则将其倾倒至笼盒底部。喂食器工作正常的标志为喂食时缠绕于螺杆上的配重绳索松开。饮食干预于小鼠植入芯片后的4周龄开始。另有339只小鼠在16月龄前持续喂食非辐照饲料，16月龄后更换为添加142ppm封装雷帕霉素的5LG6饲料（雷帕霉素购自Rapamycin Holdings，饲料中实际雷帕霉素浓度为14ppm，TestDiet, Purina）。所有小鼠均饲养至其自然死亡。动物房采用12小时光照/12小时黑暗的光周期（光照时间为早6:00至晚6:00），环境温度控制在73℉±2℉。 #### Svenson研究 我们于约4周龄时从杰克逊实验室的DO小鼠繁育种群中获取雌性小鼠，在1年内分8批次收集，并随机分配至不同饮食干预方案。小鼠以8只为一组饲养于大型单笼盒中。干预方案参照Harrison研究实施，仅存在少量差异：自由进食对照组小鼠（n=319）喂食4%辐照饲料（5K52，亦称“5KOG”，TestDiet, Purina）。饮食限制组小鼠（n=316）每日每只喂食2.2g碾碎的4%辐照饲料，饲养员于每日约7:00将饲料直接放置于笼盒底部。周五时饮食限制组小鼠获得三倍喂食量（6.6g/只），并于周一清晨再次喂食。另有一组小鼠（n=317）在16月龄前维持自由进食方案，16月龄后更换为前述添加雷帕霉素的饲料。小鼠仅需少量人工操作（每月称重、每周更换笼盒），并饲养至其自然死亡。动物房采用12小时光照/12小时黑暗的光周期（光照时间为早6:00至晚6:00），环境温度控制在70℉±2℉。 本研究原本计划收集DNA样本用于小鼠通用基因分型阵列（Mouse Universal Genotyping Array, MUGA）基因分型，如其他研究所述。但由于样本标记/处理流程存在不规范，我们对小鼠与样本之间的ID匹配准确性产生了质疑。我们通过比较基因型预测毛色与实际记录毛色（Silvers, 2012）开展质量控制评估，确认存在大量样本匹配错误（数据未展示）。因此，我们将该部分数据从遗传分析中剔除。 #### Shock研究 我们从杰克逊实验室的DO小鼠繁育种群中获取雌性（n=244）与雄性（n=240）小鼠，于2011年6月至2012年8月分5批次收集。小鼠按性别以5只为一组饲养于标准通风双笼盒中，所有小鼠均自由进食6%灭菌谷物饲料（5K52，亦称“5KOG”，TestDiet, Purina）。小鼠仅需少量人工操作（称重及其他非侵入性操作），分别于6、12、18月龄时采集3次100μl眶后静脉血样，每次采血间隔2周。所有小鼠均饲养至其自然死亡。动物房采用12小时光照/12小时黑暗的光周期（光照时间为早6:00至晚6:00），环境温度控制在70℉±2℉。 本研究所有实验方案均经杰克逊实验室动物护理与使用委员会审查批准。 ### 基因分型 本研究所有样本的基因型均通过小鼠通用基因分型阵列（MUGA）检测获得（Morgan等, 2016）。采用标准方法从尾尖提取DNA，送往Neogen Genomics（美国内布拉斯加州林肯市）进行分析。不同研究分别采用MUGA（Harrison研究）、MegaMUGA（Shock研究）或GigaMUGA（DRiDO研究）基因分型阵列进行检测。使用R/qtl2软件重构创始单体型，保留基因型检出率≥90%的样本用于后续分析。基因组坐标基于小鼠参考基因组GCRm39，基因位置信息取自小鼠基因组信息数据库（Mouse Genome Informatics, MGI）（Blake等, 2021）。 ### 数据分析 #### 生存分析 我们对各队列动物的生存情况以及研究内不同实验组的生存情况进行了比较：通过绘制Kaplan-Meier（卡普兰-迈耶）生存曲线，采用对数秩检验（log-rank test）比较各队列或实验组的生存分布等价性，同时开展整体检验（跨队列及队列内）以及各实验组与对应研究内对照组的两两比较（例如在Harrison研究中比较雷帕霉素处理组与自由进食对照组）。报告的P值未进行多重比较校正，以P<0.05作为显著性阈值。我们估算了各队列及各实验组的中位寿命。通过Cox比例风险回归模型估算饮食干预和/或性别的效应，以风险比（hazard ratio, HR）及95%置信区间（confidence interval, CI）报告结果，同样以P<0.05作为显著性阈值且未校正多重比较。生存分析采用R语言的“survival”包（Therneau和Grambsch, 2000; Therneau, 2024）完成，绘图通过“ggsurvfit”包（Sjoberg等, 2024）实现。我们采用Gompertz对数线性风险模型估算自干预开始后的死亡率倍增时间与基线风险，通过R语言的“flexsurv”包（Jackson, 2016）计算95%置信区间，并以自由进食雌性小鼠为参照计算相对百分比变化。 #### 全基因组加性扫描 所有遗传分析均采用R语言的“rqtl2”包完成（R Core Team, 2024; Broman等, 2019）。通过scan1()函数开展寿命数量性状位点（Quantitative Trait Locus, QTL）的全基因组扫描，采用混合效应模型，将寿命作为因变量，以每个个体的8个等位基因概率作为自变量。在饮食限制与Harrison研究中，将饮食干预与DO小鼠世代作为加性协变量纳入模型；在Shock研究中，纳入性别与DO小鼠世代作为加性协变量；在元分析中，纳入研究、饮食、性别与DO小鼠世代作为加性协变量。对于每一次全基因组扫描，我们对表型进行1000次随机置换后开展全基因组扫描，记录每次置换扫描中得到的最大LOD（对数似然比分值）值，以1000次置换得到的最大LOD值分布的95th百分位数作为显著性阈值（ɑ=0.05），LOD值高于该阈值的QTL被认为达到显著性水平。此外，我们还采用LOD≥6作为另一显著性阈值以鉴定寿命变异相关位点：该阈值虽保守性更低，但比此前报道的方法更为严格（Wright等, 2022）。我们报告了每个扫描得到的峰值标记附近的2-LOD支持区间（2LOD SI），即峰值位置两侧LOD值下降2的区间范围。 #### 正向回归分析 在元分析中，我们采用正向回归分析以在搜索额外寿命调控位点时，纳入全基因组显著性QTL的效应。通过scan1()函数构建混合效应模型，纳入饮食干预、性别、DO小鼠世代作为加性效应，并将亲缘关系作为随机效应。除上述协变量外，我们还将此前鉴定得到的全基因组显著性QTL作为加性效应纳入模型。QTL以数值变量编码，代表关联定位得到的最高LOD值标记处的基因型状态。仅将达到置换显著性阈值的QTL作为加性协变量纳入模型。 #### 效应量估算与变异解释率 我们采用“rqtl2”包中的scan1blups()函数计算所有QTL的最佳线性无偏预测值（Best Linear Unbiased Predictor, BLUP）及其95%置信区间，协变量与前述全基因组加性扫描部分一致。每个QTL解释的表型变异量通过以下公式计算： 1 - 10^(-(2/n)*LOD) 其中n为对应数据集的样本量，LOD为该QTL峰值标记的LOD值。 #### 变异关联与精细定位 我们在每个QTL峰值位置两侧2-LOD区间内开展精细定位。采用“rqtl2”包中的scan1snps()函数完成精细定位，协变量与全基因组加性扫描部分一致。本分析使用的变异与基因SQLite数据集可从“rqtl2”用户指南网站获取：https://kbroman.org/qtl2/assets/vignettes/user_guide.html。 #### 饮食与性别特异性位点的单QTL模型 在单个研究中，我们检验了QTL与该研究特有实验因素的交互作用：在Shock队列中检验QTL与性别的交互作用，在其他队列中检验QTL与一种或多种饮食干预的交互作用。由于“rqtl2”包无法将交互作用作为随机效应拟合，因此未开展全基因组水平的交互检验。对于每个QTL，我们采用“rqtl2”包中的fit1()函数，基于全基因组扫描鉴定得到的峰值位置处的8个等位基因概率开展检验。为评估显著性，我们构建了两种模型：①加性模型，纳入实验因素（性别、饮食干预）与DO小鼠世代作为加性协变量；②交互模型，在加性模型基础上额外纳入待检验的实验因素交互项。报告的LOD值为交互模型的LOD值减去加性模型的LOD值。为确定显著性阈值，我们对每个QTL开展1000次置换：先随机置换表型，再分别构建加性模型与交互模型，将得到的1000个LOD值排序后，以95th百分位数作为显著性阈值（ɑ=0.05）。若QTL与实验条件的交互作用LOD值高于该阈值，则认为其具有显著性。交互效应以校正加性协变量后的残差寿命值随性别的变化情况进行可视化展示。 #### 饮食与性别特异性位点的全基因组扫描 我们采用Python 3.8.16版本中的“do-qtl”软件包运行基因-环境混合效应模型（Gene by Environment Mixed Effects Model, GxEMM）（Wright等, 2022）。模型中纳入研究、性别与饮食作为固定效应，亲缘关系与饮食作为随机效应。P值基于1000次数据置换计算得到，以p≤1×10^-4作为统计学显著性位点的判定阈值（Wright等, 2022）。