Working title : Late-life fitness gains explain the absence of a selection shadow in ants

A key hypothesis for the occurrence of senescence is a decrease in the selection strength because of low late-life fitness – the so-called selection shadow. However, in social insects, aging is considered a plastic trait and senescence seems to be absent. By life-long tracking of 102 ant colonies, we find that queens increase sexual productivity in late life regardless of their absolute lifespan or worker investment. This indicates a genetically accommodated adaptive shift towards increasingly queenbiased caste ratios over the course of a queens’ life. Furthermore, mortality decreased with age, supporting the hypothesis that aging is adaptive. We argue that selection for late life reproduction diminishes the selection shadow of old age and leads to the apparent absence of senescence in ants, in contrast to most iteroparous species.

When considering social insect reproductives (e.g. ant queens), the basic assumptions underlying MA and AP may not be met (Monroy Kuhn and Korb 2016;Toth et al. 2016;Lucas and Keller 2017).
Extrinsic mortality is minimized as queens live in a sheltered fortress. Keller and Genoud (Keller and Genoud 1997) argue that low extrinsic mortality of queens maintains a high selection pressure with age, while the costs of reproduction (energy intake, brood care) are outsourced to the workers. In addition, it is posited that selection at high ages is reinforced if there is no marked decrease in fecundity after maturity (Keller and Genoud 1997), predicting that classic energy trade-offs may be shaped differently or may not even exist. This has found support in studies of the model ant Cardiocondyla obscurior (Schrempf et al. 2005;Schrempf et al. 2011;Oettler and Schrempf 2016) where queens exhibit a positive correlation between lifespan and reproduction (Kramer et al. 2015).
Furthermore, experimental manipulation of the fecundity of C. obscurior queens does not affect lifespan  and queens appear to have only a very brief period of senescence (Heinze and Schrempf 2012). Additional evidence for the lack of senescence comes from a gene expression study, which showed a lack of transcriptional ageing in old C. obscurior queens, possibly facilitated by strong selection at old age and by well-regulated anti-ageing mechanisms (Harrison et al. 2021). So how did this exceptional pattern of positive correlation between fecundity and lifespan, and the absence of senescence, evolve, and are there other trade-offs that shape queen lifespan instead?
In order to disentangle aging and the investment into somatic maintenance and sexual reproduction of ant queens it is necessary to consider the colony as a superorganism (Boomsma and Gawne 2018), comprising a soma-(i.e. workers) and a germline (i.e. queens and males), where the longevity and mortality of the individual affects the fitness of the queen (Bourke 2007;Kramer and Schaible 2013). Therefore, by manipulating the investment into caste ratio by lifelong standardization of worker number, we expected to find trade-offs at the level of the germline (i.e. lifespan and productivity of queens) and/or at the level of the colony (investment in next-generation soma-or germline), if they exist. To this end, we monitored the lifetime production of 102 individual queens in colonies, which were standardized weekly to 10, 20 or 30 workers ( Fig. S1A and Fig. S1B), corresponding to natural colony size variation (Schrader et al. 2014, Fig. S2). In order to compare queen mortality with worker aging patterns, we tracked the survival of 40 workers kept in colonies with 10 or 20 marked nestmate workers.

Results and Discussion
We hypothesized that queens which experienced a worker shortage would compensate by investing less into queen production. This assumes queen control over caste fate in C. obscurior, in contrast to many species where caste is environmentally or genetically determined (Corona et al. 2016). Indeed, queens with 10 workers (n=31) produced significantly fewer queen pupae than queens with 20 (n=34) (glmmTMB z-value = 2.81, IRR = 1.97, p = 0.005) and 30 workers (n=34) ( z-value = 2.58, IRR = 1.78, p = 0.009, Fig. 1A) with no significant differences between 20 and 30 workers (z-value = -0.49, p = 0.877). We used a correction factor (c) based on dry weight of individuals to estimate the energy allocated to castes, considering the higher costs of queen compared to worker production, following the rationale for estimating investment sex ratios (Boomsma et al. 1995). Queens' total investment was higher in colonies with 20 and 30 workers (defined as queen*c/(queen*c + worker pupae), 10 vs. 20 workers: z-value = 4.42, IRR= 0.095, p <0.001 and 10 vs. 30: z-value = 4.08, IRR=0.086, p < 0.001, Fig. 1B), with no significant differences between queens with 20 and 30 workers (z-value = -0.45, p = 0.90). The factor c takes differences in investment into the two phenotypes into account and is sensitive to a power-conversion factor for differences in respiratory rates (see Methods). The treatment did not affect total egg production ( Fig. 1C, 10 vs. 20 workers: z-value = -0.38, p = 0.70 and 10 vs. 30: z-value = -0.96, p = 0.34) or worker pupae production ( Fig 1D, 10 vs. 20 workers: zvalue = 0.09, p = 0.93 and 10 vs. 30: z-value = -0.39, p = 0.70). We explored whether the quality of workers was affected by measuring the head width of workers produced over months 3 to 6 of the queen's lifetime (~5 workers per month). Head width was 7.5 µm and 11.2 µm significantly smaller in workers produced in small colonies than in colonies with 20 (lme F6,76 = 2.36, p = 0.026) and 30 workers respectively (lme F6,76 = 3.52, p < 0.002, Fig. 2).
We could not detect a "terminal investment" strategy where queens increased their productivity until death as described for C. obscurior queens from a Brazilian population (Kramer et al. 2015), in queens from a Japanese population used for this experiment (Fig. 3A). Worker production tightly matched the curve of egg production ( Fig 4A). Importantly, queen-biased caste allocation increased late in life irrespective of treatment ( Fig 3B), in both short-lived and long-lived queens (queens with a lifespan below and above the mean lifespan of 25 weeks, Fig. 3C and D respectively). Variation in caste investment is not correlated with queen lifespan, similarly to fecundity .
To compare mortality and fecundity patterns we mean-standardized queen age-specific mortality and fecundity (Jones et al. 2014). After an earlier increase in relative mortality, C. obscurior ant queens exhibit a decrease below the average level of adult mortality late in life (Fig. 4A), indicating selection against senescence. Interestingly, the mean-standardized mortality of workers is very similar to the queen's aging pattern, where relative mortality also first increased then decreased (Fig 4B).
Furthermore, in contrast to the prediction that there is no marked decrease in queen fecundity after maturity (Keller and Genoud 1997), relative fecundity reaches a maximum (~16 weeks) before the median lifespan (~26 weeks). However, the relative investment in queen pupae reaches a maximum late in life (~28 weeks). This pattern is not due to the delay in development from egg to pupa, because queen development only lasts ~5 weeks (Schrempf and Heinze 2006). A first peak in queen-biased investment occurs at an earlier age, followed by an increasing queen bias with increasing age ( Fig   3A).
This caste ratio shift is independent of queen lifespan (Fig 3B, C), implying that queens are selected for late life productivity. Founder queens invest first in growing numbers of workers (ergonomic phase) and subsequently in the production of new sexuals, when the colony has reached the threshold required to enter the reproductive phase (Macevicz and Oster 1976;Oster and Wilson 1978;Beekman et al. 1998). Our data suggest that this switch in caste allocation is a fixed trait, independent of colony size and absolute lifespan of queens.
Senescence is defined as an increase in relative mortality with age due to a decrease in the strength of selection, and a decrease in relative fecundity (Hughes and Reynolds 2005). Strikingly, C. obscurior queens exhibit a decrease in relative mortality with age. Further, while fecundity decreases, investment into queen pupae reaches a maximum late in life. In addition, because C. obscurior is a polygynous species, effective removal of senescent individuals with diminishing fecundity from the population, and replacement with young queens, might lead to additional selection for a short senescent phase. Thus, C. obscurior queens continue to experience strong selection at high ages. In turn, the short selection shadow might have led to the absence of a distinguishable senescence phase.
Whether this holds true for social insects in general, and whether this underlies the evolution of the extraordinary lifespans found in social insects, remains to be studied.  (Table S1). Eggs Queen Pupae Worker Pupae Mortality produced offspring originated from the focal queen. Queen control over caste fate was assumed, as caste fate can be determined as early as the last embryonic stage. The number of counted eggs correlates with the production of workers, queens, and the workers and queens together ( Fig. S4.A-C, Kendall's Tau correlation test, p < 0.001: eggs-worker pupae= 0.59, eggs-queen pupae= 0.70, eggs-worker and queen pupae=0.73). Pupae might have been counted more precisely than eggs, especially when larger number of eggs were produced.
Pupae are hardly missed, compared to eggs which tend to cluster together. Eggs and worker pupae might have been counted more than once, as development lasts a median of 8 and 18 days for eggs and worker pupae, respectively. Finally, three colonies (10 worker treatment) were not considered in the analysis as they were accidentally killed.
To examine the worker aging patterns, 40 focal worker pupae were set up in individual colonies with 10 or 20 marked workers. These two treatments were selected, as no significant differences were observed between the 20 workers and 30 workers colonies in terms of queen productivity. Marking was done by clipping the tarsae of the middle right leg. The colony was set up with brood (5 larvae in the 10 workers colonies, and 10 in the 20 workers colonies), and two wingless queens to avoid a queenless period. The number of marked workers, queens and larvae was standardized weekly to the assigned treatment, and newly produced pupae produced were removed. Dead marked workers were replaced with fresh worker pupae and clipped one or two days after eclosion to avoid confusion with the focal worker.

Offspring investment
Freshly eclosed adult workers were sampled monthly for head width measurements (from the 3rd to 6th month of the queen's life, and up to five workers depending on availability).
Workers were dried, pinned, and blindly measured using a Keyence Microscope 200X. A single worker was chosen randomly and measured 10 times to obtain a proxy for measurement error (Mean = 383.61 μm, standard deviation=5.05 μm). Additionally, newly eclosed worker and queen adults from the stock population were dried at 60°C for 72h and five individuals were pooled to weigh to the nearest 0.1 μg with a fine scale (Sartorius Micro SC2) for biomass investment calculations, until 50 individuals from each were measured.
Caste allocation was measured including biomass weight adapting the formula for sexual investment as = ( ! " ) # , where q and w are the queen and worker dry weight and the power conversion power is k=0.7 (Boomsma et al. 1995), as commonly used for differences in investment among queens and males. We calculated caste investment (I) as $ = * $ + $ (Boomsma 1989) and explored differences in caste allocation for a range of k values between 0.6-1. This power of conversion might be species specific, and the exact value for C.
obscurior is unknown but is likely to be within this range.

Statistical tests
To test for significant differences between treatments, we used generalized linear mixed effects models within the R package glmmTMB (R version 3.5.2, (Pinheiro et al. 2011)) and a negative binomial distribution for count data. If the count data and caste investment ratios were log transformed, a Gaussian family distribution was used. The dependent variable was analyzed as a function of the fixed effects: treatment (Number of workers as a factor), and random effects: stock nest and box of origin, box of set up, set up date. All models were also graphically checked for consistency and model diagnostics were performed using the DHARMA package (R version 0.3.3.0, (Hartig 2020)). To test for differences in head width size, we used the average of the head width measurements of the workers per time point (each month). Data is available as external database S1, and the R-script is filed under database S2.