Body size predicts how long ant workers live - but not how they age or how they die from heat

Alana Moscardi; Rafael da Silva; Gleycon Silva

q-bio.PEAug 14, 2026

Body size predicts how long ant workers live - but not how they age or how they die from heat

Authors: Alana Moscardi, Rafael da Silva, Gleycon Silva

Organizations: Graduate Programme in Ecology, Instituto Nacional de Pesquisas da Amazônia (INPA), Manaus, Brazil · Graduate Programme in Data Science, Eastern University, St. Davids, PA, USA

Abstract

In social insects, mortality risk comprises distinct components that may not share the same predictors: lifespan duration, senescence trajectory, and thermal vulnerability. We tested these three axes in 18 Australian ant species using paired field-laboratory survival assays (2,363 cohort-day observations; 1,148 workers). Body size predicted duration (Cox HR = 0.67, p = 0.002), while colony size (p = 0.60) and the size x temperature interaction (p = 0.72) showed no detectable moderating effect. A weak but significant size x foraging-rate interaction was detected (LRT p = 0.014), suggesting that intrinsic physiology remains the most parsimonious explanation for the main size-longevity pattern, although ecological context may contribute. Senescence trajectory was associated with circadian niche rather than size: it was steepest in matinal species (Kruskal-Wallis p = 0.009; matinal vs. crepuscular p = 0.002) and was uncorrelated with body mass (Spearman p = 0.32). Thermal hazard plateaued above 20 degrees C (Delta AIC = -38; p < 0.001), with elevated thermal sensitivity in Rhytidoponera (Ectatomminae) above the plateau (5% per degree C, p = 0.015). Circadian regime and lineage identity, not body size, therefore emerge as the most climate-relevant axes, although they are strongly collinear (Cramer's V = 0.85). These results show that body size captures only one dimension of mortality risk and that size-based vulnerability indices may misrank taxa when senescence and thermal sensitivity are decoupled from body size.

Figures & tables

Figure 1: Conceptual framework: three decoupled axes of worker mortality risk. Body size predicts only the duration of life (HR = 0.67); it is statistically decoupled from the senescence trajectory ( $\rho$ –mass Spearman $p=0.32$ ) and from thermal vulnerability (size $\times$ temperature $p=0.72$ ). The trajectory axis is associated with circadian activity regime (matinal species senesce fastest; Kruskal–Wallis $p=0.009$ ), and the thermal axis is lineage-patterned (elevated thermal sensitivity in the Rhytidoponera lineage (Ectatomminae); PGLS $p=0.081$ , corroborated by individual-level Cox $p=0.015$ ). Circadian regime and lineage identity are strongly collinear in this assemblage (Cramér’s $V=0.85$ ) and cannot be fully separated at $N=18$ species; they are therefore drawn as complementary, not independent, predictors. The three climate-relevant signals converge on matinal Rhytidoponera (Ectatomminae). Each axis is labelled in text on the schematic so the diagram is interpretable independently of colour. Abbreviations: HR, hazard ratio (Cox proportional-hazards model); $\rho$ , Weibull shape parameter (senescence rate); KW, Kruskal–Wallis test; PGLS, phylogenetic generalised least squares; Cramér’s $V$ , association statistic between two categorical variables.

Analysis	Outcome
Q1. What mechanism links body size to longevity?
Claim 1. Body size predicts duration; intrinsic physiology inferred by exclusion.
Larger workers live longer (H1)	$\checkmark$ HR = 0.67, $p=0.002$
Colony size does not modulate it (H2a)	$\circ$ $p=0.60$
Foraging rate modulates it weakly (H2b)	$\checkmark$ LRT $p=0.014$ (M2)
Body size does not buffer heat (H3)	$\circ$ HR = 1.00, $p=0.72$

Table 1: Analytical roadmap: questions, claims, internal hypotheses, and outcomes. Internal labels (H1–H9) cross-reference the methods, code, and supplementary tables.

Figure 2: Representative study organism: Rhytidoponera sp. (Ectatomminae). A worker of Rhytidoponera sp., one of three ectatomminine species included in the field mark–recapture dataset (Ectatomminae: R. metallica , R. tasmaniensis , R. victoriae ). This subfamily is distinguished by notably low colony size (mean $\approx$ 170 workers) relative to myrmicine and dolichoderine species, the steepest actuarial senescence trajectories in laboratory survival assays ( $\rho_{\rm lab}\approx 1.4$ – $2.0$ ), and elevated thermal hazard sensitivity in the field. Photo: Rhytidoponera sp. by P. Kavanagh ( https://www.flickr.com/photos/patrick_k59/49413794622 ), CC BY 2.0.

Scale of inference	Factor(s) applied	Number of replicates
Species ( $N=18$ )	Body mass (log 10 , centred); circadian niche (3 groups); subfamily (5 groups)	18 species
Cohort	Daily mortality event; observation day	106 cohorts, 39 colonies
Assemblage	Daily maximum air temperature ( maxt , °C)	2,363 cohort-day observations
Individual (lab)	Body mass; colony identity	1,148 workers, 39 colonies

Replication Statement (Functional Ecology requirement)

Figure 3: Forest plot of hazard ratios from the main Cox proportional-hazards model M0 (Claims 1 and 3). Each row shows a predictor with its point estimate and cluster-robust 95 % CI (cluster = cohort_id ). Marker shape encodes effect type independently of colour: diamonds mark statistically significant predictors (Wald

p<0.05

: body size and maximum temperature); crosses mark null predictors (body size

\times

temperature interaction); circles mark subfamily baseline-hazard contrasts relative to the Dolichoderinae reference. Vertical dashed reference line at HR = 1. Model fitted on

N=2{,}363

cohort-day observations (1,826 events). M0, primary Cox model ( logw_c + maxt + C(SF) ); HR, hazard ratio; CI, confidence interval; SF, subfamily.

Covariate	$\bm{\beta}$	SE	HR	95 % CI	p
Body size (log 10 mass, centred)	$-$ 0.398	0.130	0.67	[0.52, 0.87]	0.002
Max. temperature (°C)	$+$ 0.035	0.010	1.036	[1.016, 1.057]	$<0.001$
Subfamily: Ectatomminae	$-$ 0.600	0.210	0.55	[0.36, 0.83]	0.004
Subfamily: Formicinae	$-$ 0.105	0.193	0.90	[0.62, 1.32]	0.589
Subfamily: Myrmeciinae	$-$ 0.399	0.280	0.67	[0.39, 1.16]	0.155
Subfamily: Myrmicinae	$-$ 1.205	0.231	0.31	[0.19, 0.47]	$<0.001$

Table 3: Main Cox proportional-hazards model (M0) coefficients (Claims 1 and 3, H1, H3, H7). Reference category: Dolichoderinae. HR, hazard ratio; CI, 95 % confidence interval;

z

, Wald test statistic.

Figure 4: Kaplan–Meier survival curves by body-size quartile (Claim 1, H1). Field workers are grouped into four body-size quartiles (Q1 = smallest,

N=620

; Q2,

N=563

; Q3,

N=596

; Q4 = largest,

N=584

; quartile boundaries defined from the full cohort-day field dataset). Shaded bands are pointwise 95 % confidence intervals; vertical ticks mark censored observations; the number of individuals at risk is tabulated below the time axis. The stepwise separation between quartiles illustrates the intrinsic size–longevity relationship: larger workers survive significantly longer across the full observation window (log-rank

p<0.001

; Cox HR per log 10 unit = 0.67, 95 % CI [0.52–0.87], model M0, Table 3 ).

Figure 5: Temperature–mortality relationship under the piecewise spline model (Claim 3, H8). Panel A: Log-hazard ratio (centred at the assemblage mean daily maximum temperature,

\bar{T}_{\max}=22.8

°C) as a function of daily maximum temperature ( maxt ), with the estimated breakpoint

T^{*}=20

°C marked by a dashed vertical line. The shaded band is the pointwise 95 % confidence interval derived from the cluster-robust variance matrix of the piecewise Cox model; faint background shading distinguishes the rising segment (below 20 °C; HR = 1.216/°C) from the near-plateau segment (above 20 °C; HR = 1.016/°C). The piecewise model fits substantially better than a linear model (

\Delta

AIC =

-38

; Davies permutation

p<0.001

; Table 8 ). Panel B: Partial-effect hazard curves for each body-size quartile over the same temperature range, with all other covariates held at their means. The four curves are parallel, consistent with the null size

\times

temperature interaction (H3, Table 3 ), and the plateau is equally present across all size classes.

Term	$\bm{\beta}$	SE	$\bm{t}$	95 % CI	p
Intercept	71.9	35.5	2.02	[ $-$ 1.0, 144.7]	0.053
Subfamily: Ectatomminae	10.2	44.5	0.23	[ $-$ 81.1, 101.6]	0.820
Subfamily: Formicinae	7.5	38.4	0.20	[ $-$ 71.3, 86.3]	0.846
Subfamily: Myrmeciinae	50.6	51.0	0.99	[ $-$ 54.0, 155.2]	0.329
Subfamily: Myrmicinae	51.4	46.1	1.11	[ $-$ 43.3, 146.1]	0.275
Body size (centred)	13.3	18.1	0.73	[ $-$ 23.8, 50.4]	0.470

Table 4: Colony-level OLS regression of total lifespan on body size and colony size (H2a). Interest is the body-size

\times

colony-size interaction (null hypothesis: colony dilution does not moderate the size benefit). Reference: Dolichoderinae.

Model	k	log-lik	AIC	Concordance	$\bm{\chi^{2}(1)}$	p
M0: logw_c + maxt + C(SF)	6	$-12264.88$	24541.8	0.637	—	—
M0 fr : M0 $+$ fr_c	7	$-12264.67$	24543.3	0.637	$0.42^{a}$	0.52
M2: M0 fr $+$ logw_c $\times$ fr_c	8	$-12261.63$	24539.3	0.636	$6.08^{b}$	0.014

Table 5: Cox proportional-hazards model comparison (Claim 1, H2b). All models include cluster-robust standard errors (cluster = cohort_id ), subfamily fixed effects (5 levels, Dolichoderinae reference), and daily maximum temperature. a LRT

\chi^{2}(1)

vs. M0; b LRT

\chi^{2}(1)

vs. M0 fr .

Figure 6: Weibull shape parameters

\rho

by species, estimated from laboratory survival data (Claim 2, H4, H5). Species are sorted in ascending order of

\hat{\rho}_{\rm lab}

. Horizontal bars are 95 % confidence intervals from maximum-likelihood Weibull fits to individual-level laboratory survival; sample sizes (

n

individuals per species) are annotated to the right of each estimate. The dashed vertical line at

\rho=1

separates frailty selection (

\rho<1

) from senescence (

\rho>1

). Points are coloured and shaped by circadian activity niche: matinal (blue triangle

\blacktriangle

), diurnal (orange square

\blacksquare

), crepuscular (purple circle

\bullet

) — dual encoding makes the grouping legible in greyscale and for colour-vision deficiencies. Fourteen of 18 species have

\rho>1

(actuarial senescence). The distribution of

\rho

across niches illustrates the circadian structuring of trajectory: matinal species concentrate at higher

\rho

values (Kruskal–Wallis

H=9.35

p=0.009

; matinal vs. crepuscular

p=0.002

Species	Subfamily	Niche	$\bm{\rho_{\rm lab}}$	95 % CI	$\bm{\lambda_{\rm lab}}$	Median (d)	Pattern
							( $n$ )
Iridomyrmex notialis	Dolichoderinae	D	2.51	[1.72, 3.30]	22.5	19.4	Senescence (30)
Rhytidoponera tasmaniensis	Ectatomminae	M	1.99	[1.65, 2.33]	58.5	48.7	Senescence (90)
Meranoplus fenestratus	Myrmicinae	M	1.85	[1.52, 2.17]	83.6	68.6	Senescence (90)
Crematogaster laeviceps	Myrmicinae	M	1.82	[1.30, 2.34]	35.5	29.0	Senescence (30)
Rhytidoponera victoriae	Ectatomminae	M	1.70	[1.33, 2.07]	55.6	44.8	Senescence (60)

Table 6: Weibull shape parameter

\rho_{\rm lab}

estimated from laboratory mortality data, with species ordered by descending

\rho_{\rm lab}

(Claims 2, H4, H5). Circadian niche is assigned from peak foraging hour (M = matinal, D = diurnal, C = crepuscular). The shape parameter classifies the trajectory:

\rho>1

, senescence (rising hazard);

\rho\approx 1

, constant hazard;

\rho<1

, frailty selection (declining hazard).

\lambda_{\rm lab}

is the Weibull scale parameter (characteristic lifetime, in days); Median (d) is the model-implied median lifespan. 95 % CIs are from the maximum-likelihood Weibull fit ( lifelines WeibullFitter ); the parenthetical value in the Pattern column is

n

, the number of individuals assayed per species. The two bottom rows summarise the assemblage: mean / median

\rho_{\rm lab}

, and the count of senescent species (

\rho_{\rm lab}>1

) out of 18.

Species	Niche	$\bm{\rho_{\rm lab}}$	$\bm{\rho_{\rm field}}$	$\bm{\Delta\rho}$	Direction
Meranoplus fenestratus	Matinal	1.85	3.28	$+1.43$	$\uparrow$
Rhytidoponera metallica	Matinal	1.38	0.54	$-0.84$	$\downarrow$
Rhytidoponera tasmaniensis	Matinal	1.99	2.94	$+0.95$	$\uparrow$
Rhytidoponera victoriae	Matinal	1.70	3.59	$+1.88$	$\uparrow$
Anonychomyrma nitidiceps	Matinal	1.56	2.12	$+0.57$	$\uparrow$
Chelaner cinctum	Matinal	1.46	1.34	$+0.13$	$\uparrow$

Table 7: Field amplification of mortality trajectory (H6):

\Delta\rho=\rho_{\rm field}-\rho_{\rm lab}

per species. Positive

\Delta\rho

indicates steeper senescence in the field. Species are grouped by circadian niche.

\dagger

= field

\rho

from pooled AFT fallback.

Model	AIC	$\bm{\Delta}$ AIC	LRT p	Note
Linear (M0, reference)	24,542	0	—	Baseline
Piecewise spline, $T^{*}=20$ °C	24,504	$-$ 38	$<0.001$	Best breakpoint

Table 8: Piecewise spline model of the temperature–hazard function (H8, Claim 3), in two panels. Upper panel: model comparison; the best breakpoint

T^{*}=20

°C was selected by minimising AIC across

T^{*}\in\{15,16,\ldots,34\}

°C. Lower panel: marginal thermal hazard within each spline segment.

\Delta

AIC, change in Akaike information criterion relative to the linear baseline (negative values favour the piecewise model); HR/°C, hazard ratio per 1 °C increase in daily maximum temperature;

T^{*}

, estimated breakpoint; LRT, likelihood-ratio test; a dash (—) denotes a quantity that is not applicable (the linear model is the reference).

Covariate	Test statistic	p
Body size ( logw_c )	1.854	0.173
Max. temperature ( maxt )	1.410	0.235
Subfamily: Ectatomminae	0.292	0.589
Subfamily: Formicinae	0.323	0.570
Subfamily: Myrmeciinae	0.569	0.451
Subfamily: Myrmicinae	1.529	0.216

Table S1: Proportional hazards assumption test via Schoenfeld residuals for the main Cox model M0 ( logw_c + maxt + C(SF) ,

N=2{,}363

cohort-day observations, 1,826 events, cluster = cohort_id ). Test statistic and two-sided

p

-value from lifelines.statistics.proportional_hazard_test using the rank-transformed time axis (Grambsch–Therneau test, rank variant); the rank transform reduces sensitivity to the baseline hazard shape and improves power against smooth time-varying alternatives. logw_c : log 10 body mass, mean-centred; maxt : daily maximum temperature (°C). No covariate shows evidence of a time-varying hazard ratio at

\alpha=0.05

, supporting the PH assumption for M0.

Model	Species control	HR ${}_{\text{logw}}$	p
M0	Subfamily fixed effects (ref)	0.672	0.002
M0_spp	Species fixed effects (18 dummies)	0.770	0.471
M0_strat	Stratified Cox by species	0.710	0.354

Table S2: Species-frailty robustness check for H1. Three nested Cox models are compared to assess whether the body-size hazard ratio (HR = 0.67 in M0) is robust when species identity is controlled. M0_spp adds species fixed effects (18 dummies absorbing all between-species variation); M0_strat uses species as strata, allowing species-specific baseline hazards. Both variants yield a directionally consistent but statistically non-significant HR, indicating that the M0 coefficient reflects a predominantly interspecific size–longevity relationship.

Term	Coef.	SE	t	p
Intercept (Dolichoderinae)	0.018	0.016	1.16	0.277
Ectatomminae	0.048	0.024	1.97	0.081
Formicinae	0.024	0.024	0.98	0.355
Myrmeciinae	0.034	0.036	0.95	0.367
Myrmicinae	0.035	0.024	1.44	0.185
Joint $F(4,9)=1.09$				0.416

Table S3: Phylogenetic generalised least-squares (PGLS) analysis of subfamily differences in thermal sensitivity (primary analysis for between-species comparisons, Claim 3). Response variable: per-species

\hat{\beta}_{\text{maxt}}

from individual-species Cox models (

N=14

of 18 species converged). Predictor: subfamily (reference = Dolichoderinae). Phylogenetic covariance matrix derived from the ultrametric mr_phylo.tre tree [ 29 ] , scaled by Pagel’s

\hat{\lambda}=0

(LRT

p=1.00

; no detectable phylogenetic signal), so PGLS reduces to OLS.

Term	$\hat{\lambda}$	LRT p	Conclusion
Intercept (assemblage $\beta_{\text{logw}}$ )	0.06	0.90	No phylogenetic structure; M0 retained

Table S4: Phylogenetic generalised least-squares (PGLS) for the body-size log-hazard signal (A1 robustness check). Response variable: per-species

\hat{\beta}_{\text{logw}}

from individual-species Cox models ( logw_c + maxt ;

N=14

converged). Model: intercept-only PGLS to obtain the assemblage-level phylogenetically corrected mean. Pagel’s

\hat{\lambda}=0.06

(LRT

\chi^{2}(1)=0.02

p=0.90

): no detectable phylogenetic signal, supporting retention of the M0 Cox analysis as the primary model.

Component	$R^{2}_{\rm adj}$	Interpretation
Circadian niche only	0.33	Unique niche contribution
Subfamily only	0.31	Unique subfamily contribution
Full model (niche + SF)	0.19	Combined (not additive due to collinearity)
Shared (overlap)	$-$ 0.45	Redundancy; collinearity confirmed

Table S5: Variance partitioning of

\rho_{\rm lab}

between circadian niche and subfamily (A3 robustness check;

N=18

species). Adjusted

R^{2}

for three OLS models: niche-only, subfamily-only, and combined. Negative shared fraction confirms high collinearity between predictors (Cramér’s

V=0.85

), consistent with the non-significant full model (

F_{6,11}=1.67

p=0.22

Nociception as a Control Primitive: Afferent Channels and Nociceptive Memory for Agents Deployed in One Body
Sep 28, 2026Wolfgang MaassRobotic Control Memory-Augmented Agents
Saarland University and German Research Center for Artificial Intelligence (DFKI)
Emergent Cooperative Thermoregulation in Lunar Swarm Robotics
Oct 22, 2026Anais Jeger, Harry Holt, Lennart PuckSwarm Robotics Space Robotics
Center for Bits and Atoms, MIT · Advanced Concepts Team, ESA · Planetary Robotics Laboratory, ESA
Your Agents Are Aging Too: Agent Lifespan Engineering for Deployed Systems
May 25, 2026Jianing Zhu, Yeonju Ro, John Robertson +5AI Agent Reliability AI Agent Benchmarks
The University of Texas at Austin

Sep 28, 2026J/K move · Enter open · S save

Nociception as a Control Primitive: Afferent Channels and Nociceptive Memory for Agents Deployed in One Body

Wolfgang Maass

Saarland University and German Research Center for Artificial Intelligence (DFKI)

Robotic Control Memory-Augmented Agents

Open paper AskPDF arXiv

An agent deployed in a single body cannot learn how fast that body wears, because every trial that would reveal its wear resistance wears the body it would protect. We study this epoch-one setting, in which the parameters of a fixed-weight policy are set before the body is drawn and never updated in life. The agent carries a load-gated nociceptive channel and a memory that retains what was felt. We prove that felt cost moves the allocation to the best-paid work not yet felt rather than the gentlest, that an agent without retention never sees the felt-cost constraint bind, and that the channel pays only where the threat is individually unpredictable, cheap to avoid and expensive to ignore. We measure per body, setting the agent with channel and memory against the same individual without them, where neither carries a schedule learned across lives. On

2{,}000

simulated floor-layer knees, with wear anchored to published loss rates, feeling, retaining and substituting extends the working life from age

55.2

59.6

and raises career output from

33.7

36.1

69.3\%

of bodies gain and none lose. A body that feels but retains nothing past the day gains one of the

+4.4

years, and retention carries the rest. A population-trained agent gains

+0.65

years from the same channel at

-0.54

output. The difference is what a species prior already supplies, and a single body has none. The two are related by an identity, the ablation mean reporting

(1-χ)

of the per-body value with

χ

the share a blind schedule already captures, so we report both. Where the regime map predicts value, a care robot sextuples its certified service life and a field-anchored fleet writes off

0.15

of its machines instead of

0.55

. Where it predicts none, a rover gains little over blind caution, so the map holds in both directions.

Abstract

Explore similar work

Nociception as a Control Primitive: Afferent Channels and Nociceptive Memory for Agents Deployed in One Body

Emergent Cooperative Thermoregulation in Lunar Swarm Robotics

Your Agents Are Aging Too: Agent Lifespan Engineering for Deployed Systems