2025 - 2026 M2 Master’s thesis Defending the colony against infection progression Investigating allogrooming variations and survival after fungal exposure in different ages of the clonal raider ant, Ooceraea biroi, alongside an extended automated behavioral tracking approach Drawing by Leo KORST Written by the MSc student Léo KORST Ecophysiology, Ecology and Ethology Master degree Strasbourg’s University, Strasbourg, France Supervised by Sarah Rogoz, and Yuko Ulrich PhD student, and Research group leader in Ulrich lab Max Planck Institute for Chemical Ecology, Jena, Germany

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 Acknowledgments Firstly, I want to thank Sarah Rogoz, for her supervision, her help during the experiments, our lonnnng discussions on the experimental designs that we had, and also our great teamwork during this Master's thesis. Being your first Master's student was fabulous for me: Sarah is a great manager and supervisor, and gave me enough autonomy and responded to my questions when I needed them! Discovering AI tracking and computational analysis with you was also a great opportunity for me. Secondly, I want to say thank, Yuko Ulrich, our group leader, for this wonderful opportunity you gave me to be a part of your lab team. It was an incredible adventure for me to work with your team and your ant model! Your comments and remarks on my work and on my personality in the lab gave me more enthusiasm for my future scientific research and projects! You are a great leader and researcher for this lab! Then, I want to say thank you to all the lab members: Sandra, Bhoomika, Lai Ka, Irene, Tim, Tiago, Silke, Roland, Arun, Anika, Yun, and Jonghyun! I shared beautiful moments and conversations with all of the team. I discovered great people, good lab members, and friends for me. Thank you for the positivity and all laughing moments we shared, outside and inside the lab. I will not forget! Thank you for the events inside the lab, such as the famous Fruity Friday, and outside the lab with the movie nights and others! I also want to say thank you, Hala and Naeem, who were not directly part of the ant lab, but part of the lab family! The moments I shared with you were handsome, the conversations and laughing Hala, the conversations and the new passion for wood carving with you, Naeem! All of this would not be possible without the support of my family, my parents, my brother, and my sister during these months. Thank you also to you, Chimène, who listened to me every day about the ants I painted and the long experiment days I did, and you gave me soooo much advice and support during this whole Master's thesis! I also want to say thank you to the people I met in Jena, the Urban Sketchers group, and the other people. I shared some great time outside the lab with these people, and even outside, the ants were not too far from me and my passion for drawing.

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 Contribution of author Experimental designs: 50% (collaboration with Sarah Rogoz and Yuko Ulrich) Collecting data: 60% (coordination with Sarah Rogoz on data collection between the pathogen lab and tracking chamber) Data treatment: 90% Statistical analysis: 90% Report writing: 100%

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 Abbreviations B. bassiana…………. BORIS……………… CI…………………… CNN………………… DF…………………... GLMM…………....... KS………………....... LM………………….. LMM……………….. LOA………………… LR………………....... O. biroi……………… PDA………………… Pre-E……………...... Post-E………………. SD………………....... SE…………………... SLEAP……………... Beauveria bassiana Behavioral Observation Research Interactive Software Confidence interval Convolutional neural network Degree of Freedom Generalized linear mixed model Kolmogorov-Smirnov Linear model Linear mixed model Limits of agreement Likelihood ratio Ooceraea biroi Potato Dextrose Agar Pre exposure Post exposure Standard deviation Standard error Social LEAP Estimates Animal Poses

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 Table of contents Introduction ........................................................................................................................................ 0 1. Sociality effects ........................................................................................................................ 1 2. Social immunity and mechanisms ............................................................................................ 1 3. Entomopathogenic fungi .......................................................................................................... 2 4. Case study, the clonal raider ant, Ooceraea biroi ................................................................... 3 5. Research question and hypothesis ........................................................................................... 4 Materials and Methods ...................................................................................................................... 5 1. Ant colonies maintenance ........................................................................................................ 5 2. Fungal cultures and preparation for exposure ........................................................................ 6 3. Experiment 1: Allogrooming response in dependency of age and infection stage .................. 6 4. Experiment 2: Effect of social environment on survival .......................................................... 7 5. Video recording and behavioral scoring for the experiment 1 ................................................ 7 a. SLEAP Convolutional neural network ................................................................................ 8 b. BORIS software ................................................................................................................... 8 6. Statistical analysis.................................................................................................................... 9 a. SLEAP performance and filters accuracy analysis .............................................................. 9 b. Grooming analysis with BORIS results ............................................................................. 10 c. Survival analysis ................................................................................................................ 10 Results ............................................................................................................................................... 11 1. Experiment 1 .......................................................................................................................... 11 a. SLEAP performance and grooming classification filters accuracy ................................... 11 b. Fungal pathogenicity check by survival analysis ............................................................... 12 c. Allogrooming response in ant colonies across infection progression ................................ 12 2. Experiment 2 .......................................................................................................................... 14 a. Social context ..................................................................................................................... 14 b. Age composition ................................................................................................................ 15 Discussion.......................................................................................................................................... 15 1. Extend grooming classification approach using SLEAP pose estimation ............................. 15 2. Defending the colony across the fungal infection .................................................................. 16 3. The sensitivity of the social context young ants survival ....................................................... 18 Conclusion......................................................................................................................................... 18 Future directions .............................................................................................................................. 19 Bibliography .........................................................................................................................................

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 1 Introduction 1. Sociality effects Sociality is described as a cooperative behavior among individuals in a group of the same species (Wilson, 1975). In this social context, interactions produce constraints and a personal and collective balance between costs and benefits (Lewin, 1987). An example in the foraging process demonstrates that the presence of other individuals increases the risk of visibility for predators but also increases the dilution effect and the group defense. Moreover, individuals can also help to find food. However, the members of the group can also create a food competition within the group (Lewin, 1987 ; Ward & Webster, 2016). These different balances exist in many different taxa such as mammals, birds, or insects, and differ according to their sociality level (Ward & Webster, 2016). Effectively, the social organization can be described as a gradient from solitary, with social interaction only for reproduction, to complex systems with reproduce castes, and care of offspring from the other members of the group (Ward & Webster, 2016). The highest described level of sociality is called eusociality with a specific caste for reproduction, an overlap between different generations, care towards brood or young individuals in the group (Ward & Webster, 2016). Among the highest level of social organization in animals, insect societies show an advanced system of cooperation between conspecifics and division of labor, with natural selection allowing right number of individuals to be adjusted for an ergonomic efficiency of the colony (Wilson, 1978). Local division of labor and information transfer also induces subgroups within the colony, creating a general global organization (Wilson, 1978). Nevertheless, insect societies show common characteristics, such as multiple interactions, thigh density in a colony, and genetic proximity. These features increase sanitary risks due to pathogen spread between the individuals (Cremer et al., 2007). 2. Social immunity and mechanisms To counter sanitary risks within the group, insect societies developed various collective defense strategies termed “social immunity”. These immunity defenses include personal and collective mechanisms developed by the group members against pathogens (Cremer et al., 2007 ; Meunier, 2015). The social immunity is also composed of physiological and behavioral factors at different stages (Cremer et al., 2007 ; Liu et al., 2019).

Léo KORST - Master’s thesis 2026 Figure 1. Infection process of the fungal pathogen Beauveria bassiana on insects (Designed by Leo Korst, based on Ortiz-Urquiza & Keyhani, 2013 ; Ortiz-Urquiza et al., 2015 ; Mascarin & Jaronski, 2016).

Léo KORST - Master’s thesis 2026 2 The preventive measure in social immunity is to reduce the uptake and intake of the pathogen in the colony, using special castes as guards of the entrance of the colony in bees, or by avoiding the infected area in ants and termites species (Cremer et al., 2007). Upon exposure to the pathogen, environmental hygiene reduce pathogen establishment by antimicrobial substances utilization to clean different parts of the nest ; this mechanism is also used in fungal contamination context in bees nests (Cremer et al., 2007 ; Liu et al., 2019). However, when the pathogen is already inside the colony, the colonies use reactive measures such as social organization. The latter can be modulated to reduce the risk of pathogen spread, using subgroups, which function as protecting layers to protect the center of the nest. Thus, the division of labor plays a role in who interacts with the pathogen primarily, oftentimes resulting in the older individuals outside of the nest (Cremer et al., 2007). Then, the ergonomic efficiency of social organization in the colony is remodulated by collective behaviors to avoid the transmission between the individuals, using restricting interaction between some subgroups, active isolation and collective movement towards the infected individuals to rescue them or reject them depending on the sanitary context of the colony (Cremer et al., 2007). Furthermore, the social immunity has some different levels of actions depending on the pathogen risk for the colony and the group members. Local reactions to information within the subgroups lead to a global defense reaction pattern in a social context (Ward & Webster, 2016). 3. Entomopathogenic fungi Among the pathogen risk, the pathogen type can produce a different reaction in host and also induce specialized social immunity mechanisms. In this context, the killer entomopathogenic fungi use a specific pathway to kill their host using their conidia, the asexual spores released by the fungus. There are different infection stages after fungal exposure: conidial adhesion and germination, cuticle penetration by the long cell filaments, the hyphae, and colonization until the complete invasion and death of the host (Figure 1) (Amutha & Gulsar, 2017). The generalist fungal entomopathogen, such as Beauveria bassiana or Metarhizium sp., penetrate the cuticle within 24 to 72 hours and differ depending to the properties and structures of species cuticle (Vestergaard et al., 1999 ; Ortiz-Urquiza & Keyhani, 2013 ; Amutha et Gulsar, 2017). During the fungal infection, the colonies implement a social immunity with different defense strategy types: avoidance, resistance, and tolerance (Liu et al., 2019). However, when the pathogen is already in contact with the colony, the colony use resistance strategies, such as cannibalism, aggressive behaviors towards the exposed individual, voluntary or forced isolation by the nestmates, cleaning behaviors to the exposed individuals, or others (Liu et al., 2019).

Léo KORST - Master’s thesis 2026 Figure 2. Synchronized asexual reproduction cycle in O. biroi (by Ravary & Jaisson, 2002). Letters designation: G = generation, WP = white pupae, RP = red pupae.

Léo KORST - Master’s thesis 2026 3 Among the defense strategies, one of the social mechanisms to save colony members and promote individual survival is a mutual cleaning response called allogrooming (Zhukovskaya et al., 2013). The other nestmates perceived a signal message to promote allogrooming and spore removal toward the exposed ant (Casillas-Pérez et al., 2023). Allogrooming behavior is the first behavioral barrier to prevent the spread of the fungus across members and penetration of spores into the cuticle, but it also represents a risk for pathogen spreading between individuals (Zhukovskaya et al., 2013). 4. Case study, the clonal raider ant, Ooceraea biroi Among social insects species, Ooceraea biroi, a subtropical, small ant species, blind, and known to live in small groups have a particular complex system. In contrast with the other castes in ant colonies, O. biroi species has a specific reproduction composed of a clonal reproduction cycle, parthenogenesis, and made by ant workers, with no queens inside colonies (Ravary & Jaisson, 2002 ; Kronauer, 2025). The reproduction cycle is performed by the workers through a synchronization cycle separate into two phases: a reproductive phase with a maximum of individuals in the nest taking care of the brood, and a foraging phase with a segregation between individuals into nurses and foragers (Figure 2) (Ravary et Jaisson, 2002). The genetic part does not change between the previous individuals’ generation and the new individuals’ generation, called callows, allowing a genetic match of each individual (Kronauer, 2025). Moreover, callows show behavioral differences with more attention to the larvae inside the nest than the old ants from the previous generation due to a neuropeptide, inotocin (Fetter Pruneda et al., 2021). However, regarding a defense context, a previous study showed a different division of labor between the individuals, and an effect of the group size. This is characterized by increased defense behaviors in a large colony but a conservation of functional colony dynamics in a small colony (Li et al., 2024). Thus, the interactions between individuals after a fungal exposure could be affected by the group size and the information exchanged between the previous and new generation within the same colony. The spatial segregation could create a different defense strategy used by ants depending on their age in a pathogen risk context. The age composition of the colony has an effect on the nestmates survival, with the highest risk of mortality for ant colonies composed only of old individuals, demonstrating the role of young ants in colonies survival (Alciatore, 2022). In a fungal pathogen risk context, this queenless species showed cooperative reactions towards the exposed individuals with an allogrooming response. They are also able to determine the immune status of their nestmates to adjust the behavioral response after an immune challenge based on a non-pathogenic fungal exposure, thereby increasing the centrality of the exposed individuals (Alciatore et al., 2021).

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 4 5. Research question and hypothesis Different responds mechanisms were observed varying with the infection progression, but the results remain unknown in allogrooming response in ant colonies especially. In other insects societies, a study on the infection virulence highlighted a link between the intensity of an alarm signal behavior before the grooming reaction of the nestmates, and the virulence of infection by the spores detection by the termites Reticulitermes flavipes (Bulmer et al., 2019). Similarly, in this species, another study showed that the social response depends on the infection stage of the exposed individuals, with a decreasing grooming behavior in an advanced infection stage, and a substitution with cannibalism behavior (Davis et al., 2018). In ant colonies, an analysis was conducted on the interaction between the fungus-growing species Trachymyrmex sp. and their fungal cultures within the nest, and showed an adjustment in the hygienic behaviors across infection stages. This demonstrates the influence of infection progression on the colony's culture adjustment (Bonadies et al., 2019). However, even if those two social systems share a similar complexity of sociality, the groups differ by their phylogeny, their colonial caste organization, and behavioral reaction in the same social context (Thorne & Traniello, 2003). Likewise, in O. biroi, the absence of a queen and their clonal synchronization create a big difference in the interaction between individuals and their spatial organization. Therefore, we can ask how does the infection stages of fungal-exposed individuals modify the allogrooming response of the colony? And how could individuals ages in the colony influence this behavior in different infectious stages? To address those questions, Beauveria bassiana, a common generalist entomopathogenic fungus with a wide host range, was chosen to create the pathogen risk during experiments. This pathogen can be found in the same geographical habitat as O. biroi colonies in South Asia and shares similar environmental parameters with high temperature and humidity as the clonal raider ant colonies (Thakur & Sandhu, 2010). Against the fungal pathogen, allogrooming response is a key component of the social immunity response for ant colonies and used in this study as the main measure of behavioral change of ant colonies. It is usually measured manually, which is time-consuming and prone to observer bias. Recent and advanced research in automated tracking enables the detection of discrete behaviors but are not yet developed for complex social behaviors such as allogrooming with high individuals proximity (Legan et al. 2023). Thus, the first step project is to test whether an automatic tracking allows to follow the species and if the application of grooming classification filters permits to quantify this behaviour by comparison to manual measurements. To respond this question, an automated tracking system relying on body part position estimation was used to establish the first approximation of the allogrooming classifier. We hypothesized that body part coordinates from SLEAP and grooming filters would allow extending grooming quantification.

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 5 Then, we focused on the biological part. We first asked how do the infection stage of exposed individual and colony age modify the hygienic behavioral response after a pathogen exposure? To answer this question, the first experiment focused on the effect of the infection stage on the grooming response of the colony toward the exposed individuals, also according to the age of the colony and the establishment of extended tracking approach. Regarding the capacity of the other ants to detect the quantity of spores on the cuticle, we hypothesized that, on one hand, the grooming response of the colony will be higher in the first stage of infection, just after fungal spore exposure, and lower grooming response when the infection stage is advanced. On the other hand, the higher proximity of the young clonal raider ants in the nest taking care of the brood could be an advantage after fungal exposure, with more nestmates engaged in grooming and more grooming received by the exposed individuals. In a second part of the project, we focused on varying ages of the exposed individuals and how the social environment could affect survival of exposed ants with a delayed reintroduction to create a potential high infection stage. In ant societies, the presence of nestmates after a fungal exposure increases the survival rate of the exposed individual (Hughes et al., 2002). However, in the context of an advanced infection, we asked how does the presence of individuals in a colony and the individual age group affect the survival of the exposed individual? And how does the presence of age- matched between-group and focal exposed ants play a role in the focal survival? A survival analysis was thus conducted to address these question. We hypothesized that in an advanced infection stage, the presence of a group could affect the exposed individual’s survival, increasing especially for young ants caused by their care behavior. However, when the age composition is different, with a young ant exposed in an older colony and the reverse for the old-exposed ants ; the survival of young ant could be lower due to the potential interaction differences between the old individuals. Materials and Methods 1. Ant colonies maintenance Three different colonies of O. biroi were separated for experiments with individuals aged- matched per colony and from the same genetic lineage (colony STC1, with genotype B) (Kronauer et al., 2012). Each colony contained 400 to 600 ants. In the laboratory, they were housed in a plastic box with a wet plaster floor to maintain constant humidity. The ants were fed three times per week with two different preys: Oecophylla and Tetramorium bicarinatum ; and maintained in dark at 29°C and 69% humidity. Regarding to the reproductive cycle of O. biroi, colonies were kept age pure by removing all consecutive emerging young cohorts. The colonies for experiment 1 and 2 were maintained in the same way.

Léo KORST - Master’s thesis 2026 Figure 3. Design of Experiment 1 with (A) an isolation challenge involved young ants (N = 12 for 1- hour isolation, N = 12 for 48-hour isolation) and old ants (N = 12 for 1-hour isolation, N = 12 for 48-hour isolation). The colonies contained 8 ants, and 8 larvae for the brood (gray shape). (B) A survival test was also carried out day by day with isolated individuals outside a social context with Triton or fungal exposure. Figure 4. Picture of a colony setup in a Petri dish with a humidified plaster floor, containing 8 identified ants and 8 larvae in the brood (to the right of the plate).

Léo KORST - Master’s thesis 2026 6 2. Fungal cultures and preparation for exposure B. bassiana fungus was maintained on Potato Dextrose Agar (PDA) plates, as a standard microbiological culture with nutrients, an acidic pH range, and solid support, in the dark at 28°C. 16 days later, the fungal culture was scratched and deposited on a slide for a manual cell counting by Malassez method under an inverted microscope (Leica DMi1) 2h before the exposures to determine the basal concentration. The dilutions allowed to get a final concentration of 1.109 spores*ml-1 used for all experiments, creating a high risk for the colonies (Bulmer et al., 2019). Triton 0.05% buffer was used as a control solution, which mimic the fungal viscosity and was also used on fungal dilution. 3. Experiment 1: Allogrooming response in dependency of age and infection stage To investigate the rule of the infection stage in the colony reaction and especially the grooming hygienic behavior, an isolation challenge was realized (Figure 3A). Two different ages were used for the experiment: 24 colonies with 7 days old ants called “young”, and 24 colonies with 92 days old ants called “old”. At the beginning, colonies were established over 5 days for colony settling, with 8 ants in each colony and 8 larvae to set the colony into the foraging phase. They were put in sterile Petri dishes (50 mm diameter * 9 mm height), with a 4±0.5 mm thick humidified plaster floor to maximize contrast between ants and background (Figure 4). A unique identification was established by painting tags for individual recognition with 2 different color dots on the abdomen, and thorax; our previous experiments demonstrated no effect on behavior. Then, 5 days after colony assembly, colonies were recorded for 2 h to obtain a grooming dynamic baseline pre-exposure (Pre-E). The next day, 24h±1 h after the recording Pre-E, two individuals were randomly chosen in each colony. One was exposed to a control solution (Triton 0.05%), and the other to a fungal solution (1.109 spores*ml- 1 in Triton) for 10 seconds by dipping method previously used with termites (Davis et al., 2018). The choice of two focals per colony was established to avoid the risk of colony effect between treatments. Then, the groups of old and young colonies were split again into two different groups for an isolation challenge. Among the age-matched colonies (N = 24), the focal individuals received one isolation treatment during 1 h±5min (N = 12) or 48 h±5min (N = 12), leading to a delayed reintroduction to create different infection stages. The focals were isolated and kept in identical Petri dishes mimicking the environment without group presence during this period. After isolation, the focals were reintroduced in colonies and another recording, post-exposure (Post-E) was made during 2 h (Figure 3A). In addition, a survival analysis was performed (Figure 3B) as a fungal pathogenicity check on the clonal raider ants. A batch of 25 young ants and 25 old ants was isolated in Petri dishes with Triton exposure, and 30 young ants and 30 old ants received fungal exposure (1.109 spores*ml-1). Survival daily monitoring was performed until dead of all isolated fungal exposed individuals.

Léo KORST - Master’s thesis 2026 Figure 5. Design for Experiment 2 on the survival challenge involving young and old ants (Total N = 116). Two social contexts were created: alone and in group, and within the group, two age composition were conducted: same age, and reverse age between focal and nestmates. Ants in the colonies were isolated individually for either 48h ±5min before reintroduction into their colony. The alone individuals were kept separately and exposed directly after the separation from the original batch. The hourly survival monitoring of each focal was started after 48h post-E or focal reintroduction. The dead individuals were not removed from the colonies.

Léo KORST - Master’s thesis 2026 7 4. Experiment 2: Effect of social environment on survival In the second experiment, we investigated the role of a social environment in ant colonies on the exposed-age survival with an advanced infection stage (48 h) (Figure 5). A survival experiment was conducted with two social contexts: alone or in a group with 7 other ants. Two age compositions were also added in the group context: same age and reverse age between the focal and the nestmates. All social environments were done with young ants (7 ± 1 days old) and old ants (74 ± 1 days old). The old ants were from the same colony as the young ants from the cohort of the previous experiment and were aged to 2.5 months at the time of the experiment. The young ants are from the same colony but another cohort of individuals. In total, 116 ants were monitored during the experiment in different conditions and treatments. The colonies assembly took place five days before the start of the survival monitoring, with 8 ants and 8 larvae for the brood (in gray). Then, one individual was chosen in each colony and received one of the two treatments in the same age composition: Triton 0.05% (N per age = 10) or fungus 1.109 spores*ml-1 (N per age = 12). In the reverse age composition, the focal received only a fungal exposure treatment (N per age = 12). The alone individuals were taken from their respective cohorts a few minutes before the Triton (N per age = 12) or fungus (N per age = 12) exposures. The individuals were placed in individual Petri dishes for 48 h to create an advanced infection progression, and focal individuals from the group context were reintroduced after this isolation period. All survival monitoring was conducted hourly for five days following the reintroduction of focal individuals to their colonies and 48 hours post-exposure for the isolated individuals. 5. Video recording and behavioral scoring for the experiment 1 For the first experiment, the allogrooming dynamics of the colony before and after exposure were recorded with Basler cameras (model acA20440-20gc; Ahrensburg, 124 Germany) with the objectives Edmund optics 59870 16mm/F1.6, 2464px * 2056px resolution and 10 frames per seconds parameters was selected on the LoopBio Motif (v.6) software. The first 16 minutes (10000 frames) of the 2h recording were used for the behavioral scoring as the first colony reaction to fungal exposure. The videos were analyzed by two different approaches, using 1) the Social LEAP Estimates Animal Poses (SLEAP) (Pereira et al., 2022) to train a convolutional neural network (CNN) for body part recognition of ants, and 2) Behavioral Observation Research Interactive Software (BORIS) (Friard & Gamba, 2016) for a manual annotation of grooming events as ground truth data, described using the proximity of ants and body interaction between nestmates towards the focal.

Léo KORST - Master’s thesis 2026 Figure 6. SLEAP Model preparation and grooming classification. (A) Skeleton of the clonal raider ant (Designed by Leo KORST). (B) Walking behavior and (C) Grooming behavior of the clonal raider ant with SLEAP annotations (Extracted from experiment 1 and calculated with ImageJ). (D) Metrics evaluation of SLEAP model based on node distances (ground truth vs prediction).

Léo KORST - Master’s thesis 2026 8 a. SLEAP Convolutional neural network SLEAP CNN is an open-source deep-learning framework designed for multi-animal pose tracking (Pereira et al., 2022). SLEAP works in greyscale to separate the animals from the background. To analyze SLEAP accuracy for a grooming quantification in a pathogen context within a colony, a batch of 48 videos Pre-E and 24 videos Post-E were used to compare with manual annotated videos from BORIS (page 11). These videos were analyzed in three steps: skeleton labeling, model training, and tracking inferences. At first, the ant skeleton was established as connected nodes with edges, and in this project, the thorax was the skeleton anchor to maintain skeleton stability. Then, each body part and joint were marked with nodes, as for the spine: head, thorax, dot_1, dot_2, abdomen_1, and abdomen_2. The legs were also marked with one node in each leg and two nodes for each antenna. The nodes were connected with edges to finalize the articulated skeleton of ants called “instance” by the CNN (Figure 6A). Then, a model composed of 250 annotated ants was designed with different body positions during walking and grooming in different greyscale values due to the body color of ants to ensure best performance in complex cases (Figure 6B and 6C). After model training, a metrics evaluation was created to establish model accuracy for each body part position, with a global average of 3.66 px error representing 4% of the ant length (90px) (Figure 6D). Regarding the spine, the percentage of pixel length error is node-dependent: 2.87% for head, 0.41% for thorax, 2.58% for dot_1, 3.14% for dot_2, 3.53% for abdomen_1, and 4.16% for abdomen_2. Finally, the tracking inference was realized using a top-down approach, based on the detection of the anchor first (centroid model) and then prediction of the other body parts (centered-instance model). The simple tracking approach with greedy and iou algorithms was used to associate the most similar instances between frames to create a tracking connection and avoid identity swap during inference tracking. When the tracking was realized, an approximation of allogrooming behavior was created. Based on the previous metrics evaluation, the spine was selected for grooming classification: 4 focal nodes (head, thorax, abdomen_1, abdomen_2), and 2 groomer nodes (head, thorax). Then, the single body part coordinates of the focal nodes and the groomer nodes were extracted to sort interactions and to establish grooming filters based on manual analysis of nodes distance average during a grooming event: 40 px maximum between nodes and 40 frames minimum to be considered as an allogrooming event between groomer and focal. b. BORIS software BORIS is an open-source software employed for video or live observations with manual behavioral annotation (Friard & Gamba, 2016). All results from experiment 1 were used in the biological analysis, i. e. 48 videos Pre-E and 48 videos Post-E. An ethogram with two allogrooming types was created: simple (1 or 2 groomers engaged) and cluster of grooming (>2 groomers engaged).

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 9 6. Statistical analysis All statistical analyses from both experiments were carried out in R 4.5.2 (2025-10-31). All graphic visualizations were created with the R packages ggplot2 and ggpubr. All post-hoc pairwise comparisons used the packages emmeans and multcomp to extract the p-values, and non-significant results shared a common letter (P > 0.05), and different letters in significant comparisons (P < 0.05). a. SLEAP performance and filters accuracy analysis The performance of SLEAP was analyzed by the default package stats in R. The assignment rate was calculated as the proportion of each node detected during tracking for each instance. Then, the assignment rate was finally determined as the average across the 8 instances per colony, and the global assignment rates Pre-E and Post-E were extracted. Based on the algorithm and change of behaviour in ants the continuity of recognizing all 8 ants in one colony fluctuates and was analyzed as the tracking stability. The latter was assessed by the proportion of video where SLEAP consistently tracked all 8 ants per colony or fewer during the tracking. The stability was compared to the proportion of video where a grooming event was detected across frames by the BORIS software. Based on the grooming filters applied to SLEAP data, the accuracy of grooming classification was analyzed with a Spearman rank correlation test, due to the non-parametric data, to calculate the global correlation between the two methods, and a specific correlation between Pre-E and Post-E. The similarity between the total time of grooming between BORIS and filters was also determined using the package blandr for an application of the Bland-Altman method. It’s an agreement limits method to evaluate the interchangeability between methods using the difference between the data on the global mean between the methods' data, including the concordance limit and bias, in contrast to the correlation only (Altman & Bland, 1983). The discrepancies in recognizing interactions as grooming events between the filters’ and the manual annotation were investigated to highlight the concordance of grooming events detection with the sensitivity and the specificity. The filter’s sensitivity (Se) is defined as the capacity of the filter to detect a grooming event in a frame with data overlap with BORIS. The results also emphasize the presence of a false negative rate (Fn) in the analysis. The specificity (Sp) of the filters detection was described as the filters capacity to detect the absence of allogrooming behavior in a frame. The reverse highlighted the false positives rate (Fp). 𝑆𝑒 = 𝑂𝑣𝑒𝑟𝑙𝑎𝑝𝑝𝑖𝑛𝑔 𝑔𝑟𝑜𝑜𝑚𝑖𝑛𝑔 𝑒𝑣𝑒𝑛𝑡𝑠 𝑆𝐿𝐸𝐴𝑃 & 𝐵𝑂𝑅𝐼𝑆 (𝐴𝑙𝑙 𝑒𝑣𝑒𝑛𝑡𝑠 𝑓𝑜𝑢𝑛𝑑 𝑏𝑦 𝐵𝑂𝑅𝐼𝑆 𝑜𝑛𝑙𝑦 𝑎𝑛𝑑 𝐵𝑂𝑅𝐼𝑆+𝑓𝑖𝑙𝑡𝑒𝑟𝑠) ∗ 100 𝐹𝑛 = 100 − 𝑆𝑒 𝑆𝑝 = 𝑂𝑣𝑒𝑟𝑙𝑎𝑝𝑝𝑖𝑛𝑔 𝑛𝑜 𝑔𝑟𝑜𝑜𝑚𝑖𝑛𝑔 𝑒𝑣𝑒𝑛𝑡𝑠 𝑆𝐿𝐸𝐴𝑃 & 𝐵𝑂𝑅𝐼𝑆 (𝑙𝑙 𝑒𝑣𝑒𝑛𝑡𝑠 𝑓𝑜𝑢𝑛𝑑 𝑏𝑦 𝑓𝑖𝑙𝑡𝑒𝑟𝑠 𝑜𝑛𝑙𝑦 𝑎𝑛𝑑 𝐵𝑂𝑅𝐼𝑆+𝑓𝑖𝑙𝑡𝑒𝑟𝑠) ∗ 100 𝐹𝑝 = 100 − 𝑆p

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 10 b. Grooming analysis with BORIS results Regarding the biological analysis for experiment 1, two individuals exposed to Triton were removed from the analysis by their death before the exposure day, and all data from one young colony with 1 h hour of isolation were deleted due to the presence of larvae stuck on the lid of the Petri dish, causing weird behaviours during the recording Post-E. To determine the grooming effort received by the focals, the grooming was quantified according to the type of grooming, using the formula: Total duration of grooming * type of grooming. Simple allogrooming is classified as 1 due to the presence of one or two groomers. Cluster of grooming is assigned with a value of 1.5, including the participation of minimum one individual in the grooming event. Then, the allogrooming received in continuous variable was analyzed Pre-E and Post-E to determine the modification in allogrooming baseline received by the focals in their respective colony. A generalized linear mixed model (GLMM) was executed, using the package glmmTMB, based on the residuals dispersion and positive deviation, with a Tweedie regression model family with treatment and exposure period as a fixed-effects, and the colony ID as a random effect (Brooks et al., 2017). The next part of the analysis, focused on Post- E period exposure, analyzed the grooming received as explanatory variable with a linear model (LM) according to the residuals’ normality and including treatment, isolation challenge, and age as categorial variables. 3 outliers were found in the model checking, representing 3.26% of the observations, they were considered negligible. The grooming received by the focals was also analyzed with the type of grooming, isolation challenge, and focal age as fixed effect, and colony ID as a random effect in a linear mixed model (LMM), using the package lme4 (Bates et al., 2015). The package car was used for all models application with an ANOVA analysis type “III”. The choice of model was determined by AIC step selection using the package MASS. The residual diagnostics of the models were established with the packages DHARMa and performance. c. Survival analysis In the experiment 1, one old ant with fungal exposure was removed from the analysis due to death during the exposure process. The survival analyses on fungal pathogenicity check in experiment 1 were performed with the survival package and survminer package for a global Log Rank Test, used with the presence of no death for some group, then the package coxme was added to perform an interaction cox test (Therneau, 2015). In experiment 2, one young focal ant in a reverse age composition was removed from the analysis due to the same exposure process.

Léo KORST - Master’s thesis 2026 Figure 7. Evaluation of SLEAP accuracy on clonal raider ant tracking according to exposure period. (A) Assignment rate of each body part according. (B) Stability of tracking using the number of ants found per video across frames. (C) Evolution of grooming events across frames from BORIS results. Figure 8. Evaluation of grooming filters to BORIS results. (A) Correlation ship between the Total grooming in the two methods. (B) Bland-Altman Test visualization with mean bias (red line) and limits of agreement (black lines). (C) Example of grooming patterns detected by the grooming filters and manual annotation from BORIS according to the exposure period with overlapping results (dark green), false positives (orange), and false negatives (light green) detected by the filters method.

Léo KORST - Master’s thesis 2026 11 Results 1. Experiment 1 a. SLEAP performance and grooming classification filters accuracy At first, the results showed a high global assignment rate, the Pre-E (N = 96) with a high average assignment rate (SD =1.84 ; Rate = 97.62%), and a reduction of 3% Post-E (N = 47) with larger dispersion (SD = 3.40 ; Rate = 94.37%) (Figure 7A). The SLEAP capacity to detect ants showed a decrease across frames in tracking stability Post-E, with a global average of 88.27% frames in which found 8 tracks Pre-E (N = 96), and 73.71% frames Post-E (N = 47) (Figure 7B). The tracking stability and grooming occurrences comparison showed an inverse association between tracking stability of SLEAP and the occurrence of grooming events across time (Figure 7C). The Spearman rank correlation test results showed a significant global positive correlation between the filters and manual methods (R = 0.80 ; S = 328349; P = 6.85*10-15) (Figure 8A). However, the correlation coefficient is dependent on the exposure period. The correlation is lower Pre-E (N = 96) with an overestimation of total grooming time for both treatments (Rpre = 0.65 ; S = 175180 ; P = < 2.2*10-16). In the Post-E (N = 47), the results showed a higher correlation (Rpost = 0.91 ; S = 5510.7 ; P = < 2.2*10-16) but depend to the treatment (Rpost - control = 0.51 ; S = 988 ; P = 1.36*10-2 ; Rpost - fungus = 0.79 ; S = 485.61 ; P = 4.64*10-6), showing overestimation in control individuals. Secondly, the interchangeability between filters and BORIS results was demonstrated by a Bland-Altman method (Figure 8B). The results showed a significant bias between the method (t = 8.78 ; DF = 142 ; P = 4.66*10-15). The global systematic bias (d̄) is positive with an overestimation from filters method (SD = 172.64 ; d̄ = 126.81 ; [CI 95%: 98.27 - 155.35]) and a wide limits of agreement (LOA) (Upper LOA = 465.19 ; [CI 95%: 416.30 - 514.08]) (Lower LOA = -211.57 ; [CI 95%: -260.46 - -162.68]). When the visualization is separate between exposure period, the Bland- Altman diagram showed a higher systematic positive bias in Pre-E (N = 96) (SD = 146.02 ; d̄ = 148.32 ; [CI 95%: 118.73 - 177.91]) than Post-E (N = 47) (SD = 212.22 ; d̄ = 82.87 ; [CI 95%: 20.56 - 145.18]), and a presence of 8 outliers in Post-E. Finally, the filters’ sensitivity and specificity results demonstrated the limits on grooming event detection (Figure 8C). The filters showed a global sensitivity of Se = 83.71% and a global specificity of Sp = 54.10%. However, the results are also dependent to the exposure period, with Se = 94.45% Pre-E, so a rate of false negatives Fn = 5.55%, and Se = 80.63% and Fn = 19.37% Post-E showing a high level of filters sensitivity. The specificity is Sp = 50.51% Pre-E, so a rate of Fp = 49.49%, and Sp = 59.67% Post-E with Fp = 40.33%. The results demonstrated a reduction in filters sensitivity Post-E but a higher presence of false positives Pre-E than Post-E.

Léo KORST - Master’s thesis 2026 Figure 9. Survival fungal pathogenicity check on young and old ants isolated during 5 days Post-E A Factor N Observed/Expected Chisq (X²) P value Treatment = Control Treatment = Fungus 50 59 0 / 32.1 59 / 26.9 113 <2*10-16 *** B Factors Coef HR ratio CI 95% P value Age 1.06 2.88 1.65-5.02 2.01*10-4 *** Table 1. Analysis of the fungal pathogenicity risk after a fungal exposure. (A) Log Rank Test on treatment effect. (B) Cox Test on age effect only after fungal exposure. Figure 10. Total grooming received by focals according to exposure period and treatment (Letters from post-hoc test adjusted by Tukey). Chisq DF P value (Intercept) Treatment Exposure period Treatment * Exposure period 239.59 0.80 120.22 11.10 1 1 1 1 < 2.2*10-16 0.37 < 2.2*10-16 8.62*10-4 *** *** *** Table 2. Results of total grooming received by focals according to exposure period and treatment (ANOVA Test applied on GLMM with Tweedie family) (N per exposure = 48).

Léo KORST - Master’s thesis 2026 12 b. Fungal pathogenicity check by survival analysis The fungal pathogenicity check realized during experiment 1 showed a significant decrease in survival in the fungal-exposed ants (Chisq (X²) = 113 ; DF = 1 ; P < 2*10-16) (Figure 9, Table 1A). No deaths and no differences were observed in the control treatment between ages. The control treatment was removed for Cox test application on age base. The comparison between old and young exposed to fungus demonstrated a significant decrease in survival for young ants than old ants (LR = 13.60 ; P = 2*10-4) (HR = 2.88 ; [CI 95%: 1.65 - 5.02] ; P = 2.01*10-4). The results showed a reduced and shorter survival probability in the young ants than old ants (Table 1B). c. Allogrooming response in ant colonies across infection progression The results of the GLMM in allogrooming received Pre-E and Post-E showed a significant increase of allogrooming Post-E for both treatment (X² = 120.22 ; DF = 1 ; P < 2.2*10-16) (Figure 10, Table 2). The significant interaction effect between treatment and exposure period highlighted a difference in grooming received between the treatments within the exposure periods (X² = 11.10 ; DF = 1 ; P = 8.62*10-4). The post-hoc results adjusted by Tukey showed the same grooming received by both focal Pre-E (EstimateControl pre – Fungus pre = -0.31 ; SE = 0.35 ; z = -0.90, P > 0.05) with a time average of grooming received of 65.19 seconds in control exposed ants, and 47.24 seconds in fungal exposed ants Pre-E. However, the results showed a higher grooming effort receiving on the focal exposed to fungus than control Post-E (EstimateControl post – Fungus post = 0.93 ; SE = 0.13 ; z = 7.02 ; P < 1*10-4) with an average of 302.04 seconds on control exposed ants and 762.45 seconds on fungal exposed ants, an allogrooming increase of 152.43% of grooming after a fungal exposure. The control exposed ant also received more grooming post-E than Pre-E (EstimateControl pre – Control post = -1.56 ; SE = 0.27 ; z = -5.73 ; P < 1*10-4).

Léo KORST - Master’s thesis 2026 Figure 11. Total grooming received by focals Post-E according to the isolation, age, and treatment. Chisq DF F value P value (Intercept) Treatment Isolation Age Treatment * Isolation Treatment * Age Isolation * Age Treatment * Isolation * Age Residuals 8169563 1066163 43388 85085 60042 57025 279491 237261 4495747 1 1 1 1 1 1 1 1 84 152.64 19.92 0.81 1.59 1.12 1.07 5.22 4.43 < 2.2*10-16 2.49*10-5 0.37 0.21 0.29 0.30 2.48*10-2 3.82*10-2 *** *** * * Table 3. Results of total grooming received by focals Post-E according to the isolation, age, and treatment (ANOVA Test applied on LM ) (N = 12 per group). Contrast Estimate SE t. ratio P value Old 1h fungus – Old 48h fungus Old 1h fungus – Young 1h fungus Old 1h fungus – Young 48h fungus Old 48h fungus – Young 1h fungus Old 48h fungus – Young 48h fungus Young 1h fungus – Young 48h fungus Old 1h control – Old 1h fungus Old 48h control – Old 48h fungus Young 1h control – Young 1h fungus Young 48h control – Young 48h fungus 85.0 -121.8 272.0 206.8 186.9 393.7 421.5 563.0 564.3 298.7 94.4 96.6 94.4 96.6 94.4 96.6 94.4 94.4 101.0 96.6 0.90 -1.26 2.88 2.14 1.98 4.08 4.46 5.96 5.58 3.09 0.99 0.91 0.09 0.40 0.50 2.5*10-3 6*10-4 < 1*10-4 < 1*10-4 5.21*10-2 ** *** *** *** . Table 4. Results of total grooming received by focals Post-E according to the isolation, age, and treatment (Post hoc Test adjusted by Tukey on LM)

Léo KORST - Master’s thesis 2026 13 Then, the analysis was focused on Post-E period and the effect of the focal age and the isolation challenge (Figure 11). The results in Post-E showed a significant increase in allogrooming in fungal exposed ant in all ant ages and isolation challenge (F(1,84) = 19.92 ; P = 2.49*10-5). The triple interaction was also significant and demonstrated a difference within the groups (F(1,84) = 4.43 ; P = 3.82*10-2) (Table 3). Therefore, the post-hoc test adjusted by Tukey highlighted a conservation of the highest grooming attention received by the fungal exposed ants than control ants across isolation (P < 0.05) (Table 4). Only in the young ant colonies 48 h Post-E, the effect of treatment was not significant but close to the α limit (EstimateYoung 48h control – Young 48h fungus = 298.7 ; SE = 96.6 ; t = 3.09 ; P = 5.21*10- 2). Young control ants received an average of 254.44 seconds of grooming by the nestmates 48 h Post- E, compared to the young fungal ants with an average of 553.15 seconds of grooming. Then, the difference between old and young ants in 1 h Post-E and 48 h Post-E didn’t show a difference in the grooming receiving by the fungal-exposed individual (EstimateOld 1h fungus – Young 1h fungus = -121.8 ; SE = 96.6 ; t = -1.26, P = 0.91 ; EstimateOld 48h fungus – Young 48h fungus = 186.9 ; SE = 94.4 ; t = 1.98 ; P = 0.50). However, within the colonies with the same age, the young ants exposed to fungus received more allogrooming in 1 h than 48 h (EstimateYoung 1h fungus – Young 48h fungus = 393.7 ; SE = 96.6 ; t = 4.08 ; P = 2.5*10-3). The mean grooming time received by fungal-exposed ants 1 h Post-E was 946.86 seconds, and 553.15 seconds 48 h Post-E, highlighting an increase of 71.17% of grooming towards the focals in contact with early fungal-exposed ants. In contrast to the young ants, the old colonies demonstrated no significant difference across isolation time (EstimateOld 1h fungus – Old 48h fungus = 85 ; SE = 94.4 ; t =0.90 ; P = 0.99) (Figure 11).

Léo KORST - Master’s thesis 2026 Figure 12. Type of grooming received by focals exposed to fungus Post-E according to the isolation challenge, age, and grooming type. Chisq Df P value (Intercept) Isolation Grooming type Age Isolation * Grooming type Isolation * Age Grooming type * Age Isolation * Grooming type * Age 101.86 2.46 22.98 0.21 2.22 2.22 2.64 3.12 1 1 1 1 1 1 1 1 < 2.2*10-16 0.12 1.64*10-6 0.65 0.14 0.97 0.10 0.08 *** . Table 5. Results of grooming type received by focals exposed to fungus Post-E according to the isolation challenge, age, and grooming type (ANOVA Test applied on LMM ) (N per group = 12). Figure 13. Survival probability in young (orange/ light green) and old (red/dark green) focal ants according to the treatment (control, C) or fungus (F), and the social context: alone (one ant) or in a group of 7 same-age ants (represented by two ants) (Log Rank Test and Cox Test, N per group = 12). Factors Coef HR ratio z P value Social context Age of focal Social context * Age of focal -0.18 3.48 -2.19 0.83 32.47 0.11 -0.42 5.57 -3.05 0.67 2.49*10-8 2.26*10-3 *** ** Table 6. Results of survival probability after fungal exposure according to age of focal and age composition of the group (Cox test).

Léo KORST - Master’s thesis 2026 14 The results from the type of grooming demonstrated a non-significant effect of the focal age in the type of grooming used to the focals (X² = 0.21 ; DF = 1 ; P = 0.65), and the isolation challenge (X² = 2.46 ; DF = 1 ; P = 0.12) (Figure 12, Table 5). However, a significant effect of the grooming type was highlighted (X² = 22.98 ; DF = 1 ; P = 1.64*10-16), showing an engagement of 1 or two ants for allogrooming behavior toward the focal than 3 or more ants of the colony. The triple interaction between isolation, grooming type and age was not significant (X² = 3.12 ; DF = 1 ; P = 0.08). 2. Experiment 2 a. Social context In the second experiment, the social context defined by the focal exposed individual being alone during the experiment, or in a group of 7 other ants with same age, was analyzed (Figure 13). The global effect on treatment exposure with a Log Rank Test highlighted survival decrease in the global fungal exposed ants (X² = 94 ; DF = 1 ; P < 2*10-16). The focus on the control treatment demonstrated a significant decrease in young ant survival in a solitary context (X² = 12 ; DF = 3 ; P = 7*10-3). However, the Cox test on solitary control individuals showed a non-significant difference in young and old ant survival (LR = 3.54 ; DF = 1 ; P = 0.06) (HR = 5.81 ; [CI 95%: 0.68 - 49.84] ; P =0.11). Thus, the solitary young ants were more affected by the social context, showing death after a few days without nestmates, in contrast with the old ants. In the pathogen exposure, 100% of young individuals alone and in a group died during the monitoring. In old focals, only 1 survived in group context. The Cox test on survival focused in the fungal treatment demonstrated a significant global effect of the focal age and social context interaction (LR = 40.02 ; DF = 3 ; P = 1.05*10-8) (HR = 0.11 ; z = -3.05 ; P = 2.26*10-3) (Table 6). The post hoc test adjusted by Tukey highlighted a decrease in young ants survival without a colony age-matched (HR = 10.77 ; [CI 95%: 2.51 - 46.22] ; z = 4.19 ; P = 2*10-4) in contrast of the old ants survival isolated and not in a colony (HR = 1.20 ; [CI 95%: 0.39 - 3.67] ; z = 0.42 ; P = 0.97). On average, the mean death for young-isolated ants was 14.5 hours, 48 h post-E. On the contrary, the mean death of young focal ants was 31 hours in a colony after the reintroduction, 48 h post-E. The young ants survival within the colony was also significant reduced than the old ants survival within their colony, but close to the α limit (HR = 0.28 ; [CI 95%: 0.08 - 0.99] ; z = -2.60 ; P = 4.64*10-2) and this decrease of young ant survival was kept in an isolated environment (HR = 0.031 ; [CI 95%: 0.01 - 0.15] ; z = -5.57 ; P < 1*10-4). However, the focal young ants survival in a group didn’t differ to the focal old ant alone (HR = 0.33 ; [CI 95%: 0.10 - 1.08] ; z = -2.40, P = 7.64*10-2).

Léo KORST - Master’s thesis 2026 Figure 14. Survival probability after fungal exposure in young (orange) or old (red) focal ants depending on the age composition of the group of 7 ants: same age (same group data in previous social context in a group) (similar color to the focal) or reverse age (color of the other age focal) (Cox test, N per group = 12). Factors Coef HR ratio z P value Age composition Age of focal Age composition * Age of focal -0.61 1.78 -0.62 0.54 5.94 0.54 -1.32 3.56 -0.95 0.19 3.75*10-4 0.34 *** Table 7. Results of survival probability after fungal exposure according to age of focal and age composition of the group (Cox test).

Léo KORST - Master’s thesis 2026 15 b. Age composition The global Cox test on colony age composition (same or reverse) and focal age interaction didn’t demonstrate a difference in survival 48 h Post-E (LR = 20.10 ; DF = 3 ; P = 2*10-4) (HR = 0.54 ; [CI 95%: 0.15 - 1.93] ; z = -0.95, P = 0.34) (Figure 14, Table 7). On average, the mean death time for young ants in an old colony (reverse age) was 19.45 hours and 31 hours for the young ants with young nestmates. For the old exposed ants, the average death time was 51.25 hours 48 h post-E in the age-matched colony, and 35.33 hours in a reverse age colony. Nevertheless, the absence of significant results in interaction showed an absence of survival difference in young and old focal ants in an age-matched colony, in contrast with the previous analysis on social. A statistic discordance in the colony age-matched context between focal ages with the same Cox test ([Social context analysis]Young-group vs Old-group ; P = 4.64*10-2 ; [Age composition analysis]interaction Age composition*Age of focal ; P = 0.34]). Discussion 1. Extend grooming classification approach using SLEAP pose estimation The results observed in SLEAP performance showed the model capacity to predict the body- part positions of the skeleton in an ant model, especially for a small ant species (2mm) in a colony and pathogen exposure context. Moreover, the body-part assignment rate demonstrated the strong performance of SLEAP to find the position of body-part in both exposure periods, showing that the model can adjust the estimations and connections between the body-parts according to the body position. The stability of tracking demonstrated that 8 ants in one colony can be detected simultaneously by using the greedy and iou algorithm, but the presence of fungus decreases the detection capacity, as observed with the evolution of grooming across time. The decrease in detection could be caused by more interactions during grooming event between individuals, which increased the visual occlusion during the inference tracking. We observed that the cluster of grooming forms complex shapes, non-resembling of ants which affects key point recognition in SLEAP greyscale and didn’t allow it to find the right position of the individuals, but also the number of ants engaged in the grooming event. Grooming filters using a 40 px distance maximum and 40 frames minimum showed a globally high correlation with the manual data from BORIS in total grooming time quantification, but a difference between exposure periods with an overestimation of grooming Pre-E.

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 16 The positive deviation Pre-E could be explained by the directionality of grooming. During the data visualization from SLEAP, we observed grooming events of the focals towards the other nestmates Pre-E and Post-E in both treatments. However, only BORIS includes information of the nestmates grooming toward the focal, whereas the filters used assume a symmetric interaction, that includes both grooming directions, receiving and performing grooming. Thus, the allogrooming received was overestimated by the grooming performed by the focal directly. The current parameters might are not giving grooming direction, especially observed in Pre-E and in control exposed ant, who has not shown grooming perturbation in the colony. Finally, the relation between the timing of grooming event in filters and BORIS highlighted the sensitivity of filters to detect grooming event with a higher sensitivity Post-E than Pre-E. Variation in grooming directionality might also explain the filter’s sensitivity, either through focal engaged grooming toward another nestmates in Pre-E, but excluded from BORIS detection, or misidentified by the filters as grooming events by the filters. We observed that some ants can stay close to the other ants when they perform self-grooming or rest for a long time. Therefore, these behaviors increased the risk of false positives being detected. The presence of false negatives is reduce in Pre-E and Post- E in contrast with false positives, showing the first step to use the first parameters as base for a future grooming classification. Thus, based on SLEAP accuracy and the grooming filters applied during the experiment, the pose estimation with nodes method could be an extend approach for allogrooming automated tracking quantification, but needs more information to clarify allogrooming events or no-grooming events. 2. Defending the colony across the fungal infection Regarding the biological results from the first experiment, the fungal pathogenicity between young and old ants after a fungal exposure demonstrated an increased capacity in age to survive after a fungal exposure. The fungal exposure on callows presents a high risk for the ant’s new generation and is stabilized in time for the clonal raider ant (Alciatore et al., 2021). The decreased survival in young ants could be caused by different individual factors, such as cuticle thickness, which differs and increases with age in this species and could make pathogen penetration easier (Matte et al., 2025; Ortiz-Urquiza & Keyhani, 2013). The behavioral difference of less self-grooming in young ants could also increase the risk of penetration and infection progression.

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 17 Then, the grooming response post-E demonstrated the capacity of the grooming dynamic baseline of the colony to be adjusted in a context of pathogen risk, including a grooming effort toward the exposed individuals. In the first 16 minutes, the mean was 12min42 for fungal-exposed ants receiving grooming, whereas only 5min for the control-exposed ants. This social behavior adjustment was conserved across species (Casillas-Pérez et al., 2023 ; Liu et al., 2019). Nevertheless, the allogrooming responses to the fungal-exposed ants vary depending on the infection stage. A strong allogrooming persists across infection progression in the old ant colonies. The persistence of allogrooming in O. biroi colonies differs from the termites’ reaction in advanced infection, which demonstrated a decreasing of allogrooming reaction and increasing of aggressiveness towards the infected individuals (Davis et al., 2018). Moreover, active self-exclusions, aggressive or cannibalism behaviors towards the fungus-infected ants such as Formica rufa or Lasius neglectus species were observed in contrast with the O. biroi colonies reaction (Liu et al., 2019). However, in the young ant colonies, the results highlighted a reduction of grooming across infection stages, with a grooming reaction close to the control exposure 48 h Post-E. Thus, the young ants confirmed the hypothesis with an adjusted grooming response according to the infection progression, in contrast with old ants or according to the presence of a dying individual in the colonies, resulting an avoidance behavior from the nestmates to reduce the risk of pathogen spread. In contrast with their strong proximity within the nest and the brood, the young ants modify their interaction in a pathogen context (Fetter-Pruneda et al., 2021). The large variation of grooming received by the young ant 48 h Post-E could also be due to individual factors previously explained in the fungal pathogenicity check, such as self-grooming realized by the focal, but also the capacity of the pathogen to penetrate the ant body, creating a different infection stage in young ants 48 h than in old ants 48 h. Therefore, the early death in young ants could be explained by the quickness of the infection progression in young individuals and lead to this difference in allogrooming reaction. Furthermore, the grooming in smaller or bigger groups didn’t highlight a difference between age or treatment groups, demonstrating that one or two ants engaging in a grooming event is a conservative colony reaction and rejected the hypothesis of more groomers engaged in young ants. Thus, one or two groomers are the predominant reaction of the colony towards the exposed individuals and could be explained by physical constraints of space to groom, but also by the group size (Li et al., 2024).

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 18 3. The sensitivity of the social context young ants survival To be alone could represented a risk for the individual, especially for social insects. The risk of isolation was observed in young ants exposed to control treatment with a decreasing survival but was not observed in the old ants. The difference in age survival alone could be attributed to maturation processes needed to establish reserves in young ants and the possibility of nutritional exchange between individuals in the nest (Matte et al., 2025 ; Fetter-Pruneda et al., 2021). In the pathogen context with an advanced infection, the social isolation affected only the young ants. The conspecifics presence in young exposed ants allowed to reduce the survival risk for the young ant, but didn’t allow to recover equal survival as control exposed ants in a group. This demonstrates that fungal pathogen exposure represents a higher risk for survival than the isolation role in young ant. In the old-exposed ants, social presence of nestmates didn’t allow to reduce survival after a fungal exposure. In contrast, other studies about the early infection stage showed that individuals isolated from the colony have a reduced survival rate in Acromyrmex echinatior (Hughes et al., 2002). In the advanced infection stage, the spores could already have penetrated the ant's body or be strongly attached to the cuticle, and the conspecifics' presence might not allow the removal of spores and thus reduce the fungal colonization and death of the exposed ant. The age composition of colonies demonstrated that the presence of conspecifics, no matter the age-matched or not inside the colonies, didn’t allow to modify the survival rate of young and old exposed ants after an advanced fungal exposure. The proximity between young ants inside the nest and their care specialization during their first days after emergence didn’t affect the survival directly (Fetter-Pruneda et al., 2021). Thus, the hypothesis of age-matched needed for young ants is rejected, and the presence of conspecifics only is necessary for the survival of young exposed ants. Conclusion To conclude, in the first part of the project, SLEAP demonstrated a strong capacity to track O. biroi in a small colony without pathogen context and with identity swaps in a pathogen context. The grooming filters findings indicated a potential extension approach for allogrooming behavior classification and automated quantification. The direction groomer-recipient in grooming event need requires filter adjustments with more conditions and parameters to reduce the risk of mismatches. However, this tracking approach would allow for reducing the observational bias and the time spent on manual tracking.

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 19 Finally, the infection stages and ant ages demonstrated a persistent allogrooming behavior against fungal exposure in old colonies of O. biroi and highlighted an allogrooming disruption across infection progression in young ants. This is based on individual factors such as early death, but also collective factors with less allogrooming 48 h Post-E, and with the group-living benefit when the group is composed of young ants, meaning an higher sensitivity of young ants to fungal exposure. Thus, social immunity defense to the fungal pathogen risk in young ant fluctuates more than in old ant in O. biroi. Future directions Concerning the methodological part, the future grooming classification created with SLEAP information needs additional conditions to increase the sensitivity of filter interactions and reduce the risk of false positives. One of the parameters that could reduce the risk of mismatch due to directionality could be the frequency of body position movement. We observed that the focal individuals reduce their mobility when they are in a grooming event and the groomers perform head sweeping movements on the focal during the grooming event. Thus, the frequency of body movement or the groomer's head specifically could be a new parameter to specify allogrooming behavior. Regarding the biological results obtained during this project, the new information provided by the infection progression in ant colonies demonstrated variability according to the ants' age. Therefore, the next future directions could be based on the speed of infection stages between young and old colonies to highlight the difference in infection stages 1 h and 48 h Post-E such as the colonization of fungus outside and inside the ant body to establish the stages. Different methods can be used, like a scanning electron microscope to highlight the place where the fungus is present, but also body cross-sections to demonstrate the progression or final state at timepoint of reintroduction. Group-living effect on young ants could also be explained by new experiments on spore removal efficiency in colonies of young and old ants, either directly by quantifying spores on the cuticle before and after fungal exposure or indirectly by grooming effort toward the infected individual from nestmates, according to their age. Then, testing morphological and behavioural differences and which have a stronger effect within young and old ants or also restriction in number of interactions between young and old in mixed colonies could highlight the role of each age in defense strategies against fungal pathogen.

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 Bibliography Alciatore, G. (2022). Interplay between group composition, individual behaviour and network structure in the clonal raider ant (p. 97 p.) [ETH Zurich; Application/pdf]. https://doi.org/10.3929/ETHZ-B-000558574 Alciatore, G., Ugelvig, L. V., Frank, E., Bidaux, J., Gal, A., Schmitt, T., Kronauer, D. J. C., & Ulrich, Y. (2021). Immune challenges increase network centrality in a queenless ant. Proceedings of the Royal Society B: Biological Sciences, 288(1958), 20211456. https://doi.org/10.1098/rspb.2021.1456 Altman, D. G., & Bland, J. M. (1983). Measurement in Medicine: The Analysis of Method Comparison Studies. Journal of the Royal Statistical Society. Series D (The Statistician), 32(3), 307–317. https://doi.org/10.2307/2987937 Amutha, M., & Gulsar Banu, J. (2017). Variation in Mycosis of Entomopathogenic Fungi on Mealybug, Paracoccus marginatus (Homoptera: Pseudococcidae). Proceedings of the National Academy of Sciences, India Section B: Biological Sciences, 87(2), 343–349. https://doi.org/10.1007/s40011-015-0624-8 Pereira, T. D., Tabris, N., Matsliah, A., Turner, D. M., Li, J., Ravindranath, S., Papadoyannis, E. S., Normand, E., Deutsch, D. S., Wang, Z. Y., McKenzie-Smith, G. C., Mitelut, C. C., Castro, M. D., D’Uva, J., Kislin, M., Sanes, D. H., Kocher, S. D., Wang, S. S.-H., Falkner, A. L., … Murthy, M. (2022). SLEAP: A deep learning system for multi-animal pose tracking. Nature Methods, 19(4), 486–495. https://doi.org/10.1038/s41592-022-01426-1 Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, 67, 1–48. https://doi.org/10.18637/jss.v067.i01 Bonadies, E., Wcislo, W. T., Gálvez, D., Hughes, W. O. H., & Fernández-Marín, H. (2019). Hygiene Defense Behaviors Used by a Fungus-Growing Ant Depend on the Fungal Pathogen Stages. Insects, 10(5), 130. https://doi.org/10.3390/insects10050130 Brooks, M. E., Kristensen, K., Benthem, K. J. van, Magnusson, A., Berg, C. W., Nielsen, A., Skaug, H. J., Mächler, M., & Bolker, B. M. (2017). glmmTMB Balances Speed and Flexibility Among Packages for Zero-inflated Generalized Linear Mixed Modeling. The R Journal, 9(2), 378–400. https://doi.org/10.32614/RJ-2017-066 Bulmer, M. S., Franco, B. A., & Fields, E. G. (2019). Subterranean Termite Social Alarm and Hygienic Responses to Fungal Pathogens. Insects, 10(8), 240. https://doi.org/10.3390/insects10080240 Casillas-Pérez, B., Boďová, K., Grasse, A. V., Tkačik, G., & Cremer, S. (2023). Dynamic pathogen detection and social feedback shape collective hygiene in ants. Nature Communications, 14(1), 3232. https://doi.org/10.1038/s41467-023-38947-y Cremer, S., Armitage, S. A. O., & Schmid-Hempel, P. (2007). Social Immunity. Current Biology, 17(16), R693–R702. https://doi.org/10.1016/j.cub.2007.06.008

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 Davis, H. E., Meconcelli, S., Radek, R., & McMahon, D. P. (2018). Termites shape their collective behavioural response based on stage of infection. Scientific Reports, 8(1), 14433. https://doi.org/10.1038/s41598-018-32721-7 Fetter-Pruneda, I., Hart, T., Ulrich, Y., Gal, A., Oxley, P. R., Olivos-Cisneros, L., Ebert, M. S., Kazmi, M. A., Garrison, J. L., Bargmann, C. I., & Kronauer, D. J. C. (2021). An oxytocin/vasopressin-related neuropeptide modulates social foraging behavior in the clonal raider ant. PLOS Biology, 19(6), e3001305. https://doi.org/10.1371/journal.pbio.3001305 Hughes, W. O. H., Eilenberg, J., & Boomsma, J. J. (2002). Trade-offs in group living: Transmission and disease resistance in leaf-cutting ants. Proceedings of the Royal Society of London. Series B: Biological Sciences, 269(1502), 1811–1819. https://doi.org/10.1098/rspb.2002.2113 Kronauer, D. J. C. (2025). Clonal raider ants. Current Biology, 35(1), R7–R8. https://doi.org/10.1016/j.cub.2024.11.034 Kronauer, D. J. C., Pierce, N. E., & Keller, L. (2012). Asexual reproduction in introduced and native populations of the ant Cerapachys biroi. Molecular Ecology, 21(21), 5221–5235. https://doi.org/10.1111/mec.12041 Legan, A. W., Vogt, C. C., & Sheehan, M. J. (2023). Postural analysis reveals persistent changes in paper wasp foundress behavioral state after conspecific challenge. Ecology and Evolution, 13(9), e10436. https://doi.org/10.1002/ece3.10436 Lewin, R. (1987). Social Life: A Question of Costs and Benefits. Science, 236(4803), 775–777. https://doi.org/10.1126/science.3576197 Li, Z., Wang, Q., Knebel, D., Veit, D., & Ulrich, Y. (2024). Division of labour underlies efficient colony defence in a clonal ant (p. 2024.02.16.580644). bioRxiv. https://doi.org/10.1101/2024.02.16.580644 Liu, L., Zhao, X.-Y., Tang, Q.-B., Lei, C.-L., & Huang, Q.-Y. (2019). The Mechanisms of Social Immunity Against Fungal Infections in Eusocial Insects. Toxins, 11(5), 244. https://doi.org/10.3390/toxins11050244 Mascarin, G. M., & Jaronski, S. T. (2016). The production and uses of Beauveria bassiana as a microbial insecticide. World Journal of Microbiology and Biotechnology, 32(11), 177. https://doi.org/10.1007/s11274-016-2131-3 Matte, A., Billen, J., Shit, P., Heinze, J., & Bernadou, A. (2025). Delayed maturation of the exoskeleton and muscle fibres in the ant Platythyrea punctata. Biological Journal of the Linnean Society, 144(2), blae025. https://doi.org/10.1093/biolinnean/blae025 Meunier, J. (2015). Social immunity and the evolution of group living in insects. Philosophical Transactions of the Royal Society B: Biological Sciences, 370(1669), 20140102. https://doi.org/10.1098/rstb.2014.0102

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 Ortiz-Urquiza, A., & Keyhani, N. O. (2013). Action on the Surface: Entomopathogenic Fungi versus the Insect Cuticle. Insects, 4(3), 357–374. https://doi.org/10.3390/insects4030357 Ortiz-Urquiza, A., Luo, Z., & Keyhani, N. O. (2015). Improving mycoinsecticides for insect biological control. Applied Microbiology and Biotechnology, 99(3), 1057–1068. https://doi.org/10.1007/s00253-014-6270-x Pereira, T. D., Tabris, N., Matsliah, A., Turner, D. M., Li, J., Ravindranath, S., Papadoyannis, E. S., Normand, E., Deutsch, D. S., Wang, Z. Y., McKenzie-Smith, G. C., Mitelut, C. C., Castro, M. D., D’Uva, J., Kislin, M., Sanes, D. H., Kocher, S. D., Wang, S. S.-H., Falkner, A. L., … Murthy, M. (2022). SLEAP: A deep learning system for multi-animal pose tracking. Nature Methods, 19(4), 486–495. https://doi.org/10.1038/s41592-022-01426-1 Ravary, F., & Jaisson, P. (2002). The reproductive cycle of thelytokous colonies of Cerapachys biroi Forel (Formicidae, Cerapachyinae). Insectes Sociaux, 49(2), 114–119. https://doi.org/10.1007/s00040-002-8288-9 Thakur, R., & Sandhu, S. S. (2010). Distribution, occurrence and natural invertebrate hosts of indigenous entomopathogenic fungi of Central India. Indian Journal of Microbiology, 50(1), 89–96. https://doi.org/10.1007/s12088-010-0007-z Therneau, T. M. (2009). coxme: Mixed Effects Cox Models (p. 2.2-22) [Dataset]. https://doi.org/10.32614/CRAN.package.coxme Thorne, B. L., & Traniello, J. F. A. (2003). COMPARATIVE SOCIAL BIOLOGY OF BASAL TAXA OF ANTS AND TERMITES. Annual Review of Entomology, 48(Volume 48, 2003), 283– 306. https://doi.org/10.1146/annurev.ento.48.091801.112611 Vestergaard, S., Butt, T. M., Bresciani, J., Gillespie, A. T., & Eilenberg, J. (1999). Light and Electron Microscopy Studies of the Infection of the Western Flower ThripsFrankliniella occidentalis(Thysanoptera: Thripidae) by the Entomopathogenic FungusMetarhizium anisopliae. Journal of Invertebrate Pathology, 73(1), 25–33. https://doi.org/10.1006/jipa.1998.4802 Ward, A., & Webster, M. (2016). Sociality. In A. Ward & M. Webster (Eds.), Sociality: The Behaviour of Group-Living Animals (pp. 1–8). Springer International Publishing. https://doi.org/10.1007/978-3-319-28585-6_1 Wilson, E. O. (1975). Sociobiology The New Synthesis. http://archive.org/details/sociobiology-the- new-synthesis-edward-osborne-wilson-1975 Wilson, E. O. (1978). The Ergonomics of Caste in the Social Insects. The American Economic Review, 68(6), 25–35. Zhukovskaya, M., Yanagawa, A., & Forschler, B. T. (2013). Grooming Behavior as a Mechanism of Insect Disease Defense. Insects, 4(4), 609–630. https://doi.org/10.3390/insects4040609

Léo KORST - Master’s thesis 2026

Léo KORST - Master’s thesis 2026 Author: Léo KORST Master: Ecophysiology, Ecology, and Ethology, Strasbourg, France Host institution: Max Planck Institute for Chemical Ecology, Jena, Germany Supervisors: Sarah Rogoz and Yuko Ulrich Title: “Defending the colony against infection progression: Investigating allogrooming variations and survival after fungal exposure in different ages of the clonal raider ant, Ooceraea biroi, alongside an extended automated behavioral tracking approach” Key words: Automated tracking, grooming, fungal infection, social immunity, O. biroi Summary: Animal sociality produces costs and benefits, particularly in insect societies due to genetic proximity and interaction frequency, creating a sanitary risk for the group members. However, they developed a social immunity composed of physiological and behavioral strategies to prevent the risks, including against fungal pathogens. In that case, the ant colonies use allogrooming behavior directly after exposure to remove the spores. Allogrooming is usually analyzed manually, causing observer bias. Moreover, the task specialization and mixed age composition of the colony, especially in the clonal raider ant, Ooceraea biroi, create interaction variations in ant colonies after a fungal infection, but the effect of these social contexts and allogrooming response was unknown during infection progression. In this study, we investigated an extended automated approach by body parts pose estimation SLEAP to establish an allogrooming classification by filters for a future quantification of colony allogrooming reactions on one side ; and the allogrooming response of the colony across infection progression and colony age, as well as the survival of exposed individuals in different social environments after an advanced fungal infection on the other side. The results demonstrated a possibility for SLEAP to track the O. biroi ants in a colony and pathogen context. The grooming filters demonstrated great grooming event detection sensitivity but low specificity, highlighting the complexity of grooming behavior and the information requirements and conditions for an allogrooming classification. Then, biological investigation demonstrated the enduring allogrooming response across infection stages in old colonies. However, younger ant colonies showed decreasing grooming responses with infection stages, likely due to the rapid progression of infection in these individuals, indicated by early mortality. Furthermore, in the advanced infection stage, survival rates diverge based on socio-environmental factors, particularly among young ants exposed to fungus, necessitating group living but no age-matched individuals to enhance survival.