Space nematode experiments constitute a main scientific experiment aboard China Space Station, where non-contact behavioral analysis via images is indispensable. Instance segmentation proves particularly crucial in this context by providing quantitative posture characterization essential for microgravity and radiation effect studies. However, conventional deep learning methods face significant challenges when directly applied to this task due to the unique physiological characteristics of nematodes such as their slender structure and high transparency. These inherent physiological characteristics frequently lead to segmentation fractures and weak boundaries under overlapping conditions. Meanwhile, space experimental images introduce additional interference factors such as bubbles, uneven illumination, and low contrast. The lack of annotated space datasets has hindered the identification of key challenges and the development of effective computational tools. To address these challenges, we propose the first space C. elegans instance segmentation (SpaceAnimal-CIS) dataset and benchmark, and a customized method tailored to C. elegans structures and space image characteristics. Our method introduces a Worm Convolution that dynamically extracts C. elegans features and constructs a Topology Preservation Attention to enhance the extraction of topological continuity, mitigating segmentation fractures. We also propose a Posture Point Loss to handle complex boundary conditions of overlapping C. elegans. The validation experiments on our built data demonstrate the effectiveness of our proposed method with 68.1% AP, which is 8.9% higher than current related methods, and in the more complex overlapping scenario, our method reaches 57.1%, which is 10.2% higher than the current related method.