Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disorder characterised by progressive motor impairment and motor neuron (MN) degeneration. However, the early cellular and molecular events underlying its pathogenesis are not fully understood. There is mounting evidence that points to synaptic loss as an early hallmark of ALS, but the mechanisms that trigger and sustain this process are still unclear. Given the near-universal presence of TDP-43 pathology in ALS, here we used genetically tractable Drosophila TDP-43 models to analyse circuit-level impairments, glial contributions and evolutionarily conserved pathways of synaptic elimination. Behavioral and anatomical analyses revealed that humanized ALS-associated TDP-43 mutant flies exhibited impaired motor control in the absence of neuromuscular junction (NMJ) defects. Instead, synaptic structural analysis uncovered a preferred loss of inhibitory (INH) synapses between premotor interneurons and MNs in the ventral nerve cord, whereas excitatory (EXC) connections were preserved. Our results pointed to a CNS origin for motor impairments, aligning with the idea that ALS involves selective vulnerability of INH synapses, which disrupts excitation/inhibition (E/I) circuit balance. We identified Draper, a glial phosphatidylserine (PtdSer) receptor, as a candidate effector of the observed synaptic loss. Functionally analogous to mammalian phagocytic receptors such as MEGF10, MEGF12, TREM2, and MERTK, Draper mediates glial engulfment of synapses. Draper-mediated engulfement is regulated through a balance between its pro-phagocytic Draper-I and anti-phagocytic Draper-II isoforms. The anti-phagocytic Draper-II isoform was reduced in both TDP-43 loss-of-function and humanized ALS models, and its glial re-expression partially restored motor performance. These findings suggested that chronic, aberrant glial pruning of synapses via the PtdSer-Draper axis underlied INH synapse loss and associated motor impairments. Building on these insights, a strategy was designed to prevent inappropriate synaptic elimination by masking exposed PtdSer in the central nervous system and thereby blocking ectopic PtdSer-dependent glial phagocytosis. In Drosophila, this approach unexpectedly enhanced pruning and worsened phenotypes, which may reflect technical limitations such as overexpression effects. To evaluate the strategy in a mammalian context, we designed an adeno associated virus (AAV) based gene therapy approach enabling astrocyte-mediated delivery of PtdSer-masking peptides. In the widely used SOD1G93A ALS mouse model, intrathecal administration of the AAV contruct improved locomotor performance, supporting the therapeutic potential of targeting aberrant glial synapse elimination. To summarize, this work establishes INH synapse loss as an early and selective feature of ALS models, independent of MN degeneration or NMJ disruption. It further implicates PtdSer-dependent glial engulfment pathways as mechanistic drivers of this process, while providing proof of concept that synapse-targeted interventions may hold therapeutic promise. These findings refine our understanding of ALS as a disorder of circuit E/I imbalance and underscore both the promise and the challenges of targeting glial-synapse interactions in neurodegeneration.