Spiking Neural Networks (SNNs) offer a promising path toward energy-efficient, eventdriven computation inspired by the human brain. However, conventional training approaches in SNNs - such as unsupervised Spike-TimingDependent Plasticity (STDP) and surrogate gradient backpropagation—often face a trade-off between biological plausibility and task-specific performance. This paper introduces a novel biohybrid learning framework that integrates local STDP-based learning with global error-driven backpropagation to overcome this limitation. The proposed Adaptive STDP (A-STDP) algorithm dynamically adjusts synaptic updates using both temporal spike relationships and gradient feedback. Experiments on the MNIST dataset demonstrate that A-STDP achieves 93.2% classification accuracy while consuming 38% less energy compared to standard ANN models. Furthermore, hardware simulations on Intel’s neuromorphic chip, Loihi, confirm the scalability and energy efficiency of the approach. This hybrid model presents a practical and biologically grounded step toward deploying highperformance SNNs in real-world, low-power edge computing environments.