Virtual staining has emerged as a transformative technology in digital pathology, enabling the efficient and environmentally friendly generation of stained pathological images for diagnosis. However, current approaches for H&E to IHC conversion often fail to fully capture cross-channel relationships and neglect independent modeling of channel-specific features, limiting the preservation of key pathological semantics. To tackle these challenges, we propose a novel Multi-channel Correlation Learning (MCCL) framework. Considering that the H channel (Hematoxylin), E channel (Eosin), and D channel (DAB) respectively correspond to the staining of nuclei, cytoplasm (and extracellular matrix), and cell membranes, while IHC images only contain information for the H and D channels, we introduce the E channel from H&E images and calculate the cross-channel correlations among E, H, and D channels. This approach achieves comprehensive alignment of the semantic relationships among cellular structures. Furthermore, we utilize a pre-trained pathological foundation model to perform channel-level feature distillation on the H and D channels, enhancing the preservation of channel-specific pathological features while maintaining model compactness and computational efficiency. Experimental outcomes on the BCI and MIST datasets show that MCCL achieves state-of-the-art performance in quantitative metrics such as PSNR, SSIM, PCC, and FID. The framework generates high-quality and reliable virtual stained images, providing strong support for automated pathological diagnosis.