Existing deep networks for histopathology image synthesis cannot generate accurate boundaries for clustered nuclei and cannot output image styles that align with different organs. To address these issues, we propose a style-guided instance-adaptive normalization (SIAN) to synthesize realistic color distributions and textures for different organs. SIAN contains four phases, semantization, stylization, instantiation, and modulation. The four phases work together and are integrated into a generative network to embed image semantics, style, and instance-level boundaries. Experimental results demonstrate the effectiveness of all components in SIAN, and show that the proposed method outperforms the state-of-the-art conditional GANs for histopathology image synthesis using the Frechet Inception Distance (FID), structural similarity Index (SSIM), detection quality(DQ), segmentation quality(SQ), and panoptic quality(PQ). Furthermore, the performance of a segmentation network could be significantly improved by incorporating synthetic images generated using SIAN.