Abstract
Fertilization plays a vital role in replenishing soil nutrients, shaping microbial community composition, and enhancing agricultural productivity. Castor bean cake (CBC) is a nitrogen- and carbon-rich by-product increasingly used as an organic amendment, yet its effects on soil microbiomes remain unclear. Here, we compared CBC with a compound chemical fertilizer (CF) and a manure-based organic fertilizer (OF) across dose gradients using 16S rRNA sequencing and multi-level ecology analyses (α/β diversity, co-occurrence networks, and community assembly models). The results revealed that CBC increased bacterial richness and phylogenetic breadth relative to the unfertilized cultivated control, whereas OF showed dose-dependent declines in richness and CF maintained relatively stable richness with slight reductions in evenness at higher doses. Phylum-level composition shifted strongly with fertilizer identity: Bacillota decreased, whereas Pseudomonadota and Acidobacteriota increased under fertilization, with the largest compositional changes under CBC. CBC strengthened nutrient-enzyme-microbe coupling and generated increasingly complex, highly connected, and robust co-occurrence networks along the dose gradient, outperforming high-dose OF in network complexity and robustness, while OF maintained higher modularity. Null-model partitions (βNTI/RC_bray, NST, NCM, iCAMP) indicated that stochastic processes dominated community assembly across treatments; along the CBC gradient, dispersal limitation decreased from CBC1 to CBC2 and drift remained dominant, indicating increasing stochastic stabilization at moderate-high doses. Together, CBC promoted microbiome recovery and ecological resilience and represents a promising amendment for soil health.