Cryptic genetic variants exert minimal phenotypic effects alone but are hypothesized to form a vast reservoir of genetic diversity driving trait evolvability through epistatic interactions1–3. This classical theory has been reinvigorated by pan-genomics, which is revealing pervasive variation within gene families, cis-regulatory regions and regulatory networks4–6. Testing the ability of cryptic variation to fuel phenotypic diversification has been hindered by intractable genetics, limited allelic diversity and inadequate phenotypic resolution. Here, guided by natural and engineered cis-regulatory cryptic variants in a paralogous gene pair, we identified additional redundant trans regulators, establishing a regulatory network controlling tomato inflorescence architecture. By combining coding mutations with cis-regulatory alleles in populations segregating for all four network genes, we generated 216 genotypes spanning a wide spectrum of inflorescence complexity and quantified branching in over 35,000 inflorescences. Analysis of this high-resolution genotype–phenotype map using a hierarchical model of epistasis revealed a layer of dose-dependent interactions within paralogue pairs enhancing branching, culminating in strong, synergistic effects. However, we also identified a layer of antagonism between paralogue pairs, whereby accumulating mutations in one pair progressively diminished the effects of mutations in the other. Our results demonstrate how gene regulatory network architecture and complex dosage effects from paralogue diversification converge to shape phenotypic space, producing the potential for both strongly buffered phenotypes and sudden bursts of phenotypic change.
