In this paper, we propose a new formulation of gradient-enhanced universal Kriging that uses sparse polynomial chaos expansion (PCE) as trend functions. In this regard, the gradient information is used to improve both the trend function and the Gaussian process part. The optimal set of polynomial terms is selected based on the least angle regression algorithm. We tested the performance of the proposed gradient-enhanced polynomial chaos Kriging (GEPCK) in several algebraic and non-algebraic test cases and compared it with ordinary Kriging (OK) and ordinary gradient-enhanced Kriging (GEK). Results show that GEPCK consistently outperformed other methods or at least competitive to the best performing method on both algebraic and non-algebraic problems. This indicates that the performance of the conventional GEK can be further improved by incorporating sparse PCE as trend functions.