It has been known for nearly a half century that human tumors, including those derived from the nervous system such as glioblastomas, medulloblastoma, and neuroblastomas are much more sensitive than normal tissues to l-methionine (l-Met) starvation. More recently, systemic l-Met depletion by administration of Pseudomonas putida methionine-γ-lyase (MGL) could effectively inhibit human tumors xenografted in mice. However, bacterial-derived MGLs are unstable in serum (t(1/2) = 1.9 ± 0.2 h) and highly immunogenic in primates. Since the human genome does not encode a human MGL enzyme, we created de novo a methionine degrading enzyme by reengineering the structurally homologous pyridoxal phosphate-dependent human enzyme cystathionine-γ-lyase (hCGL). hCGL degrades l-cystathionine but displays no promiscuous activity toward l-Met. Rational design and scanning saturation mutagenesis led to the generation of a variant containing three amino acid substitutions (hCGL-NLV) that degraded l-Met with a k(cat)/K(M) of 5.6 × 10(2) M(-1) s(-1) and displayed a serum deactivation t(1/2) = 78 ± 5 h (non-PEGylated). In vitro, the cytotoxicity of hCGL-NLV toward 14 neuroblastoma cell lines was essentially indistinguishable from that of the P. putida MGL. Intravenous administration of PEGylated hCGL-NLV in mice reduced serum l-Met from 123 μM to <5 μM for over 30 h. Importantly, treatment of neuroblastoma mouse xenografts with PEGylated hCGL-NLV resulted in near complete cessation of tumor growth. Since the mode of action of hCGL-NLV does not require breaching the blood-brain barrier, this enzyme may have potential application for sensitive tumors that arise from or metastasize to the central nervous system.