Hydrogen and oxygen isotopic compositions of natural spa waters from hot and/or mineral-rich springs (including flowing/pumping wells) potentially provide important clues about water cycle in Earth's interior, because some of the waters are deep-seated and non-meteoric. However, the isotopic compositions of these waters are typically altered by contamination with meteoric water and are thus poorly known. To reconstruct their original (i.e., before these waters mix with meteoric water) isotope compositions, here we develop an isotopic evolution model of the lithosphere-enclosed waters and derive the “ocean-origin lithospheric water curve” (OLWC; δ2H = 60 / (δ18O – 11)) as a typical evolutionary track starting from ocean water. Then we apply them to spa waters from central Japan, a globally unique geotectonic region where double subduction of slabs occurs. The OLWC explains well observed isotope data of subseafloor pore waters, submarine mud volcano pore waters, coastal oil-field brines, and subduction zone volcanic steam. Mixing analysis indicates that lithospheric water accounts for 19–95 % of each spa water we investigated. The nonlinear relationships of the reconstructed water δ18O and Cl or Li concentrations justify our hypothesis that the lithospheric waters are undergoing evolution from ocean water to magmatic water by rock-water isotopic exchange and that they are not simply mixtures of the two. A comparison of the reconstruction results with model predictions under slab subduction settings suggests that most spa waters in the southern to central part of the study region contain progressively evolving lithospheric waters released from the subducting Philippine Sea slab, and that a few at western edge of the region contain waters released from the Pacific slab. The others in the northern and eastern part of the region contain waters at the early-to-mid evolutionary stage, which are discharged from Tertiary marine sedimentary rocks (including tuffs). All these waters were enclosed in the lithosphere for more than 1–5 Ma. Our approach provides useful tools for tracing sources of non-meteoric water and highlights water cycle through lithosphere for more than one million years.