Metabolic interaction is a fundamental feature of co-existing microbial populations, yet current detection methods are usually slow and tedious, due to the inability to rapidly reconstruct population structure and profile species-resolved metabolic states in cocultures. Here we propose a D2O-probed single-cell Raman spectra (SCRS) based approach, and use it to probe the interactions between two co-inhabiting, closely related oral Streptococcus species of S. mutans (Sm; a cariogenic pathogen) and S. sanguinis (Ss; a symbiont) at single-cell resolution. Monocultures of each species reveal that D2O incubation enhances SCRS' ability to distinguish both between species and among growth phases. Thus, using a reference ramanome database of 14,650 cells from Sm and Ss monocultures, each under two conditions (50 % D2O and H2O) and at six time points, species plus metabolic state can be predicted with 99.7 % accuracy from a monoculture-derived SCRS under 50 % D2O, and this is further validated by mock two-species mixtures. Therefore, by temporally resolving both species and metabolic states (i.e., growth phases) via SCRS, cocultured Sm (cSm) and cocultured Ss (cSs) exhibit opposite patterns of oscillation in abundances; however, cSm grows slower than Sm, with cSs being the opposite. Moreover, for both cSm and cSs, within-population metabolic heterogeneity temporally increases. The coculture promotes nucleic acid and protein synthesis in cSm while suppresses them in cSs, and shows opposite effects in C-D bond (suppressed in cSm and promoted in cSs) and concurrent C-H bond reduction in both strains. Therefore, due to its ability to rapidly and label-freely profile both species and metabolic state at single-cell precision, this approach can be a valuable tool to study species interactions.