Defining the source of HIV-1 RNA in cerebrospinal fluid (CSF) will facilitate studies of treatment efficacy in the brain. Four antiretroviral drug-naive adults underwent two 48-hr ultraintensive CSF sampling procedures, once at baseline and again beginning on day 4 after initiating three-drug therapy with stavudine, lamivudine, and nelfinavir. At baseline, constant CSF HIV-1 RNA concentrations were maintained by daily entry of at least 10(4) to 10(6) HIV-1 RNA copies into CSF. Change from baseline to day 5 ranged from -0.38 to -1.18 log(10) HIV-1 RNA copies/ml in CSF, and from -0.80 to -1.33 log(10) HIV-1 RNA copies/ml in plasma, with no correlation between CSF and plasma changes. There was no evidence of genotypic or phenotypic viral resistance in either CSF or plasma. With regard to pharmacokinetics, mean CSF-to-plasma area-under-the-curve (AUC) ratios were 38.9% for stavudine and 15.3% for lamivudine. Nelfinavir and its active M8 metabolite could not be accurately quantified in CSF, although plasma M8 peak level and AUC(0-8hr) correlated with CSF HIV-1 RNA decline. This study supports the utility of ultraintensive CSF sampling for studying HIV-1 pathogenesis and therapy in the CNS, and provides strong evidence that HIV-1 RNA in CSF arises, at least in part, from a source other than plasma.