Antiretroviral drugs and the central nervous system.

Abstract

Introduction With potent antiretroviral drug combinations it is possible to reduce the number of HIV RNA copies in plasma to undetectable levels [1]. Persistent low levels of HIV RNA in plasma are correlated with an improved prognosis [2]. However, for several sites in the infected host, it is as yet unclear how efficacious combination treatment is in suppressing viral replication. One of these sanctuary sites is the central nervous system (CNS). The presence of HIV and the level of HIV replication in the brain parenchyma are difficult to study, and we have to rely on post-mortem examination of brain tissue, animal studies, and cerebrospinal fluid (CSF) levels of HIV RNA. For clinical practice, CSF viral load measurements are relatively easy to obtain and are potentially useful. HIV RNA is detectable in the CSF of most HIV-infected individuals. The CSF HIV load varies widely (median, 3 log10 copies/ml; range, 0–6 log10 copies/ml), but is generally lower than the plasma viral load [3–10]. Patients in whom CSF viral load has exceeded plasma viral load have been reported [5,11–14]. Conflicting results have been found in the correlation between plasma and CSF viral load. In many studies a positive correlation has been noted [4,8,10,15–17], but in others no correlation was found [5–7,9,13,14,18]. A further intriguing finding is that in several studies a correlation was found between CSF viral load and the extent of damage to the blood—CSF barrier or CSF white cell count [12,14,15,19], whereas in others this correlation has been disputed [3,6,7,20]. Different study populations and the influence of antiretroviral therapy may explain some of these differences. Several groups of patients can be described (Table 1). The first group consists of patients with low CD4 cell counts and CNS opportunistic infections (e.g., patients with cryptococcal meningitis usually have high CSF viral loads) [6,14]. An elevated CSF white cell count probably contributes to this increased level of HIV replication [6,14]. A second group consists of patients with low CD4 cell counts and AIDS dementia complex or HIV encephalitis. In these patients, CSF HIV load probably reflects increased CNS (but not plasma) viral replication [6,13]. High CSF viral loads correlate with the presence and degree of cognitive impairment [3,6,9,10] or neuropathological abnormalities [13]. Furthermore, a correlation has been found between brain and CSF levels of HIV RNA in patients with AIDS dementia [10]. The third and largest group comprises neurologically asymptomatic patients who may have detectable CSF viral loads [8,9,12,21].Table 1: . Cerebrospinal fluid (CSF) viral load studies.The CSF viral load increases with decreasing CD4 cell count [10]. In antiretroviral-naive patients, viral replication within the plasma and CSF compartments occurs independently [12]. To our knowledge, no data are available on CSF viral RNA in patients with primary HIV infection. It is, however, known that HIV can be cultured from the CSF (and not from the blood) during seroconversion [22,23]. HIV antigen can sometimes be detected in the CSF (before it is present in the blood) from such patients [24]. HIV load within the brain parenchyma can only be determined by post-mortem examination. In two studies [10,25], brain tissue was examined for the presence of viral RNA. In the first study, HIV RNA could be detected in 50% of the examined brains [25]. Quantitative data were given in another study, but non-demented patients did not differ from demented patients [10]. Although provirus was detected in the brain of one patient who had been infected 15 days earlier, generally no substantial amounts of provirus could be detected in asymptomatic patients. In most patients with AIDS, proviral DNA has been found, with the highest levels in patients with HIV encephalitis [26–29]. Does antiretroviral treatment suppress HIV replication in the CNS? Patients with HIV dementia may improve with zidovudine (ZDV) treatment, and the widespread use of ZDV has contributed to the decrease in the incidence of HIV dementia [30–32]. HIV-specific neuropathological abnormalities are less frequent in patients who used ZDV until death, compared with patients who discontinued the drug 1 month (or earlier) before death [33]. CSF viral load was reduced in most patients after the start of ZDV monotherapy [34]. These data suggest that ZDV suppresses HIV replication within the CNS. Data for all the other antiretroviral drugs are sparse. Reduction of CSF viral load has been observed in some patients on didanosine (ddI) monotherapy [34]. In several nucleoside-containing combination studies, CSF viral load has been reported to be reduced to undetectable levels [8,12,35,36]. In one study using ultrasensitive quantitative PCR, the viral load decreased to values below 20 copies/ml [21]. In two patients who had been given triple therapy, the decrease in CSF viral load lagged behind the decrease in plasma viral load [37]. Ritonavir-saquinavir combination treatment, and other saquinavir or indinavir-containing regimens (sometimes including nucleoside analogues) have also resulted in undetectable CSF viral loads in several patients [17,38–41]. With an increasing number of available antiretroviral drugs, the penetration of these drugs in the various body compartments has become an important issue in the selection of drug combinations. We have reviewed all the available CNS pharmacokinetic studies conducted with currently available antiretroviral drugs, both in animals and humans. The anatomy of the barriers, transport across the barriers, and pharmacokinetic methodology are discussed and an overview of the pharmacokinetic studies is presented. Some recommendations for the optimal treatment of HIV replication within the CNS are given. The barriers in the CNS The blood-brain barrier (BBB; between blood and brain interstitial fluid) and the blood—CSF barrier (BCB; between blood and CSF) both protect the brain from transient changes in the composition of the blood. Because the surface area of the brain capillaries is 5000-fold larger than that of the choroid plexus, transport across the BBB represents the principal route of entry for most molecules into the CNS [42–44]. The blood-brain barrier The BBB is formed by brain capillary endothelial cells that are fused together by tight junctions. These continuous tight junctions, the lack of intercellular pores, the paucity of pinocytosis, and the large mitochondrial content to fuel the transport pumps of the endothelium form the unique characteristics of the BBB. Small lipidsoluble molecules such as ethanol can cross the BBB by simple diffusion in the lipid layers of the endothelial cell membrane. Lipid solubility and molecular weight largely determine brain endothelial permeability. Several lipid-insoluble molecules, such as glucose, are actively transported through carrier systems [44,45] (Fig. 1A).Fig. 1: . (A) The blood-brain barrier (BBB). The BBB is formed by brain capillary endothelial cells (BCEC) that are fused together by tight junctions (TJ). The brain capillary endothelial cells and pericytes (P) are surrounded by a basement membrane (BM). Astrocyte foot processes (AFP) almost completely surround the brain capillary. CL, Capillary lumen; BECF, brain extracellular fluid. (B) The blood—cerebrospinal fluid (CSF) barrier. The choroid plexus endothelial cells (CPEC) form fenestrated capillaries. They are surrounded by a basement membrane (BM). The capillaries are covered by a single layer of choroid epithelial cells (CEC). These choroid epithelial cells are sealed together by tight junctions (TJ). One side of the choroid epithelial cell is in contact with the blood filtrate within the extracellular space of the choroid plexus (ECS). The opposite side is in contact with the CSF in the ventricles. (C) The CSF-brain barrier. The CSF is separated from the brain extracellular fluid (BECF) by the loosely linked ventricular ependymal cells (VEC). These cells do not form an anatomic barrier. Diffusion of compounds may occur in both directions. The brain extracellular fluid surrounds the astrocytes [A; with astrocyte foot processes (AFP)] and neurons (N). Neurons have dendrites (D) and an axon (Ax). One capillary is shown surrounded by astrocyte foot processes.The permeability of the BBB is the result of the ratio of transfer constants (k) of a substance for influx and efflux. If these transfer constants are equal, free concentrations within plasma and brain tissue will eventually become equal in a steady-state situation. For higher values of k, the steady state will be reached sooner. However, the continuously circulating CSF may act as a sink to the brain, keeping the concentration in the brain lower than in the plasma. A further asymmetry in transport may be caused by carrier mechanisms, protein or tissue binding, and metabolic processes [46,47]. Concentrations of molecules within the brain are not homogeneous. The concentrations depend on differences in vascularity between, for example, the white and grey matter and in the distance from the CSF compartment [46]. The blood-CSF barrier The choroid plexuses and the arachnoid membrane constitute the BCB. The arachnoid membrane has a merely passive role. The choroid plexuses, however, actively regulate the composition of the CSF and are therefore often considered as ‘miniature’ kidneys. The choroid plexuses consist of capillaries surrounded by a single layer of epithelial cells. Adjacent choroid epithelial cells are sealed together by tight junctions and these form the anatomical basis of the BCB. One side of the epithelial cell is in contact with the blood, and the opposite side is in contact with the CSF in the ventricles (Fig. 1B). The most important function of the choroid plexuses is the secretion of CSF. The total volume of CSF (140 ml) is replaced 4–5 times daily. Only 25 ml flows through the ventricles; the major part moves through the subarachnoid space by bulk flow and eventually drains into the venous system through the arachnoid granulations in the superior sagittal sinus. Entrance via the choroid plexus through active transport systems present in the choroid plexuses only is the principal route of penetration into the CNS for some solutes (e.g., vitamins), which can reach high CSF concentrations. The choroid plexus also contains ‘exit’ pumps [42,48]. The permeability of the BCB is also defined by the ratio of the transfer constants (k) for influx and efflux of a compound. It is important to note that differences in concentrations between ventricular and spinal CSF may exist for slowly penetrating substances [46]. The CSF-brain barrier The CSF is separated from the extracellular fluid of the brain by the loosely linked ependymal cells of the ventricles. These cells do not form an anatomical barrier. Diffusion of solutes therefore occurs in both directions [46] (Fig. 1C). The distributed model for drug delivery addresses the relationship between brain tissue and CSF drug concentrations after systemic drug delivery [47]. According to this model, CSF concentrations will be less than brain tissue drug concentrations during plasma steadystate concentration. Slowly penetrating drugs have the steepest gradients within the brain parenchyma. It has often been suggested that intrathecal administration of slowly penetrating drugs might circumvent the BBB. Unfortunately, brain tissue levels rapidly decrease with increasing distance from the ventricular surface [48,49]. Therefore, intrathecal administration is far less efficient for treating brain parenchymal disease (e.g., AIDS dementia complex) than for intrathecal disease (e.g., lymphomatous meningitis). Antiretroviral drug entry into the CNS Most experimental studies of antiretroviral drug entry into the CNS have focused on nucleoside analogues. The nucleoside analogues undergo intracellular phosphorylation to the active triphosphate metabolites. It would be ideal to measure the concentrations of these triphosphates in the different infected cells in vivo, although from a practical point of view this is very difficult. Therefore, most approaches have centred on the concentrations of the parent drugs. In early reports it was found that none of the nucleoside analogues were measurably transported across the BBB, although most drugs had measurable CSF concentrations. Therefore, it was hypothesized that nucleoside analogues would enter the brain tissue interstitial fluid via the CSF [50]. In more recent studies it was found that ZDV crosses the BBB (4 versus 1%) and BCB (2.7 versus < 1%) significantly better than truly inert polar molecules such as mannitol [51]. CNS entry probably occurs purely by diffusion. Nucleoside transporters do exist, but modification at the 3′-position of the sugar moiety (as in ZDV) greatly reduces the affinity of the nucleoside analogues for this transport mechanism [51]. A further argument against an important role for nucleoside transporters is the absence of competitive inhibition between nucleoside analogues [51,52]. Of all nucleoside analogues, ZDV is the most lipophilic molecule (Table 2). From the few studies in which influx transfer constants were calculated, it appears that diffusion of ZDV into the CSF (k = 0.04) is indeed about 20-fold greater than that of ddI (k = 0.002), consistent with a 20-fold higher lipophilicity [53–56].Table 2: . Characteristics of antiretroviral drugs that are important for penetration into the central nervous system (CNS).Because actual CSF concentrations are also dependent on efflux, differences in the influx—efflux ratio may explain the discrepancies between CNS concentrations and lipophilicity, as found for stavudine (D4T) [52]. An active efflux transport system for ZDV, ddI, and probably D4T is present in both barriers [57]. It is inhibited by probenecid, resulting in increased CSF [55] and brain concentrations [57,58] of ddI and increased CSF [59–63] and brain concentrations [57–59,64] of ZDV. For ddI, this transport system was found to be identical to the one removing benzylpenicillin [65]. Very few studies have concentrated on the influx and efflux mechanisms of other antiretroviral drugs. The protease inhibitors were recently found to be a substrate for P-glycoprotein present in the BBB. P-glycoprotein is responsible for actively pumping back several large (>500 Da) lipid-soluble compounds into the blood, thereby preventing their entry into the brain tissue [66]. P-glycoprotein therefore limits the penetration of protease inhibitors into the brain tissue. This is an important field of research, because such efflux mechanisms may cause lower brain tissue penetration than expected in view of other drug characteristics. Table 2 summarizes the characteristics of the antiretroviral drugs that are important in CNS penetration. CNS penetration is linearly related to lipid solubility divided by the square root of molecular weight [48,67]. Substances with a molecular weight above 500 Da are often considered physically impeded to cross the cell membrane [66]. The oil/water partition coefficient is a measure of lipid solubility for neutral compounds. For acids and bases, lipid solubility is further determined by their degree of ionization, which is pH-dependent. For this reason, the partition coefficients of such drugs must be determined at physiologic pH and these are the distribution coefficients. The optimal oil/water partition coefficient is about 100. Drugs with very high oil/water partition coefficients (> 1000) have lower diffusibility because they do not easily diffuse from the lipid layer into the extracellular fluid of the brain. Furthermore, such drugs often have high serum protein binding [43,56,68,69] (manufacturer's information: Zerit, Bristol-Myers Squibb, Princeton, New Jersey, USA). Because only the free (non-protein bound) portion of drug can enter the CNS, high serum protein binding (> 90%) restricts drug entry into the CNS [48]. Table 2 also includes median inhibitory concentration (IC50) values, which range according to the assays, cell lines, virus strains, and duration of treatment. The IC50 values for clinical isolates tested in peripheral blood cells were selected for Table 2 [70–78] (manufacturers' information: Retrovir, Glaxo Wellcome, Research Triangle, North Carolina, USA; Crixivan, Merck Sharpe and Dohme, Haarlem, The Netherlands; Viramune, Boehringer Ingelheim, Ridgefield, Connecticut, USA). Quantification of drug transport across the barriers Transport of drugs across the BBB Quantitative measurement of drug delivery to the brain has been hindered by the complexity of experimental techniques. The invasive nature of brain tissue sampling generally limits these experiments to animal studies and to only one measurement per animal. A single determination of the ratio of brain tissue to plasma concentration, however, will not usually represent the dynamics of drug entry into the brain [47,48]. Several in vivo and in vitro methods for transport across the BBB have been described [43,46,79,80] (Table 3).Table 3: . Models for quantifying drug transport across the blood-brain barrier (BBB).Brain uptake index The brain uptake index (BUI) technique represents a single-pass method using intracarotid administration of a radiolabelled drug and a highly diffusible reference compound (such as butanol or water) and immediate (after 5–15 sec) decapitation of the animal. The BUI determines the proportion of drug removed from the blood in a single passage compared with the reference substance; BUI can only take values from 0 to 1. For values below 0.04, the accuracy of BUI is limited [80]. In situ brain perfusion The in situ brain perfusion method estimates the brain parenchymal uptake of radiolabelled tracers from the perfusion fluid after arterial inflow. Because this method extends the time of exposure of the brain to the drug of interest to 20–30 min, this method has far greater sensitivity for the uptake of slowly moving drugs than the BUI [51]. Intravenous administration BBB permeability can also be assessed by determining the brain tissue—plasma concentration ratio after intravenous administration of the drug. Animals receive either a constant rate infusion (resulting in steady-state plasma drug concentrations), or an intravenous bolus (resulting in declining plasma drug concentrations), and after a variable period of time, the animal is decapitated and the brain tissue is homogenized for the determination of drug concentrations. Using mathematical models, transfer rate constants (k) for influx and efflux into the brain (or CSF) can be calculated. The unit of k is ml/minXg brain tissue, but because the specific gravity of brain is assumed to be 1.0, k is expressed in time−1. The ratio between these transfer constants for influx and efflux determines the BBB permeability. The linear two-compartment model (plasma and brain), which assumes no drug exchange between the CSF and brain tissue, is most often used. The results are model-dependent. In humans, the intravenous technique can be extended to measure regional differences by using positron emission tomography [79]. Either radiolabelled drug is administered, and the brain is counted for total radioactivity, which includes (in)active metabolites, or drug concentrations are measured by high performance liquid chromatography (HPLC) or radioimmunoassays, which are specific for the unchanged drug [81]. Microdialysis A different method used to study in vivo transport is microdialysis. A probe is implanted into a specific region of the brain and often a second probe is implanted in the ventricle. The tip of the probe is in contact with the interstitial fluid. This fluid will then be continuously sampled for drug concentrations (the concentration—time curve in a single animal can be constructed). A further advantage is that no correction for intravascular drug is necessary. The major disadvantage is the possible disintegration of the BBB caused by the procedure, resulting in an overestimation of brain tissue penetration. Essentially the same mathematical models are necessary to interpret the data as for the intravenous method [43,79]. The in vitro model consists of a bovine brain endothelial cell monolayer, whereby values for penetration across the monolayer can be calculated [80]. In vivo intracerebral microinjection The in vivo intracerebral microinjection technique is a novel method to measure brain—blood transport [58]. A microinjection needle is inserted into the brain through which the radioactively labelled drug of interest is administered. The animal is decapitated after the sampling of CSF, and brain tissue samples are collected for counting of radioactivity. An efflux rate constant is determined. By using this technique, it is possible to study the efflux transport systems at the BBB in regions distant from the ventricles. However, in brain regions near the ventricles, removal of drug at the BCB after diffusion into the CSF is too important to be neglected. In summary, the simplest method is the BUI, which can act as a screening procedure for gross estimation of affinity for the brain. The in situ brain perfusion method is superior for drugs with low diffusibility because of greater sensitivity. However, no steady-state ratios can be obtained by either of these methods. If the goal of the study is to estimate transfer rate constants, then the intravenous method or microdialysis should be used. By using one of the first three methods, the brain tissue is homogenized and includes extracellular fluid, intravascular fluid and intracellular fluid. The measured content of drug in the brain should therefore always be corrected for the quantity of drug present in the vascular compartment. This correction is especially important for poorly penetrating drugs, which have significant intravascular fractions of total brain content. A major disadvantage of the intravenous method is that one concentration—time curve has to be constructed from data of different animals [43,79,80]. Transport of drugs across the BCB Many studies have dealt with CSF penetration. In the majority of these studies, a CSF sample was taken at variable intervals after bolus administration. The ratio between the concentrations in a single plasma—CSF pair is generally taken as a measure of CSF penetration. This method is flawed because ratios tend to increase in time due to the different shape of concentration—time curves in the plasma and the CSF. Because the elimination from the CSF compartment is generally slower, ratios may rise above 1.0 for drugs with short plasma half-lives [47,48,67,68]. For these reasons, the CSF—plasma (p) ratio is only a suitable method in steady-state situations. In the absence of steady state, the ratio between the area under the curves (AUC) for CSF and plasma should be taken (Fig. 2). In the steady-state situation, small lipophilic drugs will have CSF concentrations (almost) equal to free (equalling protein unbound) plasma concentrations. With all other drugs, CSF concentrations will be lower than plasma concentrations [47,68]. The AUCCSF/AUCp is considered the gold standard for characterizing drug entry [67].Fig. 2: . An example of a concentration—time curve in the cerebrospinal fluid (CSF) and plasma after an intravenous bolus administration of a drug has been given. The plasma concentration decreases rapidly after intravenous bolus administration. The CSF concentration—time curve lags behind the plasma concentration—time curve. The maximum CSF concentration is reached later than the maximum plasma concentration. The CSF—plasma ratio changes over time, and in this example becomes greater than 1.0 after 3 h. A single CSF—plasma ratio is therefore an inadequate method to estimate CSF penetration. The areas under the concentration—time curves (AUC) from the start of the infusion to the last concentration measured in the plasma can be calculated and extrapolated to infinity. This CSF-plasma ratio of the AUC is a far more precise estimate for the availability of a drug in CSF. The AUCplasma is shown in yellow and orange. The AUCCSF is shown in red and orange. The orange area is the area that is part of both AUCplasma and AUCCSF. The yellow part of AUCplasma shows the time course during which plasma concentration is greater than CSF concentration. The red part of AUCCSF shows the time course during which CSF concentration is greater than plasma concentration.In animal studies, the ventricular CSF has been examined most often. Two animals were evaluated for both lumbar and ventricular drug concentrations: lumbar CSF—plasma AUC ratio (0.41) was higher than ventricular CSF—plasma AUC ratio (0.08) [82]. Exchange of drugs between the CSF and brain tissue The ventriculocisternal perfusion technique measures the efflux from CSF to brain (and plasma) during steady-state ventricular drug administration. The radioactively labelled compound of interest is perfused through the ventricular system, the animal is sacrificed after an interval up to several hours, and the brain tissue is subsequently sampled [80]. An alternative method to study steady-state ventricular drug administration is intracerebroventricular administration. In this case the microdialysis ventricular probe is used to instillate the drug. The brain parenchymal probe is used to measure interstitial fluid drug concentrations [52,83]. Pharmacokinetic studies of antiretroviral drugs Brain tissue penetration studies A review of studies on the penetration of antiretroviral drugs into the brain tissue has been conducted, and studies were classified according to the methodology used. Just as single determinations of CSF—plasma ratios are considered inadequate, single brain tissue—plasma ratios do not represent the dynamics of drug concentrations in the CNS. The discussion will therefore focus on those studies in which brain—plasma ratios in a steady state after continuous infusion were reported, or in which AUC ratios after intravenous bolus administration were calculated (Table 4).Table 4: . Brain tissue penetration studies.BUI studies Few studies have used BUI [50,84,85]. For dideoxyadenosine (which is degraded to ddI), zalcitabine (ddC) and ZDV, a very low uptake (< 0.05) was found, comparable to methotrexate [46]. BUI is generally considered inaccurate for such low values [51]. For indinavir, BUI was found to be higher (0.33) [85]. Because different reference compounds were used in the above-mentioned studies, the results are not entirely comparable. In situ brain perfusion The in situ brain perfusion technique results in a time-dependent brain tissue—plasma ratio and CSF-plasma ratio of the amount of radioactivity [51,86]. One study compared ZDV and D4T. A significant difference for all timepoints until the end of the experiment (20 min) was found for penetration into brain tissue, and ZDV (ratio of 3.5%) was found to be superior to D4T (ratio of 1.1%) [86]. Intravenous studies Several studies evaluated single brain tissue—plasma ratios [56,76,87–99]. The following antiretroviral drugs were investigated by this inadequate method only: abacavir (1592U89), nevirapine, delavirdine and ritonavir [73,76,96,98,99]. In other studies, however, brain—plasma ratios in steady-state or AUC ratios were given [53–55,81,85,100–104]. All studies used HPLC (or radioimmunoassay) for drug concentration measurements, except for two studies that measured total radioactivity [85,104]. In these two studies, the major source of radioactivity is probably represented by inactive metabolites. Therefore, one should be cautious in interpreting their results. The values for brain tissue—plasma ratios of a single drug vary widely between studies (Table 4). However, ZDV appears to penetrate better into brain tissue than ddI. Furthermore, indinavir appears to be superior to both nucleoside analogues. CSF penetration ratios are given for comparison in Table 4. In intravenous studies that assessed both brain and CSF penetration, brain—plasma ratios were greater than CSF—plasma ratios. This difference disappeared when a correction for the intravascular content was made [53]. Microdialysis studies Brain tissue penetration of ZDV was most often evaluated by using microdialysis techniques [52,59,64,83,105–108]. By using microdialysis it is possible to directly and continuously measure the free extracellular concentrations of drugs in tissue and blood. This is an important difference compared with intravenous models in which total drug concentration is measured. Therefore, results from microdialysis studies can be compared with one another, but for comparison with data from intravenous studies, the protein binding should be accounted for. Only two other drugs were evaluated by microdialysis: ddC [109], and D4T [52]. Brain penetration of D4T was greater than ZDV (Table 4); furthermore, brain penetration of ZDV was greater than ddC. The results for D4T and ddC cannot be compared directly, because instead of a brain—plasma ratio, a brain—muscle ratio was determined for ddC [109]. In vitro models Three in vitro studies have been published [61,73,110]. Relative values for penetration were given in one study [110]. Relative penetration across the monolayer for nevirapine, ZDV and indinavir was 10 : 2 : 1. Penetration of delavirdine was virtually absent [110]. Preliminary data show that abacavir and ritonavir may be superior to ZDV when measured using this method [73]. CSF penetration studies Animals Only studies in which either the CSF—plasma ratio in steady state (by continuous infusion) or CSF—plasma AUC are considered in this analysis. These studies involved either primates [82,95,111–114], rodents [52–55,59,61–63,83,104,105,107] or dogs [90,100,115,116] (Table 5). CSF—plasma ratios in ventricular fluid were low for ddC (0.03), ddI (0.05), and lamivudine (3TC; 0.08) in primates. Ratios were much higher for ZDV (0.21) and D4T (0.50). D4T was not tested in primates but in rats. Ratios for specific drugs did not differ much among species.Table 5: . Cerebrospinal fluid (CSF) penetration after intravenous drug administration in different species.Other studies considered only a single CSF—plasma ratio [60,73,99,114,117–121]. Three drugs were evaluated by this method only: abacavir, efavirenz and 524W91 [73,99,120,121]. Ratios were 0.17, 0.60, and 0.04, respectively, at various timepoints after dosing. Therefore, interpretation of these ratios is difficult. Humans For data in humans, two different studies were reported [122–124]. Both studies evaluated ZDV. A steady-state CSF—plasma ratio of 24 ± 9% was found in children, and an AUC ratio of 75 ± 26% was found in adults. Absolute CSF concentrations as a function of time after last administered dose were compared [12,37,39–41,78,125–152] (manufacturers' information: Zerit, Bristol-Myers Squibb; Retrovir, Glaxo Wellcome; Videx, Bristol-Myers Squibb). This comparison focused on CSF concentrations in patients who had chronic oral dosing schedules comparable to current schedules [12,37,39–41,78,125,127–140]. Table 6 shows CSF concentrations in humans after chronic oral dosing regimens of all antiretroviral drugs. Most studies used HPLC methods for the analysis, with the exception of three studies, which used radioimmunoassay for D4T [128] or ZDV [12,137]. CSF concentration—time curves in chronic oral dosing are far more stable than plasma concentration—time curves [12,136,137] (Fig. 2). Therefore, these concentrations could be present for most of the time, and can be compared with IC50 values (Table 2).Table 6: . Cerebrospinal fluid (CSF) concentration in humans during chronic oral dosing regimens.The absolute CSF concentrations (expressed in μmol/1) of the different nucleoside analogues in humans did not differ more than two—threefold [12,73,127–130,132,133,135–137]. CSF concentrations of protease inhibitors appeared to be below the detection limits of the drug assay, except for indinavir [37,39–41,125,138,140]. Of the non-nucleoside reverse transcriptase drugs, some data were available on atevirdine, nevirapine, and efavirenz [78,134,139]. Nevirapine showed by far the most promising CSF drug concentration. CSF—brain exchange Diffusion of ZDV and D4T from the CSF to the brain extracellular fluid was found to be considerable during steady-state intraventricular drug administration using microdialysis [52,83]. Using the ventriculocisternal technique, intraventricular administration appeared to be very inefficient because only small amounts of the drug diffused to brain tissue [51]. Intraventricular administration has been applied to humans, but its usefulness should be clearly proven before considering this technique in clinical practice [52,83,153]. Discussion In the course of HIV infection, the presence and replication of HIV in the CNS is important, because in a minority of advanced-stage patients, HIV replication in the CNS leads to HIV dementia, and because the CNS is probably the most important sanctuary site in HIV-infected individuals. Important questions are as follows: Is it possible to suppress viral replication within the CNS as successfully as observed systemically with the current available antiretroviral drugs? Which drugs should be used for this purpose and how should their efficacy be measured? Drug entry into the CNS is hampered by the BBB and the BCB. CNS entry probably occurs purely by diffusion, because currently no antiretroviral drug transporters have been found [51]. Drugs with a very high protein binding (i.e., most protease inhibitors) are impeded in crossing these barriers [48]. Lipid-soluble drugs are known to have better CNS penetration; however, an optimum lipid solubility exists for drugs with an oil/water partition coefficient of 2 log10. The nucleoside analogues therefore have lipid solubility that is too low, and most protease inhibitors have lipid solubility that is too high. Nevirapine appears to be have the best drug characteristics. Many drugs with a molecular weight above 500 Da (including protease inhibitors) have been found to be a substrate for P-glycoprotein, which is responsible for actively removing several lipid-soluble compounds from the brain parenchyma [66]. Different efflux mechanisms seem to exist for several nucleoside analogues, further impeding effective CNS drug concentrations [57]. Even though knowledge on CNS entry of antiretroviral drugs in animals and humans is limited, the following summary may be given. Considering brain tissue penetration of the nucleoside analogues, ZDV has brain tissue levels that are ∼20% and D4T has levels that are ∼30% of concomitant plasma levels. Both drugs appear to be superior to ddI (∼2–4%). For 3TC no data on brain tissue penetration are available. Two studies comparing D4T and ZDV yielded conflicting results [52,86]. ZDV was found to be superior to D4T using the brain perfusion technique, but inferior using microdialysis at steady state. This difference may however be explained by a slower, but not necessarily inferior CNS penetration of D4T. The steady-state results should probably be given more weight. CSF penetration studies show that ddC, ddI and 3TC have low CSF penetration, comparable to methotrexate and cephalosporins [68]. ZDV and D4T have much higher CSF-plasma ratios. However, after chronic oral dosing all nucleosides reached comparable absolute CSF concentrations (expressed in μmol/1) in humans. Because it is believed that the amount of time in which the (CSF) drug concentration is above IC50 levels is important, these CSF concentrations should be compared with IC50 levels [137,154]. Because of large variability in published IC50 values, one should be cautious in the comparison of drug levels with IC50 values. 3TC and ZDV appear to have the most favourable CSF concentration-IC50 ratios. The data on protease inhibitors and non-nucleo-side reverse transcriptase inhibitors are still too sparse to draw firm conclusions. Indinavir and nevirapine appear promising as they show the best CSF concentration to IC50 ratio in humans and in experimental studies [85]. Preliminary findings on in vitro penetration showed the most favourable results for nevirapine and ritonavir [73,110]. Efficacy measured as a reduction in CSF viral load show promising results for ZDV monotherapy, 3TC-ZDV, 3TC-D4T, and other nucleoside-containing multiple drug regimens [8,12,21,35,36]. It has thus been demonstrated that it is possible to reduce the number of HIV RNA copies in CSF to undetectable levels with antiretroviral drug regimens that cross the CNS barriers. It is still not clearly established whether CSF viral load reflects viral load within the brain parenchyma during early stages of HIV infection, but CSF viral load was found to correlate with brain parenchymal virus load in patients with AIDS dementia [10]. Because viral load within the CSF and blood compartment often do not correlate, a lack of response within the CNS concomitantly with a good response in plasma (so-called ‘CNS escape’) may be possible, and has indeed been previously described [155]. Therefore, CSF sampling, even though an invasive procedure, may become important in routine follow-up of patients with undetectable plasma viral loads to establish whether a CNS response has also occurred. More data on this subject are urgently needed. Considering all the available information, at present we would recommend either 3TC-ZDV or 3TC-D4T-containing drug combinations for optimal treatment of HIV replication within the CNS. When more data on other drugs are available, these recommendations may be broadened to include other therapies. Acknowledgement The authors would like to express their appreciation to Delmar Molenaar for providing the illustrations in Fig. 1.