Drugs and human memory (part 2). Clinical, theoretical, and methodologic issues.

Abstract

THE skeleton of the memory experiment should have three phases: a study or acquisition phase, a retention interval, and a test or retrieval phase. Testing acquisition versus retrieval is a common experimental manipulation. For example, subjects might be required to learn one or more lists of words before drug administration and then asked to recall the material during the period of drug action (table 6). For most drugs, recall would not be impaired, even if the subjects seem to be very drowsy and sedated. In contrast, recall of word lists learned after drug administration would be greatly reduced.Another manipulation is to test for state-dependent memory or to control its effects. The most common design has been the 2 × 2 (table 7), in which subjects learn material in either a drug or a placebo state and later try to recall the information in either the same or the opposite state.82There would thus be four groups of subjects assigned to the following treatment conditions during acquisition and recall: drug–drug, drug–placebo, placebo–drug, and placebo–placebo. Symmetrical state-dependent memory would be demonstrated if the drug–drug and placebo–placebo groups recalled better than the drug–placebo and placebo–drug groups. Asymmetrical state-dependent memory would be demonstrated if the drug–placebo group recalled less than the drug–drug group. The subject is further complicated by the sensitivity of state-dependent effects to the type of memory tasks used. 117,118For drug studies, the definitive standard design is the randomized, prospective, concurrent assignment of subjects to the drug and placebo groups, under double-blind conditions, in which neither the subjects nor the researchers can determine which treatment is being used. Unfortunately, there are circumstances in which this strategy may not be feasible. It may not be possible to “blind” patients to some treatments that have recognizable effects, e.g. , treatment with general anesthetics. It may not be ethical to use a placebo group, e.g. , in surgical and invasive procedures that require a sham procedure, or if there is a risk of exacerbation of illness.Investigation of drug effects has one significant design advantage over many studies of cognitive impairments: the possible use of pretreatment and posttreatment comparisons . Premorbid assessment is usually not available when impairment is caused by trauma or disease. In studying the effects of drugs, however, it is possible to compare the behavior of the subject, both before and after administration of the drug, allowing unambiguous attribution of behavioral changes to the influence of the drug. A second fundamental design component is the use of a nondrug (placebo) control sample in which subjects receive identical treatment except for administration of the drug. Both design elements are essential. Pretreatment–posttreatment comparisons alone are inadequate because practice on experimental tasks, environmental influences, fatigue, and a host of other factors can change behavior over time and affect the comparison of performance before and after drug administration. Comparison of treatment and control groups alone is also inadequate unless it can be established that the groups are equivalent before treatment. Otherwise, an observed difference could have existed regardless of treatment or a true difference could have been masked by different baseline measurements between groups.Inclusion of a placebo control group is particularly important in assessing the influence of a drug on learning and performance. In several of our studies, we have noted little or no difference in performance between pretreatment and posttreatment with an active drug.95These failures to find significant differences might be incorrectly attributed to a lack of a treatment effect except that the control group performance showed marked improvement in the same task from pretreatment to posttreatment. For example, figure 10shows performance in learning sequences of 15 digits. Placebo subjects demonstrated an immediate improvement from their first test to their second, with no further improvement. For diazepam-treated subjects, the improvement was delayed, with greater delays for higher doses. In other words, the drug suppressed the usual performance improvement that occurs with repeated practice, thereby showing a reduction in new learning caused by the drug. Thus, a placebo-controlled design is essential to assess practice effects. Otherwise, drug effects may be confounded with practice effects.An “active” control group, e.g. , a group treated with a benzodiazepine when investigating a new potential amnesic agent, may also be included in the design. This would be advantageous when the sensitivity of the tests used has not been established, as a safeguard against false-negative results, and as a standard for comparison with the new drug–induced effects.Methodologic flaws are common in studies of the effects of drug abuse on cognition. The majority of the studies have not included measures of premorbid cognitive function, raising the possibility that differences between drug users and controls existed before the onset of drug use, rather than being caused by drug use. Some studies have not included a control group of nonusers. To convincingly demonstrate cognitive deficits in drug users, comparison with an appropriately matched control group is essential. Many studies have involved sample sizes too small to provide valid conclusions. These methodologic flaws can be avoided by measuring pre-morbid cognitive function, including a control group, and using a large sample size. To control for the possibility that drug abusers were poorer mentally and intellectually before starting their abuse, Block et al. 40pioneered matching drug users and nonusers on their previous scores during the fourth grade on the Iowa Test of Basic Skills achievement tests. 119Another methodologic constraint in studies of recreational drug users is the fact that most subjects use more than one drug. It is therefore important when investigating one specific drug to take a careful drug history and set strict maximal limits on the frequency and quantity of use of other drugs when recruiting subjects.One of the most elementary considerations in pharmacology is the relation between the size of the dose administered and the size of the measured behavioral response. However, the simple assumption that larger doses result in greater effects than smaller doses may not be true. For example, midrange doses of physostigmine exert positive effects on memory performance, whereas higher and lower doses impair it. 120,121Other “memory-enhancing” drugs, such as epinephrine and other endogenous stress hormones, may also show similar “inverted-U” dose–effect curves. 122–125This shape of the curve has been variously explained as being due to the high doses inducing hyperstimulation effects, producing state dependency, or facilitating learning of other interfering material. 126Other drugs may show a biphasic action on behavior, with small doses improving and larger doses impairing behavior. 127Tolerance has been defined as a shortened duration and decreased intensity of drug effects after repeated administration. Short-term tolerance to psychoactive drugs may develop within the time course of a single dose. Behavioral impairment may recover toward baseline levels while the plasma concentrations of the drug remain relatively high. This has been demonstrated for many drugs, including barbiturates, benzodiazepines, caffeine, and cocaine. 128–130The rapid distribution of a drug in and out of the brain may produce the same effects as short-term tolerance. Experiments with steady state blood concentrations may be needed to distinguish between distribution effects and short-term tolerance.With repeated administration, long-term tolerance to the behavioral effects of psychoactive drugs can develop. 131,132The opposite effect to tolerance has occasionally been reported. Repeated administration of cocaine may produce sensitization or heightened responses. 133,134The relative ease of measuring a psychotropic drug concentration in blood (or other body fluids) compared with objective dynamic measurements of memory or other central nervous system (CNS) effects has led many to assume that blood concentrations are synonymous or linearly related to drug effects, which may not be true. If a continuous or repeatable discrete measure of a drug effect can be obtained with concurrent measurement of drug blood concentrations, it is possible to develop pharmacokinetic–pharmacodynamic (PK-PD) modeling concepts to characterize relevant parameters that quantify drug effects. 135There are several advantages for studying PK-PD relations 136,137: (1) It allows more complete understanding of the determinants of drug action, including phenomena such as distributional delay of effect, formation of active metabolites, and short-term tolerance. (2) It quantitates the effects of the drug on the brain by calculating values for parameters such as Cp50AMNand Cp50SED, which represent the plasma drug concentrations required to produce one half of maximal amnesia and sedation. 138As valid measures of intrinsic drug potency and brain sensitivity within an individual, those parameters allow exploration of the psychotropic differences between drugs and explanations of effects of factors such as aging and drug–drug and drug–disease interactions on the drugs’ actions. (3) The information would make it possible to design optimal infusion schemes for drugs during conscious sedation and anesthesia or during investigations of their behavioral effects. (4) It provides a rationale for monitoring drug plasma concentrations as indicators of clinical efficacy or toxicity and use for medicolegal purposes.Several steps are involved in studying the PK-PD relation and evaluating drug action: (1) Pharmacokinetics describe and predict the time course of concentrations in body fluids, usually blood (fig. 11). Arterial blood sampling allows for the calculation of accurate data during drug distribution and the rate of blood–brain equilibration. 139It is the preferred site because most of the studies in the literature evaluate the effects of single bolus doses or relatively short infusions and are performed during the distribution/redistribution phase. The issue of plasma protein binding is also of importance 140,141because the unbound (free) drug in plasma is presumed to represent the drug fraction that is available for transport across the blood–brain barrier. (2) Pharmacodynamics describes the time course and intensity of drug effects (fig. 11). This is the difficult step and is the reason for the deficiency of adequate studies of the pharmacokinetic–amnesic relation for drugs. The behavioral tests must be short, amenable to frequent repetitions, and sensitive to low drug doses and concentrations. The brevity of the tests reduces subjects’ fatigue, and the test sensitivity allows determination of memory function over a wide range of drug concentrations. We developed in our laboratory the use of a 15-digit number serial learning task, repeated over three trials for such studies.95The task is sensitive, short (approximately 3.5 min), and can be administered as frequently as desired to correspond to changing drug concentrations. The task may also be administered several times before the actual study to reduce improvement in performance over time. There is virtually no limit to the numbers that can be generated by a computer, unlike words or pictorial lists. To compensate for any residual practice effects, one may use a placebo correction. Changes over baseline scores after administration of active medication are corrected by subtraction of scores at corresponding times after placebo administration. (3) PK-PD modeling describes the relation between the dose (concentration) and its effects. Data should be obtained from repeated and simultaneous sampling over a wide range of drug concentrations. A mathematical model is developed that fits the data and allows inference of the effect site concentrations based on plasma concentrations. Various PK-PD models may be used. 135–137,142The most appealing is the sigmoid Emaxmodel, because of its similarity to the receptor binding model. Interpretation of the concentration–effect relation can be complicated by the lack of a temporal relation between the two variables, so-called hysteresis. Two types of cognition–blood drug concentration curves may be found (fig. 11). The drug effect may decrease with time for the same drug concentration, described as clockwise hysteresis as shown by the arrows in figure 11. This may be caused by tolerance (short-or long-term), progressive learning of the task, and the presence of active antagonistic metabolites. 143,144It is not possible to separate tolerance from learning without a placebo control. The formation of active antagonistic metabolites is rare, but there are a few examples of metabo-lites that alter the dynamics of the parent drug by modifying its kinetics, e.g. , 5-hydroxy-pentobarbital. 144The presence of clockwise hysteresis has some important practical applications. Medicolegally, blood concentrations may not adequately predict impairments from these drugs.Another type of drug concentration–effect curve can demonstrate anticlockwise hysteresis. The effect of the drug increases with time for a given drug concentration, which, when taken sequentially, produces a direction that is counterclockwise. A common cause is the delay for a drug to be transported from the systemic circulation (sampling site) to its site of action and then to elicit a measurable response. This type of hysteresis may be missed because of infrequent early sampling and assay of the drug in venous rather than arterial blood. 139,142Another cause is the production of active metabolites from the parent drug. These would have maximum concentrations and a combined peak activity at some later time compared with the parent drug concentration. 145Other uncommon causes are delayed drug action, drugs working through a cascade reaction, and short-term sensitization or up-regulation of receptors.The applicability of mathematical models to describe the pharmacodynamic response becomes questionable when hysteresis occurs. The hysteresis must be collapsed or removed. One frequently used approach assumes an effect compartment 146to correlate memory changes with changes in the blood concentrations of drug. It can be thought of as the kinetically defined biophase of the CNS actions of the drug. The drug effect is directly related to its concentration at the receptor site. A link model 142describes the transfer between the plasma and effect compartments. The equilibration delay between the compartments is characterized by the rate constant ke0with units of reciprocal time, which governs the transfer of drug.All of the drugs currently available for human use that are capable of producing amnesia also cause sedation. There is no drug that only affects memory. For theoretical and clinical reasons, it is important to separate the effects on memory systems from impairments in attention, arousal, or mood. It is also important when investigating potential memory-enhancing drugs to separate effects on alertness, attention, and fatigue from genuine effects on learning and memory. The general consensus is that drug-induced amnesia is independent of sedation. Table 8summarizes the approaches that have been used to dissociate the effects on memory and sedation. One method is to study two or more drugs that produce the same effects on sedation but different effects on memory. For example, Green et al. 147compared chlorpromazine with lorazepam in doses that produced equal degrees of sedation but found that memory was impaired only by lorazepam. Curran et al. 148compared the effects of diphenhydramine with those of scopolamine and lorazepam. In the doses used, the three drugs produced similar levels of sedation, but the antihistamine did not impair memory. It should be noted, however, that because tests of sedation and memory may vary in difficulty, dissociations of this kind do not provide compelling evidence for independence between the two behaviors.Another method of demonstrating the specificity of the memory effects of drugs is to study the rates of development of tolerance to the actions of the drug. Overall, the evidence is that tolerance develops to sedative effects much faster than it develops to memory effects. For example, tolerance develops to the sedative effects of diazepam after its 3-week administration to healthy volunteers but not to its amnesic effects. 149Tolerance develops to the memory effects of alprazolam after 8 weeks of treatment in patients 150and at least 6 months after treatment with other benzodiazepines. 151,152An alternative way of dissociating the two effects would be to show differential reversal of amnesic and sedative effects by an antagonist. Use of small doses of flumazenil 153or pretreatment with flumazenil before administration of a benzodiazepine 154results in reduction of sedative effects without relief of memory impairment. A fourth method of dissociation is through demonstration of different dose–response curves for sedation and amnesia. 155–157Using the auditory event-related potential with different groups of drugs that produced equivalent sedative but differing amnesic effects, Curran et al. 148and Veselis et al. 158(also more recently, Veselis RA, Reinsel RA, Feshchenko AV, Johnson R: Thiopental and propofol effects on memory are dissociable by event related potentials. Poster presented at the 50th Annual Meeting of the Association of University Anesthesiologists, Milwaukee, Wisconsin, May 1–3, 2003) reported that early components of the event-related potential were affected similarly by sedatives, whereas later components were affected more by amnesics. Statistical methods are also important for showing this dissociation. Analysis of covariance can be used to separate effects attributable to sedation. However, covariance assumes a linear relation between variate and covariate, and the relation between memory and sedation maybe more complex than that. 159,160More recently, Veselis et al. 138used several statistical scaling procedures, normalization of drug concentration levels, and arbitrary standards of memory and sedation to compare memory performance after equisedative doses of four drugs (midazolam, propofol, thiopental, and fentanyl). These drugs exhibited very different sedation and amnesia relations for the same criteria of felt sedation and objective memory impairment. For example, propofol at low serum concentrations showed a high likelihood of exceeding the criterion of memory impairment well before it met the criterion of sedation. In contrast, fentanyl exceeded the sedation criteria and showed low probability of amnesia for the same concentration range (fig. 12). Finally, Eger’s group has demonstrated chemical compounds that suppress learning without causing sedation in animals 161–163and shown that the two functions need not be inseparable.Functional neuroimaging opens a window to view the brain at work. It provides a unique in vivo opportunity to study the neurobiology of human memory and its functional and neural architecture. It is also a rapidly developing, highly interdisciplinary and complex technical field, requiring multidisciplinary teams of scientists (in physics, radiologic science, mathematics, statistics, computer programming, engineering, cognitive neuroscience, and medicine). 164Brain imaging has been used relatively recently to investigate several areas of memory, including the nature and function of components of the memory systems and regional cerebral blood flow changes associated with performance of memory tasks under the influence of drugs. 165–169Many new insights have been gained, and these in turn promise a deeper understanding of the foundations of memory.The two major techniques are positron emission tomography (PET) and functional magnetic resonance imaging (fMRI). Both measure neuronal activity by assessing changes in local cerebral blood flow. For the PET method, a radioactive tracer is injected immediately before the start of a cognitive task. The radiotracer accumulates in the brain in direct proportion to the local blood flow. For the most widely used fMRI method, called BOLD (blood oxygen level dependent), images are generated through changes in blood oxygenation that accompany neuronal activity without the need for a radioactive tracer. When neural activity increases, local blood flow and oxygen consumption increase, but the former increases more than the latter, resulting in a local increase in the amount of oxygenated blood and a net decrease in deoxyhemoglobin. Deoxyhemoglobin is paramagnetic, resulting in local magnetic field changes that provide the imaging contrast. 170At least two issues need to be considered when planning neuroimaging studies, as discussed here.If the researcher wants to construct, for example, an episodic memory retrieval task in which the subject recalls orally during scanning the words of a list learned earlier, changes in blood flow should be the result of memory retrieval and not due to other mental activities. A common strategy is to use a paired image subtraction design. In addition to the scan during the word list recall, another scan is taken during a control condition that shares the same mental operations except for those of explicit retrieval. For example, we asked the subjects in our laboratory 171to repeatedly count “1, 2, 3, . . .” aloud at a rate of approximately 1 number/s, which is expected to match the rate of verbal output during the memory test. This repetitious rehearsal in short-term memory of a vastly over-learned and automatized sequence should minimize episodic memory retrieval. Subtracting the blood flow maps during the control state, which accounts for speech activity, from those during the activation state would identify the regions that are involved in the desired memory task. This subtraction method has been criticized. There is no guarantee that the performance in the experimental task will differ from the control state in only one way. Also, the addition of the extra processing component per se in the experimental task may affect processes common to the experimental and control tasks. If so, it would not be possible to subtract them out. 172Nonetheless, the majority of results from studies of memory have been generated by this method, and robust and reliable patterns of activation have been demonstrated. 173Some researchers also use a resting state as a baseline. Subjects lie quietly without specific instructions regarding mental activities. Critics argue that the variability in the mental state during such a condition is such that it may not serve a useful purpose. 164In our laboratory, we ask the subjects immediately after the period to describe what they had been thinking to discern differences in mental states between subjects in the experimental and control groups. 174The characteristics of the stimulus, its mode of delivery, its timing, and its timing and duration in relation to the scanning periods must be precisely controlled. 175A widely used PET radiotracer is oxygen-15–labeled water (H215O), which has a half-life of approximately 2 min, allowing a series of injections to be performed every 12–15 min. For each injection, the cognitive task and scanning are performed during the time that the labeled blood perfuses the brain. It provides a 40-s window on brain activity, with a spatial resolution of approximately 6–10 mm. The advantages of PET include relatively silent scanning, accessibility of the patient for monitoring, and the ability to provide quantitative as well as relative measures of blood flow. The latter is important in studies with drugs that may affect global cerebral blood flow, either directly or indirectly, e.g. , via changes in arterial carbon dioxide tension (Pa CO2). The advantages of fMRI compared with PET include the avoidance of exposure of subjects to ionizing radiation and improved spatial and temporal resolution. Its limitations are confining the subject inside the scanner, with its risks of limited monitoring and claustrophobia in some individuals, acoustic noise, and signal artifact at the base of the brain. 164Both PET and fMRI have high spatial but poor temporal resolution. Conversely, electroencephalography, event-related potentials, and magnetoencephalography rapidly measure the current flows induced by synaptic activity. Electroencephalography and event-related potentials quantify electric potentials with electrodes at the scalp. Magnetoencephalography is a newer technique in which the magnetic fields associated with current flow within neurons induce a current in a detection coil on the scalp. To pick up these small signals, the detection coils are coupled to a superconductive device within a magnetically shielded room. 164However, the accurate localization of neuronal current flows based on data generated by these methods alone is problematic. Recently, techniques have been developed that use both hemodynamic and electromagnetic measures to arrive at estimates of brain activation with high spatial and temporal resolutions. These methods range from simple juxtaposition to simultaneous integrated techniques. 176Images are reconstructed before statistical analysis. They are corrected for sources of noise in the signal due to scanner drift or artifacts, are realigned to correct for slight head movement, and may undergo spatial smoothing. 164Usually the subject’s functional results are displayed on his/her own structural magnetic resonance imaging scan; otherwise, images are transformed to a stereotactic coordinate space, based on a common template. 177This is done to counteract individual differences in brain size and gyral anatomy and facilitates group analyses, as well as the communication of results across laboratories. Typically, the comparison of blood flow maps associated with the cognitive task and its control is performed using a t test, regression, or multivariate statistical approaches. 178,179Tulving et al. 180proposed the hemispheric encoding/ retrieval asymmetry model. According to this model, the prefrontal regions in the left hemisphere tend to be differentially activated during episodic encoding and semantic retrieval, whereas the right prefrontal regions tend to be differentially involved during episodic memory retrieval (fig. 13). Considerable evidence supports this model, 181although some critics have argued that this hemispheric asymmetry seems to depend to some extent on the type of stimuli used. 182,183The latest version of the model acknowledges that the right pre-frontal lateralization of episodic retrieval seems less complete than originally proposed. 184A second general observation of the neuroimaging literature is that prefrontal regions seem to interact with posterior brain regions during memory encoding and retrieval. 173Episodic encoding usually involves activation of the left prefrontal, left temporal, and anterior cingulate regions. The left hippocampus is usually involved with verbal material, and the right hippocampus is involved with nonverbal materials. 185–187There are two functional neuroimaging studies that demonstrate that activation of the amygdala at encoding is correlated with later recall of emotional material. 188,189Episodic retrieval usually activates the right prefrontal region, the anterior cingulate region, the cerebellum, and the hippocampus. 190,191Semantic retrieval is usually associated with activation of the left prefrontal, left temporal, and anterior cingulate regions. 190,192For working memory , the central executive is typically associated with activation of prefrontal regions, the phonologic loop is associated with the parietal regions (for storage) and the Broca area (for rehearsal), and the visuospatial sketch pad is associated with the occipitotemporal, occipitoparietal, inferior prefrontal, and superior prefrontal regions. Object maintenance tends to be left lateralized, and spatial maintenance tends to be to be right lateralized. 193,194Priming is accompanied by reductions in the amount of neural activation relative to naive or baseline task performance (fig. 14). Decreased activation bilaterally in occipitotemporal cortical areas is usually associated with perceptual priming, and the left inferior frontal cortex is usually associated with conceptual priming. 191,195,196Last, aversive conditioning is associated with activation of the amygdala. 197,198Table 9summarizes these results. It should be emphasized, however, that there are discrepancies and uncertainties about precise anatomic localization of various memory processes. For example, in a review of verbal working memory by Ivry and Fiez, 199Broca area activation was found in only 9 of 12 data sets by different groups of investigators. Neuroimaging is a “noisy” technique, and results obtained in one study may not be replicated in a second. Assumptions that the cognitive tasks used in different studies evaluate the same memory processes may not be certain, and teasing apart the different operations involved in complex mental functions is far from easy.The standard subtraction approach to analyzing functional neuroimaging data can be used to identify the brain regions active in certain tasks. However, it does not indicate the functional interrelations between such regions and regions that do not show differential activity but may still be part of the specific functional network. The network approach complements the subtraction approach in characterizing the functionally specialized brain regions and their interactions. 200–202Several