We present a novel PET imaging approach for capturing dopamine fluctuations induced by cigarette smoking. Subjects smoke in the PET scanner. Dynamic PET images are modeled voxel-by-voxel in time by lp-ntPET, which includes a time-varying dopamine term. The results are ‘movies’ of dopamine fluctuations in the striatum during smoking.
We describe experimental and statistical steps for creating dopamine movies of the brain from dynamic PET data. The movies represent minute-to-minute fluctuations of dopamine induced by smoking a cigarette. The smoker is imaged during a natural smoking experience while other possible confounding effects (such as head motion, expectation, novelty, or aversion to smoking repeatedly) are minimized.
We present the details of our unique analysis. Conventional methods for PET analysis estimate time-invariant kinetic model parameters which cannot capture short-term fluctuations in neurotransmitter release. Our analysis – yielding a dopamine movie – is based on our work with kinetic models and other decomposition techniques that allow for time-varying parameters 1-7. This aspect of the analysis – temporal-variation – is key to our work. Because our model is also linear in parameters, it is practical, computationally, to apply at the voxel level. The analysis technique is comprised of five main steps: pre-processing, modeling, statistical comparison, masking and visualization. Preprocessing is applied to the PET data with a unique ‘HYPR’ spatial filter 8 that reduces spatial noise but preserves critical temporal information. Modeling identifies the time-varying function that best describes the dopamine effect on 11C-raclopride uptake. The statistical step compares the fit of our (lp-ntPET) model 7 to a conventional model 9. Masking restricts treatment to those voxels best described by the new model. Visualization maps the dopamine function at each voxel to a color scale and produces a dopamine movie. Interim results and sample dopamine movies of cigarette smoking are presented.
Despite overwhelming evidence of the medical risks, tobacco smoking is still a major health problem. It is simply very hard to quit smoking. Over 20% of the adult U.S. population continues to smoke and most smokers who attempt to quit relapse within the first month 10. Unfortunately, there are few available treatments to aid in smoking cessation and/or reduce nicotine dependence. In our lab, we are interested in using PET imaging to understand addiction and dependence in order to aid in the development of new medications for smoking cessation and other drug-taking.
The rapid rise of dopamine in the striatum is believed to encode the addictive liability of drugs and behaviors 11 and the rapid return of dopamine to baseline may be related to withdrawal and subsequent drug-seeking. For some addictive substances and behaviors like cigarette smoking, the elevation of striatal dopamine is very short-lived (minutes); the magnitude of the rise is not large (1-2X baseline); and the spatial extent of these responses may be limited to small sub-regions of the striatum.
Animal experiments clearly demonstrate that nicotine causes dopamine release in the nucleus accumbens of rats 12. But early attempts -using conventional analyses – to estimate dopamine changes in humans during or following nicotine or smoking have yielded unreliable and contradictory results 13-18. Some of these studies allowed smokers to smoke outside the scanner. Others delivered only nicotine to the subject. To best study addiction to cigarettes, we set out to develop better imaging protocols and complement them with advanced analyses that would allow us to capture the brain’s response to a quasi-natural smoking behavior.
Positron Emission Tomography (PET) is unique among brain scanning techniques in its ability to probe the neurochemistry of the human brain in vivo. Many PET tracers exist to track dopamine receptors and many are susceptible to competition with endogenous dopamine. Unfortunately, the conventional methods of PET image analysis estimate the steady state ratio of bound to free tracer, known as binding potential (analogous to in vitro methods), from dynamic PET images. An apparent change in the steady state ratio (e.g. from the baseline to the smoking condition) is taken to indicate dopamine change. But the dopamine changes relevant to addiction are inherently transient so estimates of a steady-state quantity are flawed. Furthermore, the typical region-of-interest analysis averages the tracer concentration over large anatomically-defined regions and is likely to miss highly localized brain responses – such as those we expect from cigarette smoking. Previous PET studies of smoking may have also suffered from movement of the smokers’ heads during smoking in the scanner.
Functional MRI (fMRI) offers the necessary spatial and temporal resolution that would be needed to capture events occurring in sub-regions of the striatum on the minute time-scale but fMRI lacks the molecular specificity of PET. The BOLD signal derives from changes in blood flow and is therefore neuronally and molecularly nonspecific. Thus, we utilized PET – but in a new way. The goal of this protocol was to estimate the brief and localized dopamine responses to smoking because they are believed to underlie the neurochemical manifestation of craving and drug-seeking behavior.
To estimate dopamine transients that are captured in dynamic PET images made with dopamine-receptor ligands, we previously introduced a series of kinetic models, collectively referred to as “ntPET” for neurotransmitter PET 1,5,6,19, that were based on the conventional two-tissue compartment model but were augmented by terms for the time-variation in dopamine and the interaction between dopamine and the tracer (i.e. competition). These models have been validated against a gold standard. Specifically, we have demonstrated that our models predict dopamine concentrations over time from PET data in rats that are in good agreement with simultaneously acquired microdialysis measurements 4,7. Advantages: The most recent of our models have been either linear and non-parametric (np-ntPET) 1 or linear and parametric (lp-ntPET)7. The latter model derives from an earlier linear model introduced by Alpert et al. 20. Linearization is a key development because it assures that applying the models to dynamic data at the voxel level is computationally simple. In a recent proof-of-concept paper, we were able to create dopamine movies of a human subject performing a motor task 3 and show that the movies were sensitive to the timing of the motor task as would be expected. Movies are representations of the time course of dopamine levels at every voxel in an image. Voxel-by-voxel methods in PET generally suffer from low signal to noise ratio, so to minimize the noise inherent in voxel-based time-activity curves (TACs), we apply an innovative spatial filter, ‘HYPR’, 8 as a pre-processing step. This step preserves key temporal characteristics of the responding voxels while reducing noise.
Smoking is more than nicotine delivery. Cigarettes contain 4,000 chemicals in addition to nicotine. While nicotine is thought to be primarily responsible for the initial addictive effects, all the other cues and sensory components of smoking become reinforcing to a habitual smoker. We chose to study the entire behavior of smoking which meant that we needed to be able to image smokers smoking while inside the PET scanner. Unfortunately, with smoking comes head motion. To eliminate head motion artifacts in our images, we use the Vicra motion-tracking system (NDI Systems, Waterloo, Canada) and event-by-event motion correction as part of an iterative, resolution recovery reconstruction algorithm 21.
Our new scanning and analysis methods are designed to elicit and capture brief and localized dopamine transients that are the unique signatures of the brain’s response to addictive drugs and behaviors. Performed voxel-by-voxel, our models produce a dynamic set of images of striatal dopamine fluctuations – i.e. “dopamine movies”. These movies represent a new spatio-temporal biomarker of addiction and could serve as a direct, multi-dimensional indicator of risk for addiction and/or indicator of treatment efficacy.
An outline of the entire procedure, described below, for producing multi-slice dopamine movies is summarized in the flow chart in Figure 1.
Figure 2. The effects of two different HYPR spatial filters on the smoothness of the time-activity data at a single striatal voxel. Top row: 11C-raclopride PET emission images from a 3 minute frame centered at 46.5 minutes (not filtered, filtered with a 3 x 3 x 3 voxel kernel, filtered by a 5 x 5 x 5 voxel kernel). Middle row: 11C-raclopride PET emission images from a 3 minute frame centered at 61.5 minutes (not filtered, filtered with a 3 x 3 x 3 voxel kernel, filtered by a 5 x 5 x 5 voxel kernel). Bottom row: Corresponding time activity curves from the same single voxel location in the left dorsal caudate. Note that the apparent dip in the 11C-raclopride uptake (due to release of dopamine) at the time of smoking is preserved although the noise is diminished with greater filter size.
Figure 3. A selection of representative dopamine response functions that were pre-computed for fitting the lp-ntPET model to the PET time-activity data at each voxel according to Normandin et al. 7. In the case of our smoking paradigm, cigarette smoking commences 45 minutes after tracer injection begins. Even if striatal dopamine responses encode anticipation of smoking – e.g. due to handling of the cigarette or other cues that foretell smoking – we reasoned that the response functions could safely be limited to curves that take off from baseline no earlier than 5 minutes prior to smoking (a). Similarly, curves were limited to take-off times no later than 15 minutes after start of smoking. Curves with take-off times at 40 minutes represent possible dopaminergic responses due to expectation. (b) Representative response functions all taking off from baseline at 45 minutes; the time when smoking commences. 500 different plausible response functions are generated. For illustration, plots in (a) and (b) show only a sampling of curve shapes and take-off times.
Figure 4. (a) The operational equation for the lp-ntPET model. The model is linear in parameters (R1, k2, k2a, γ) which allows fast computation of parameter estimates at each voxel within the striatal mask. (b) Parametric images of (R1, k2, k2a, γ) for a single coronal brain slice for a single subject. Although γ alone is the parameter that encodes the magnitude of a dopamine response, simultaneous estimation of all 4 tracer parameters is necessary to fit the model to the time-activity data at each voxel.
Figure 5. Fits of the conventional (MRTM) and new (lp-ntPET) models to the time-activity data from a voxel in the left caudate. MRTM fit is in blue. lp-ntPET fit is in red.
Figure 6. (a) shows the weighted sum of squared residuals (WSSR) from MRTM and (b) from lp-ntPET fits to the data at each striatal voxel. The two WSSR images produced from the same data are compared to produce a map of the F-ratio at each voxel (i.e. an F-map), shown in (c). (d) The F-map is thresholded at p < 0.05 to produce a binary significance map (see step 2.10 of the protocol). For n timeframes, and 4 parameters of the lp-ntPET model, the threshold for the F-statistic corresponding to a probability level, p < 0.05 (for 90 minutes of data binned in 3-minute frames, the threshold is 4.23) (e) The significance map is filtered with a morphological filter (an “opening”) to eliminate tiny clusters of voxels that are most likely to represent noise. The Final Significance Mask preserves only those voxels in the striatum whose TACs are better fit (statistically) by the lp-ntPET model as opposed to the conventional MRTM model and thus are believed to contain a dopaminergic response to smoking. This threshold does not correct for multiple comparisons. Instead, to guard against false positive findings, we create Final Significance Masks for a control condition as well (see Figure 7 and Protocol Steps 1.8 – 1.10).
Figure 7. (a) shows one coronal slice of the Final Significance Mask for the smoking condition in a single subject. Figure (b) shows the Final Significance Mask for corresponding subject and slice in the baseline condition. The presence of clusters of retained voxels in the mask of smoking as opposed to the near-complete absence of clusters in the mask of control supports the contention that the dopamine movies (see below) are not simply chance events or events related to noise in the data. (Note: the injected activity – and thus the signal to noise ratio – in the baseline and smoking conditions were comparable.)
Figure 8. The dopamine movie of a single slice of brain in the coronal orientation shows the frame-by-frame dopamine level relative to the basal (resting) dopamine level. (a) shows the movie of the baseline condition and (b) shows the movie of the smoking condition. The dopamine levels are encoded in color. Specifically, the colors – shown in the color bar with corresponding numerical values – represent the change in dopamine above the basal level as a percent of basal. Again, the dopamine levels are shown only for voxels in the Final Significance Mask that exceed the p < 0.05 significance level.
Figure 9. The multi-slice, multi-condition dopamine movie for the same subject as in Figure 8 with all slices of the ventral striatum displayed simultaneously for baseline and smoking conditions.
Figure 1. Flowchart of experiment and image analysis procedures (a-c). Click here to view larger figure.
Figure 2. Effects of HYPR filters of different kernel size on images (top and middle) and on time activity curves (bottom) at a single voxel. Click here to view larger figure.
Figure 3. Examples of dopamine response functions that take-off at (a) 40 min or (b) 45 min post-tracer injection.
Figure 4. Parametric images generated by fitting the lp-ntPET operational equation (a) to the PET data. (b) Images corresponding to the 4 parameters of the model, R1, k2, k2a, γ, are evaluated for the striatum and shown overlaid on the corresponding MR slice. Click here to view larger figure.
Figure 5. Fits of MRTM (blue) and lp-ntPET (red) models to time-activity data from a single voxel.
Figure 6. Parametric images of the WSSR for (a) MRTM and (b) lp-ntPET. The respective WSSR maps are compared to create the F-map (c), which in turn is thresholded to a binary mask (d) and then filtered to produce the Final Significance Mask. Click here to view larger figure.
Figure 7. Comparison of Final Signifcance Masks for smoking (a) and control (b) conditions for the same subject.
Figure 8. Single-slice dopamine movie for single subject in control (‘Rest’) and smoking conditions. Click here to view Figure 8.
Figure 9. Multi-slice dopamine movie for single subject in (top) smoking and (bottom) control (‘Rest’) conditions. Click here to view Figure 9.
Findings in the PET literature on the dopamine response to smoking are inconsistent 13-18. There may be many reasons for this. Various methodological difficulties arise with any attempt to image cigarette smoking. At the very least, one must contend with possible motion artifacts in the data, second-hand smoke exposure for researchers, modest and short-lived changes in dopamine that cause only subtle alterations to the uptake and retention of the tracer, 11C-raclopride.
Artificial induction of a large and sustained response of dopamine might be possible by administering an IV injection of a large dose of nicotine. However, this would be contrary to our underlying aims in creating dopamine movies of cigarette smoking. Our intent was to examine as carefully as possible the dopaminergic response to the entire behavior of smoking. In addiction research, an important distinction is made between passive administration of drugs to a subject and self-administration. Our aim was to image self-administration – a smoker smoking his/her own preferred brand of cigarette – in order to capture and characterize the brief dopaminergic response to smoking. PET analyses typically assume that the effects of a drug or other challenge are long-lived relative to the scan duration. Imaging smoking thus required innovations in modeling and in experimentation with PET.
Critical Steps in our protocol
Facilitating smoking in the scanner
Maximizing sensitivity of 11C-raclopride uptake to small changes in dopamine levels
Limitations to the interpretation of the presented result
Future
We have developed a new model of PET tracer uptake in the presence of a short term fluctuation in endogenous neurotransmitter level. Because the model is linear in parameters, it can be computed quickly and easily at many voxels. The endpoint of fitting such a model to the PET data on a voxel-by-voxel basis is a "movie". For studies with the D2 receptor tracer, 11C-raclopride, the endpoint is a dopamine movie. Dopamine is the key neurotransmitter involved in the brain's processing of rewarding stimuli that leads to addiction. Because some stimuli (most notably cigarettes and alcohol) produce only mild and probably short-lived dopamine changes, the movies may have their greatest potential for studying the abuse of these two stimuli. If we can use our dopamine movies to identify spatial and temporal patterns of dopamine release that are indicative of dependence or risk of abuse, then these patterns could serve as markers of disease, risk of disease, and – assuming the patterns are reversible – indicators of (pharmacological- or cognitive-) treatment efficacy.
There is nothing about our movies that restricts them to the dopamine system. All that is needed is a PET tracer for a target of interest that is sensitive to (i.e. easily displaceable by) fluctuations in the endogenous ligand for the same target. To date, there has been halting progress to identify PET tracers that are reliably sensitive to endogenous neurotransmitters other than dopamine. A review of the serotonin literature in 2010, for example, painted sobering picture of our current limited ability to detect serotonin release with PET 32. Recently, there have been some encouraging developments. A number of publications have reported sensitivity of serotonin tracers to elevations in endogenous serotonin in non-human primates 33-36 but the field awaits similar demonstrations in humans. As we have discussed elsewhere 37, sensitivity to changes in endogenous neurotransmitter concentration appear to be comprised of an optimal rate of displacement from the receptor combined with an ease of efflux of the tracer from tissue to blood. Once serotonin ligands have been validated and shown to have such properties, then serotonin movies will also be possible.
Presently, most PET studies with receptor-tracers lead to the generation of parametric images. A parametric image is a map of a given tracer kinetic model parameter evaluated at every voxel in the object (i.e. the brain). Application of conventional models such as SRTM 38,39 or the one- or two-tissue compartment model yields parametric images of Ri, the regional flow parameter, or BP, the regional binding potential value. Both of these parameters are physiological constants that are believed to represent processes that are in steady state. Sometimes, however, the system and/or the process of interested are unsteady. That is, they are transient. Such is the case with the short-lived response of dopamine to cigarette smoking. In such circumstances, it is not possible to characterize the dopamine transient with a single parametric image. Nor is it appropriate to model the data with a model that is strictly time-invariant in parameters. There is a need for a model with a time-varying term to describe dopamine concentration changes in the striatum in response to smoking. The natural output of such a model when used with a dopamine tracer, is a movie of dopamine. This is a new form of functional image output that likely will spur and require new forms of analysis to maximize its utility.
The authors have nothing to disclose.
The authors thank the members of the Yale PET Center chemistry team for tracer synthesis, the imaging team for tracer injection and image acquisition and Ms. Sheila Huang for expert flow-chart design.
Much of the development of the ntPET techniques was supported by R21 AA15077 to E. Morris. K. Cosgrove is supported by K02 DA031750.
Name of Reagent/Material | Company | Catalog Number | Comments |
Vicra | NDI Systems, Waterloo, Canada | ||
HRRT | Siemens | ||
Air Filter | Movex, Inc, Northampton, PA | LFK 175 | With extractor and clear hood |
11C-raclopride | prepared at Yale PET Center from O-Desmethyl precursor | ||
O-Desmethylraclopride | ABX advanced biochemical compounds, Radeberg, Germany | Product #1510 | Precursor of 11C-raclopride |
Table 1. Materials used. |