Optimizing Experimental Protocols for Quantitative Behavioral Imaging with ¹⁸F-FDG in Rodents

MATERIALS AND METHODS

To optimize a quantitative behavioral imaging protocol in which ¹⁸F-FDG uptake occurs outside the tomograph and the animal is subsequently anesthetized and scanned, one group of animals served as a fully quantitative reference group. This group received an intravenous ¹⁸F-FDG injection and fully dynamic list-mode data were obtained, along with a complete input function. This same dataset was also analyzed as though the animals were outside the tomograph for the first 50−55 min of uptake, and the last 30 min of the dynamic scan (from 60−90 min) were isolated and analyzed for the SUV using input from several points on the arterial blood curve. These data were then compared with the fully quantitative metabolic rates obtained from the same studies. Identical experiments and the same quantitative strategies (SUV calculated from several different arterial blood samples and full kinetic analysis) were performed on another group of animals after intraperitoneal ¹⁸F-FDG administration. A third group of animals received intraperitoneal ¹⁸F-FDG outside the tomograph and a full arterial blood curve was obtained. These animals were anesthetized after 60 min of uptake and, again, the data were quantified using several points from the arterial curve as input. These studies were designed to investigate the applicability of intraperitoneal routes of radiotracer administration and also to investigate whether the SUV would be proportional to the metabolic rate of glucose (MRGlu) after intraperitoneal or intravenous ¹⁸F-FDG injections.

Ten 7- to 8-wk-old male Sprague–Dawley rats (mean ± SEM, 298 ± 11 g body weight) with preimplanted carotid artery catheters (polyurethane tubing, 0.64-mm inner diameter, 1.02-mm outer diameter, 10-cm length) were purchased from Taconic Farms. The catheters were implanted into the right carotid artery and subcutaneously tunneled to the back of the animals between the shoulder blades. All animals were examined once and fasted for at least 2 h before the start of the study. Animals were divided into 3 groups (Table 1): the first group was anesthetized and received intravenous injections of ¹⁸F-FDG during a dynamic acquisition (group 1, n = 4). The second group (group 2, n = 6) was identical to the first group, except that these animals received an intraperitoneal injection of ¹⁸F-FDG. The third group of animals received intraperitoneal ¹⁸F-FDG and were awake for a 60-min uptake period, followed by ketamine/xylazine anesthesia and a 30-min static acquisition (group 3, n = 4). In all 3 groups, full arterial blood curves were obtained from the tracer injection until the end of the scan. Data from the dynamic studies (groups 1 and 2) were quantified in 3 separate ways: (i) using the full blood curve and dynamic data to calculate the MRGlu; (ii) using an image-derived input function obtained from an ROI over the left ventricle of the heart, which was also in the field of view (Fig. 1); and (iii) using blood sampled from single time points to calculate 4 different SUVs using several points from the input function curve. These values were directly compared with fully quantitative MRGlu values using a correlation analysis. Data from the awake studies (group 3) were quantified with the SUV, using the same time points as groups 1 and 2.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1. 

Strategy for comparing image-derived or measured ¹⁸F-FDG input function on estimates of MRGlu. (A) Radioactivity distribution of ¹⁸F-FDG is shown in sagittal slice from animal with a carotid artery catheter. Both the brain (†) and the heart (*) are in the field of view. The time–activity curve of ¹⁸F measured from an ROI over heart is presented with that sampled from carotid artery (B), and time–activity curve of cortical brain ¹⁸F-FDG uptake is given in C. Inset in B depicts expanded view of early time points illustrating the temporal resolution of scanner vs. manual measurements. This difference can be attributed to difficulty in rapidly sampling the blood in initial seconds after bolus injection.

RESULTS

Plasma glucose levels ranged from 4.01 to 14.75 mmol/L in both groups and did not significantly differ between animals given intraperitoneal or intravenous ¹⁸F-FDG. Over the time course of the scan (90 min), glucose levels rose but were not significantly different at the 90-min time point relative to the initial sample (7.5 ± 3.0 vs. 11.1 ± 4.9 mmol/L [mean ± SD] from the initial and final glucose samples, respectively). When all animals were pooled with data from the last 2 frames of the emission scan, there was a significant negative correlation between glucose levels obtained during the scan (mean of 45- and 90-min points) and whole-brain ¹⁸F-FDG uptake (R² = 0.752, y = −0.0015x + 12.6; P < 0.001). Thus, brain ¹⁸F-FDG uptake was inversely proportional to plasma glucose levels.

Figure 1 depicts the strategy used here for comparing measured input function versus image-derived input function. The latter used an ROI over the left ventricle of the heart (Fig. 1A). With this approach, the peak concentrations of ¹⁸F-FDG occurred at 2.5 s in the image-derived input function (Fig. 1C). From the same animal, the measured peak concentration of ¹⁸F in plasma occurred 10 s after the start of injection. This was consistent with all 4 animals given intravenous ¹⁸F-FDG (Fig. 2), where there was a significant delay in the measured versus the image-derived peak. Although the image-derived concentration curve appears to more accurately catch the initial activity after a bolus of ¹⁸F-FDG, there is poor agreement at the tail of the distribution, or at later time points. In fact, at later time points, the 2 curves oppose each other; where there is continuously decreasing activity in the measured blood curve, activity in the image-derived curve increases (Fig. 1). It is at these later time points that the Patlak method relies on the slope of the input function and tissue concentration of ¹⁸F-FDG in a given ROI. This may be one explanation for the low and sometimes negative values for MRGlu obtained from image-derived time–activity curves.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2. 

Time–activity curve of measured ¹⁸F in plasma after intraperitoneal (IP) or intravenous (IV) injection of ¹⁸F-FDG. Each point represents mean ± SE for 4 animals injected intravenously and 6 animals injected intraperitoneally. Arrows indicate time points chosen for the input function in calculation of SUV.

The measured arterial input function (normalized by the injected dose of activity) is depicted in Figure 2. The peak concentration of measured ¹⁸F after intravenous administration was 10 s. With intraperitoneal ¹⁸F-FDG, peak plasma concentrations occurred 15 min after injection. Twelve brain regions were examined (location and volumes are given in Table 2) to assess (a) the differences in intraperitoneal versus intravenous MRGlu and (b) the optimal time point for obtaining a single blood sample after each route of administration. In this case, optimal is defined as the plasma time point (concentration in plasma or Cp) that yielded results that most closely matched the MRGlu obtained using the Patlak quantitative method. For animals given intraperitoneal ¹⁸F-FDG (Table 3), average cortical values for MRGlu determined using the Patlak method correlated highly with those of animals given intravenous ¹⁸F-FDG (values are given in Table 4; R² = 0.3472, P = 0.044). In Figures 3 and 4, the correlation between Patlak-derived MRGlu and SUVs are given for each individual animal after an intraperitoneal (Fig. 3) or intravenous (Fig. 4) injection. Figures 3A and 4A depict the SUV calculated using plasma ¹⁸F concentrations at the peak of the blood curve, which was 10.2 s in intravenously injected animals and 15 min in animals given intraperitoneal ¹⁸F-FDG. These times are indicated by arrows in Figure 2. Figures 3B and 4B depict the correlation between Patlak-derived MRGlu and SUV, where the SUV was calculated from a plasma point taken 45 min after injection for intraperitoneal (Fig. 3) or intravenous (Fig. 4) ¹⁸F-FDG. For both routes of administration, the correlation was significant at P < 0.01. For intraperitoneally injected animals, the correlation between MRGlu and SUV is strongest when calculated from a blood sample taken 60 min after injection (Figure 3C). For intravenously injected animals, the strongest relationship between MRGlu and SUV occurred with plasma taken 45 min after injection (Figure 4B). At later time points, the correlation coefficient decreases, presumably as the single plasma point used for the SUV calculation becomes less representative of the actual input. If the injected dose is used instead of plasma points (Eq. 5), the correlation coefficient relating SUV to MRGlu is 0.61 for intravenous ¹⁸F-FDG injections (significant at P < 0.01) and 0.43 for intraperitoneal administration (P < 0.05).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3. 

MRGlu and ¹⁸F-FDG uptake after intraperitoneal (IP) ¹⁸F-FDG. Plots of ¹⁸F-FDG uptake at several time points vs. glucose metabolism measured by fully quantitative Patlak plots are shown. Each symbol represents data from a different animal (indicated in legend). Whereas SUVs at later time points are significantly correlated with MRGlu, SUVs calculated from earlier time points or from injected dose are not, thus suggesting that behavioral neuroimaging experiments that use intraperitoneal routes of administration should obtain at least 1 blood point at least 60 min after injection.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4. 

MRGlu and ¹⁸F-FDG uptake after intravenous (IV) ¹⁸F-FDG. Plots of ¹⁸F-FDG uptake at several time points vs. rate of glucose metabolism measured by fully quantitative Patlak plots are shown. Each symbol represents data from a different animal (indicated in legend). Here, the most significant correlations are obtained from earlier time points, with the optimal time after injection being 45 min. Unlike intraperitoneal injections, here there was also a highly significant correlation between glucose metabolic rate and SUV calculated using the injected dose of ¹⁸F-FDG (R² = 0.61, P < 0.01).

Table 5 gives the SUVs calculated using the 4 different time points in animals that were awake for uptake and received an intraperitoneal injection of ¹⁸F-FDG. At each time point, these values were significantly higher than the same regional SUV calculations obtained from anesthetized animals (P < 0.001 at all time points, Student t test).

These studies also show that the heart is not a useful input region because of spillover from the surrounding myocardium; a single plasma point from the input function taken at 60 min after injection for intraperitoneal injections or at 45 min after injection for intravenous injections provides a measure of relative glucose metabolism that is highly correlated with measures of MRGlu that were obtained from a fully quantitative input function.

DISCUSSION

The premise of behavioral neuroimaging with ¹⁸F-FDG in rodents relies on the fact that the animals are outside the tomograph during radiotracer uptake and are anesthetized only for a short duration at the end of the uptake period, eliminating the possibility of capturing an image-derived input function or using a β-sensitive probe (18) or arteriovenous shunt (19) to measure the intravascular radioactivity online. This is possible because the 2-deoxy derivative of glucose (2-deoxyglucose) is transported and phosphorylated by hexokinase to the 6-phosphate but is not metabolized further. Glucose-6-phosphatase does not recognize ¹⁸F-glucose-6-phosphate because of the missing hydroxyl group and so the labeled molecule is trapped and can be imaged late after the injection. Thus, whereas human subjects can be performing tasks in the tomograph in a fully conscious state, small-animal imaging studies require that the animal perform a given task outside the tomograph and then be anesthetized and scanned.

Behavioral neuroimaging is also predicated on the ability to simplify quantitative image analysis to a single static uptake parameter (20,21). One criticism of the SUV is that it is subject to too many sources of variability that have not been well controlled or even considered in reported experiments. Here, we have attempted to find those parameters that lead to the most quantitative approach for behavioral imaging, without the requirement of a chronically placed cannula, full arterial curve, or intravenous radiotracer injection.

The present study was undertaken to investigate the feasibility of intraperitoneal ¹⁸F-FDG injections for behavioral neuroimaging studies and to establish the optimal parameters for quantitation of these data. Intraperitoneal radiotracer injections are experimentally more feasible for behavioral imaging experiments, as they do not require a chronically placed catheter. However, there are few studies of possibilities for quantitating intraperitoneal injections of ¹⁸F-FDG, and before this route of administration can be routinely used it is important to optimize experimental protocols relative to more established routes of administration. For example, if a single blood point is used for the SUV input function, the optimal sampling time for intravenous ¹⁸F-FDG is 45 min, whereas the optimal sampling time from an animal given intraperitoneal ¹⁸F-FDG would be 60–90 min. These studies have also shown that, under these conditions, a single blood point can be used as input for the SUV and yield similar values as MRGlu calculated using the Patlak method (Tables 3 and 4).

The rate of uptake was different between intravenously and intraperitoneally injected animals. Intravenously injected animals took up the ¹⁸F-FDG in <10 min, but it took 30 min for the intraperitoneally injected animals to attain the same level of brain uptake. At the time of peak plasma ¹⁸F concentrations, there appears to be a 10-fold difference in the calculated SUV for intravenous versus intraperitoneal routes of administration, whereas at later sampling times these values do not differ (Tables 3 and 4). In addition, values for animals injected in the awake state are much higher than those for anesthetized animals (Table 5), consistent with previously reported results (22). Most likely this difference is due to the lower values for plasma radiotracer concentrations at the peak after an intraperitoneal injection (0.09%–0.16% of the injected dose at the 15-min peak) versus after an intravenous injection (6%–8% of the injected dose at the 10-s peak). Thus, after an intraperitoneal injection, peak plasma concentrations never reach the vicinity of peak concentrations after an intravenous injection. This makes sense, considering the fact that the former must filter through the peritoneal cavity before reaching the bloodstream, whereas the latter is injected into the bloodstream directly. Finally, the observed difference in the rate of uptake allows the investigator to choose the route of injection according to the time profile of the stimulus. For example, in pharmacologic studies where drug-induced changes occur in a slow time sequence, it might be advantageous to use the intraperitoneal route of administration over intravenous administration. Alternatively, for stimulation experiments that are immediate cause-and-effect situations, one might prefer to use the intravenous route of administration so that stimulation can be maintained for a shorter time period. Nevertheless, the present studies and others (23) show that at later time points, which are relevant to behavioral imaging protocols, the 2 routes of administration provide virtually identical quantitative values (Tables 3 and 4).

One difference between the 2 routes of administration is the correlation between SUV calculated with the injected dose (SUV_ID) and MRGlu. When given intravenously, the correlation between SUV_ID and glucose metabolism is near unity and the relationship between the 2 is highly significant. However, although intraperitoneal ¹⁸F-FDG yields only a slight correlation between SUV_ID and the calculated metabolic rate, both routes give significant correlations when the SUV is calculated from plasma concentrations. We propose that the SUVs are different because of the variability inherent in intraperitoneal injections. Intravenous injections introduce the radioactivity into the bloodstream directly, from which it is delivered to the brain and other organs. Intraperitoneal injections deliver the radioactivity to the peritoneal cavity, from which it must enter the bloodstream by passive diffusion through the peritoneal wall into capillaries and vessels. The amount of activity that reaches the brain and other organs becomes dependent on the rate and the amount of radioactivity transported from the peritoneal cavity to the bloodstream. The reason for this most likely lies in the variability inherent in intraperitoneal injections — some animals are injected with a mCi, only a fraction of which enters the bloodstream while other animals have much higher penetration from the peritoneum. Thus, using only the injected dose to calculate uptake values can lead to spurious results, especially if the volume of tracer is different (e.g., soon after delivery yields a high radioactivity concentration in a low volume, whereas animals injected with the same delivery later in the experiment may receive an identical dose in a much larger volume). Thus, it is advised to always inject the tracer in the same volume and to maintain this volume between 0.5 and 0.8 mL for intraperitoneal routes of administration. Higher tracer concentrations in smaller volumes may lead to accumulation in organs surrounding the peritoneal cavity, diminishing the amount of tracer available to the bloodstream and altering the kinetics of the tracer.

Because the SUV measures total activity in an ROI at a single time point and the Patlak method incorporates the time course of this activity in the brain and in the blood, variability due to cardiac output or other cardiac parameters may affect the ability to use the SUV from a single time point. The Patlak slope uses the integral under the arterial input function (i.e., the sum of all of the available ¹⁸F-FDG to the ROI, or the “available dose”) for normalization. The SUV approximates this integral from either the injected dose or a single plasma point, divided by the body weight of the animal. Changes in cardiac output would be accounted for by the dynamic Patlak model but would be obscured by the SUV model. The 2 approaches do not yield identical values, even at the “optimal” blood sampling time point demonstrated here (60 min in Fig. 3 or 45 min in Fig. 4). Agreement between SUV and Patlak has been shown to be higher in peripheral tumors if the integral of plasma input is used for calculation of the SUV (24). Using the integral of plasma input would account for variability in cardiac parameters and perhaps tighten the correlation between calculated SUV- and Patlak-modeled data in the brain as well.

In anesthetized animals, the average whole-brain MRGlu calculated in the present study using the Patlak method and a full arterial curve was 29.7 ± 0.95 μmol/100 min/mL for intravenous ¹⁸F-FDG and 27.3 ± 1.78 μmol/100 min/mL for intraperitoneally injected animals (Tables 3 and 4). This is in good agreement with the results from earlier quantitative rodent PET studies (22,25), albeit lower than values obtained using autoradiographic methods (26). In animals that were not anesthetized for uptake, calculated SUVs were significantly higher than the SUV calculated in the anesthetized animals (Table 3 vs. Table 5). Two reasons for this are that the ketamine/xylazine cocktail directly interacts with plasma glucose (27) or that, globally, anesthesia decreases the rate of brain glucose metabolism (22).

In general, the image-derived time–activity curves yielded higher peak maxima than the well counter blood sample data (Fig. 1), suggesting an underestimation of the peaks even when rapid blood sampling techniques were used. The external blood sampling and the measured input curve derived from the list-mode image automatically differ because of the sampling and the calibration, as has been shown previously (5,27–29). In particular, Wu et al. (30) compared the input function derived from an ROI in the left ventricle of small primates with the input function derived from blood sampling and well counting. These authors found that a factor-analysis–derived input function combined with a blood sample from a late time point provided accurate measurements of the rate of glucose metabolism derived from either a compartmental or Patlak graphical analysis. However without the factor analysis, for which the entire time course of activity is needed, the spillover activity due to the overlap of the blood pool and myocardium cannot be ignored. Our thrust is behavioral neuroimaging, where animals receive the ¹⁸F-FDG tracer injection away from the tomograph while performing a behavioral task, so factor analysis or other dynamic extraction methods are not practical. Because the static images are acquired at a late time point after injection, and because data show that the cardiac blood pool is contaminated by myocardial spillover at these time points, an image-derived input function is not practical either. Other sources for image-derived input functions could potentially include the carotid artery, which is not biased by extremely active surrounding muscle. However, resolution limits the use of this region as well.

We opted for the Patlak analysis because this method gives a quantitative value that, for the most part, relies on time–activity data obtained during the last few time frames of a PET experiment—that is, the ratio of tissue to plasma ¹⁸F concentrations yields a curve that eventually approaches a straight line, the slope of which is proportional to the rate of glucose metabolism. The time points at which the Patlak plot became linear were those time points used for the SUV analysis. Whereas a nonlinear least-squares kinetic analysis of the complete tissue uptake data allows determination of the individual rate constants, we used a quantitative method that isolated the time points of interest for a behavioral imaging experiment. We maintain that the initial time frames can either be spent inside the tomograph (anesthetized) or outside the tomograph, awake and behaving. Therefore, of the available quantitative methods, we chose the one that most closely reproduced the awake and behaving experimental condition. This encompassed the last 30 min of the scan, which was divided into 6 time points (6 frames of 5 min each).

One limitation of the Patlak method and of the SUV analysis proposed is that both assume no dephosphorylation of ¹⁸F-FDG-6-phosphate by the enzyme hexose phosphatase (k₄ = 0). This assumption is based on 2 pieces of evidence, both highly debated: (a) low values for brain hexose phosphatase activity in the rat (26,31)—although see (32)—and (b) autoradiographs made 17 and 24 h after an intravenous injection, which indicated half-lives for regional 2-deoxyglucose phosphate loss of 6–10 h (26). However, the negligible k₄ assumption was called into question by a report from Hawkins and Miller, who reported half-lives for tissue deoxyglucose phosphate loss of between 70 and 100 min (33). The controversy surrounding the influence of k₄ on measurements of the MRGlu (34) has been largely resolved with the general consensus that, if it is a concern, it can be readily incorporated into quantitative models (35). Our own previous experiments found values for k₄ in the rat of 0.008 min⁻¹ during a 60-min scan (13). Because the fractional rate of dephosphorylation becomes increasingly important as time proceeds, shorter scanning durations (e.g., 10 instead of 30 min) should be considered for behavioral imaging experiments.

Because glucose competitively inhibits ¹⁸F-FDG transport across the blood–brain barrier and its phosphorylation by hexokinase, plasma glucose levels can influence estimations of MRGlu and even ¹⁸F-FDG uptake. Our findings are consistent with this, showing that as plasma glucose levels decrease, brain ¹⁸F-FDG uptake increases. This observation supports the importance of including plasma or whole-blood glucose levels in any estimations of standardized ¹⁸F-FDG uptake. In an animal study examining the effects of hyperglycemia, Wahl et al. (36) reported that as plasma glucose is increased the ¹⁸F-FDG uptake in the brain is decreased. It has also been shown that certain types of anesthesia, such as ketamine/xylazine, have effects on plasma glucose levels (37).

CONCLUSION

Minimal blood sampling is a requirement for rodent imaging, and maintaining chronically implanted catheters can be problematic, especially in application to these types of behavioral experiments. The SUV as a quantitative outcome measure provides the advantage of requiring only a single blood point; however, it is based on the underlying assumption that tissue radioactivity is a good measure of the rate of glucose use. This may be the case only if appropriate variables are taken into account and if certain experimental parameters are controlled. Common examples applied to SUV calculations include failure to standardize measurement times after injection, failure to correct for body weight, and failure to correct for plasma glucose levels. Therefore, we recommend that the time from injection to measurement must be held constant to eliminate, or at least minimize, the time dependency of the SUV. This time should be 60 min for an intraperitoneal injection or 45 min for an intravenous injection. Appropriate corrections for body weight and plasma glucose levels should be applied, and a single ¹⁸F blood point should be used in favor of the injected dose for SUV input. Where structures are small and the anatomic geometry is complex, as in the rodent brain, absolute determinations of tracer uptake—be it ¹⁸F-FDG or a specific neuroreceptor ligand—should be viewed with caution.