Evaluation of [¹¹C]CS1P1 in Healthy Young and Older Adults

Abstract

Background and Purpose

Sphingosine-1-phosphate receptor 1 (S1PR1) is a key regulator of neuroinflammation and plays a crucial role in multiple neurodegenerative diseases. [¹¹C]CS1P1 is a novel PET tracer for measuring expression levels of S1PR1 in humans. Before widespread application, its quantification must be established and evaluated in healthy young and old adults through characterization of binding topographies, kinetics, and tracer metabolism rates.

Materials and Methods

We acquired dynamic [¹¹C]CS1P1 emission data from 29 healthy controls and investigated the topography of [¹¹C]CS1P1 uptake, radiolabeled metabolites of the tracer, an image-derived input function estimation, and tissue compartment modeling.

Results

The image derived input function approximated the arterially sampled input function. Further, radiolabeled metabolites of the tracer accumulated linearly throughout the scan and demonstrated consistency across participants. A two-tissue compartment model fitted the observed emission data well, consistent with previously reported nonhuman primate studies. Kinetic modeling using the image derived input functions, corrected by population estimates of tracer metabolism, provided a good fit for tissue activity curves. Graphical Logan analysis reliably estimated volume of distribution (Vt), and Vt closely reproduced S1PR1 distribution in the brain.

Conclusions

In this study, we have established a quantitative [¹¹C]CS1P1 PET processing approach using a two-tissue compartment model and imaging-derived input function with population metabolite correction. [¹¹C]CS1P1 PET reflects S1PR1 topography and supports its use for investigating neuroinflammation in humans.

SUMMARY

Previous Literature

The Sphingosine-1-phosphate receptor 1 (S1PR1) plays an important role in regulating neuroinflammation. It has been implicated in multiple neurodegenerative diseases, including multiple sclerosis, Alzheimer’s disease, and Parkinson’s disease. A recently developed radiotracer, CS1P1, has been shown to bind S1PR1 and can be used to quantify its expression and distribution in human brains.

Key Findings

CS1P1-PET shows high uptake in the cortex and subcortical grey structures, with lower levels in white matter. A two-tissue compartment model with reversible binding best fits observed data. Importantly, an image-derived input function can replace invasive arterial blood sampling. We found no significant difference between young and older healthy adults.

Knowledge Advancement

Our description of the binding patterns of CS1P1 and development of tools for quantitative imaging of S1PR1 opens the door to studies in clinical populations with neuroinflammatory and neurodegenerative diseases. These studies will investigate the common role of neuroinflammation across multiple diseases.

Introduction

Cerebral inflammation contributes to immune-mediated (e.g., multiple sclerosis [MS], sarcoidosis) and infectious CNS disorders¹. Recognition of the role of inflammatory processes² in neurodegenerative diseases such as Alzheimer’s disease³, cerebral small vessel disease⁴, and Parkinson’s disease⁵ has motivated the development of novel radiotracers targeting inflammatory processes to investigate pathophysiology, improve diagnosis, measure target engagement of anti-inflammatory therapies, and serve as a biomarker of therapeutic efficacy.

The sphingosine-1-phosphate receptor family consists of five membrane bound G-protein coupled receptors (S1PR1 – 5) broadly expressed in the body, including the CNS⁶. S1PR1 is the most abundant and regulates egress of T lymphocytes from lymph nodes⁷, formation of vascular structures⁸, and maintenance of endothelial barriers⁹. S1PR1 is expressed on astrocytes, microglia, and infiltrating immune cells, and to a lesser extent on neurons, oligodendrocytes, and oligodendrocyte precursor cells¹⁰. S1PR1 plays a significant role in MS, where it is overexpressed on astrocytes within active and chronic MS lesions¹¹ and is required for astrogliosis¹². Four FDA-approved S1PR1 modulators (e.g., fingolimod) target S1PR1 with varying levels of specificity and are highly effective at preventing MS relapses¹³. In the periphery, these drugs cause S1PR1 involution on T lymphocytes thereby sequestering them within lymph nodes¹⁴. This mechanism may impact autoimmunity more generally, also reducing disability due to inflammatory bowel disease¹⁵. This evidence supports a crucial role of S1PR1 in the mediation of inflammatory disease.

While S1PR1 modulation impacts the peripheral immune system, S1PR1 modulation within the CNS is also relevant to several immune mediated conditions. Deletion of S1PR1 from astrocytes reduces the severity of experimental autoimmune encephalomyelitis (EAE), a common animal model of MS, and negates the benefit of fingolimod treatment, an S1PR1 modulator¹⁶. This demonstrates that CNS S1PR1 is also important in the context of MS. Beyond MS and its model system EAE¹⁷, S1PR1 is increased in response to infection¹⁸, atherosclerosis¹⁹, and vascular injury²⁰. S1PR1 is also highly relevant to T cell mediated damage in Alzheimer disease²¹, further demonstrating the importance of S1PR1 to neuroinflammation mediated neurological disease.

[¹¹C]CS1P1, formerly known as [¹¹C]TZ3321, has good selectivity for the S1PR1 vs S1PR2–5 both in vitro and in vivo²². [¹¹C]CS1P1 uptake is increased in animal models with focal inflammation, specifically EAE and vascular injury²³. In ex vivo MS tissue samples, [¹¹C]CS1P1 uptake is increased within white matter lesions, consistent with histological assessment¹⁷. Rodent toxicity studies²⁴ and subsequent human studies²⁵ demonstrated favorable dosimetry and safety profile for CS1P1.

Despite the great potential of [¹¹C]CS1P1 PET as a novel imaging biomarker for neuroinflammation in various neurological diseases, its quantification has not been thoroughly investigated. In this study, we aim to establish a [¹¹C]CS1P1 PET quantification approach by examining the effects of radiolabeled metabolite, image derived input function, and kinetic modeling. Additionally, we evaluate whether [¹¹C]CS1P1 topography in the brains of healthy adults reflects expected S1PR1 expression.

Methods

Participants

Participants were recruited from the university community using outreach websites and word of mouth. All protocols were approved by the Human Research Protection Office and our institution’s Radioactive Drug Research Committee and followed FDA IND regulations. Each participant provided informed consent prior to participating, and none had any known cognitive, neurological, or vascular impairments. The study included a young cohort (ages 21–37) and an older cohort (ages 54–81) (Table 1). Older adults had a negative amyloid PET scan and were cognitively normal. A subset of young participants (N=5) returned for a second imaging session after approximately three months. The TRIPOD checklist is included in supplemental materials.

TABLE 1.

Demographic information

	Young	Old
N	12	20
# of scans	17	20
Sex F (%)	88	95
Age (SD)	28 (5.7)	65.1 (7.7)
Weight [kg]	73.4 (15.4)	71.9 (15.9)
Height [cm]	163.8 (6.5)	161.9 (8.1)
BMI	8.6 (1.7)	8.4 (2.6)

[¹¹C]CS1P1 PET Acquisition and Reconstruction

We collected all imaging data using a Siemens Vision PET/CT scanner, including low-dose head and neck CT for attenuation correction. [¹¹C]CS1P1 was produced in our cGMP radiopharmaceutical production facilities as previously described²⁶. Injected doses ranged from 4.2 to 13.8 mCi and were followed by a saline flush. No adverse reactions were observed. After injection, 90 minutes of dynamic emission data were collected. The protocol comprised ten 6-second frames, six 20-second frames, four 30-second frames, five 60-second frames, five 120-second frames, eight 300-second frames, and two 900-second frames. Participants requiring a break were taken off the scanner between 60 and 90 minutes post-injection and repeat CT scan was performed at reinitiation of data collection. PET images were reconstructed on the scanner console using TrueX + Time of Flight reconstruction with 8 iterations and 5 subsets. Image size was 440 with a zoom factor of 2.0. Images were corrected for scatter, randoms, attenuation, and deadtime.

MRI Acquisition

Participants underwent MRI, including 3D T1W MPRAGE and 3D FLAIR for image registration and brain segmentation using FreeSurfer v5.3²⁷.

Blood Sampling

For 9 scans (2 old, 7 young from 6 participants), the participant underwent cannulation of the radial artery for arterial blood sampling, which consisted of drawing uncollected blood to purge the line, collecting 2–3 mL of arterial blood, and then flushing the line contents. Manual sampling proceeded every 6 seconds during the first 2 minutes of the scan and then progressively less frequently. Twenty participants had a contralateral venous line placed in lieu of arterial sampling, due to participant preference or inability to place an arterial canula. Venous blood was obtained at 8 time points (5, 10, 15, 20, 30, 45, 60, and 90 minutes post injection). For both access schemes, larger quantities of blood were drawn at approximately 5, 20, 30, 45, and 60 minutes post injection to measure radiolabeled metabolites.

Blood Processing

Radioactivity was measured from blood (1 mL) and plasma (0.4 mL) samples using an automated gamma counter and corrected for volume, background, and isotope decay to quantify the input function.

Radiometabolite measurements

We measured radiolabeled metabolites using a column-switching HPLC system equipped with an HLB online capture column (80 Å 15 mm, 3.9 mm x 20 mm) and analytical C18 column, following a published protocol²⁸. The blood was centrifuged for 5 min at 3,000 rpm (2967 x g), and the plasma supernatant was collected and mixed with solid molecular biology certified urea to eliminate protein binding and allow back-flushing of parent compounds and metabolites from the capture column to the analytical column. The plasma-urea mix was filtered and injected into a 5 mL HPLC loop concatenated with the capture column. The enriched radioactivity was reverse-flowed into the analytical column, and the eluent fractions were collected with an automated fraction collector in 1 minute intervals. Signals from the detector (UV detector wavelength: 254 nm) were collected and analyzed using Chromeleon 7, and they were used to integrate the peak areas decay-corrected to the elution time of the parent compound. The inline radioactive detector generated radioactive chromatograms to provide the percentage of each radioactive peak, including the radioactive metabolite(s) and parent compound [¹¹C]CS1P1.

PET processing

We corrected for motion during PET image acquisition using rigid registration. Frames were individually registered to a mean image of five middle frames, except for the first five frames (first 30 seconds), which all used the registration of their mean image, and the second five frames (30–60 seconds), which were first registered to their mean image. This was necessary due to significant noise and vascular signal immediately after injection. Finally, we registered the motion-corrected PET volume to an MPRAGE.

Calculation of Image-Derived Input Function (IDIF)

The petrous segment of the carotid artery was used for IDIF calculation as it is the largest segment which is embedded in bone and thus fixed in position relative to the brain and relatively free of signal spill-in. The average petrous carotid diameter in this sample was M = 5.3mm SD = 0.37mm and the functional resolution of the Siemens Vision PET scanner is 3.5mm. We manually selected an ROI including the petrous carotid arteries and selected the brightest ten percent of voxels during the initial tracer bolus. From these voxels, we found that selecting the top three percent consistently identified a group of voxels near the center of the petrous carotid arteries, similar to other approaches²⁹. A subset of participants (n=8) had arterial blood sampling at ~40 points during PET imaging, enabling direct comparison of image-derived and blood-derived input functions. We calibrated IDIFs by offsetting the entire IDIF to match blood samples between 30- and 60-minutes post-injection.

ROI and Time-Activity Curve (TAC) creation

We created regional TACs by masking the registered dynamic PET image using Freesurfer segmentations representing cerebral cortex, subcortical gray structures, normal appearing white matter (NAWM), cerebrospinal fluid (CSF), choroid plexus, and non-brain, and then applying partial volume correction for these ROIs using the symmetric geometric transfer matrix method³⁰.

Kinetic Modeling

All kinetic modeling was performed with the kinfitr package (v0.7.0) in the R Statistical Software (v4.2.2). We fitted both one- and two-tissue compartment models and evaluated the sensitivity of parameter estimates to the use of either arterial sampling-based or image-based input function and individual versus overall population tracer metabolism rates. We allowed kinfitr to evaluate K1, k2, k3, and k4 estimates ranging from near zero to well above the expected distribution. Additionally, starting parameters were varied across multiple iterations (n=60) to avoid local minima. Tissue blood volume (vB) was also fit and estimated ranged conformed with prior PET studies³¹.

Results

[¹¹C]CS1P1 uptake topography reflects S1PR1 expression.

High [¹¹C]CS1P1 uptake in the cortex and subcortical grey matter (Fig. 1) is consistent with postmortem evaluation of S1PR1 expression³². Areas with the highest binding were located within subcortical grey matter structures, namely the amygdala, caudate, putamen, and thalamus. The hippocampus had significantly lower binding than other subcortical grey structures and was comparable to cerebral cortex. This finding is sporadically reported in human postmortem examinations and was seen in non-human primate microPET with [¹¹C]CS1P1²². We compared CS1P1 SUV with the GTEx Portal gene expression data for S1PR1 and found a strong correlation (r = 0.58; Table 2) across the 10 brain regions samples. Importantly, CS1P1 uptake was not correlated with S1PR2–5 expression.

FIGURE 1.

Mean SUV [11C]CS1P1 images derived from 60–75 minutes post injection. Highest signal is observed in subcortical grey structures (top and right white arrows), followed by cortex, and lowest signal is observed in the white matter. Extracranial uptake was high in the salivary glands (bottom white arrow).

TABLE 2.

Gene expression vs. CS1P1 SUV Correlation

	r	p
S1PR1	0.584536	0.098322
S1PR2	−0.50946	0.161234
S1PR3	−0.1933	0.618275
S1PR4	−0.24271	0.529193
S1PR5	−0.64609	0.060116

Extracranially, we observed high uptake in the salivary glands and lymph nodes (Fig. S1A). Images, including MR, were reviewed for incidental findings. The most common incidental finding was non-specific white matter hyperintensities which demonstrated reduced CS1P1 uptake in their core (Fig. S1B).

No adverse events were observed during scanning, nor were any reported in the 24hr post-scan follow-up calls.

Calibrated IDIF recapitulates the mean arterial input function.

Eight participants (2 old, 6 young, with 9 scans total) underwent arterial blood sampling during PET acquisition. We correlated IDIF and AIF within participants by interpolating IDIF values to match each blood sampling timepoint (between 28 and 43 timepoints per scan) (Fig. 2A; mean (n=9) Pearson’s r = 0.980, SD = 0.012). Calibration shifted the IDIF to minimize the difference between IDIF and blood samples collected 30–60 minutes post-injection. These offsets were small, between −0.079 and +0.034 mCi/mL (M = −0.031, SD = 0.030), and were applied prior to performing all other analyses. Six scans had sampling frequency sufficient for modeling the initial influx of tracer, but difficulties with line access delayed sampling during the first 60 seconds for three scans. This initial bolus peak was often lower and broader in AIFs, likely due to the dispersion effects inherent in radial artery sampling. However, the width of the peak at half its maximum (FWHM) was not significantly different between image- and blood-derived input functions (IDIF: M = 23.03 s, SD = 5.06 s, AIF: M = 24.86 s, SD = 5.20 s, paired t-test t(5) = −1.827, P = 0.10). Similarly, the area under the curve (AUC) (Fig. 2B) for IDIF (M=17.05 min*mCi/ml, SD=7.20) and AIF (M=16.25, SD=8.17) using a paired samples t-test found no significant difference, t(5) = 1.64, P = 0.16. There was no correlation between the difference in image- and blood-derived input function AUC and the diameter of the carotid, suggesting that partial volume effects were mitigated by the present approach.

FIGURE 2.

A) Comparison of activity measures using image-derived and arterial blood sampling methods during the first five minutes (left) after radiotracer injection and during the remainder of the scan (right). B) Comparison of area under the curve for the two input function estimation methods.

Blood Plasma activity is approximately 20% higher than whole blood.

We found the ratio of plasma to whole blood activity stayed near 1.2 for the duration of the scan for both venous and arterial samples (Fig. S2). Excluding four early outlier values (ratio over 5 and sampling time under 10 seconds post-injection), the remaining 396 samples had a median ratio of 1.21, with 95% between 1.09 and 1.38. The ratio was not associated with sampling method (venous vs arterial), time after injection, age, sex, tracer metabolic rate, or tracer dose. Radiotracer free fraction was measured in a subset of older participants and found to be consistent across individuals (mean = 63.5% [SD = 3.3%]).

Tracer metabolism rates are not related to age, sex, or tracer dose.

We considered tracer metabolism models following zero order and first order kinetics. Both models fit observed group averages well, and we used a linear model for all analyses. Radiolabeled metabolite accumulation rates were between 0.2% and 0.8% per minute, with 90% within 0.3–0.7% per minute (Fig. 3). Independent sample t-tests indicated no difference in rates between young (M = 0.541% / min, SD = 0.121) and old (M = 0.534% / min, SD = 0.138) groups (t(19) = −0.11, P = 0.91). Exclusion of any one participant from the mean estimate changed the population mean by 0.0053%, or by less than 1% of the mean value. Two young participants had a repeat scan approximately 3 months later. Tracer metabolism rates between scans differed by 0.034% and 0.033% per minute. An ANOVA showed that tracer metabolism rates were not associated with biological sex, age, or tracer dose (F(1, 15) = 0.586, P = 0.456; F(1, 15) = 0.070, P = 0.795; F(1, 15) = 0.462, P = 0.51, respectively).

FIGURE 3.

Individual and group tracer parent fractions as measured by HPLC during the first 60 minutes of PET acquisition. No difference was observed between young and older groups (group means in bold lines).

Two-tissue compartment models better fit [¹¹C]CS1P1 data than one-tissue compartment models.

Time Activity Curves (TACs) from representative regions of interest are shown in Figure 4A. Parenchymal values showed increased uptake in cortex and subcortical gray matter compared to NAWM (Fig. 4A, inset). We fitted kinetic models to TACs using one-tissue and two-tissue compartment models featuring either reversible or irreversible binding. Prior animal and ex vivo work supports a two-tissue compartment model²⁵. An example two-tissue compartment fit is shown in Figure 4B. We evaluated the qualities of fit of the models using Akaike information criteria (AIC) (Fig. 4C,D). The two-tissue compartment model with reversible binding outperformed the one-tissue compartment model (AIC_1tcm-AIC _2tcm-rev M=10.0, range = −22.3 – 80.8, SD = 21.2) and outperformed the two-tissue compartment model with irreversible binding (AIC_2tcm-irrev-AIC _2tcm-rev M=7.4, range = −17.8 – 62.3, SD = 15.3). Mean microparameters are shown in Table 3. All participants contributed 90 minutes of emission data, but a subset (N = 7) had longer data collection. In those patients, we found that there was minimal change in tracer uptake after 90 minutes (Fig. S3), supporting the analysis scheme here.

FIGURE 4.

A) Time activity curves for ROIs, including the image derived input function (IDIF), cerebral cortex, normal appearing white matter (NAWM), Subcortical grey structures, cerebrospinal fluid (CSF), and choroid plexus. The main panel highlights the initial influx of tracer, with the heights of the initial peak activity in descending order being IDIF, choroid plexus, cortex, subcortical grey structures, NAWM, and CSF. The inset shows the remainder of the scan, with the final values being highest in the cortex, subcortical grey, and NAWM, followed by choroid plexus, IDIF, and CSF. B) Measured regional values for an example region (cortex) and the fitted values from a two-tissue compartment model. The lower graph shows the difference between measured and estimated values. C) Comparison of Akaike Information Criterion (AIC) values for two-tissue compartment models featuring reversible and irreversible binding. D) Comparison of AIC values for two-tissue compartment and one-tissue compartment models.

TABLE 3.

Mean two tissue compartment model fit microparameters

Region	K1	k2	k3	k4	vB	Vt
Cortical Grey Matter	0.0584	0.4004	1.3229	0.0180	0.0484	6.0113
White matter	0.0140	0.8945	1.8000	0.0095	0.0038	4.3018
Subcortical Grey Matter	0.0311	0.0791	1.2383	0.0236	0.0295	5.7928

Vt estimates modestly impacted by choice of input function, but not by the use of population metabolic rate.

Arterial sampling is invasive; venous sampling facilitates more broad application. Measurement of radiolabeled metabolites of the tracer requires additional time, equipment, and personnel. We asked whether substituting venous-calibrated IDIF and population metabolic rates would significantly impact kinetic modeling. A paired-samples t-test revealed that IDIF provided slightly higher Vt estimates than AIF (between −19% and +27%; M = +7%, SD = 16%) (t(23) = −2.65, P = 0.014) (Fig. 5A). However, when comparing IDIF-based Vt estimates using individual versus population tracer metabolism rates, our paired-samples t-test did not show a difference (t(41) = 1.21, P = 0.23) (Fig. 5B).

FIGURE 5.

A) Scatter plot (left) and Bland-Altman plot (right) comparing Vt estimates for cortex and NAWM using arterial blood sampling and calibrated IDIF. B) Scatter (left) and Bland-Altman (right) plots comparing Vt estimates for cortex and NAWM using individual and population tracer metabolism rates. C) Scatter plots showing similar cortex to NAWM Vt ratios when substituting IDIF for AIF (left) and population for individual tracer metabolism rates (right).

We next asked whether the topography of Vt estimates within participants was robust to variations in processing methodologies. We calculated the ratio of cortex Vt to NAWM Vt and found it was consistent regardless of choice of AIF vs. IDIF or measured vs. population metabolite estimation (Fig. 5C). Critically, we are not suggesting that NAWM be used as a reference tissue, but rather demonstrating that the bias due to choice of input function or metabolite curve manifests as a multiplicative constant. This constant does not change the topography of the quantified tracer uptake.

Cortex and subcortical gray Vt are higher than NAWM Vt and are not significantly associated with age, and Vt estimates display test–retest reliability.

We next fit two-tissue compartment models using a calibrated IDIF and the population tracer metabolism rate to compare regional Vt estimates across the entire study population (Fig. 6A). Vt estimates were higher in cortex than in NAWM and higher in subcortical grey matter than NAWM (cortex M = 6.39, SD = 3.04; subcortical M = 5.41, SD = 2.62; NAWM M = 2.93, SD = 1.50; cortex-NAWM paired samples t-test t(36) = 10.10, P < 0.001 and subcortical-NAWM paired samples t-test t(36) = 9.03 P < 0.001). Five participants from the young group had a repeat scan after approximately 3 months. We found that Vt estimates at follow-up were highly correlated with the first scan (r(17) = .85, P < 0.001) (Fig. 6B). Intraclass Correlation Coefficient (ICC) between first and follow-up scans were high in grey matter and lower in white matter ( ICC[cortex] = 0.91 [ 95% C.I. = 0.357 – 0.99 ] ; ICC[subcortical grey] = 0.83 [ 95% C.I.= 0.054 – 0.981 ]; ICC[nawm] = 0.17 [ 95% C.I. = −0.832 – 0.913 ] ).

FIGURE 6.

A) Comparison of Vt estimates between young and older participants for the Cortex, NAWM, and Subcortical gray structures. The difference between young and old groups was not significant for any region. B) Comparison of Vt estimates from initial and follow-up scans (n=6) for the same regions. Model fitting failed for NAWM for one scan (point not shown).

Two-Tissue Compartment Model Vt is comparable to Logan analysis Vt.

Compared to two-tissue compartment models, Logan graphical analysis is more straightforward, allows for shorter scanning intervals, and is more broadly applicable. We compared Vt estimates from the two-tissue compartment model (Fig. 4B) and Logan graphical analysis (Fig. 7A). Vt estimates from the two methods were highly correlated (r(109) = .63, P < 0.001) (Fig. 7B). Estimates from two-tissue compartment models (M = 4.91, SD = 2.85) were slightly higher than those from Logan plot analysis (M = 4.35, SD = 2.62) (paired samples t-test (t(110) = 2.51, P < 0.05), with a mean difference of 24% (Fig. 7C). Bland Altman analysis shows insignificant systematic change in estimate bias across the range of measured Vt ( Pearson r(109) = .09, P = .33 ).

FIGURE 7.

A) Example of a Logan plot for a typical cortex ROI, showing best fit line for the last 8 frames of the scan with a Vt estimate (slope) of approximately 6.4. B) Comparison of Vt estimates between two-tissue compartment models and Logan plot analysis. C) Bland Altman plot showing differences between the two methods.

Vt estimate topography is similar to previous investigations of S1PR1 distribution in the brain³². We estimated the Vt in 5 lobar cortical regions, 7 subcortical grey structures, and normal appearing white matter. Vt was highest in the cortex and subcortical regions and generally low in the white matter (Fig. S4). Within the cortex, binding in the occipital lobe was numerically higher than other regions. In the subcortex, thalamic and basal ganglia uptake was higher than medial temporal lobe structures, including the hippocampus. Vt evaluated within each region of interest in the young and old cohort is shown in Supplemental Table 1.

Discussion

This study describes the dynamic brain imaging characteristics of [¹¹C]CS1P1 in a cohort of young and old healthy adults. We have reported several results important to future application of this tracer in clinical populations. First, we described and quantified the impact of heuristic approaches to quantitative kinetic modeling and found that the IDIF and population metabolite estimates produce results consistent with invasive sampling. Second, we have shown that kinetic modeling of this tracer can be reliably achieved with two-tissue compartment models or approximated with Logan graphical analysis. Finally, we showed that Vt estimates have excellent test–retest reliability. Most importantly, the measured topography of [¹¹C]CS1P1 matches closely the known distribution of S1PR1 expression. These results demonstrate that this tracer is suitable for application in patients with neurological diseases involving neuroinflammation.

Kinetic Modeling of [¹¹C]CS1P1 Uptake Recovers Known Receptor Expression Topography

The S1P signaling pathway is known to play a critical role in demyelinating lesions in MS and may be important for other neurological, psychiatric, or general medical conditions. [¹¹C]CS1P1 reliably enters the human brain and binds in a consistent pattern similar to post mortem measures of S1PR1 expression in humans³². Nishimura and colleagues studied 20 autopsy samples using immunohistochemistry targeting S1PR1. They found uptake was highest in the cortex. Similar to our results, no large differences were found across cortical regions. A slight difference in our findings is that we found reduced binding in the hippocampus. Nishimura et al also found high binding in the striatum and thalamus and low binding in the white matter, consistent with our observations. This pattern also agrees with ex vivo [¹¹C]CS1P1 studies in human brain tissue and non-human primate microPET²². S1PR1 is broadly expressed both topographically and within a large range of cell types. This distributed expression, captured by [¹¹C]CS1P1 emphasizes the broad impact this signaling pathway may have on brain disease (e.g., in MS) but also complicates analysis. For example, it is not clear that there is a suitable reference region and the broad expression may include non-specific binding. Future studies in clinical populations and with blocking drugs will clarify these limitations. In summary, our study demonstrates that [¹¹C]CS1P1 reliably captures known S1PR1 topography.

We studied both younger and older controls and found similar [¹¹C]CS1P1 Vt values across tissue classes. Despite a relatively small sample size, our study suggests that age alone may not affect tracer behavior or binding. Future studies and larger sample sizes are needed to determine if diseases alter [¹¹C]CS1P1 uptake. This observation is consistent with animal models wherein S1PR1 expression is dysregulated in AD but not in normal aging²¹.

Utility of IDIF vs. AIF in Kinetic Modeling of [¹¹C]CS1P1 Uptake

Arterial blood sampling is invasive and burdensome for study participants. However, accurate modeling of tracer kinetics requires an input function which reflects radioactivity levels in arterial blood plasma. If this information can be derived from PET data or a combination of PET data and venous blood samples, the study complexity, number of potential failure points, necessary staff and equipment, and study cost will be reduced. We found that the signal from [¹¹C]CS1P1 in the petrous segment of the carotid artery is highly correlated with arterial blood sampling, and correlation was further improved by calibration to venous blood samples. The mean venous blood calibration factor was small (<0.04 mCi/ml), however there was significant variability at the single subject level. Thus, the IDIF provides a reasonable proxy for arterial blood derived input function at the group level, but variability at the single subject level may impact sample size considerations. Of note, IDIF can be affected by radiotracer and PET scanner spatial resolution. The PET/CT Vision used here has a spatial resolution 3.5mm³. Studies on older PET scanners with lower resolution and radiotracers with different positron ranges may need more sophisticated IDIF approaches.

Impact of Radiometabolites on Tracer Kinetic Quantification

Many brain PET radiotracers require complex modeling of tracer metabolism to account for their influence in model tissue compartments. The radiometabolite of [¹¹C]CS1P1 does not cross the blood brain barrier to confound PET measurements²⁸. Thus, tracer metabolism can be entirely accounted for within the input function, reducing model complexity. We used 5 sampling points from the first 60 minutes post injection to determine that [¹¹C]CS1P1 metabolism is essentially linear over the time scale of a [¹¹C]CS1P1 PET study. Critically, substituting population tracer metabolic rates for individual rates in kinetic modeling did not differentially affect cortex and white matter. Thus, if individual dynamic metabolism measures are not available, accurate within-subject comparisons of [¹¹C]CS1P1 binding between brain regions can still be made with a smaller number of samples or using the population average metabolic rate.

Limitations and Future Directions

Some limitations impact the conclusions of this study. Arterial sampling and free fraction analysis are only available for a fraction of the cohort. This study does not contain known neurological diseases, but unrecognized neurological or inflammatory diseases may confound the estimates in healthy controls. To this end, future work with this tracer is underway, examining the role of S1PR1 in MS, Alzheimer’s disease, and other degenerative conditions. A comparison of this tracer to the frequently used TSPO tracer family is also key to placing these results into a larger context.

Conclusion

Here, we characterize [¹¹C]CS1P1 binding properties in young and old healthy individuals, demonstrating that [¹¹C]CS1P1 has multiple desirable PET radiotracer characteristics. [¹¹C]CS1P1 readily crosses the blood brain barrier and binds in a pattern consistent with S1PR1 expression. We confirmed that a two-tissue compartment model best fits observed tracer activity. We determined that the tracer is metabolized at a nearly linear rate similarly across age groups and sex. Importantly, we showed that using a venous blood-calibrated image-derived input function and a population tracer metabolic rate did not significantly change the estimated volume of distribution within ROIs covering the cerebral cortex, cerebral white matter, or subcortical grey matter structures. These characteristics greatly simplify data acquisition by reducing reliance on arterial blood sampling and processing multiple blood samples by HPLC. We conclude that [¹¹C]CS1P1 is a promising radiotracer for mapping S1PR1 expression in the human brain to investigate neuroinflammation.