Simultaneous quantitative imaging of two PET radiotracers via the detection of positron–electron annihilation and prompt gamma emissions

Abstract

In conventional positron emission tomography (PET), one radiotracer is imaged at a time, because all PET isotopes produce the same two 511-keV annihilation photons. Here, we describe an image-reconstruction method for the simultaneous in vivo imaging of two PET tracers and thereby the independent quantification of two molecular signals. The method leverages the 650-to-700-keV range to maximize the capture of 511-keV annihilation photons and prompt gamma-ray emission in the same energy window, hence eliminating the need for energy discrimination or for signal separation beforehand. We used multiplexed PET to track, in mice with subcutaneous tumours, the biodistributions of intravenously injected [¹²⁴I]I-trametinib and 2-deoxy-2-[¹⁸F]fluoro-D-glucose, [124I]I-trametinib and its nanoparticle carrier [⁸⁹Zr]Zr-ferumoxytol, and the prostate-specific membrane antigen (PSMA) and infused PSMA-targeted chimeric antigen receptor T cells after systemic administration of [⁶⁸Ga]Ga-PSMA-11 and [¹²⁴I]I. Multiplexed PET gives new uses to prompt-gamma-ray-emitting isotopes, reduces radiation burden by omitting the need for a computed-tomography scan, and can be implemented on preclinical and clinical systems without any modifications in hardware or image-acquisition software.

An image-reconstruction method leveraging the capture of 511-keV annihilation photons and prompt gamma-ray emission in the same energy window allows for the simultaneous in vivo imaging of two radiotracers for positron emission tomography.

Positron emission tomography (PET) in combination with computed tomography (CT) is a gold-standard imaging technology in both clinical and preclinical molecular imaging. The traditional workhorse radiotracer, 2-deoxy-2-[¹⁸F]fluoro-D-glucose ([¹⁸F]FDG or ¹⁸FDG) is used in PET/CT for identifying metabolically active cancers¹, although uptake could result from inflammation, infection or physiologic uptake. Research in molecular imaging has led to increasingly specific radiotracers targeting disease-specific cellular markers, and molecular medicine is now sequencing entire tumour genomes for most patients. However, PET imaging remains a monochromal modality, focused on one radiotracer per imaging session. Sequential PET scans to detect additional signatures with other tracers could be scheduled, but this is costly and dependent on several factors, such as the sufficient decay of one tracer over time² and patient compliance. Repeated imaging also increases radiation exposure from each accompanying CT scan, which can be as high as twice that of a single PET scan³. Furthermore, sequential PET scans have a temporal lag time between the two recorded parameters, decreasing biological continuity as well as patient convenience and compliance². Alternatively, blood samples lack the spatial information of PET imaging. Many markers, such as immune cells, reside in the tissues and not in the blood, implying that blood samples can provide only little information about the actual events in tissues⁴. Here, multiplexed PET (mPET) expands radiotracer density during a PET/CT scan, allowing for the synchronization of two compatible radiotracers from known imaging workflows to the same scan session, provided that the isotope pair is mPET compatible.

Multiplexed imaging^5,6 broadly applies different imaging modalities⁷, fluorescent emissions^8,9, or $\gamma$-rays in the case of single-photon emission computed tomography¹⁰ (SPECT) to identify each distribution. Previous PET multiplexing methods have used specialized sequential scanning with short lived isotopes¹¹, compartmental modelling from prior single-tracer scans¹², multiple list-mode reconstruction¹³, or a preferential decay of one isotope over another. Sequential scanning increases the gap time required for the decay of one tracer before being able to image the other, and also increases the dose from the repeated CT scans. The main disadvantage of compartment modelling is the requirement of information from prior single scans, yielding a method that is radiotracer-specific and organ-specific. Improvements in PET-acquisition-and-reconstruction technology have included exceedingly fast processing by GPU systems¹⁴, as well as corrections for highly energetic positrons^15,16. Many PET isotopes have additional $\gamma$-rays that require further corrections¹⁷, including background compensation^18,19, patient-specific corrections²⁰, and geometric scatter corrections for 3D reconstruction²¹. New deep-learning²² and whole-body tracking methodologies²³ are further reducing the PET dosage required. Despite the wealth of possible corrections and dose improvements, PET remains a single radiotracer system using mostly ¹⁸F.

Once a positron (${\text{β}}^{+}$) is emitted, it rapidly annihilates with an electron, yielding two 511 keV $\gamma$-rays in opposite directions, where they can be detected in coincidence (~3.5 ns) in a detector ring surrounding the object. Between the two detectors that identified the coincidence, a virtual line connects the coincidence; this is known as the line of response (LOR), and provides a probable origin of the positron. These LORs can be summed into a sinogram, and then reconstructed into a tomographic image. Current PET scanners are designed to detect and record these coincidences (doubles) from purely ${\text{β}}^{+}$-emitting isotopes (Table 1). However, when ${\text{β}}^{+}\gamma$ emitting isotopes are used (Table 2), the prompt $\gamma$-rays emitted per decay can be detected together with the doubles giving rise to triples, or even multiple coincidences^24,25. The relative abundance of these events mostly depends on the decay scheme of the isotope (number of prompt gamma rays emitted per positron emission and their energy), the energy of the $\gamma$-rays, and the sensitivity of the scanner²⁴. Most PET scanners store triples in the list-mode data as a set of contiguous double coincidences. These double coincidences are not reconstructed, as they provide two incorrect origins of the radioisotope, reducing the perceived quality of the reconstructed image from these isotopes^17,20,26,27 with the solution often to reduce the detector energy window to omit the prompt gamma or to use another isotope entirely. It is important to note that mPET needs the combination of a prompt gamma and a pure emitting positron for the separation to occur, and that the prompt gamma can be detected on the scanner and has a relatively high prompt-emission rate to that of the positron (Table 2). mPET leverages the prompt gamma emission as the identifier for the second isotope, (as a ‘tag’), without requiring any energy discrimination.

Table 1 |.

Positron emitting isotopes.

Doubles	Half-life	${\text{β}}^{+}$ yield (%)
15 O	2.04 minutes	100
13 N	9.96 minutes	100
11 C	20.36 minutes	100
68 Ga	68 minutes	89
18 F	109.7 minutes	97
64 Cu	12.7 hours	17.6
89 Zr	3.3 days	23

Table 2 |.

Prompt-gamma-emitting isotopes.

Triples	Half-life	${\text{β}}^{+}$ yield (%)	Main prompt $\gamma$ [keV] & (${\text{β}}^{+}\gamma /{\text{β}}^{+}\text{yield}$)
82Rb	1.27 minutes	95	777 (13%)
52mMn	21.1 minutes	97	1434 (96%)
60Cu	23.7 minutes	93	1333 (88%)
94mTc	52.0 minutes	70	871 (96%)
110mIn	1.15 hours	62	658 (99%)
120I	1.35 hours	46	560 (72%)
44Sc	3.97 hours	94	1157 (100%)
86Y	14.7 hours	33	1080 (83%), 627 (33%)
76Br	16.2 hours	26	559 (58%)
72As	1.08 days	88	834 (79%)
124I	4.18 days	23	602 (51%)
52Mn	5.59 days	29	744 (19%), 1434 (100%)

Results and discussion

Here we propose a general method²⁵ for separating and reconstructing double coincidences from triple coincidences (Fig. 1) by implementing an additional independent reconstruction method (Supplementary Fig. 1a) solely for list-mode double coincidences, generating two separate datasets for each PET scan (Fig. 2, Supplementary Fig. 1b and 1c). These datasets can be refined with iterative reconstructions using the abundance ratios from prior cylinder-sensitivity and uniformity-phantom studies (Supplementary Fig. 1d) to produce quantitative²⁸ mPET images (Supplementary Fig. 1e and 1f) with similar reconstruction performance to traditional LOR reconstruction methods with sufficient iterations (Supplementary Figs. 1g and 1h) at a total reconstruction time around 30 minutes on a CPU for a two-isotope method without subsets or approximations (Methods). Because many scanners do not record the energy associated with the individual event, the detected prompt $\gamma$-ray triple is differentiated instead by the type of LOR used for the reconstruction. Reconstruction and correction methods for standard PET are focused on coincidences using a single LOR. When three coincident $\gamma$-rays are detected, three LORs are possible; however, predominantly only two LORs are detected and stored as double coincidences in the list-mode. In this reconstruction approach, the two LORs previously omitted from a triple coincidence can be combined as a V-shaped LOR (VLOR), better approximating the origin of the radioisotope (Supplementary Fig. 1a). VLOR reconstruction has been widely implemented in multiplexed pinhole SPECT imaging²⁹, and has yet to be used as part of PET reconstruction. Thus, acquiring list-mode PET scans containing both a double-emitting and triple-emitting radioisotope and reconstructing triple coincidences as VLORs allows two PET isotopes to be effectively separated simply on the basis of the type of coincidence recorded and without the need for energy discrimination of the $\gamma$-rays. To the best of our knowledge, this is currently not possible on standard commercial PET scanners. Therefore, this method represents a fundamental advancement over previous proposed methods^13,30–32, and opens most preclinical and clinical PET scanners to the simultaneous acquisition of two rationally paired radiotracers.

Fig. 1 |. Overview of mPET using a pure positron and positron–gamma radionuclide pair.

PET uses 511 keV annihilation photons from an emitted positron for imaging, recording the pair of photons as a coincident event called a ‘double’ (blue ${\text{β}}^{+}$). Other positron-producing radionuclides can emit a positron and an additional gamma particle detected within the traditional coincidence-timing window, called a ‘triple’ (red ${\text{β}}^{+}$). Traditionally, triple events are considered spurious and not reconstructed, and energy windows of triple-emitting radionuclides often are adjusted to limit triple events. By increasing the coincidence energy window to include the prompt gamma while also providing a reconstruction method for triple coincidences, two-isotope distribution can be reconstructed from the list-mode events via multiplexed PET (mPET). This method can reconstruct both isotopes solely on how events are recorded: as double or triple coincidence events. For detectors with sufficient energy resolution that register energy with event coincidence, additional isotope parings could be further multiplexed with each unique prompt gamma energy.

Fig. 2 |. mPET processing workflow in comparison with traditional PET. Top,

Traditional PET image workflow processes coincident events for normalization, sinogram formation, and 2D or 3D reconstruction. Bottom, multiplexed PET image processing expands the reconstruction to account for traditional coincidences, triple coincidences, as well as for the proportion of random prompts to both categories. This iteration is isotope-specific and system-specific, including factors such as scatter within the detector field of view (see Methods for processing details). Machine and isotope factors can be determined a priori with uniformity and mixture phantoms.

The maximum-likelihood expectation–maximization (MLEM) reconstruction of VLORs in list-mode corresponds to the standard expression for iterative reconstruction of VLORs (equation 1):

$${\text{x}}_{\text{j}}^{\text{n}+1}={\text{x}}_{\text{j}}^{\text{n}}\frac{{\sum }_{\text{v}}{\text{a}}_{\text{vj}}\left(\frac{1}{{\text{P}}_{\text{v}}}\right)}{{\text{s}}_{\text{j}}}$$ Equation (1):

where $x_j^n$ is the value of voxel $j$ on the current reconstruction at iteration $n$, $s_j$ represents the sensitivity of voxel $j$, and $a_{vj}$ is the probability that an emission at voxel $j$ is detected at VLOR $v$. As each VLOR $v$ consists of two joined LORs, the projection in a VLOR $P_v$ can be simply estimated as the sum of the projections in each of the two LORs ($P_v=P_{v1}+P_{v2}$). VLOR reconstruction can be considered similar to standard reconstruction but with two-times larger LORs. If reconstruction with sinograms is preferred, it can be obtained from VLOR data using the following procedure: (1) define the relative contribution or weight $w_{v_i}$ of each LOR to the VLOR as $w_{v_1}\equiv P_{v_1}/\left(P_{v_1}+P_{v_2} \right)$ and $w_{v_2}\equiv P_{v_2}/\left(P_{v_1}+P_{v_2} \right)$. In the initial iteration, $w_{v_1}=w_{v_2}=0.5$; (2) rewrite the numerator in the MLEM expression as

$${\sum }_{\text{v}}{\text{a}}_{\text{vj}}\left(\frac{1}{{\text{P}}_{\text{v}}}\right)={\sum }_{\text{v}}{\text{a}}_{\text{vj}}\left(\frac{1}{{\text{P}}_{\text{v}1}+{\text{P}}_{\text{v}2}}\right)={\sum }_{{\text{v}}_1}{\text{a}}_{{\text{v}}_1\text{j}}\left(\frac{{\text{w}}_{{\text{v}}_1}}{{\text{P}}_{{\text{v}}_1}}\right)+{\sum }_{{\text{v}}_2}{\text{a}}_{{\text{v}}_2\text{j}}\left(\frac{{\text{w}}_{{\text{v}}_2}}{{\text{P}}_{{\text{v}}_2}}\right)={\sum }_{{\text{v}}_{\text{i}}}{\text{a}}_{{\text{v}}_{\text{i}}\text{j}}\left(\frac{{\text{w}}_{{\text{v}}_{\text{i}}}}{{\text{P}}_{{\text{v}}_{\text{i}}}}\right)$$

where $i$ is the LOR index. Therefore, one can work with the individual LORs from the VLORs just by using the weights $w_{v_i}$ (instead of ‘1’, as it is done in standard list-mode PET) that correspond to the relative value of their projections; (3) use the weights of each LOR from the VLOR. The weighted data is stored in a standard sinogram, and the sinogram is updated before each new iteration, following these three steps.

Taken together, we show that the mPET-reconstruction method can be applied to data acquired in list-mode format in commercially available preclinical and clinical PET/CT systems. We demonstrate first in phantoms the principle and quantitative capacity of mPET separation. Next, we show several examples using previously published and validated targets to highlight the possibilities that simultaneous PET isotope imaging may offer. We demonstrate in several animal models the ability of mPET to easily identify two radiolabelled small-molecule inhibitors for MEK 1/2, [¹²⁴I]I-trametinib and [¹⁸F]FDG. We also use mPET for the evaluation of nanoparticle-mediated drug delivery with [¹²⁴I]I-trametinib loaded passively onto [⁸⁹Zr]Zr-ferumoxytol (a clinically used iron oxide nanoparticle capable of carrying select small molecules in its coating³³). Lastly, prostate-specific membrane antigen (PSMA)-targeted CAR T cells were engineered to express the sodium-iodide symporter (NIS) to render them traceable in vivo by PET³⁴. We employed mPET to simultaneously visualize the distribution of NIS-expressing CAR-T cells to the PSMA-expressing tumours by ¹²⁴I-afforded and ⁶⁸Ga-PSMA-11-afforded PET, respectively. mPET is an image-reconstruction method that enables true multiplexed PET imaging from standard PET scanners without the need of any hardware or acquisition-software modification through a general PET-reconstruction method.

Imaging of phantoms

To test the feasibility of the mPET method on a preclinical scanner, phantom experiments were performed on a PET/CT system capable of acquiring list-mode data. Capable PET/CT systems include the preclinical Inveon PET/CT (Siemens Medical Solutions, PA, USA, which we used) and the SuperArgus (Sedecal, Madrid, Spain), as well as the clinical Biograph MCT PET/CT (Siemens Medical Solutions, PA, USA), amongst other potential list-mode acquisition scanners. To set up the system for mPET acquisition, a new PET protocol was configured by expanding the upper energy range from 650 keV to 700 keV to maximize the capture of the 511 keV annihilation photons and the prompt gamma in the same energy window. The full energy range of was not selected, so as to reduce random event noise and multiple coincidences from Iodine-124 (¹²⁴I). As PET detectors possess energy resolutions on the order of 20% of the energy of the photopeak, additional energy range is necessary beyond the 511 keV and prompt gamma to capture the majority of emitted photopeaks⁷. Cylindric uniformity and sensitivity phantoms containing 1 L with 9.25 MBq (250 μCi) of each isotope separately were acquired with the new energy settings to determine detector uniformity, and used to estimate the double-to-triple sensitivity ratio (Supplementary Fig. 2a). These phantom calibrations are specific to the isotope mixture and scanner used, and can be applied after mPET image acquisition. To test the separation performance of mPET, a hollow body (for background activity) and four distinct cavities were prepared containing different mixtures of two isotopes, Zirconium-89 (⁸⁹Zr) and Iodine-124. ⁸⁹Zr is considered a double-emitting isotope, whereas ¹²⁴I is a triple-emitting isotope (Table 1). Despite the prominent emission of the additional 909 keV photopeak by ⁸⁹Zr, this emission is not within the coincident timing window (3.6 ns), and is thus only viewed by a PET detector as a double-emitting isotope. The phantom as mixed was acquired with a 350–700-keV energy window for 30 minutes. The subsequently acquired list-mode file was processed using the standard histogram and OSEM2D-reconstruction algorithm as well as independently with the mPET reconstruction method. As seen in Fig. 3a, processing with the traditional PET OSEM2D reconstruction method yields a phantom cavity as well as cavity 1 through 4 with a uniform distribution in the cavity, yet a reduced signal of ¹²⁴I as many events are discarded as scatter (triples) and not reconstructed. Using the mPET reconstruction method, we observed the separation of both ⁸⁹Zr (Fig. 3b) and ¹²⁴I (Fig. 3c) in the phantom cavities, and show that separation was possible on a preclinical PET/CT scanner. In Fig. 3d the activity measured with mPET in each cavity agreed with prior activities measured during phantom preparation, confirming the accuracy of the measurement and separation.

Fig. 3 |. Phantom performance of mPET on preclinical and clinical PET/CT systems.

a, A phantom containing 9.25 MBq (250 μCi) ⁸⁹Zr in the cavity with an additional mixture of 2.22 MBq (60 μCi) ⁸⁹Zr and/or ¹²⁴I in four exterior cavities was measured on a Siemens Inveon preclinical PET/CT. PET image as acquired using the 350–700-keV energy window and reconstructed with the traditional OSEM2 method, which attenuates the ¹²⁴I signal partially as scatter, as seen by the reduced intensity in cavity 4. b, mPET separation and reconstruction of ⁸⁹Zr coincidences recapitulates the traditional PET image seen in Fig. 2a without ¹²⁴I coincidence events. c, Using triple coincidences from the list-mode, mPET recovered the ¹²⁴I titration in the phantom. Differences in image quality for the cavities in a and b result from the lower number of triple events available for reconstruction in a, as well as from the increased positron flight of ¹²⁴I compared with that of ⁸⁹Zr. d, Quantitative recovery of counts in each of the four cavities matches the activity measured during the preparation of the phantom, which shows that the reconstruction method can support the quantitation of two radiotracers simultaneously. e, A human-scale liver phantom with 21.8 MBq (590 μCi) ⁸⁹Zr dispersed in the 1.4 L liver, with a separated 6.5 mL spherical cavity containing 20.7 MBq (560 μCi) ¹²⁴I. PET image as acquired using the 350–700-keV energy window on a clinical Siemens Biograph mCT clinical PET CT and reconstructed with the traditional OSEM2D method. f, mPET separation of ⁸⁹Zr coincidences residing in the liver phantom, with slightly less signal where the spherical cavity phantom was located. g, mPET reconstruction of ¹²⁴I triple coincidences in the tumour sphere located on top of the liver. Results from a single preclinical or clinical phantom. Values reported in d represent mean μCi in each spherical cavity measured from voxel intensities in inside the 1 mL cavity region of interest. Error bars in d represent Capintec dose calibrator error of 5 percent from single activity measurements before addition to the phantom, while mPET error bars represent the SD of all of the voxel intensities in the region of interest. In a–c, %IA/CC represents percent injected activity per cubic centimetre; in e–g, SUV means ‘standard uptake value’.

We further tested the performance of mPET reconstruction by using a mouse bearing ¹²⁴I in the thyroid, and obtained similar images with a standard reconstruction as well as mPET (Supplementary Fig. 2b). Furthermore, the activity concentration and pixel position across the mouse thyroid could identically differentiate thyroid lobes with mPET as with the standard reconstruction (Supplementary Fig. 2c). The mPET reconstruction produces similar images to the standard reconstruction, even with multiple subjects present in the scanner (Supplementary Fig. 2d). Additional testing of the linearity of the mPET reconstruction method was done with increasing phantom wells for ¹²⁴I, in the range of 3.7–44 MBq (100–1200 μCi), and produced identical images for the standard reconstruction (Supplementary Fig. 3a), doubles by mPET (Supplementary Fig. 3b) and triples (Supplementary Fig. 3c) by mPET, where the ratio of triples to doubles did not significantly deviate across the scanner range (Supplementary Fig. 3d). The sensitivity and noise of the mPET method was tested as a function of total counts, with decreasing fractions of the total counts acquired, with 10% of the total counts producing similar images than full counts yet with increased noise (Supplementary Fig. 4a). A change in activity reconstructed was only observed below 20% of the total counts, as the number of triples approached that of the random triples observed in the scanner (Supplementary Fig. 3b). Overall, mPET provided increased image information density with little bias (<10 %), similar spatial resolution and contrast, and a small relative noise increase (<10 %) compared with that obtained with single-tracer PET acquisitions (Supplementary Figs. 2–4).

To test the effectiveness of our method in a clinical scanner, a Siemens Biograph mCT was used to record mPET images of a 3D-printed liver containing a background of ⁸⁹Zr ~22 MBq (~600 μCi) while a focal ‘liver lesion’ was filled with ¹²⁴I ~20 MBq (~550 μCi). Traditional PET reconstruction (Fig. 3e) yielded a ‘liver’ containing a diffuse amount of ⁸⁹Zr with a hotspot ¹²⁴I ‘lesion’. Upon separation and reconstruction, the ‘liver’ (Fig. 3f) and ‘lesion’ (Fig. 3g) signals could be easily differentiated in each channel, where the mPET reconstruction method could be applied to clinical scanning systems and workflows in addition to preclinical systems. We found that both preclinical and clinical systems could acquire data suitable for mPET separation, with preclinical performance nearly identical to that of the standard reconstruction methods, while providing two simultaneous isotope images. With mPET configured on both preclinical and clinical systems, we next prepared several preclinical in vivo experiments with mPET to address a series of biological questions.

Imaging two tumour markers with mPET

Molecular imaging has benefited the development of radiotracers as analogues to clinically approved drugs. The success of [¹⁸F]FDG as a molecular imaging tool in oncology exploits the Warburg effect, using the high GLUT1 activity to bring the radiotracer into a glycolytically active cell where it is entrapped¹. However, not all tumours are highly metabolically active, requiring additional molecular imaging agents for identification and staging. [¹²⁴I]I-trametinib is an imaging isotopologue of the MEK1/2 inhibitor trametinib³⁵, targeting proliferating tissue, and distinct from [¹⁸F]FDG targeting glycolytic activity. With traditional PET imaging, the combination of [¹²⁴I]I-trametinib with [¹⁸F]FDG traditionally would require different imaging sessions owing to isotope-decay overlap. Here mPET reconstruction can bring both imaging agents together to simultaneously determine the GLUT1 and MEK 1/2 targeting in melanoma tumours undergoing therapy. [¹⁸F]FDG was administered intravenously and imaged one hour post injection with a prior injection of [¹²⁴I]I-trametinib 24 hours prior. Upon separating the doubles ([¹⁸F]FDG, bottom row) from the triples ([¹²⁴I]I-trametinib, top row), two distinct biodistributions were observed (Fig. 4a). [¹²⁴I]I-trametinib uptake in untreated mice (left column) was seen in the GI tract as well as in the right-flank B16F10 tumour. Mice receiving cold trametinib prior to radiotracer injection (Fig. 4b) saw blocking in the abdomen and tumour, whereas treatment with a BRAF inhibitor upstream of MEK, vemurafenib, (Fig. 4c) yielded a partial reduction in [¹²⁴I]I-trametinib uptake in the tumour and GI tract. Combination therapy with both cold trametinib and vemurafenib (Fig. 4d) yielded a combination of [¹²⁴I]I-trametinib uptake from both the vemurafenib and trametinib treatment arms, with low uptake in the tumour and with some clearance in the GI tract. As melanomas are known to be [¹⁸F]FDG-avid, high tumour uptake was seen in all mice across treatment groups (Supplementary Fig. 5) alongside the heart, retina, bladder and brain. Quantitative ROI analysis of the tumours treated with trametinib show increased [¹⁸F]FDG uptake compared with untreated tumours and with tumours treated with vemurafenib and trametinib (Supplementary Fig. 5f), with axial slices of the mice shown in Fig. 4 (Supplementary Fig. 6). The increase in [¹⁸F]FDG uptake occurs when BRAF wildtype tumours are given BRAF/MEK inhibitor, leading to an increase in GLUT1 expression³⁶ (although this study only showed MEK inhibition). Additional mPET images of other mice in each arm can be found in Supplementary Fig. 5, and highlight the versatility and the unbiased separation of the reconstruction method. Overall, we found that dual small-molecule-drug imaging with mPET could be used to track tumour inhibition of MEK during therapy while also maintaining a standard PET [¹⁸F]FDG image for glycolytic activity, providing deeper insight into tumour metabolism in situ.

Fig. 4 |. mPET of two small-molecule radiotracers for enhanced therapy monitoring.

Mice bearing B16F10 melanoma tumours on the right flank were administered ~14.8 MBq (~400 μCi) [¹²⁴I]I-trametinib and ~7.4 MBq (~200 μCi) [¹⁸F]FDG 24 hours and 1 hour prior to PET imaging, respectively. mPET reconstruction separated both radiotracers. a, [¹²⁴I]I-trametinib in the tumour with some clearance in the gastrointestinal tract for an untreated mouse. [¹⁸F]FDG was observed in the tumour, bladder, heart and brain. mPET separated the two radiotracers present in the same mouse, quantitating MEK-inhibitor binding and glucose metabolism. b,c, In mice treated therapeutically with unlabelled nonradioactive trametinib (b) or with vemurafenib [¹²⁴I]I-trametinib (c), uptake was blocked and reduced, respectively, as expected for on-target and upstream MEK inhibition. d, Combination therapy of vemurafenib and trametinib showed a blend of [¹²⁴I]I-trametinib distribution from both individual therapy arms, although tumour uptake was also blocked as expected. By using mPET, the additional [¹⁸F]FDG or [¹²⁴I]I-trametinib information was available for each mouse. Quantitative coronal slices can be found in Supplementary Fig. 6. ROI analysis revealed higher [¹⁸F]FDG uptake (Supplementary Fig. 5f) in mice undergoing MEK-inhibitor therapy, compared with untreated, BRAF or BRAF+MEK0-treated mice (Supplementary Fig. 5). Here mPET enabled the non-invasive assessment of MEK and GLUT1 under therapy in the individual tumours. n=4 mice per treatment group, n=16 mice total. %IA/CC represents percent injected activity per cubic centimetre.

Imaging a drug and nanoparticle carrier with mPET

Nanoparticles have been widely used as formulation enhancers, carrying insoluble drugs, but also as agents to alter drug biodistribution, ideally lowering off-target delivery and its associated side effects. Examples are Doxil and nab-paclitaxel, and most recently mRNA delivery for SARS-CoV2 vaccination³⁷. Many studies rely on tracking just one component of the drug-loaded nanoparticle to assess delivery, usually the radiolabelled nanoparticle, and assume that the distribution of the drug in the nanoparticle is analogous to the distribution of the nanoparticle. With mPET, both drug and nanoparticle can be noninvasively monitored and quantified. Ferumoxytol, an iron oxide nanoparticle approved by the FDA for anaemia, can be loaded with various small-molecule cargo in its coating³³ and used in lymph-node-mapping studies³⁸ after delivery to the tumour. Ferumoxytol has been radiolabelled using a chelate-free and heat-induced labelling method, allowing PET imaging of ferumoxytol distribution in vivo^39,40. Trametinib (Mekinist) while effective, patient use is discontinued owing to on-target toxicity in the form of severe rashes and gastrointestinal distress⁴¹. Using mPET with [⁸⁹Zr]Zr-ferumoxytol loaded with [¹²⁴I]I-trametinib may elucidate the contribution of each component in nanoparticle delivery. We loaded [¹²⁴I]I-trametinib onto [⁸⁹Zr]Zr-ferumoxytol using a variety of methods, achieving a loading yield of around 20–30% with a bovine-serum-albumin coating and via co-loading with nonradioactive trametinib (Supplementary Fig. 7), allowing for direct quantitation of [¹²⁴I]I-trametinib post-loading and purification. Loaded [¹²⁴I]I-trametinib onto [⁸⁹Zr]Zr-ferumoxytol was then administered into a mouse bearing a B16F10 melanoma on the left dorsal foot pad, and imaged at 0.25, 2, 6, 12, 24, 48 and 96 hours using mPET. It became clear that soon after injection the [¹²⁴I]I-trametinib distribution (Fig. 5) did not match that of the carrier nanoparticle [⁸⁹Zr]Zr-ferumoxytol, meaning that, despite in vitro loading and stability suggesting the contrary, most of the drug had dissociated rapidly from the nanoparticle in vivo. Subsequent imaging timepoints revealed accelerated clearance of [¹²⁴I]I-trametinib by 24 hours and its absence at 96 hours (Extended Data Fig. 1). Although passive loading with this drug–nanoparticle combination was unsuccessful in vivo, mPET could separate the signals from the drug and the nanoparticle, allowing for quick confirmation of the instability of the loaded combination. Hence, mPET provides the ability to track both nanoparticle and payload, a process that currently cannot be done with one radiotracer alone in the same PET image and that is often only confirmed with a fluorescent or colorimetric dye acting as a drug surrogate.

Fig. 5 |. Visualizing nanoparticle delivery with mPET.

Biodistribution of ~7.4 MBq (~200 μCi) [¹²⁴I]I-trametinib loaded onto ~9.3 MBq (~250 μCi) ⁸⁹Zr-ferumoxytol in a mouse bearing a B16F10 melanoma on the right dorsal foot pad through 96 hours. Top, Traditional PET reconstruction shows distribution over 24 hours to mainly the liver and spleen, with some uptake in the dorsal foot pad and popliteal lymph node. Further imaging between 24 and 96 hours shows no noticeable change in biodistribution (Extended Data Fig. 1). Middle and Bottom, Initial mPET imaging identifies the distinct biodistribution of [¹²⁴I]I-trametinib and [⁸⁹Zr]Zr-ferumoxytol, suggesting that [¹²⁴I]I-trametinib was not retained on the ferumoxytol surface. Middle, [¹²⁴I]I-trametinib was observed to clear rapidly via hepatobiliary excretion, with no appreciable activity remaining after 24 hours. [¹²⁴I]I-trametinib loading was facilitated with cold trametinib; thus, the release of trametinib could also block the uptake of [¹²⁴I]I-trametinib in the dorsal foot pad, reducing the delivery of free [¹²⁴I]I-trametinib to the tumour. Bottom, Delivery of only [⁸⁹Zr]Zr-ferumoxytol to the dorsal foot pad tumour appears to have occurred by mPET. We used one mouse, and n=7 imaging timepoints in mPET. %IA/CC represents percent injected activity per cubic centimetre.

Tracking of CAR T cells and PSMA with mPET

There is increasing focus on ImmunoPET radiotracers to track immune populations such as T cells, to observe their infiltration or activation during therapy. Engineering of chimaeric antigen receptor (CAR) T cells to target a specific antigen and to include reporter genes represents a versatile imaging therapeutic system. We retrovirally transduced human T lymphocytes to express a tricistronic cassette (Extended Data Fig. 2a) comprising the PSMA-targeting second-generation CAR, the human sodium iodide symporter (NIS)³⁴ and the membrane-anchored Cypridina (maCluc) reporter genes⁹, thus enabling non-invasive in vivo therapeutic cell tracking by PET and bioluminescence imaging, respectively. CAR-T-cell antitumour activity and reporter function were assessed in vitro (Extended Data Fig. 2a–h, Supplementary Fig. 8), followed by in vivo bilateral tumour measurements post-administration of the CAR T cells (Extended Data Fig. 2i). Here mPET was used to separate the distribution of CAR T cells targeting PSMA-positive cells via ¹²⁴I while simultaneously measuring PSMA-positive tumour location and expression with [⁶⁸Ga]Ga-PSMA-11. Mice were administered ¹²⁴I two hours prior and [⁶⁸Ga]Ga-PSMA-11 one hour prior to mPET imaging. The mice had a PSMA-positive tumour (left shoulder) with an adjacent PSMA-negative tumour (right shoulder) as a control. With standard reconstruction, uptake was seen in the PSMA-positive tumour along with the organs endogenously expressing NIS (such as the thyroid and the stomach) and drug clearance (such as the kidneys and the bladder) (Fig. 6a). The separation of [⁶⁸Ga]Ga-PSMA-11 and ¹²⁴I shows that there were distinct distributions of [⁶⁸Ga]Ga-PSMA-11 in the PSMA-positive tumour, kidneys and bladder (Fig. 6b), whereas ¹²⁴I was found also in the PSMA-positive tumour, thyroid, kidneys, stomach and bladder (Fig. 6c). Individual axial slices through the PSMA-positive and PSMA-negative tumours show the distribution of ⁶⁸Ga-PSMA-11 and ¹²⁴I; both are located in the tumour, although with different intratumoural distributions. Bioluminescence imaging (BLI) of mice prior to mPET imaging agree with the CAR-T-cell targeting seen by ¹²⁴I imaging and with PSMA-positive tumour imaging with [⁶⁸Ga]Ga-PSMA-11. An expanded view containing unseparated and mPET reconstructed images for each mouse in the imaging cohort can be seen in Supplementary Fig. 9a, with overall distribution for each isotope and tumour slices (Supplementary Fig. 9b). A video overlaying [⁶⁸Ga]Ga-PSMA-11 and ¹²⁴I clearly shows the difference in distribution at two intensity levels (Supplementary Videos 1 and 2) whereas a geometric mean representation of the image product identified more similarities in radiotracer distribution than in the original separated differences (Supplementary Fig. 10). A bicistronic cassette without NIS was also generated using ⁸⁹Zr-Oxine for ex vivo CAR-T-cell labelling in combination with [⁸⁶Y]Y-DOTA-PSMA (Extended Data Fig. 3), showing similar PSMA-positive targeting by CAR T cells. Here the mPET reconstruction provided a method to simultaneously track distribution and targeting of CAR T cells to PSMA-positive tumours while confirming PSMA expression in vivo during the same PET scan.

Fig. 6 |. Tracking CAR-T cell therapy with mPET.

Mice bearing PSMA-positive (left) and PSMA-negative (right) tumours were administered CAR T cells bearing a tricistronic construct hNIS.P28z.exCLuc targeting PSMA and expressing the NIS reporter gene (Extended Data Fig. 2, Supplementary Fig. 8). At day-7 post CAR-T administration, BLI was performed to identify PSMA targeting to tumours; subsequently, ~10.4 MBq (~280 μCi) [¹²⁴I]I was added to identify the NIS-positive cells (that is, the distribution of CAR T cells). One hour after [¹²⁴I]I administration, ~13.7 MBq (~370 μCi) [⁶⁸Ga]Ga-PSMA-11 was administered to identify PSMA-positive tumour tissue, at which point mPET imaging was conducted. Traditional PET imaging (uncalibrated) shows high uptake in the kidneys, bladder and thyroid, owing to the normal distribution and clearance of both [¹²⁴I]I and [⁶⁸Ga]Ga-PSMA-11, with uptake in the PSMA-positive tumour. mPET imaging reveals CAR-T-cell distribution via [¹²⁴I]I and PSMA expression within the PSMA-positive tumours. Individual mice and isotope separations can be seen in Supplementary Fig. 9. Furthermore, an orthogonal CAR-T-cell mPET experiment using ex-vivo-labelled [⁸⁹Zr]Zr-oxine CAR T cells in combination with [⁸⁶Y]Y-DOTA-PSMA also confirmed targeting to PSMA-positive tumours (Extended Data Fig. 3). We used 5 mice for tricistronic mPET imaging of CAR T cells (data from individual mice are shown in Supplementary Fig. 9). We used one representative mouse in Extended Data Fig. 3. %IA/CC represents percent injected activity per cubic centimetre. BLI radiances are shown in p/s/cm²/sr (photons per second per centimetre per steradian).

Dual immunoPET identifying T-cell exhaustion with mPET

Although CAR T cells and clinically approved peptides or small molecules are of great utility for PET imaging, the use of radiolabelled antibodies for immunoPET can provide exquisite sensitivity and specificity as molecular imaging agents of membrane-bound antigens. These immunoPET agents often require days to circulate for maximal uptake in the tumour, although newer labelling systems such as the inverse electron demand Diels-Alder reaction (IEDDA) have decoupled antibody circulation with imaging distribution^42,43 to reduce off-target radiation exposure. As a molecular imaging tool, antibodies or fragments targeting the immune system and CD8⁺ T cells⁴⁴ have been developed to identify immune infiltration, known as pseudoprogression, in solid tumours during PD-1 immunotherapy⁴⁵. Newer targets defining subpopulation and activation states of CD8⁺ T-cells have proposed defining the state of T-cell exhaustion with tumour burden. More recently, the functional differences between progenitor-exhausted and terminally exhausted TILs have been shown to have distinct functional properties, such that under PD1 therapy a high ratio of progenitor exhausted TILs confers a greater overall survival⁴⁶. Progenitor T cells highly express Slamf6 (Ly108) and are absent of CD39, a marker of terminal exhaustion of TIM-3⁺ T cells. Over the course of PD1 therapy, exhausted T cells switch to a low expression of Slamf6 and high CD39 beyond 7 days⁴⁷. We aimed to test mPET imaging with the radiolabelled antibodies [⁸⁹Zr]Zr-DFO-CD39 and [¹²⁴I]I-Ly108 in a HKP1 immunocompetent lung-cancer model to classify the state of T-cell exhaustion, which may potentially make for a PET diagnostic for checkpoint-blockade immunotherapy⁴⁶. BLI imaging was done at day 3 to bracket mice into low, medium and high lung-tumour burden by s-rank for day-7 and day-14 post HKP1 implantation and mPET imaging. Day-7 imaging by mPET (48 hours after immunoPET administration) identified two different antibody distributions, with higher uptake of [⁸⁹Zr]Zr-DFO-CD39 in the liver and spleen, alongside expected dehalogenation of [¹²⁴I]I-Ly108, with free iodine in the unblocked thyroid (Extended Data Fig. 4a). Little to no differentiation in PET uptake was seen in the lungs between CD39 or Ly108 or by disease burden initially. However, by day 14 (Extended Data Fig. 4b), increased [⁸⁹Zr]Zr-DFO-CD39 uptake was seen in the lungs. [¹²⁴I]I-Ly108 uptake was also increased at day 14 compared to day 7 in the liver and the spleen and in the abdominal cavity. Within the day-14-imaged mice, [¹²⁴I]I-Ly108 uptake in general decreased in the thoracic cavity with increasing tumour burden by BLI. Although individual tumour nodules could not be readily identified with these radiotracers owing to the inherent lower resolution of the PET system, lung tumour nodules could only be resolved by CT starting at day 14. Further studies are needed to refine CD39 and Ly108 with other immunoPET pairs to improve immune-checkpoint diagnostics. Yet mPET provided dual immunoPET radiotracer imaging in vivo showing degrees of T cell infiltration and exhaustion with lung-cancer burden. The refinement of immunoPET alongside new labelling methods for isotope pairs may lead to improvements in mPET for the stratification of responders and non-responders undergoing chemotherapy or immunotherapy.

Outlook

Multiplexed PET is a generalized reconstruction method to separate two PET isotopes with a single PET acquisition solely on the basis of the type of gamma emission. mPET provides additional temporal information via the addition of a second tracer, and the method would greatly improve visualization when used with synchronized molecular imaging agents. By differentiating the additional isotope with the prompt gamma emission, mPET imaging can leverage many isotopes previously seen as problematic owing to their additional gamma emissions. In this study we have shown that both preclinical and clinical PET/CT systems can be easily adapted to perform mPET experiments while maintaining the quantitative power of PET. This is done without requiring any energy windowing to identify either isotope (as done with SPECT). We have used this method to address drug-loading and binary-distribution questions in addition to showing dual immune-marker quantitation. The mPET reconstruction method may allow researchers and clinicians to increase the density of information extracted from a single-PET imaging study. With increasing energy discrimination, mPET could lead to several-colour PET imaging, which would provide even richer images than the current single-colour state of PET imaging.

Methods

mPET imaging

Preclinical PET imaging was conducted on a Siemens Inveon PET/CT with a target acquisition count of greater than 60 million events per radiotracer per mouse, using an in house constructed four position mouse hotel. Acquisition times were adjusted between 20 and 30 minutes depending on the activity injected (typically 4.6–5.2 MBq per mouse total at time of imaging). The Inveon PET/CT scanner consists of a 20×20 array of LSO crystals with a spatial resolution of 1.4 mm. The maximum energy window was typically set to 700 keV to maximize the detection of high-energy prompt gamma rays for Iodine-124 and match the initial uniformity phantoms. The Inveon energy range can be expanded up to 814 keV to capture the largest number of events including the highest prompt gamma coincidence fraction⁴⁸, but would also include more random scatter events. Clinical PET Imaging was performed with the generosity of the Cornell Citigroup Biomedical Imaging Center and their Siemens Biograph mCT. Uniformity and Phantoms were acquired for 2400 seconds under 64-bit mode. mCT acquisition produced the following: Double Prompt = 282.7 million counts, Double Randoms = 29.3 million counts, Triple Prompts = 10.39 million counts, and Triple Randoms = 4.55 million counts. Attenuation was estimated from the coregistered CT image, scatter and spurious background with the fast Monte Carlo method MCGPU-PET which can be found online at Github https://github.com/DIDSR/MCGPU-PET. List-mode data was processed producing doubles and triples datasets (Supplementary Fig. 11). Double coincidences were reconstructed using the standard ML-EM algorithm (Supplementary Fig. 1b), while the triples used a modified version of ML-EM (Supplementary Fig. 1c) available online at Github at https://github.com/jlherraiz/GFIRST. The separated images were obtained with an iterative algorithm based on the calibrated sensitivity of detecting double and triple coincidences for each radionuclide (Supplementary Fig. 1d), resulting in the final separated double (Supplementary Fig. 1e) and triple images (Supplementary Fig. 1f). Schema of the mPET separation process can be found in Fig. 2.

Reconstruction quantitation

Quantitative PET²⁸ imaging applied corrections in attenuation, scatter, random, and spurious background from each isotope. Corrections are applied first to the sinogram of triple coincidences with an identical correction then applied to the doubles sinogram. Attenuation correction is obtained via the coregistered CT image while scatter and spurious background with the fast Monte Carlo method MCGPU-PET, and random triples are obtained directly from the delayed coincidences provided by the scanner as non-prompt-gamma coincidences are treated as random. With the energy resolution and energy window used, contribution of triples caused by inter-detector scatter²⁴ was low, and scatter effects from contiguous detectors were not considered. Typically, only a few percent of the coincidences are triples, but significantly higher abundance than random triples in the scanners tested. To mitigate the possible noise from a low triple coincidence rate, a bilateral guided filter was also applied during the triples reconstruction. Finally, separated images of each isotope can be recovered considering the differences in sensitivity of triples vs doubles in each image voxel as estimated prior in MCGPU-PET.

Computation time

List-mode data processing takes about 5 minutes for a 2 Gb file in 1 CPU. Next the generated double and triple sinograms are reconstructed using an adapted version of an image reconstruction code written in C++/CUDA called GPU-FIRST. 60 iterations are used with no subsets, requiring 5 minutes per sinogram (i.e., 10 minutes in total for double and triple sinograms). In this implementation, an initial reconstruction with only attenuation and randoms correction (i.e., without scatter and background correction) is applied, and then the simulator MCGPU-PET is used for the estimation of scatter and background. Therefore, the total reconstruction time is about 30 minutes per acquisition. Reconstruction time may be reduced by using subsets, estimating the scatter and background in the middle of the reconstruction (as it is commonly done in clinical scanners), instead of performing two reconstructions and performing the list-mode data processing with multiple CPUs.

Uniformity calibration and mouse phantom

A 1 L cylinder was used for uniformity measurement containing 5.2 MBq (140 μCi) of one of the following isotopes: ⁶⁸Ga, ¹⁸F, ⁸⁹Zr, ⁸⁶Y, ⁷⁶Br, and ¹²⁴I. Uniformity phantoms were measured for a minimum of two hours. The mouse phantom was 3D printed by the Memorial Sloan Kettering Cancer Center department of Medical Physics containing an empty body cavity in addition to four external 1 mL tumour cavities. 9.25 MBq (250 μCi) of ⁸⁹Zr was inserted into the main body cavity, while mixtures of ⁸⁹Zr and either ⁷⁶Br, ⁸⁶Y, or ¹²⁴I we added at the following amounts 0 / 2.22, 0.74 / 1.42, 1.42 / 0.74, and 2.22 MBq / 0 MBq. For the ¹²⁴I or ⁸⁶Y and ⁸⁹Zr combinations, the mPET energy range was increased from 300–650 keV to 300–700 keV, and the phantom imaged for 30 minutes.

Small-molecule mPET and animal handling

[¹⁸F]FDG was provided as need by the Radiochemistry and Molecular Imaging Probes Core and delivered at a concentration greater than 1 mCi per mL. Synthesis of [¹²⁴I]I-trametinib was prepared as described previously³⁵. Briefly, [¹²⁴I]I-trametinib was prepared from ~200 μg boronoato-trametinib precursor with 800 μg copper (II) chloride, between 37–444 MBq (1–12 mCi) of Iodine-124, 1mg of 1,10-phenanthroline in 4:1 methanol:water and heated for 30 minutes at 80 C. The reaction mixture was purified by HPLC (Supplementary Fig. 12a) on an Atlantis T3 4.6×150mm C18 analytical column over a 45 minute 5–95 % acetonitrile: water gradient with 0.01 % trifluoroacetic acid. Synthesis and separation yielded a specific activity greater than 100 MBq per μmol, and HPLC fraction buffer exchanged via sep-pack light C18 cartridge for elution in pure ethanol for subsequent dilution for injection in saline. c57bl/6j mice bearing melanomas on their right flank were administered ~14.8 MBq (400 μCi) of [¹²⁴I]I-trametinib intravenously and imaged at 24 and 48 hours later with the additional administration ~7.4 MBq (200 μCi) of [¹⁸F]FDG respectively one hour prior to imaging. c57bl/6j mice bearing B16F10 tumours (NCI-DTP Cat# B16F10, RRID:CVCL_0159) were separated into therapy groups given via intraperitoneal injection of saline, trametinib 0.6 mg kg⁻¹, vemurafenib 1.2 mg kg⁻¹, and trametinib 0.6 mg kg⁻¹ with vemurafenib 1.2 mg kg⁻¹ once daily for three days prior to radiotracer administration. Mice were fasted 10 hours before [¹⁸F]FDG administration. n=4 mice per group. The energy window used was 350–700 keV, and the mice were imaged for 30 minutes. The experimental procedures were approved by the Institutional Animal Care and Use Committee of Memorial Sloan Kettering Cancer Center under protocol 08–07-014, and all animals were housed and cared for by Veterinary Services with attention to animal care and research ethics. NOD SCID gamma male mice and c57bl/6j female mice were obtained from Jackson Laboratories, while athymic nude female mice (outbred) were obtained from Charles River Laboratories.

[¹²⁴I]I-trametinib and [⁸⁹Zr]Zr-ferumoxytol

Iodinated trametinib as described previously³⁵ was made from a bora-pinacol trametinib precursor via copper mediated insertion of ¹²⁴I. [⁸⁹Zr]Zr-ferumoxytol was produced via an adapted heat induced radiolabelling method⁴⁹ where neutralized ⁸⁹Zr from oxalate is mixed with purified ferumoxytol (600 μg per mouse/ 20 μL stock) from AMAG Pharmaceuticals, Inc. (Waltham, MA, USA) in 10 mM MES pH 8 buffer at 95 C for 2 hours. Subsequent purification and EDTA stripping removed unbound ⁸⁹Zr and purity was determined by ITLC to be greater than 94 % (Supplementary Fig. 12c). Individual radiotracers were injected in saline at an activity concentration of ~14.8 MBq (200 μCi) [¹²⁴I]I-trametinib, and ~9.3 MBq (250 μCi) ⁸⁹Zr labelled onto 900 μg ferumoxytol. Loading of [¹²⁴I]I-trametinib onto [⁸⁹Zr]Zr-ferumoxytol was achieved through the solvent diffusion method where [¹²⁴I]I-trametinib in DMSO is added dropwise into an aqueous solution of [⁸⁹Zr]Zr-ferumoxytol under rapid vortexing. PET acquisition for mPET was acquired using a 350–700 keV energy range for 30 minutes. Experimental procedures were approved by the Institutional Animal Care and Use Committee of Memorial Sloan Kettering Cancer Center under protocol 08–07-014, and all animals were housed and cared for by Veterinary Services with attention to animal care and research ethics.

T-cell exhaustion via mPET

Antibodies against murine CD39, clone Duha59, or slamf6 (Ly108), clone 330-AJ were obtained from Biolegend (San Diego, CA, USA). CD39 antibody was conjugated with p-SCN-Deferoxamine at pH 8.5–8.9 from Macrocyclics Inc.(Plano, TX, USA) and subsequently purified via PD-10 column in Chelex grade PBS and concentrated for injection with an Amicon Ultra-0.5 30 kDa concentrator. Conjugated antibodies were subsequently radiolabelled with Zirconium-89 after neutralization in 1 M HEPES and heated for 1 hour at 37 C and characterized by ITLC in an EDTA mobile phase. Labelled antibodies were found to have a radiochemical purity exceeding 99 % (Supplementary Fig. 12e) with a specific activity of greater than 19 MBq/100 μg. Iodination of slamf6 with Iodine-124 was achieved via the IODOGEN method. Radiolabelling after PD-10 purification yielded an ITLC purity greater than 95 % (Supplementary Fig. 12f) with a specific activity greater than 19 MBq/100 μg. Mice were administered 1.5 × 10⁵ HKP1 cells⁵⁰ intravenously for lung seeding and BLI tumour measurement at day 3 before S-rank sorting into groups for mPET imaging at day 7 and 14. mPET images were acquired at 24- and 48-hours post injection of 100 μg of [⁸⁹Zr]Zr-DFO-CD39 and [¹²⁴I]I-Ly108. The mPET energy range was acquired at 350–700 keV and mice imaged for 30 minutes. The experimental procedures were approved by the Institutional Animal Care and Use Committee of Memorial Sloan Kettering Cancer Center under protocol 08–07-014, and all animals were housed and cared for by Veterinary Services with attention to animal care and research ethics.

CAR-T tracking of PSMA-positive tumours

NOD SCID gamma immunocompromised mice were subcutaneously implanted on the top shoulder flank with PSMA-positive (left) and PSMA-null (right) PC3 human prostate cancer cells. CAR-T cells and PSMA-positive and PSMA-negative cancer cells were previously engineered with three independent bioluminescence reporters, namely membrane anchored Cypridina (maCluc), Click Beatle Green (CBG) and Renilla (Rluc) luciferases, respectively, thereby enabling the long-term in vivo BLI of all three populations within the same animal⁹. Characterization of CAR-T can be seen in Extended Data Fig. 2, with flow cytometry gating strategy in Supplementary Fig. 13. At 25 days post implantation, 1 × 10⁶ hNIS.P28z.exCLuc CAR-T were administered intravenously. In vivo tumour BLI was performed at day −4 and 18, whereas CAR-T BLI was performed 1- and 18-days post CAR-T injection and allowed a reliable assessment of CAR-T targeting and homing at the tumour. At 8 days post CAR-T administration 10.4 MBq (280 μCi) Iodine-124 (MSKCC radiochemistry and molecular imaging probes core or 3D imaging) was administered 2 hours prior to mPET imaging, while 13.7 MBq [⁶⁸Ga]Ga-PSMA-11 was administered in addition 1 hour prior to mPET imaging. [⁶⁸Ga]Ga-PSMA-11 was produced using Gallium-68 from an Eckert & Ziegler manual generator and radiolabelled with PSMA-11 as described previously⁵¹. The mPET energy range was acquired at 350–700 keV with the mice imaged for 30 minutes. ex-vivo labelling of CAR-T cells was performed based on previous ex vivo labelling methods for $\gamma \delta \text{T}$ and other immune cells^52,53 using a [⁸⁹Zr]Zr-Oxine labelling strategy of live cells. Briefly, Zirconium-89 was neutralized with sodium carbonate and diluted with water for addition of 10 mg/mL 8-hydroxyquinoline in chloroform as vortexed at maximum speed for 5 minutes. Solution was expanded with 450 μL of new chloroform and vortexed for an additional 10 minutes. The chloroform layer was extracted and dried under nitrogen at 60 C before reconstitution into 20 μL of DMSO. ITLC with ethyl acetate shows [⁸⁹Zr]Zr-Oxine product traveling with the mobile phase and a purity greater than 97 % (Supplementary Fig. 12d). NOD SCID gamma immunocompromised mice were implanted on the top shoulder flank with PSMA-positive (left) and PSMA-null (right) PC3 prostate cancer cells. At 4 weeks post implantation, [⁸⁹Zr]Zr-Oxine labelled CAR-T cells (709 model, 0.74 MBq (20 μCi) per 1 × 10⁷ cells) were administered by IV and imaged 24 hours post. Yttrium-86 (MD Anderson Cyclotron) was neutralized in 150 mM ammonium acetate buffer pH 5 before addition of DOTA-PSMA precursor and subsequent heating at 90 C for 30 min. ⁸⁶Y-DOTA-PSMA was purified by HPLC (Supplementary Fig. 12b) using an Atlantis C18 analytical column over a 5–95 % acetonitrile: water gradient with 0.1 % trifluoroacetic acid. HPLC fraction was buffer exchanged on an Oasis HLB cartridge before elution with ethanol, drying with argon, and resuspending in PBS. ⁸⁶Y-DOTA-PSMA 2.4 MBq (65 μCi) / 4 μg was administered 2 hours prior to the 24-hour CAR-T imaging timepoint via retroorbital injection. mPET imaging was conducted in a single 20-minute PET/CT scan using the modified energy window of 350–814 keV. Unique biological materials are available from the authors upon reasonable request.

Reporting Summary.

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Extended Data

Extended Data Fig. 1. Extended mPET imaging of nanoparticle biodistribution through 96 hours.

Including data from Fig. 5 and extending through 96 hours. No change in biodistribution seen by standard reconstruction or mPET between 24 and 96 hours. %IA/CC represents percent injected activity per cubic centimetre.

Extended Data Fig. 2. Characterization of the tricistronic CAR T-cell PET-reporter system.

Gene map of PET and reporter genes for CAR T cells containing the sodium Iodide symporter (hNIS), PSMA targeting scFv (P28z), and a BLI reporter system (exCLuc) to produce the CAR-Tcell hNIS.P28z.exCLuc. b, Flow plot of PC3 cells engineered lacking PSMA. c, PSMA-positive engineered cells. d, Flow plot of CAR T cells expressing the anti-PSMA-scFv. e, In vitro activation of the BLI reporter in tricistronic CAR T cells shows functional incorporation of the genes. f,Addition of 124I to wild-type or tricistronic CAR T cells shows an increase with only tricistronic CAR T cells, with mild blocking with sodium perchlorate, confirming the specific uptake of 124I through hNIS. g, Confocal imaging of wild-type and hNIS tricistronic CAR-T cells for nuclear staining, hNIS and WGA (wheat germ agglutinin). Scale bar, 5 μm. h, In vitro cytotoxicity of PSMA targeting CAR-T cells reduces cell population in PSMA-positive co-cultures, whereas PSMA-null cells were unaffected through 48 hours by the co culture with CAR-T cells. I, In vivo delivery of CAR T cells targeting PSMA-positive tumours led to reduced tumour growth in PSMA-positive tumours, whereas PSMA-negative tumours continued to grow. In e, f, h, I, the lines denote the mean with error bars representing the s.e.m. b-d, Representative flow-cytometry contour quadrant plot to show cell-intensity characteristics. The gating scheme can be found in Supplementary Fig. 11. e, Box plot representing min and max values, with the middle line as the mean. In f, h, n = 3 technical replicates per condition. In i, n = 5 mice per arm. In e,h, the P values are significant, using a multiple-comparison unpaired t-test and assuming the same s.d. in the population. %IA/CC represents percent injected activity per cubic centimetre.

Extended Data Fig. 3. Alternative ex vivo labelling of CAR T-cells with [89Zr]Zr-Oxine and PSMA imaging with [86Y]Y-DOTA-PSMA.

[86Y]Y-DOTA-PSMA / [89Zr]Zr-oxine CAR T-cell mPET was found to have uptake of both PSMA- 11 and CAR T -cell tracers in the PSMA-positive tumour (left), whereas minor to no activity was observed in both tracers for the PSMA-null tumour (right). Distribution of the [86Y]Y-DOTAPSMA tracer can be seen in the PSMA-positive tumour, in the ocular cavity where injected, and in the bladder during excretion. [89Zr]Zr-oxine CAR T-cells were observed in the PSMA-positive tumour as well as in the liver and bone. Images were calibrated to a maximal 2.5 %IA/CC for [86Y]Y-DOTA-PSMA and 7.5 %IA/CC for [89Zr]Zr-oxine CAR T-cells. We used one mouse. %IA/CC represents percent injected activity per cubic centimetre.

Extended Data Fig. 4. mPET for developing immunoPET detection of T-cell exhaustion.

Mice bearing HKP1 lung tumours were monitored for terminal and progenitor T-cell exhaustion using antibodies against CD39 or Ly108 during disease progression. BLI of lung tumour burden was measured prior to injection, with representative CT slice in thoracic cavity showing potential branch occlusions (bright spots off alveolar bifurcation). Ly108, a marker of effector T cells, decreased with disease burden whereas CD39, a marker of severely exhausted T cells, increased, according to flow-cytometry data. mPET imaging with [89Zr]Zr-DFO-CD39 and [124I]I-Ly108 at day 7 and day 14 with three levels of tumour burden. At day 7 no discernible difference between low, medium or high tumour burden in mice is seen with either [89Zr]Zr- DFO-CD39 or [124I]I-Ly108. By day 14, BLI shows an appreciable increase in tumour burden in all mice, with tumour occlusions seen on CT. By mPET there was a general increase in [89Zr]Zr- DFO-CD39 and [124I]I-Ly108 uptake in the lungs compared to the day-7 group. Less lung uptake was visible with [124I]I-Ly108, with increasing tumour burden in the day-14 imaged mice. mPET again could separate two radiotracers in vivo and could be used with further radiotracer engineering to define T-cell exhaustion. N = 3 mice imaged per week (n = 6 total cohort), with each mouse receiving simultaneously [89Zr]Zr-DFO-CD39 and [124I]I-Ly108 by intravenous injection 48 hours prior to mPET. %IA/CC represents percent injected activity per cubic centimetre.

Data availability

The main data supporting the results in this study are available within the paper and its Supplementary Information. All data generated in this study, including source data for the figures, are available from figshare with the identifier https://doi.org/10.6084/m9.figshare.21816069.