Head-to-head comparison of different classes of FAP radioligands designed to increase tumor residence time: monomer, dimer, albumin binders, and small molecules vs peptides

Purpose Fibroblast activation protein-α (FAP)-targeting radioligands have recently demonstrated high diagnostic potential. However, their therapeutic value is impaired by the short tumor residence time. Several strategies have been tested to overcome this limitation, but a head-to-head comparison has never been done. With the aim to identify strengths and limitations of the suggested strategies, we compared the monomer FAPI-46 versus (a) its dimer (FAPI-46-F1D), (b) two albumin binders conjugates (FAPI-46-Ibu (ibuprofen) and FAPI-46-EB (Evans Blue)), and (c) cyclic peptide FAP-2286. Methods 177Lu-labeled ligands were evaluated in vitro in cell lines with low (HT-1080.hFAP) and high (HEK-293.hFAP) humanFAP expression. SPECT/CT imaging and biodistribution studies were conducted in HT-1080.hFAP and HEK-293.hFAP xenografts. The areas under the curve (AUC) of the tumor uptake and tumor-to-critical-organs ratios and the absorbed doses were estimated. Results Radioligands showed IC50 in the picomolar range. Striking differences were observed in vivo regarding tumor uptake, residence, specificity, and total body distribution. All [177Lu]Lu-FAPI-46-based radioligands showed similar uptake between the two tumor models. [177Lu]Lu-FAP-2286 showed higher uptake in HEK-293.hFAP and the least background. The AUC of the tumor uptake and absorbed dose was higher for [177Lu]Lu-FAPI-46-F1D and the two albumin binder conjugates, [177Lu]Lu-FAPI-46-Ibu and [177Lu]Lu-FAPI-46-EB, in HT1080.hFAP xenografts and for [177Lu]Lu-FAPI-46-EB and [177Lu]Lu-FAP-2286 in HEK293.hFAP xenografts. The tumor-to-critical-organs AUC values and the absorbed doses were in favor of [177Lu]Lu-FAP-2286, but tumor-to-kidneys. Conclusion The study indicated dimerization and cyclic peptide structures as promising strategies for prolonging tumor residence time, sparing healthy tissues. Albumin binding strategy outcome depended on the albumin binding moiety. The peptide showed advantages in terms of tumor-to-background ratios, besides tumor-to-kidneys, but its tumor uptake was FAP expression–dependent. Supplementary information The online version contains supplementary material available at 10.1007/s00259-023-06272-7.


Introduction
The tumor microenvironment (TME) is a complex fundamental part of solid tumors [1] whose composition is different among patients. Nevertheless, there are common phenotype analogies among individuals [2,3]. Stromal cells and extracellular matrix are the main component of the TME, in which cellular infiltrates such as lymphocytes, macrophages, adipocytes, and fibroblasts are present [1]. Cancer-associated fibroblast (CAF) is one of the most abundant cell type in the TME, heavily contributing to the whole tumor mass [4]. CAFs are characterized by the expression of fibroblast activation protein-α (FAP), which is a type II transmembrane serine protease found in more than 90% of epithelial tumors such as breast, lung, colorectal, pancreatic, and ovarian cancer. FAP expression in healthy tissues and in non-malignant tissues surrounding the tumor is very limited, as confirmed by immunohistochemistry [5,6]. Thus, FAP has recently been identified as a pan-tumoral agent. A class of small molecule-based radioligands targeting FAP has emerged in the last few years for imaging of solid tumors [7,8]. The value of these radioligands has been illustrated in more than one hundred patients with unprecedented tumor-to-organ selectivity. Thus, FAP-targeting radioligands have been recently dubbed "potential novel molecule(s) of the century" [8,9].
While their potential as imaging agents is undeniable, their potential for therapy is harmed by the short retention in the tumor, leading to suboptimal tumor radiation doses and, thus, limited efficacy [7,[10][11][12][13]. A promising strategy to improve the tumor retention is via the increase of radioligand's avidity for its target by dimerization of the binding moiety [14][15][16]. Another strategy involves the introduction of an albumin binder moiety, such as Evans Blue, which increase the exposure of the tumor to the radioligand due to its higher blood circulation [17,18]. Alternatively, first-in-human results of the cyclic peptidic structure [ 177 Lu]Lu-FAP-2286 showed high and persistent uptake in primary and metastatic tumor [19,20].
While many preclinical and first-in-human clinical data have been generated with these radioligands, a comprehensive, comparative study to understand the strengths and limitations among the mentioned strategies has never been performed.
Here, we compared head-to-head representative FAPtargeting radioligands from each strategy that was proposed to prolong tumor residence time. More specifically, using [ 177 Lu]Lu-FAPI-46 as the reference small molecule, we compared it with (a) a dimeric version of it, (b) two conjugates of it with different albumin binders, and (c) the [ 177 Lu] Lu-FAP-2286, as the representative peptide-based radioligand. Head-to-head in vitro and in vivo assessments were performed using two cell lines characterized by low and high FAP expression, respectively. Our aim was to identify the strengths and limitations of the different strategies, namely, dimerization, albumin binder conjugation, and peptides vs small-molecule monomers, for the development of FAPtargeting radiotherapeutics.

Cell lines
HT-1080 and HEK-293 cells were transduced with the human FAP (hFAP). The production of lentiviral particles used for the transduction and the FACS gating strategy for the selection of monoclonal cell lines expressing FAP are provided in the "Supplementary information." HT-1080. hFAP, a polyclonal cell line with heterogeneous and low FAP expression, and HEK-293.hFAP, a monoclonal cell line with high FAP expression ( Supplementary Fig. 2), were used for the in vitro and in vivo evaluation. The two wildtype cell lines HT-1080.wt and HEK-293.wt were used to assess specificity.
Upon thawing, the cell lines were kept in culture in MEM supplemented with fetal bovine serum (10%) and penicillinstreptomycin (1%) at 37 °C and 5% CO 2 .

Affinity determination and in vitro cellular uptake
The IC 50 of all the ligands against isolated hFAP protein was assessed by an inhibition assay following published protocols [15,22] (Supplementary Information and Supplementary Fig. 3).
Cellular uptake and distribution were assessed in FAPpositive cell lines at different times points (15 min, 1 h, and 4 h) after exposure to the radioligand at 37 °C. Wild-type cells were used to assess unspecific uptake. Details are provided in the "Supplementary information" (Supplementary Fig. 4 and Supplementary Table 2).

Animal studies
All animal experiments were conducted in accordance with Swiss animal welfare laws and regulations under the license number 30515 granted by the Veterinary Office (Department of Health) of the Canton Basel-Stadt. Female athymic nude-Foxn1 nu /Foxn1 + mice (Envigo, Netherlands), 4-6 weeks old, were used for generating FAP( +)/FAP( −) dual xenografts. Mice were implanted subcutaneously with 5-12 × 10 6 FAP( +) and FAP( −) cells suspended in 100 μL PBS on the right and left shoulder (imaging studies) or right and left flank (biodistribution studies), respectively. The tumors were allowed to grow until reaching a volume of 100-200 mm 3 .

SPECT/CT imaging studies
SPECT/CT images were acquired using a dedicated nano-SPECT/CT system (Bioscan, Mediso, Budapest, Hungary). Mice were injected intravenously via the tail vein with ~ 9-15 MBq (500 pmol) of the radioligand and euthanized after 4 h. Details on image acquisition and reconstruction parameters are described in the "Supplementary information."

Biodistribution studies and AUC analysis
Mice were randomized (4-5/group), injected intravenously with the radioligand (100 µL/500 pmol/0.8-1 MBq), and euthanized at different time points (4 h, 24 h, 72 h, 120 h for HT-1080 and 4 h, 24 h, 72 h for HEK-293 xenografts, respectively) by CO 2 asphyxiation. Organs of interest and blood were collected, rinsed of excess blood, blotted dry, weighed, and counted in a γ-counter. The samples were counted against a suitably diluted aliquot of the injected solution as the standard, and the results were expressed as the percentage of the injected activity per gram of tissue (%I.A./g) ± standard deviation (SD).
The area under the time-activity curves (AUC) in the tumors were generated from the biodistribution data and expressed as (%I.A./g)*h. AUC of tumor-to-critical-organs ratios were also generated. The calculations were performed using GraphPad Prism 9. Ninety-five percent confidence interval (95% CI) and statistical analysis (p values) of the AUC data are presented in the "Supplementary information."

Dosimetry
Non-decay corrected mice biodistribution data were used to generate time-activity curves for each radioligand. OLINDA/EXM 1.0 was used to integrate the fitted timeactivity curves and to estimate the tumor doses and organ doses using the whole-body adult female model, as previously described [23]. For all calculations, the assumption was made that the mouse biodistribution, determined as the %I.A./organ, was the same as the human biodistribution.
All 177 Lu-labeled ligands were prepared with apparent molar activities ranging from 8 up to 36 MBq/nmol, depending on the study, and radiochemical purity ≥ 93%. The
All FAPI-46-based radioligands were internalized (35 up to 80% of the applied radioactivity at 4 h), with a minimum amount remaining on the cell surface (0.9 up to 4% at 4 h). [ 177 Lu]Lu-FAP-2286 displayed different cellular distribution, with rather low internalization (20-30% of the applied radioactivity at 4 h), and high cell surface binding  Table 2).

SPECT/CT imaging
The visual assessment of SPECT/CT images at 4 h p.i.

Biodistribution and AUC
Biodistribution results generated by gamma-counting harvested organs are provided in Tables 1, 2  The AUC of the tumor uptake over time was assessed from the biodistribution data as a surrogate of the radiation dose delivered to the tumors (Fig. 3A  The AUCs of the tumor-to-critical-organs ratios were assessed as indicators of the therapeutic index (Figs. 4 and 5). With the exception of the tumor-to-kidneys ratio, [ 177 Lu]Lu-FAP-2286 presented the most favorable tumorto-critical-organs ratios over time in both tumor models, as indicated by the AUC of tumor-to-blood, tumor-to-liver, and tumor-to-femur ratios.

Dosimetry
The dosimetry estimates for the critical organs and the FAP-expressing tumors are reported in Table 6

Discussion
The therapeutic value of FAP-targeting radioligands is harmed mainly due to their short tumor residence time [7,11,24]. Three different strategies have been proposed for prolonging tumor residence time: (a) multimerization of the FAP-binding moiety [15,16,25], (b) conjugation of an albumin binder [17,18,[26][27][28], and (c) peptide-based structures as an alternative to small molecules [19,20]. Nevertheless, a direct comparison among them is still missing. We, therefore, synthesized and tested head-to-head a panel of FAP radioligands representing all the above-mentioned strategies, including the new albumin-binder conjugate [ 177 Lu] Lu-FAPI-46-Ibu. Our aim was to identify the strengths and limitations of the different strategies that need to be considered in the development of FAP-targeting radiotherapeutics.
In vitro, all radioligands showed very high affinity to hFAP, with a certain variation among them, and IC 50 values in the picomolar range. This allowed a fair comparison in vivo regarding FAP-targeting. Differences were observed in their cellular distribution. All [ 177 Lu]Lu-FAPI-46-based radioligands were almost entirely internalized, while [ 177 Lu] Lu-FAP-2286 remained mainly on the cell surface.
Focusing on the first strategy of dimerization, [ 177 Lu] Lu-FAPI-46-F1D presented a higher and more persistent uptake in the tumor, compared to the monomer [ 177 Lu]Lu-FAPI-46, independent of the tumor model. This is in line with the recently published studies on FAP-targeting dimers, such as BiOncoFAP [15], DOTAGA.(SA.FAPi) 2 [16,29,30], DOTA-2P(FAPi) 2 [25,31], and ND-bisFAPI [14]. Undoubtfully, this observation supports the use of multimers for FAP-targeting radiotherapeutics per se. Evidently, total body distribution and pharmacokinetics are just as important as tumor uptake. In our study, it was shown that dimerization doubled the radiation dose delivered to the tumor, but also increased the dose to non-targeted organs, especially in blood, femur, liver, and kidneys, and the overall background activity, suggesting higher toxicity.  healthy organs (e.g., colon and kidneys) [16]. No data about the therapeutic efficacy are available so far [16].
Focusing on the second strategy of the albumin binder conjugation, our study indicated that the outcome heavily depends on the albumin binder moiety of choice.   [20]. The same study showed that the tumor half-life of [ 177 Lu] Lu-FAP-2286 is shorter than the above-mentioned approved radiotherapeutics, even though longer compared to the FAPI-based small molecules [20].
Last, but not least, we tried to understand the discrepancy between HT-1080.hFAP and HEK-293.hFAP on the in vivo uptake of [ 177 Lu]Lu-FAP-2286. Our initial hypothesis was that a saturation level was reached in HT-1080.hFAP tumors with the injected mass of 500 pmol used in the study, given the low expression level of FAP. We, therefore, evaluated the biodistribution of [ 177 Lu]Lu-FAP-2286 and of [ 177 Lu] Lu-FAPI-46 in HT-1080.hFAP xenografts using tenfold less amount (Supplementary Table 11). The results using 50 pmol instead of 500 pmol indicated that no saturation was reached. To determine the saturation effect on the two tumors, an ex vivo blocking study was performed for [ 177 Lu] Lu-FAP-2286 and [ 177 Lu]Lu-FAPI-46 in both tumor models. In each case, 60-fold excess of the non-labeled ligand was administered 5 min before the injection of the corresponding radioligand. While in the HT-1080.hFAP tumors a complete inhibition of the radioligand uptake was observed, in the high-expressing FAP cell line HEK-293.hFAP tumors the inhibition was lower, still significantly lower when compared to the radioligand uptake without the blocking (Supplemental Fig. 5 and 6). These results indicate that the HEK-293. hFAP model has available amounts of hFAP that require more than 30.5 nmol of FAP ligands to be completely occupied. The in vitro autoradiography performed on HT-1080. hFAP and HEK-293.hFAP tumor slides after incubation with [ 177 Lu]Lu-FAPI-46 and [ 177 Lu]Lu-FAP-2286 with and without the presence of 10,000-fold excess of the non-labeled ligand (Supplemental Fig. 7) corroborated these results. In addition, the autoradiography confirmed the difference observed in vivo between the [ 177 Lu]Lu-FAP-2286 and [ 177 Lu]Lu-FAPI-46 in the two tumor models. Our second hypothesis was that the two classes of the studied ligands, FAPI small molecules and a cyclic peptide, may present different binding sites as they are structurally very different. To test this hypothesis, we performed some preliminary in vitro experiments on cell membranes. We observed lower blocking efficiency when FAPI-46 was used to block the binding of [ 177 Lu]Lu-FAP-2286, compared to its efficiency to block [ 177 Lu]Lu-FAPI-46 (data not shown). This is an indication that the two ligands might present different binding sites. However, further and more sophisticated experiments have to be designed for testing this hypothesis.
The presented results on two cell lines with distinct expression and homogeneity levels of FAP (polyclonal vs high-expressing monoclonal, Supplementary Fig. 3) underlined the importance of the tumor model in assessing FAPtargeting ligands. Different target density on the cell surface may have a profound impact on the receptor occupancy, affecting the total uptake of the radioligand [32]. Moreover, since FAP is known to be active upon homodimerization, a higher receptor density may promote oligomerization, affecting the radioligand binding [33]. Furthermore, it is known that the glycosylation pattern can vary among different cell lines expressing the same protein, rendering the binding site of radioligands less accessible [34]. Complementary to our second hypothesis, we may speculate that homo/oligo-merization and/or glycosylation pattern is more relevant for the binding of the FAP-targeting peptide-based structures than the quinoline-based small-molecule inhibitors. This might explain why the uptake of the [ 177 Lu]Lu-FAP-2286 was significantly impacted by the FAP-expression level and density, which was not the case for the FAPI-46-based radioligands. Nevertheless, as far as we know, no data are available in the literature to support this hypothesis. Finally, using cell lines with distinct characteristics and FAP expression levels may elucidate the interactions of structurally different radioligands with FAP.
To the best of our knowledge, this is the only study so far providing a fair comparison among the different structural designs. We choose representative radioligands from each strategy with very similar behavior to corresponding radioligands reported in the literature [15,18,19,21]. The results captured the typical features of each strategy design that impart to the targeting ligand and give hints for the design of FAP-targeting radio-therapeutics.
In conclusion, this head-to-head comparison indicated that dimerization of the FAPI small molecules and the cyclic peptide are two very promising strategies for enhancing tumor radiation dose, compared to FAPI monomers. In addition, the present study indicated that the therapeutic outcome of using albumin binders heavily depends on the selection of the albumin binding moiety. Considering the combination of tumor radiation dose (tumor uptake and residence), in vivo specificity, and tumor-to-background ratios (therapeutic index), the peptide showed certain advantages. However, the discrepancy of its performance between the different tumor models needs further investigation for concluding on any overall superiority compared to the other strategies and to FAPI small molecules.