Development of PET Radioligands Targeting COX-2 for Colorectal Cancer Staging, a Review of in vitro and Preclinical Imaging Studies

Colorectal cancer (CRC) is the second most common cause of cancer death, making early diagnosis a major public health challenge. The role of inflammation in tumorigenesis has been extensively explored, and among the identified markers of inflammation, cyclooxygenase-2 (COX-2) expression seems to be linked to lesions with a poor prognosis. Until now, COX-2 expression could only be accessed by invasive methods, mainly by biopsy. Imaging techniques such as functional Positron Emission Tomography (PET) could give access to in vivo COX-2 expression. This could make the staging of the disease more accurate and would be of particular interest in the exploration of the first metastatic stages. In this paper, we review recent progress in the development of COX-2 specific PET tracers by comparing the radioligands' characteristics and highlighting the obstacles that remain to be overcome in order to achieve the clinical development of such a radiotracer, and its evaluation in the management of CRC.


INTRODUCTION
Colorectal cancer (CRC) is one of the most frequent types of cancer, ranking second in most developed countries, and has the second highest mortality rate (1,2). This rate is especially high for late diagnosis and advanced stage disease. Thanks to an increase in CRC screening, the incidence rate has been decreasing for the last two decades (2). However, this high mortality rate draws early detection of CRC and prediction of recurrences and metastases as two major public health challenges, in order to initiate the appropriate treatment as early as possible.
Inflammation seems to play a key role in CRC physiopathology. Among the many known markers of inflammation, cyclooxygenase-2 (COX-2) was identified as having a crucial role from the first stages of tumorigenesis (3,4). Reviews of over 30 epidemiologic studies showed that regular consumption of non-steroidal anti-inflammatory drugs (NSAID) was associated with a 30-50% reduced incidence of several cancer types, including CRC (5,6). Interestingly, patients with familial adenomatous polyposis receiving COX-2 inhibitors as a preventive treatment developed fewer adenomas than patients receiving a placebo (7). COX-2 is an enzyme that intervenes in the first steps of prostaglandin E2 synthesis from arachidonic acid and can be induced by various pro-inflammatory signals. COX-2 overexpression has been identified in different malignant neoplastic tissues, especially in up to 85% of adenocarcinomas (8)(9)(10) and is also associated with the potential for progression and recurrences of colorectal tumors. Molecular biology studies on CRC tissue samples concluded that COX-2 expression is significantly correlated with invasive (11) and metastatic phenotypes (12,13). Moreover, high COX-2 expression in patients treated with radical surgery is a prognostic factor for recurrences, mainly because of undetected metastasis (13,14). The COX-2 expression level is high in CRC cells. On the contrary, COX-2 has a low basal expression in healthy colon epithelial cells (15), making COX-2 a relevant biomarker for molecular imaging, especially for Positron Emission Tomography (PET).
The correlation of elevated COX-2 expression with the first stages of tumorigenesis and cancer progression suggests that COX-2 could be a target for the early imaging of pre-cancerous colorectal lesions. Since inflammation seems to be a hallmark of malignant CRC, PET imaging could allow the early detection of CRC dissemination and thus spot metastasis. In addition, in other cancerous pathologies (breast, prostate), COX-2 expression increases with the stage of the disease (16,17). PET-scanning could therefore be a tool for CRC staging. Furthermore, positive therapeutic results in cohorts of patients with different types of solid tumors receiving COX-2 inhibitors in addition to their respective standard chemotherapy (18)(19)(20) suggest that COX-2 molecular imaging could provide stratification of the patients, enabling individualized treatment approaches. Notably, it could help to decide whether neoadjuvant chemotherapy is needed or determine whether patients would respond to COX-2 inhibitors therapy.
Up to now, numerous radioligands targeting COX-2 have been developed for SPECT or PET applications. Reviews by Laube et al., de Vries et al., and Pacelli et al., summarized their structures and synthesis methods and highlighted the challenges encountered in the development of COX-2 radiotracers (21)(22)(23). In the present review, we explore the recent progress of COX-2 molecular imaging in CRC, by comparing the radioligands' characteristics and highlighting the obstacles that remain to be overcome in order to achieve the clinical development of such a radiotracer.
The PubMed database was screened using pre-defined search dates (January 1995-January 2021). The search terms used were as follows: "colorectal cancer" and "COX-2" or "cyclooxygenase 2" and "PET." It yielded 78 results. We screened preclinical results performed only on colorectal cancer cell lines. Title-, abstract-or full text-reading led to the exclusion of 70 papers because they did not focus on CRC or did not include in vivo PET radioligand evaluation or did not target COX-2. Eight original papers fulfilled these criteria and were thus included in the scope of the present review. Chemical structures of the PET radioligands are presented. The main conditions and results of these PET imaging preclinical studies assessing COX-2 radiotracers in CRC xenograft models are summarized in Table 1.

COXIB FAMILY DERIVATIVES
Celecoxib 1 (Figure 1) is a well-known COX-2 specific inhibitor used as an anti-inflammatory drug. Its specificity and selectivity for COX-2 over COX-1 makes it a natural candidate for PET applications. In 2005, Prabhakaran et al. synthesized 11 C-Celecoxib 2 (Figure 1) in 8 efficient steps including a palladium catalyzed radiomethylation (32), but this radioligand has not yet been tested in a CRC preclinical model. A biodistribution study in baboons showed unsatisfactory pharmacokinetic properties. 11 C-2 underwent a fast metabolism (80% after 30 min in plasma samples) and was quickly cleared out of the organism by urinary excretion (33). Therefore, several series of celecoxib derivatives were proposed to work around the limitations of celecoxib. In 2011, Uddin et al. developed a series of celecoxib [ 18 F]-fluorine derivatives (24). Synthesis in 7 steps involving an aliphatic nucleophilic substitution led to the best candidate. Diethylaminosulfurtrifluoride (DAST) mediated fluoration of its alcohol precursor gave the fluoromethyl derivative 18 F-3 ( Figure 2). In inhibition assays using purified COX-2, 18 (Figure 3). Derivative 4 exhibited interesting in vitro properties on human recombinant COX-2 (log P = 3.18; IC 50 = 360 vs. 40 nM for 1) but suffered from a lack of specificity. Indeed, cellular uptake experiments in human CRC cells expressing COX-2 (HCA-7) showed an accumulation of the radioligand that could not be inhibited by a celecoxib or rofecoxib pre-treatment. In vivo, NIH-III mice xenografted with HCA-7 cells exhibited a maximum tumor to muscle ratio 10 min post injection (p.i.) of 1.4, slowly decreasing over time. Dynamic acquisitions showed a rapid elimination of the radioactivity through the intestinal tract. In vitro results were not convincing, and specificity was not investigated in vivo.
Other teams focused on different coxib-like derivatives.      Biodistribution results showed a %ID/g of 2.12% in HCA-7 tumors, scaled down to 1.04% after a celecoxib pretreatment. However, radiotracer uptake was similar in HCA-7 and HCT-116 tumors. Expression of COX-2 in HCT-116 tumors was confirmed by immunohistochemistry, highlighting a cell line drift and questioning the relevance of the HCT-116 cell line as a COX-2 negative model. They concluded that 18 F-pyricoxib 6 was a very promising candidate for a "first in human" study (27). In 2018, Tietz et al. deepened their understanding of this chemical entity by performing pharmacomodulation experiments (28). For the O-analog 18 F-7 the radiolabeling was carried out on a preoxidized iodyl precursor. The authors started from 4chloro-2-(methylsulphonyl)-6-(trifluoromethyl)pyrimidine after 5 steps including radiolabeling with 4-[ 18 F]fluorobenzylamine. 18 F-8 radiosynthesis was similar to that of 18 F-pyricoxib 6, using sulfonyl precursors and a 4-[ 18 F]fluorobenzylamine (FBA) as building block (Figure 5). The three tested probes 18 F-6-8 exhibited similar in vitro properties (lipophilicity, affinity, specificity for COX-2). However, even slight modifications of the structure led to radically different pharmacokinetic, biodistribution and uptake profiles ( Table 1). One of the radiotracers was slightly metabolized and excreted via the urinary tract but did not accumulate in the tumors, while the other was rapidly eliminated but had a better T/M ratio.
The same team developed a pyricoxib derivative to reduce this unspecific binding previously observed with 18 F-pyricoxib 6 (29). They designed 18 F-triacoxib 9 which was obtained from the Celecoxib 1 structure by replacing the CF 3 by a CH 3 group and the acidic sulfonylamide by a simple methylsulfonyl. Nevertheless, the best innovation in this series consisted in the insertion of a triazole between the pyrazole and the phenyl ring, which required building an azidopyrazole as key intermediate. Noteworthy, 18 F-fluorination was achieved using a pinacol boronic ester (in the presence of a copper catalyst [Cu(OTf) 2 (py) 4 ]) as leaving group (Figure 6). Compound 9 displayed a satisfactory in vitro affinity and selectivity for COX-2 (IC 50 = 90 vs. 70 nM for 1; IC 50 COX-1 > 100 µM for both compounds, HCA-7 cells). In vitro binding experiments with increasing doses of non-radiolabeled compounds (celecoxib or triacoxib) resulted in an inhibition of the radiotracer uptake by 47 to 63%. In vivo experiments on BALB/c mice bearing HCA-7 xenografts suggested an improved stability of 18 F-triacoxib 9 compared to 18 F-pyricoxib 6 (respectively, 90 vs. 75% intact radiotracer in plasma samples 60 min p.i.). PET acquisitions confirmed a higher uptake in the tumors than in the muscle, within a similar range as 18 F-pyricoxib 6 (tumor-to-muscle ratio of 1.49 vs. 1.47, respectively, 60 min p.i.). In Celecoxib 1 pretreated mice, a partial inhibition of 18 F-triacoxib 9 uptake was observed (20%). Unspecific binding was also observed in lipid rich tissues. 18 F-triacoxib 9 did not seem to be a better candidate than 18 F-pyricoxib 6 as a diagnostic tool for CRC.

2,3-DIARYL INDOLES
In 2003, Hu et al. synthesized a large series of original 2,3-diaryl indoles (36). Structure-Activity Relationships (SAR) were established from about thirty derivatives obtained from anthranilic acids in 5 or 6 steps including indole formation by McMurry cyclization (Figure 7). Derivative 10 exhibited high affinity and selectivity for COX-2 (IC 50 = 20 vs. 520 nM for 1) over COX-1 (IC 50 >10 µM for both compounds), and better anti-inflammatory properties than celecoxib in a rat carrageenan-induced foot pad edema assay. In 2012, Kniess et al. radiolabeled this promising compound using the nucleophilic substitution of a trimethylammonium salt by 18 F-fluorine (30). They investigated the determination of PGE2 levels in cell culture supernatants, which is therefore a direct measure of the COX

INDOMETHACIN DERIVATIVES
Indomethacin 11 (Figure 8) is a COX ligand that binds to both COX-1 and COX-2. Kalgutar et al. modified its chemical structure by adding an amide group (37). After SAR analysis and the building of nearly twenty analogs, they clearly improved the selectivity of the derivatives for COX-2, which was inhibited in the nanomolar range. Following this work, Morgenroth et al. synthetized radio iodinated analogs from stannyl precursors (31). Derivative 125 I-12 (Figure 9) was found to be the most suitable tracer. In vitro studies on CRC cell cultures HT-29 and HCT-116 showed a significantly higher concentration of 12 in COX-2 positive (HT-29) cells (COX-2+/COX-2-ratio of 2.5 and 7.6 for the two tested radiotracers). Blocking experiments were performed on human embryonic kidney (HEK) and human umbilical vein endothelial cell (HUVEC) cell lines, showing a partial decrease of the uptake in cells pre-treated with celecoxib 1. These affinities and selectivities were confirmed using both in vivo microPET and bio-distribution studies. The compound with the highest uptake was radiolabeled with 124 I and used for PET imaging of HT-29 and HCT-116 xenografted mice. Due to the lipophilicity of this compound (log D = 4.41), the radioactivity was mainly localized in the liver and the gastrointestinal tract. No tracer uptake was observed in HCT-116 tumors whereas HT-29 tumors displayed a significant uptake. Measurement of the remaining activity in each organ with a gamma counter confirmed a COX-2+/COX-2-ratio of 5. The T/M ratio was over 50 times higher in HT-29 tumors than in HCT-116 tumors.
In 2011, Uddin et al. also synthesized a series of fluorinecontaining indomethacin derivatives. These probes were satisfactory COX-2 inhibitors with an IC 50 in the range of the nanomolar both in inhibition assays using purified COX-2 and intact cells. However, they could not be successfully radiolabeled due to a lack of stability in the severe reaction conditions required (24).

DISCUSSION
This review focuses on COX-2 radiotracers tested in CRC models. Other COX-2 radiotracers have also been developed for other oncologic indications as well as applications in the field of neuroinflammation (22).
These eight studies on CRC xenografts display heterogeneous results. Kaur  However, for all the tested compounds, most of the radioactivity uptake was located in the digestive tract. The T/M ratios were deemed satisfactory, ranging from 1.4 to 5.0, but the tumor-to-intestine ratio was not always mentioned in these studies, and, when available, was well above the T/M ratio. The tumors were detectable as xenografts in mice flanks, but these T/M ratios would probably not be sufficient to detect in situ CRC tumors (Figure 10). Thus, the challenge to develop the ideal COX-2 radiotracer for CRC staging remains to date. Two areas of work can be explored to achieve this radiotracer development for CRC imaging.

Radiotracer-Linked Properties
All the tested compounds display a high lipophilicity (log P or log D ǫ [1.7-4.4]), required to reach their intracellular target COX-2, but also responsible for hepatobiliary excretion that could mask CRC tumors. This assumption questions the relevance of COX-2 as a CRC imaging target and could be tested using in situ CRC preclinical models. 11 C-celecoxib was deemed unfit for clinical applications but benefited from a renal excretion that would be a serious advantage in CRC imaging. To increase the affinity for COX-2, pharmacomodulations on this parent compound led to more lipophilic molecules. There is still a need to balance affinity and lipophilicity with ideal pharmacokinetic properties. The use of innovative vectors could prove to be the solution to selectively deliver radiolabeled coxib derivatives into cancerous cells. In particular, the use of the RGD sequence (a tripeptide composed of arginine, glycine and aspartic acid) (38), dendrimers (39) or dehydropeptides (40) proved their potential to carry COX inhibitors into inflammatory or cancerous cells. Other pharmacokinetic properties should be taken into account, especially the metabolization rate and the radiotracer half-life. To be used routinely, the ideal radiotracer should be more than 50% intact 1 h p.i. and the half-life of the vector should not be shorter than 30 min. Other well-known PET tracers, such as 18 F-FDG or 18 F-FDOPA comply with these requirements (respective renal excretion of 20% 2 h p.i. 1 and 50% 45 min p.i. 2 ).
The second parameter to consider is the radioisotope. Due to its 110 min half-life, the use of 18 F is widespread in PET imaging, and most of the COX-2 binders were labeled with 18 F. Morgenroth et al. obtained the maximum T/M with a radio-iodinate derivative (31). However, due to both the higher energy of emitted beta particles and the longer half-life of 124 I (4.2 days), radio-iodinate derivatives exhibit a poorer dosimetry profile compared to 18 F-compounds. On the contrary, the shorter half-life of 11 C makes it less accessible for routine use (onsite cyclotron required) but would endow it with a better dosimetry profile.
Finally, pharmacomodulations were performed to increase the specificity for COX-2 compared to COX-1, but other targets are known for coxibs. For instance, celecoxib or valdecoxib bind significantly (IC 50 in the order of nM) to carbonic anhydrase isoenzymes (CA) (41). In addition, celecoxib is able to bind to PDK1 (42,43), a cell survival regulation enzyme via the Akt/PKB pathway, with an IC 50 in the range of µM (44). Likewise, coxibs bind to the transmembrane protein SERCA (45)(46)(47), a pump that can induce cellular apoptosis by increasing the intracellular calcium concentration. These different proapoptotic and anti-oncogenic targets have been confirmed by an increasing number of reports indicating that celecoxib does not require the presence of COX-2 to exert its anti-tumoral activity (44,48,49). Even more striking, it has been shown that structural analogs close to celecoxib, devoid of any COX-2 inhibitory activity, were able to mimic the anti-tumor properties of celecoxib studied so far, not only in vitro but also in 1 Résumé des Caractéristiques du Produit FDG. Available online at: http://agenceprd.ansm.sante.fr/php/ecodex/rcp/R0187540.htm (accessed June 9, 2020). 2 Résumé des Caractéristiques du Produit FDOPA. Available online at: http:// agence-prd.ansm.sante.fr/php/ecodex/rcp/R0157439.htm (accessed October 6, 2020). various in vivo xenograft models (43,50,51). These data question the relevance of radiolabeled coxibs as specific COX-2 binders.

In vivo Xenograft Models
Xenografts are known to lead to derangement of the normal tumor architecture and nearby healthy tissues and vasculature, and cause altered drug-sensitivity (52,53). To address these possible biases, the use of orthotopic xenograft or genetically engineered mice expressing human colorectal cancer genes would mimic in situ CRC more accurately (54), easing the evaluation of the impact of unspecific intestinal uptake. Imaging the xenografts with 18 F-FDG prior to the radiolabeled COX-2 inhibitor would also inform about the accessibility of tumors for PET radioligands, notably regarding their perfusion.

CONCLUSION
Ultimately, most of these radioligands exhibit promising affinity and specificity in vitro but fail to prove their efficiency in vivo on xenograft models. Chemical screening and pharmacomodulations have yet to work out the ideal COX-2 radiotracer. The use of innovative vectors should be considered to selectively deliver radiotracers in the tumors. Given the hepatobiliary excretion of most of the known coxib derivatives, in situ CRC models should be considered for future explorations. In vivo preclinical studies on in situ CRC models would be decisive to conclude whether COX-2 is a relevant target in CRC imaging.
More conclusive in vivo results are required before conducting a "first-in-human" study.

AUTHOR CONTRIBUTIONS
CD wrote the present paper and contributed to the design of the review. FA contributed to the redaction of the manuscript and to the prior bibliographic study. YT contributed to the clinical rational of the work. FB and SR wrote chemical sections of the manuscript. MC contributed to manuscript revision and improvement. NA designed the objectives of the review and supervised its preparation and redaction. All authors have substantially contributed to this work. The manuscript has been re-read and approved by all authors, and they have all contributed to its scientific improvement.