Development of a fluorescent probe library enabling efficient screening of tumour-imaging probes based on discovery of biomarker enzymatic activities

Fluorescent probes that can selectively detect tumour lesions have great potential for fluorescence imaging-guided surgery. Here, we established a library-based approach for efficient screening of probes for tumour-selective imaging based on discovery of biomarker enzymes. We constructed a combinatorial fluorescent probe library for aminopeptidases and proteases, which is composed of 380 probes with various substrate moieties. Using this probe library, we performed lysate-based in vitro screening and/or direct imaging-based ex vivo screening of freshly resected clinical specimens from lung or gastric cancer patients, and found promising probes for tumour-selective visualization. Further, we identified two target enzymes as novel biomarker enzymes for discriminating between tumour and non-tumour tissues. This library-based approach is expected to be an efficient tool to develop tumour-imaging probes and to discover new biomarker enzyme activities for various tumours and other diseases.


Introduction
Complete surgical resection of tumour lesions is critical for the patients' prognosis. However, the ability to detect tiny tumour foci and to accurately delineate the border between tumour and non-tumour tissues with the unaided human eye is quite limited, and thus recurrence due to tumour cells le behind aer surgery is a serious problem. In order to ensure complete resection of tumour lesions, intraoperative uorescence imaging has attracted considerable attention, because of its high sensitivity, high spatiotemporal resolution, low cost and real-time capability. 1,2 So far, various types of uorescent imaging probes have been developed by targeting well-validated tumour biomarkers, and there are several successful examples of tumour visualization in mouse models and/or patients' specimens. [3][4][5][6] In particular, activatable uorescent probes (uorogenic substrates) targeting enzymes characteristic of tumours are promising, 7,8 since they have a high tumour-tonormal intensity ratio (T/N). We and other groups have focused on cancer-associated proteases as imaging targets, [9][10][11][12][13] since they exhibit altered expression levels in the pathological context, and some of them are known to be related to malignant tumour phenotypes such as reconstruction of extracellular matrixes, 14 generation of peptide messengers, 15 and utilization of amino acids from degradation of proteins. 16 For example, uorescent probes have been developed targeting g-glutamyl transferase (GGT), 17,18 which has been shown to visualize breast, 19,20 hepatic 21 and oral cancers. 22 However, available biomarker enzymes are still limited, and nding new biomarker enzymes is important to broaden the applicability of intraoperative tumour imaging.
To uncover new biomarker enzymes, it is important to evaluate altered enzyme activities between tumour and non-tumour tissues, rather than just altered protein expression levels, since the activities are required for uorescence activation. Considering that the enzyme activities in living systems are dynamically modied by many factors, including protein-protein interactions and posttranslational modications, 23,24 we thought that an efficient strategy would be to prepare a combinatorial library of uorogenic probes for proteases that would be suitable not only for in vitro screening, but also for ex vivo screening of patients' specimens to directly evaluate altered enzyme activities under pathophysiological conditions. So far, aminocoumarin-based uorogenic substrate libraries for proteases have been developed for in vitro screening of specic substrates or optimal inhibitors. [25][26][27][28][29] However, their use is mainly limited to in vitro biochemical assay due to the relatively short wavelengths of the uorescence emission, and especially they cannot be applied for imaging-based screening on living samples.
In this study, we aimed at developing new uorogenic substrate libraries for proteases making use of our previously reported scaffold uorophore, HMRG (hydroxymethyl rhodamine green) (Fig. S1 †). HMRG is an excellent scaffold uorophore for developing uorogenic substrates for proteases because high activation ratios can be achieved upon one-step enzymatic reaction, and the hydrolysis product, HMRG, emits bright uorescence in the visible wavelength region. 30,31 Thus, it is suitable for both in vitro screening and ex vivo screening using patients' specimens. Further, the selected probes can be directly used to provide intraoperative uorescence guidance for tumour-selective imaging. Here, we used solid-phase synthesis to construct a uorescent probe library composed of 380 HMRG-based probes for aminopeptidases and proteases, and we performed in vitro/ex vivo screening with resected specimens from lung or gastric cancer patients. We found promising probes for tumour-selective visualization of lung and gastric cancers, and identied two main target enzymes as novel biomarker enzymes for discriminating between tumour and non-tumour tissues in the local tumour environment.

Results and discussion
Construction of a uorescent probe library for aminopeptidases and proteases In order to construct the uorescent probe library based on the HMRG scaffold, we set out to install various amino acids or dipeptides onto HMRG by using solid-phase peptide synthesis (SPPS) ( Fig. 1 and Scheme 1). We focused on aminopeptidases and dipeptidylpeptidases as target enzymes because they generally have high reaction rates, and we selected dipeptide structures as candidate substrates because they provide sufficient substrate diversity for a screening library. We rst planned to synthesize compound 3 with a reduced xanthene ring in order to improve the nucleophilicity of the amino group 32 and included TBDMS protection on the hydroxyl group to reduce side reactions. This compound was linked to 2-chlorotrityl chloride resin, and coupled with Fmoc-amino acid (Fmoc-AA-OH). However, the condensation reaction between compound 3 on the resin and Fmoc-AA-OH did not proceed efficiently under any of the conditions we tried, probably due to insufficient reactivity of the amino group on the resin (data not shown). Thus, we next tried the approach illustrated in Scheme 1, in which we rst incorporated a variety of Fmoc-AA-OH onto compound 3 in the liquid phase to prepare compounds 4a-4y, which were linked to the resin as starting materials for SPPS. Aer oxidation with p-chloranil to form a xanthene uorophore form, second amino acids were incorporated, followed by cleavage from the resin with TFA/TES/H 2 O and simple purication by ether precipitation (if the purities determined from the absorbance at 254 and 490 nm in LC-MS analysis were lower than 80%, we further puried the compounds by HPLC). Aer optimizing each step, we synthesized 380 peptidyl-HMRGs with various substrate moieties as a probe library for aminopeptidases and proteases ( Fig. 1b and Table S1 †). The average amount of the synthesized probes was approximately 5 mmol and the average HPLC purities determined by monitoring at 254 and 490 nm were 93% and 95%, respectively. The prepared dipeptidyl-HMRGs were designated as P2-P1-HMRG, where P1 and P2 represent the amino acid at the corresponding position. To ensure the diversity of the library, we selected 20 proteogenic amino acids and 6 unnatural amino acids (D-alanine, D-aspartate, D-serine, b-alanine (bAla), N-methylglycine (MeGly) and methionine sulfoxide (MetO)) as P1 amino acids. As P2 amino acids, we selected six representative natural amino acids with different properties (glycine (no side chain), glutamate (acidic), lysine (basic), tyrosine (aromatic), leucine (aliphatic/ hydrophobic), and proline (cyclic)), together with the abovementioned unnatural amino acids except methionine sulfoxide. In addition, we synthesized N-terminally acetylated derivatives of a part of the probe library to target endopeptidases ( Fig. 1 and Table S1 †). The prepared uorescence probe libraries for aminopeptidases and proteases were dispensed into each well of 384 microplates to prepare stock plates for high-throughput screening.

Screening of imaging probes for lung cancer with the library and clinical specimens
With the constructed libraries in hand, we next performed screening of probes suitable for detecting lung cancer (Fig. 2), which has been the leading cause of cancer-related death in the world. Although we previously examined the ability of our developed uorogenic probe for GGT, gGlu-HMRG, to visualize lung cancer, its sensitivity and specicity were not sufficient. 33 Therefore, uorescent probes targeting other biomarker enzymes are needed.
As a rst screening, we performed high-throughput screening of probes using tissue lysate of resected specimens from lung cancer patients. All specimens, including lung tumour and normal lung tissues, were obtained from the Department of Thoracic Surgery, Graduate School of Medicine, University of Tokyo. Before the study, all the patients provided written informed consent for this ex vivo lung cancer uorescence imaging study. The Research Review Board at our institution examined and approved the research protocol, which was in accordance with the Declaration of Helsinki. We prepared ve pairs of lysates of tumour and adjacent non-tumour tissues from ve patients with lung adenocarcinoma (the most common subtype of lung cancer) and examined their reactivities with the probe libraries by measuring the time-dependent uorescence increase aer application to the 384-well plates. In order to perform tumour-selective imaging, we thought it important that the reactivity of probes in tumour lysate should be sufficiently strong and higher than that in non-tumour lysate. Thus, we set two criteria for selecting candidate probes: (1) more than 10% of the probe was converted to the hydrolysis product HMRG aer  one-hour incubation (the conversion rate was calculated by dividing the uorescence increase of each probe by the uorescence intensity of HMRG at same concentration); (2) the tumourto-normal ratio of the conversion rate was more than 1.5. Considering tumour heterogeneity, we picked up 16 probes that met the criteria in at least three pairs of lysates out of the ve. Among these 16 candidate probes, we further selected 7 probes for the second screening from the viewpoints of similarity of substrate peptide sequences: KK-, GP-, EP-, PP-, KR-, KA-and KM-HMRG (Fig. 2, S2 and S3 †). (Important note: X-MeGly-HMRGs (X ¼ amino acids except bAla) were found to be chemically unstable in aqueous buffer at physiological pH, probably due to the intramolecular nucleophilic attack of the N-terminal amino group to release HMRG).
Next, we performed ex vivo imaging-based screening with the 7 candidate probes. We applied the probes to small pieces of lung tumour (adenocarcinoma) and non-tumour tissues, and monitored the uorescence increase ex vivo. Aer testing at least 7 specimens from different patients, we found 5 probes (KK-, GP-, PP-, KR-, KA-) having the ability to distinguish tumour tissues with relatively high accuracy (Fig. 3, Table S2 and Fig. S4 †). We also examined the reactivities of these probes with squamous cell carcinoma (the second-most common subtype of lung cancer), aiming at nding a probe applicable for detecting a wide range of lung tumours (Fig. 3, Table S2 and Fig. S4 †). As a result, KK-HMRG showed the best ability to distinguish tumour and non-tumour tissue both in adenocarcinoma and squamous cell carcinoma; the sensitivity, specicity and AUC were calculated to be 0.727, 0.848 and 0.842, respectively ( Fig. 3 and Table S2 †). (Important note: the uorescence intensities at reddish non-tumour tissues tend to be undervalued, meaning that the high accuracy of KK-HMRG results not only from the difference in enzyme activities, but also from the difference in colour between tumour and non-tumour tissues.) In order to fully characterize KK-HMRG, we carried out a liquid-phase synthesis and measured the 1 H and 13 C NMR spectra and HRMS data (Fig. S5 †). KK-HMRG synthesized by solid-phase synthesis and KK-HMRG synthesized by liquid-phase synthesis were identical by LC-MS analysis (Fig. S6 †).

Target identification of KK-HMRG
We next set out to identify the target enzyme of KK-HMRG in lung tumour specimens. For this purpose, we used our previously developed zymography method termed diced electrophoresis gel (DEG) assay 34 (Fig. S7 †). In the assay, we obtained only one uorescent spot for lung adenocarcinoma lysate (Fig. 4a) and for squamous cell carcinoma lysate (Fig. S8 †), and these seemed to be located at the same position. Peptide mass ngerprinting analysis of the uorescent spot yielded a list of candidate proteins. Among them, we looked for proteins with peptidase activities in protein/ enzyme databases and focused on four candidate enzymes based on the reliability of the MS/MS analysis and biochemical knowledge of substrate specicities: they were puromycin-sensitive aminopeptidase (PSA), calpain-2, dipeptidylpeptidase-3 (DPP-3) and leukotriene A4 hydrolase (Table S3 †). To identify the main contributor to the hydrolysis of KK-HMRG, we next examined the reactivity of KK-HMRG with lung tumour lysate in the presence of appropriate inhibitors (puromycin for PSA, 35 SNJ-1945 for calpain-2, 36 3,4-dichloroisocoumarin for DPP-3 37 or SC-57461A for leukotriene A4 hydrolase 38 ). Puromycin strongly inhibited the uorescence increase of KK-HMRG with tumour lysate in a concentration-dependent manner (Fig. 4b), while the other inhibitors had little or no inhibitory effect on the uorescence increase ( Fig. S9 †), suggesting that PSA is likely the responsible enzyme. We also conrmed that KK-HMRG was hydrolyzed to HMRG by recombinant PSA protein by means of uorescence measurement and LC-MS analysis ( Fig. 4c and S10 †), and that the two lysine residues of KK-HMRG were hydrolyzed in two steps upon reaction with PSA and with tissue lysate of lung adenocarcinoma ( Fig. S10 and S11 †). We also conrmed that puromycin inhibited the ex vivo uorescence increase of resected lung adenocarcinoma tissue (Fig. 4d), and that PSA expression was upregulated in tumour tissues compared to non-tumour tissues in 4 out of the 5 adenocarcinoma lysates (Fig. 4e). These results strongly suggest that main target enzyme of KK-HMRG in lung cancer is PSA. Thus, the screening with our probe libraries led to the discovery of PSA activity as a biomarker in lung cancer. Our results suggest that KK-HMRG can be used as a lung tumourselective imaging probe.

Screening of imaging probes on ESD samples of gastric cancer
To conrm the versatility of this approach, we next applied our probe libraries to endoscopic submucosal dissection (ESD) specimens from gastric cancer patients. In contrast to the case of lung cancer specimens, it was impossible to prepare tissue lysate of ESD specimens due to their small size. Therefore, we selected several probes as a focused library, and applied them to ESD specimens to perform ex vivo imaging-based screening. To evaluate the reactivities of a few probes simultaneously on one specimen, we used Tetra-PEG gels 39 or medical gauzes that were pre-soaked with probe solutions, and measured the uorescence increase at tumour regions and non-tumour regions ( Fig. 5a and S12 †). As a result, we found that KH-HMRG showed a selective uorescence increase at non-tumour regions in 3 of 4 specimens (Fig. 5a, S13 and S14 †). Since a uorescent probe enabling negative staining of tumour region could also be a useful diagnostic tool, we next examined whether negative staining of tumour regions could be achieved by spraying KH-HMRG onto ESD specimens. In 12 specimens out of the tested 25 specimens, tumour regions were clearly visualized ( Fig. 5b and S15 †). In 11 cases out of the 13 failed cases, no uorescent signal activation was observed in non-tumour regions, meaning that there is heterogeneity of enzyme activity in nontumour regions (Fig. S16 †). To identify the target enzyme of KH-HMRG, we performed DEG assay using lysate of differentiated HL60 cells, 40 which were found to have enzyme activity to hydrolyze KH-HMRG (Fig. S17 †). (Note: as mentioned above, lysates of clinical ESD specimens were not available.) We observed one strongly uorescent spot and one weakly uorescent spot, and PMF analysis of the strongly uorescent spot identied aminopeptidase N (APN) and Xaa-Pro aminopeptidase as candidates (Fig. 5c and Table S4 †). Judging from their substrate specicity, we focused on APN, which is a membranebound alanylaminopeptidase with broad substrate specicity, 41 whereas Xaa-Pro aminopeptidase cleaves N-terminal amino acids when the amino acid at P1 0 is proline. 42 We conrmed that recombinant APN protein hydrolyzes KH-HMRG to HMRG (Fig. 5d). Further, the uorescence increase of KH-HMRG in dHL60 lysate was inhibited by the APN inhibitor bestatin 41 in a concentration-dependent manner (Fig. S18 †). LC-MS analysis of the reaction solution of KH-HMRG with APN or with dHL60 cell lysate revealed that KH-HMRG was hydrolyzed in two steps (Fig. S19 †). Additionally, to examine the selectivity of KH-HMRG for APN, we performed live-cell imaging of APN activity in HT1080 (APN-positive) and HEK293 (APN-negative) cells (Fig. S20 †). As expected, a uorescence signal was detected only from HT1080 cells when KH-HMRG was applied (Fig. S21 †). Interestingly, when we applied A-HMRG (alanyl-HMRG, alanine is a standard sequence for uorogenic substrates of APN 12 ), uorescence activation was observed with both HT1080 and HEK293 cells (Fig. S22 †). This result demonstrated that Lys-His is a highly selective sequence for APN when using HMRG as a uorophore. Finally, we conrmed that the expression level of APN was higher in non-tumour regions, especially on the apical membrane, by immunohistochemistry (Fig. 5e). In addition, the uorescence increase of KH-HMRG on ESD samples of gastric non-tumour tissue was inhibited by bestatin (Fig. S23 †). These data strongly suggest that the responsible enzyme for the uorescence increase in nontumour regions was APN. It should be noted that inammation and intestinal metaplasia were observed in non-tumour regions (Fig. 5e), which might be related to this partial upregulation of APN activity in these regions. Further studies are underway. The present results demonstrate that our librarybased approach is a versatile tool for discovering novel uorescent probe/biomarker enzyme activity pairs for tumourselective imaging.

Conclusions
We have established a library-based approach for efficient screening of uorescent probes suitable for tumour imaging and novel biomarker enzyme activities. To construct the uorescent probe library based on the HMRG scaffold, we devised a simple synthetic scheme using SPPS to prepare 380 peptidyl-HMRGs with various substrate moieties. With the library in hand, we performed screening of probes for lung and gastric cancer detection. For lung cancer, we found several probes with good ability to distinguish tumour and non-tumour tissues, and among them, KK-HMRG targeting PSA exhibited particularly high accuracy. For gastric cancer, we identied KH-HMRG targeting APN as a negative-staining probe due to its selective uorescence increase in non-tumour regions. The reactivities of these probes were similar or superior to those of the aminomethyl coumarin-based substrates (Fig. S24, Tables S4 and S5 †). The greatest advantage of this strategy is the ability to nd new biomarker enzyme activities for tumour imaging directly from clinical specimens. Furthermore, the selected probes can be directly used for intraoperative uorescence imaging guidance. Unfortunately, the sensitivity/specicity of the selected probes in this study are not high enough for clinical use, probably due to the heterogeneity of both tumour and nontumour tissues. However, we expect that we would be able to overcome the heterogeneity by targeting multiple enzyme activities 11,31,43,44 that are altered at tumour lesions, and these could also be determined by the devised library-based screening. In addition, this methodology should be applicable not only for nding imaging probes for other diseases, 45 but also nding probes to detect a specic subset of cells. 46 Therefore, we believe that our approach will accelerate research into characteristic enzyme activities in specic environments, with many potential medical and biological applications.

Author contributions
Y. K., M. K. and Y. U. co-wrote and reviewed the manuscript. Y. K., A. N. and K. T. synthesized and analysed the compounds. Y.