|Home | About | Journals | Submit | Contact Us | Français|
We recently developed activatable cell-penetrating peptides (ACPPs) that target contrast agents to in vivo sites of matrix metalloproteinase activity, such as tumors. Here we use parallel in vivo and in vitro selection with phage display to identify novel tumor-homing ACPPs with no bias for primary sequence or target protease. Specifically, phage displaying a library of ACPPs were either injected into tumor-bearing mice, followed by isolation of cleaved phage from dissected tumor, or isolated based on selective cleavage by extracts of tumor versus normal tissue. Selected sequences were synthesized as fluorescently labeled peptides, and tumor-specific cleavage was confirmed by digestion with tissue extracts. The most efficiently cleaved peptide contained the substrate sequence RLQLKL and labeled tumors and metastases from several cancer models with up to 5-fold contrast. This uniquely identified ACPP was not cleaved by matrix metalloproteinases or various coagulation factors but was efficiently cleaved by plasmin and elastases, both of which have been shown to be aberrantly overexpressed in tumors. The identification of an ACPP that targets tumor expressed proteases without rational design highlights the value of unbiased selection schemes for the development of potential therapeutic agents.
Molecular targeting of contrast and therapeutic agents to tumors is a fundamental goal of cancer research and treatment. Targeting based on tumor-specific molecules has largely relied on binding of antibodies, peptides, or vitamins to antigens or receptors expressed at high levels on tumor cells or their associated vasculature. The drawback to such mechanisms is that they rely on stoichiometric binding that lacks amplification, so either the target must be very highly expressed, or the readout must be extremely sensitive. Activatable cell-penetrating peptides (ACPPs)2 are a new strategy in which enzymatic amplification drives in vivo accumulation of peptide-conjugated cargo within tissues containing active extracellular proteases (1,–6). An ACPP consists of three domains: a polycationic cell-penetrating peptide (7,–9), a cleavable linker, and a polyanionic inhibitory domain. Before cleavage of the linker, the polyanion pairs with the polycation and prevents cell adhesion and internalization; after cleavage, the polyanion dissociates, unleashing the cell-penetrating peptide to adhere immediately to adjacent cells, followed by cellular internalization. The first ACPPs were designed to be activated by matrix metalloproteinases (MMPs), which are frequently up-regulated in tumors and play a crucial role in extracellular matrix degradation, tissue invasion, and metastasis (10,–15). Although an MMP-activated ACPP has been shown to effectively target tumors in vivo, this rational design of an ACPP is limited by the availability of known protease substrates and the tissue specific expression of the target protease. For instance, there is currently no known peptide substrate that is perfectly selective for MMPs, and no MMP is exclusively expressed in tumors. Therefore, to improve the applicability of ACPPs, it is important to explore other classes of proteases and substrates with methods that might provide greater specificity and flexibility than a priori design.
Previously, phage display technology has been used extensively in vitro to identify peptides, proteins, and antibodies based on specific binding or catalytic activity (16,–18). In addition, phage display has been used in vivo to identify linear or circularized peptides that stoichiometrically bind and accumulate in tumors, synovium, and vascular tissues (19, 20). In this report, we have used phage display to identify peptides that accumulate within tumors in vivo as a result of proteolytic modification. This method identifies novel protease substrates that are cleaved within tumors without bias for class of protease or specific substrate sequence. To examine the efficacy of this approach, we performed parallel in vivo and in vitro selections with phage displaying a library of unique ACPPs. For in vivo selection, phage were injected into tumor-bearing mice followed by isolation of cleaved phage from homogenized tumors. For in vitro selection, phage were sequentially exposed to normal versus tumor tissue extract, followed by isolation of uncleaved or cleaved phage, respectively. Phage selected ACPPs were resynthesized as fluorescently labeled peptides, injected into tumor-bearing mice, and shown to accumulate within primary and metastatic tumors in a protease-dependent manner.
A phage library was generated by PCR, using 5′-primer ggccaagagctccatcatcatcatcatcatctc, 3′-primer ggccagactagtccgtcgccgtcgccgtcg, and oligonucleotide template ccatcatcatcatcatcatctcgaggaggaggaggaggaggaggaggagnnknnknnknnknnknnkcggcgacggcgacggcgacggcgacggactag, where n denotes any nucleotide and k indicates a mixture of g and t. PCR was performed for 25 cycles with 52 °C annealing and standard reaction conditions. PCR products were digested with restriction enzymes SpeI and SacI for directional cloning into pComb M13 phage expression vector (17). The phage library contained 5 × 106 primary transformants, which were converted to infectious phage using helper phage superinfection. Phage libraries were then titered by infection of XL1blu cells and plated on ampicillin plates. For in vitro selections, 109 phage were incubated for 3 h at 37 °C with mixed liver/kidney tissue extracts equivalent to 2% (2 mg of each tissue per 100 μl). Ni2+-NTA-agarose (Qiagen) was added and incubated for 15 min, followed by three washes with PBS. The Ni2+-NTA-agarose coupled to the uncleaved phage was then exposed to 2% tumor tissue extract at 37 °C for 3 h. Unbound cleaved phage were eluted from the Ni2+-NTA with PBS and plated on XL1blue E. coli to recover phage. For in vivo selections, 109 phage in 100 μl of PBS were injected into the tail veins of 10–13-week-old transgenic mice that had 5–15-mm mammary tumors due to expression of PyMT driven by the long terminal repeat of mouse mammary tumor virus. Tumors were excised and homogenized in PBS in the presence of protease inhibitor containing 100 μm 4-(2-aminoethyl)-benzenesulfonyl fluoride-HCl, 80 nm aprotinin, 5 μm bestatin, 1.5 μm E-64, 2 μm leupeptin hemisulfate, and 1 μm pepstatin A (EMD Biosciences mixture III). Uncleaved phage were removed by absorption with Ni2+-NTA-agarose, and cleaved phage from the supernatant were reamplified and reinjected into different PyMT tumor-bearing mice.
Peptides were either obtained from the W. M. Keck Facility Small Scale Core (Yale University, New Haven, CT) or synthesized in our laboratory on a Perceptive Biosystems Pioneer automatic peptide synthesizer using standard protocols for Fmoc solid-phase synthesis. All peptides were of the general format N terminus-(d-glutamyl)9-(6-aminohexanoyl)-(6-amino acid cleavage site)-(d-argininyl)9-(either Cy5-d-cysteinamide or ϵ-(fluorescein-5(6)-carbonyl)-l-lysinamide) (see supplemental Fig. 1 for complete details). Carboxyfluorescein-labeled peptides were generated by Fmoc synthesis with Lys-5(6)-carboxyfluorescein and were acetylated at the N terminus. Albumin-reactive peptide, Mal-e9-ahx-RLQLK(Ac)L-r9-c(Cy5)-NH2, was generated by reacting 3-maleimidopropionic acid pentafluorophenyl ester (Molecular Biosciences) with the free N terminus of Cy5-labeled peptide (see supplemental Fig. 1 for detailed structures of all peptides). The negative control maleimide (Mal) peptide had -NH(CH2CH2O)2CH2CO-, an uncleavable (mpeg) linker (4), in place of the 6-aminohexanoyl and substrate sequence. All Cy5-labeled peptides were purified to greater than 95% purity using C-18 reverse phase HPLC with a 20–50% acetonitrile gradient in 0.1% TFA and the mass was confirmed by mass spectrometry.
Tissue extracts were generated by dissecting fresh tissue and gently homogenizing in PBS, while on ice, to minimize the release of intracellular proteases. Homogenates were then microcentrifuged at 14,000 × g for 1 min, and supernatants were recovered and used for in vitro testing. All cystic fluids were aspirated with a 28-gauge syringe from surgically exposed mammary tumors of PyMT mice containing fluid filled nodules. In vitro cleavages were done with 5 μm peptide in 20 μl of PBS containing 1 μm ZnCl2 plus 2 μl of 20% tissue extract (20 mg of tissue homogenate per 100 μl) or 2 μl of undiluted cystic fluid, followed by incubation at 37 °C. Aliquots were removed at time intervals from 20 min to 18 h with data being shown for the 2-h time point. For cleavage with purified enzymes, a 50 nm concentration of each enzyme was incubated with 5 μm peptide in the manufacturer's recommended buffer for 4 h. Enzymes and inhibitors were obtained from commercial sources in their active form (a complete table of enzymes and inhibitors is shown in supplemental Fig. 2). Tricine sample buffer was added, and treated peptides were boiled for 1 min prior to electrophoresis on 10–20% Tricine acrylamide gels (Invitrogen). Gels were quantitatively imaged on a UVP Biospectrum AC imaging system with external Biolite light source with excitation at 628 ± 20 nm and emission at >700 nm. Quantitation of enzyme cleavage was done using UVP Biospectrum software.
For mass spectrometry determination, each peptide was purified from homogenized cell extract by two steps of reverse phase HPLC. A 30-μl sample of 5 μm ACPP digested with tissue extract for 2 h was diluted to 2.5 ml with solvent A, which contained 2% acetonitrile and 0.14% trifluoroacetic acid. Samples were immediately injected into a Dionex BioLC HPLC system. The autosampler (Gilson model 231) had a 2-ml sample loop, so two 1-ml injections were used for each HPLC chromatogram. The column was a protein C4 from Vydac with a 4.6-mm inner diameter and 25-cm length. The column temperature was 35 °C, and the flow rate was 0.8 ml/min. After an 11-min delay (12% solvent B) for the injections, the gradient was from 12 to 56% solvent B (90% acetonitrile, 0.08% trifluoroacetic acid) in 11 min, followed by a 3-min step of 80% B. The peak eluted at 20 min. Detection was by fluorescence (Dionex RF2000 fluorescence detector, excitation at 640 nm, and emission at 680 nm), and the peak was collected with a Gilson FC203 fraction collector.
For the second step of reverse phase HPLC, the fraction was reconcentrated to 200 μl by centrifuge evaporation (Savant SpeedVac) and then diluted to 600 μl with solvent A again. A Michrom Magic HPLC system was used for liquid chromatography-mass spectrometry, and the column output was detected by a Jasco 920 fluorescence detector and an LTQ Orbitrap XL mass spectrometer. A Michrom sample loop microtrap was used to preconcentrate the entire 600-μl sample for injection. The column was a 1-mm inner diameter, 15-cm Michrom C18AQ, and the fluorescence detector settings were similar to those in the first step. The solvents, column temperature, and gradient were also similar, and the flow rate was 80 μl/min. The MS was set for ~0.5-s scans with the high resolution Orbitrap from 400 to 2000 m/z. The entire column output went into the mass spectrometer without splitting, using the electrospray interface.
For each tumor model, Cy5-labeled peptide was injected intravenously via the tail vein. Mice were imaged at time intervals up to 6 h using a Maestro imager (CRI Inc.) with excitation at 640 ± 24 nm, emission at >700 nm. For in vivo testing, we used either PyMT mice with spontaneous progressive mammary tumors or nude mice with 5–7-mm MDA-MB-435 clone M4A4 (gift of D. Tarin, UCSD) xenograft tumors generated by subcutaneous injection of 106 tumor cells into the mammary fat pads (mice 5–8 weeks of age; Charles Rivers Laboratories). Animals were anesthetized with a 1:1 mixture of 80 mg/kg ketamine and 5 mg/kg midazolam, weighed, and injected with 100 μl of 100 μm free peptides or 100 μl of 30 μm albumin-reactive peptide. After imaging, the animals were euthanized with isoflurane, and tissues were harvested and weighed for standardized uptake values (SUVs). To measure SUVs, 30 mg of each tissue was added to 100 μl of Tris buffer (pH 7.6) with 1% SDS. The tissue was homogenized, heated to 85 °C for 10 min, microwaved for 10 s, centrifuged at 20,500 × g for 15 min, and imaged on the Maestro imager. Four sets of tissue-specific standards (tumor, liver, kidney, and muscle from non-injected animals, added to known amounts of ACPP and processed as above) were used to calibrate fluorescence intensity in terms of peptide concentration. Tissue-specific standards were necessary because different tissues vary in their ability to quench or conceal Cy5 fluorescence. From this calibration, the quantity of peptide in 30 mg of tissue for each organ was calculated. SUVs were calculated as the molality of peptide in the tissue divided by the total injected dose as mol/kg of body weight. In previous experiments with the MMP-cleavable ACPP, SUVs determined using this method were similar to SUVs determined using technetium chelates (1, 4,–6, 21). Means are given ± S.D.; statistical significance of differences was determined by two-tailed t tests.
To generate mice with lung metastases, 5 × 105 4T1 tumor cells (ATCC) were injected into the tail veins of BALB/c mice 8 days prior to injection with imaging agent. Lungs were removed and imaged either 6 or 48 h postinjection, depending on the peptide construct. Average fluorescent intensity was determined for metastatic lung nodules and compared with adjacent non-malignant tissue or lungs from a tumor-bearing animal that had no lung metastases. We analyzed nine selected regions of 80–200 pixels of each type of tissue, and all tissues were imaged using identical parameters.
In order to generate a phage library displaying ACPPs, amino acid sequences of the format Met-His6-E9-X6-R9 (where X6 represents 6 randomized amino acids) were fused to the N terminus (0–4 copies per phage) of a truncated form of the M13 phage gIII coat protein (17, 22). This library contained 5 × 106 putative protease cleavage sites as determined by primary colony-forming units after helper phage infection and dispersal on plates containing antibiotics. This represents about 8% of the potential diversity of 6.4 × 107 for X6. The X6 region of the library was generated using NNK(G/T) degenerate oligonucleotides. A hexahistidine motif was added at the N terminus to allow separation by immobilized metal affinity chromatography of unmodified phage from phage whose peptides had been cleaved releasing the hexahistidine tag.
In vivo selections were performed by injecting a library of phage displaying recombinant ACPPs intravenously into PyMT mice (Fig. 1). This transgenic breast cancer model produces spontaneous heterogeneous tumors, which can be cystic and will progress to metastasis in older animals (23,–25). Six hours after injection, animals were sacrificed, and their tumors were dissected out and homogenized in the presence of protease inhibitors to prevent secondary cleavage of contained phage. Uncleaved phage in the tumor vasculature or nonspecifically localized to tumor at the time of dissection were removed by absorption with Ni2+-NTA-conjugated agarose. This purification step was necessary because, as previously reported, phage particles can be retained for long periods in tissues even in the absence of any selection (26). Purified phage that had been cleaved within the mouse were reamplified and injected into different PyMT tumor-bearing mice. This selection was continued for seven iterative cycles with 10–15 individual phage clones being sampled by DNA isolation and sequencing after each round of selection. Five sequences, RLQLKL, RTRYED, RIPLEM, QFDEPR, and TSAVRT, were each isolated multiple times during the in vivo selection (Table 1).
For in vitro selection, the ACPP phage library was pre-exposed to extracts of liver and kidney, tissues that typically have high nonspecific uptake of drugs and imaging agent, to allow cleavage by proteases contained within these extracts. Ni2+-agarose was then used to isolate uncleaved phage that had retained their hexahistidine tags. These preselected phage were then exposed to PyMT tumor extracts and selected for peptide cleavage and loss of their hexahistidine tags. Isolated phage were reamplified and subjected to six additional rounds of both negative (no liver/kidney cleavage) and positive (tumor cleavage) selection, with representative phage being isolated and sequenced after each round. This in vitro selection identified five sequences isolated in the in vivo selection scheme as well 15 additional sequences (Table 1). Only one sequence, RLQLKL, was identified in duplicate in both the in vivo and in vitro selection schemes.
To test if isolated peptides corresponding to the phage-selected sequence could be proteolytically cleaved by tumor tissue extracts, 14 of the 21 selected phage sequences were synthesized as peptides in the format acetyl-e9-(ahx)-X6-r9-5(6)-carboxyfluorescein-labeled l-lysinamide using solid-phase synthesis (for complete structures, see supplemental Fig. 1). The (ahx) denotes 6-aminohexanoyl, a flexible linker that is included in ACPPs to give the cleavable X6 domain the ability to adopt a more extended conformation within the protease cleavage domain. The polyglutamate and polyarginine segments consisted entirely of d-amino acids (denoted by lowercase letters) in order to maximize their resistance to proteolysis in vivo. Four peptides, containing X6 = RLQLKL, RTRYED, GMMYRS, and RWRTNF, were cleaved >80% by tumor tissue extract after 16-h digestions at 37 °C (Table 1).
These top four peptides were resynthesized in the format H2N-e9-(ahx)-X6-r9-(Cy5-labeled d-cysteinamide), followed by purification to greater than 95% for in vitro and in vivo testing. Each of these peptides was tested for differential cleavage by tumor versus mixed liver/kidney tissue extracts after a 2-h digestion compared with the 16-h digestion shown in Table 1 (Fig. 2A). All four peptides were cleaved >50% by tumor tissue extract, with RLQLKL being the most rapidly cleaved, >90% after 4 h (full time course not shown). Each peptide except for GMMYRS showed increased cleavage by tumor extract relative to mixed liver/kidney extracts. A control peptide with all d-amino acids, X6 = rlqlkl, was not cleaved by any of the tissue extracts.
In order to compare in vivo tumor labeling, these four Cy5-labeled ACPPs were separately injected intravenously into PyMT mice with spontaneous mammary tumors. Six hours postinjection, mice were imaged by fluorescence to show contrast of tumor compared with adjacent normal tissue. The most striking visual contrast for tumor uptake was obtained from mice injected with the RLQLKL ACPP (Fig. 2B). Consistent with this high visual contrast, the SUVs in the tumor, defined as the molality of Cy5 in the specific tissue divided by the molality of the injected dose for the total weight of the animal, was the highest for the RLQLKL ACPP at 0.39 ± 0.14 (n = 8). This compared with the uncleavable all-d-amino acid peptide, rlqlkl ACPP, which had a SUV of 0.13 ± 0.03 (n = 3) with p = 0.014. The decreased tumor accumulation in the mice injected with the uncleavable d-amino acid probe of identical molecular weight and hydrophobicity strongly suggests that tumor uptake is dependent on protease susceptibility rather than nonspecific adhesion or enhanced vascular permeability of tumors. This ratio of 3 between protease-sensitive and relatively insensitive peptides compares with a ratio of 3.6 for our previous ACPP, in which X6 = PLGLAG and targets matrix metalloproteinases (1). PLGLAG ACPP had SUV values of 0.51 ± 0.05 (n = 5) versus 0.14 ± 0.06 (n = 5), p < 0.01, for its d-amino acid control. As with all free peptide ACPPs, there is labeling of cartilaginous portions of the anterior ribcage (4, 5). One advantage of RLQLKL ACPP versus PLGLAG ACPP is that skin uptake was significantly less, improving contrast during noninvasive visualization in live animals (see supplemental Fig. 3 for images).
To explore whether RLQLKL ACPP could be used as a contrast agent for tumor models other than the PyMT mice, Cy5-labeled RLQLKL ACPP and its all-d-amino acid control were tested in mice bearing MDA-MB-435 M4A4 human xenografts. This tumor model (also known as M14 melanoma (27)) was stably transfected with GFP for unambiguous identification of xenograft versus host tissue. Tumors of mice injected with the RLQLKL ACPP showed far red (Cy5) fluorescence, which co-localized with the GFP fluorescence (Fig. 2C). Mice injected with the uncleavable d-amino acid (rlqlkl) control showed much less Cy5 fluorescence within the tumors, which were still clearly visible by GFP fluorescence. Tumor SUVs 6 h after injection were also significantly higher for the RLQLKL ACPP 0.70 ± 0.27 (n = 6) compared with 0.34 ± 0.16 (n = 6) for rlqlkl control peptide (p = 0.022).
To characterize the protease activity responsible for tumor uptake, the effect of various protease inhibitors on tissue extract mediated cleavage of RLQLKL ACPP was tested. Unfortunately, cleavage of RLQLKL ACPP by crude tissue extracts is quite resistant to inhibition by protease inhibitors, which may be due to high protease activity in tissue extracts or the multiplicity of proteases released upon tissue disruption. We therefore switched to cystic fluid obtained from fluid-filled nodules of PyMT tumors because this fluid probably contained proteases that had been secreted by adjacent tumor tissue and was less likely than tumor extracts to be contaminated with intracellular proteases. The use of similar tumor cystic fluids (although obtained from pancreatic and not mammary tumors) for cancer protease profiling has been reported elsewhere (28, 29). Cystic fluid cleaved RLQLKL ACPP similarly to tumor tissue extracts, and this cleavage was partially inhibited by two serine protease inhibitors, aprotinin and leupeptin (Fig. 3A). This inhibitor profile supported the involvement of a serine protease in the in vivo cleavage and tumor uptake of RLQLKL ACPP.
Cleavage of RLQLKL ACPP was further investigated using a panel of candidate proteases. The serine proteases plasmin, trypsin, and neutrophil elastase were the only enzymes found to efficiently cleave RLQLKL ACPP (Fig. 3B and supplemental Fig. 4). Each of these proteases has been reported to be up-regulated in cancers. There was also minimal cleavage of RLQLKL ACPP by cathepsin G, MMP-1, MMP-7, and MMP12, which at times was only visible upon gel overexposure. MMPs are not likely to be responsible for in vivo tumor uptake because they exhibit low cleavage rates. Additionally, cleavage by cystic fluid was unaffected by MMP-specific inhibitors (GM6001, SB3CT, prinomastat) and Zn2+ chelators (DTPA and TPEN). Plasmin has been reported to be associated with tumor vasculature, and urokinase plasminogen activator is known to be preferentially activated in PyMT tumors (30,–32). The plasmin-specific inhibitor serpin F2/α2-antiplasmin, however, did not inhibit the protease activity of cystic fluid or tumor extract on RLQLKL ACPP, although it did inhibit plasmin cleavage, as expected (Fig. 4A). This result indicated that plasmin was probably not responsible for tumor extract cleavage or in vivo uptake of RLQLKL ACPP. Another candidate protease for in vivo cleavage of RLQLKL ACPP, trypsin, is normally produced by the pancreas as a zymogen (trypsinogen) and is later modified into its active form by enterokinase. Trypsin and trypsin-like protease expression has been reported in malignant non-pancreatic cells and has been shown to stimulate cancer cell growth in culture and in nude mice (33,–36). However, trypsin cleavability correlated poorly with in vivo uptake because trypsin, unlike elastase, efficiently cleaved RTRYED and GMMYRS ACPPs, but these peptides showed little fluorescence contrast in tumors relative to adjacent normal tissue (data not shown). Elastase, a component of breast milk, has been reported to be expressed in breast cancer cell lines and has been implicated as a factor in both the prognosis and progression of breast cancer (37,–39). Although this protein is expressed ubiquitously, high circulating levels of its natural inhibitor α1-antitrypsin (up to 50 μm) prevent its activity in the bloodstream. We found that α1-antitrypsin inhibited the protease activity of cystic fluid and was also the only inhibitor tested that significantly inhibited tumor extract-dependent cleavage of RLQLKL ACPP (Fig. 4B, top). This inhibition increased as the concentration of tumor extract was reduced from 2 to 0.25%. Additionally, both the highly related pancreatic form of elastase and leukocyte elastase (ELA-2 from a different source) effectively cleaved RLQLKL ACPP.
To further characterize the proteolytic cleavage and means of tumor uptake of RLQLKL ACPP, mass spectrometry was used to compare the proteolytic cleavage products produced by exposure to liver, kidney, and tumor extracts to those produced after cleavage by plasmin, trypsin, and elastase (supplemental Fig. 5). The cleavage products produced by tissue extract cleavage are inherently variable, depending on time of digest, batch, and concentration of tissue extract. However, it was determined that cleavage of RLQLKL ACPP by either kidney or tumor extracts produced in differing ratios three detectable cleavage products, corresponding to (LKLr9c-(Cy5), KLr9c-(Cy5), and Lr9c-(Cy5)). KLr9c-(Cy5) was the most abundant cleavage product after treatment with tumor extracts, whereas each of the three cleavage products was represented about equally when cleaved by kidney. In contrast, cleavage by liver produced only one product Lr9c-(Cy5), which suggests that liver cleavage was due to different endoproteases from kidney and tumor or alternatively that liver was much more efficient at removing leucine and/or lysine from either LKLr9c-(Cy5) or KLr9c-(Cy5). Digestion of RLQLKL ACPP with either plasmin or trypsin produced the single cleavage product Lr9c-(Cy5), suggesting that these enzymes may be responsible for liver but not tumor cleavage. In contrast, like tumor extract, cleavage by either pancreatic or neutrophil elastase produced as their major product KLr9c-(Cy5), supporting elastases as a good target enzyme for in vivo tumor uptake of RLQLKL ACPP.
To improve tumor-selective cleavage of RLQLKL ACPP, a modified version was generated in which the ϵ-amino group of lysine was acetylated. We hypothesized that because lysine is the site of cleavage by liver extracts and the non-target enzymes trypsin and plasmin, acetylation at this residue should increase tumor selectivity by reducing off-target cleavage. Indeed, this new RLQLK(Ac)L ACPP was not cleaved by plasmin, although it was efficiently cleaved by tumor extracts (Fig. 4C). As predicted, cleavage of this acetylated ACPP by liver and kidney tissue extracts was also significantly reduced compared with the parent RLQLKL ACPP, thereby further enhancing selectivity for tumor cleavage. Interestingly, tumor extract cleavage of RLQLK(Ac)L ACPP produced two bands by electrophoresis gel analysis (Fig. 4B, bottom). The generation of one of these bands upon digestion with tumor extract could be efficiently inhibited by the addition of α1-antitrypsin, whereas production of a lower molecular weight cleavage product showed little or no inhibition. This further supported elastase as a target enzyme but highlights the presence of at least one additional protease within tumor extracts that can cleave and potentially enhance tumor uptake of RLQLK(Ac)L ACPP. The presence of neutrophil elastase within PyMT tumor extracts was confirmed by Western blot analysis (supplemental Fig. 6). Mass spectrometry was used to analyze cleavage products produced by tumor extract in the presence and absence of α1-antitrypsin, but variability between tumor extract preparations, incomplete inhibition, and heterogeneity of cleavage products prevented the identification of a specific cleavage product that was inhibited by α1-antitrypsin. We were, however, able to confirm that as with RLQLKL ACPP, cleavage of RLQLK(Ac)L ACPP by tumor extracts or elastases produced the same major product, K(Ac)Lr9c-(Cy5) (supplemental Fig. 5).
To test in vivo uptake, Cy5-labeled RLQLK(Ac)L ACPP was injected into PyMT tumor-bearing mice. This optimized ACPP was taken up in tumors to levels equal to or slightly greater than the unmodified parent RLQLKL ACPP (SUV = 0.39 ± 0.14 (n = 8) for RLQLKL and 0.48 ± 0.04 (n = 3) for RLQLK(Ac)L, p = 0.33), supporting the hypothesis that elastase but not plasmin could be responsible for tumor uptake in the PyMT model (Fig. 5A, top left). In addition to improved specificity, acetylation of the RLQLKL ACPP simplifies attachment of dyes or other functional groups to the N-terminal amine of the peptide by blocking cross-reactivity to the lysine within the protease cleavage site.
To increase tumor-specific uptake relative to background tissues, RLQLK(Ac)L ACPP was additionally modified by conjugating a maleimide (Mal) group to the N-terminal amine, thereby generating an albumin-reactive peptide (40, 41). The rapid covalent attachment of this thiol-reactive peptide, Mal-RLQLK(Ac)L ACPP, to a cysteine on serum albumin increases the molecular size of the imaging agent, thereby increasing the blood circulation time while decreasing renal and synovial filtration. The albumin-reactive ACPP was injected into PyMT tumor-bearing mice and imaged 48 h postinjection before and after skin removal. Tumor to non-tumor contrast was 5.1-fold for the Mal-RLQLK(Ac)L ACPP, compared with less than 3-fold for free peptide RLQLK(Ac)L ACPP (Fig. 5A). SUVs were 2.3 ± 0.6 (n = 3) for the albumin-reactive ACPP compared with 0.48 ± 0.04 (n = 3) for free peptide RLQLK(Ac)L ACPP. Background labeling of cartilage and bone was also greatly decreased, consistent with previously reported data for macromolecule-conjugated ACPPs (5).
In addition to identifying primary tumors, RLQLKL ACPP was used for the detection of lung metastases in PyMT mice. These spontaneous lung metastases (which occur sporadically in older animals) were clearly visible as bright fluorescent spots compared with adjacent normal lung tissue (Fig. 5B, top). These zones were confirmed as metastases by hematoxylin and eosin staining of paraffin-embedded sections (supplemental Fig. 6). Lungs from similar animals without metastases (also from animals with primary tumors) were uniformly dim (supplemental Fig. 6) and were 4.2-fold (p < 0.01) less bright than metastatic nodules on average. Nonmalignant tissue adjacent to metastases appeared brighter than tissue from lungs with no metastases, possibly because some of the fluorescence from metastases scattered into adjacent lung tissue or because some protease-secreting cells, protease, or cleaved peptide spread from the metastasis into the surrounding tissue. For comparison, the rlqlkl control peptide showed much reduced contrast compared with the RLQLKL ACPP (Fig. 5B, bottom). The small amount of contrast observed for metastasis from mice that had been injected with rlqlkl ACPP may be due to the differential cell density of lung metastasis versus normal lung tissue. Unfortunately, lung metastases were present in only a fraction of PyMT tumor-bearing mice, so data were obtained from only a few animals.
To further study the labeling of metastases by RLQLK(Ac)L ACPP, a new experimental lung metastasis model was generated by injecting 4T1 tumor cells intravenously and allowing them to seed in the lungs. Lungs from mice that had been injected with either Mal-RLQLK(Ac)L ACPP or uncleavable Mal-mpeg-ACPP were removed and imaged. Again, the cleavable probe highlighted fluorescent nodules, which were confirmed by hematoxylin and eosin staining to be metastases (Fig. 5C, top row). The uncleavable control showed little or no uptake in metastases (Fig. 5C, bottom row). Lung sections viewed at higher magnification showed that the fluorescence from the albumin-reactive cleavable ACPP was distributed throughout the metastatic nodules and originated from numerous 10–50-μm punctae, which could represent labeling of individual or small clusters of cells. These results conclusively demonstrated specific labeling of lung metastases with this newly identified elastase-cleavable ACPP.
Phage display identified the RLQLKL ACPP as a promising new tool that has the ability to detect protease activity in tumors and could be used to direct therapeutic or imaging agents to cancer cells in vivo. This selected ACPP was modified using a non-native amino acid, Nϵ-acetyl-lysine, to further enhance its enzyme selectivity and specific tumor uptake. Modification of the lysine residue within the protease cleavage site facilitates attachment to a large molecular weight carrier, such as albumin. Further modification of our current best substrate RLQLK(Ac)L ACPP by genetic or synthetic iteration around the RLQLKL cleavage sequence (e.g. holding some residues constant while exhaustively varying the remainder) could provide additional optimization. In vivo properties could be further modulated by attachment to other high molecular weight carriers for improved pharmacokinetics or increased valency of the ACPP to enhance tissue specific uptake.
In addition, because this method identifies a specific ACPP cleavage sequence (in this case RLQLKL and the acetylated derivative), traditional biochemical techniques, including inhibitor and enzyme assays, extraction methods, or activity-based protease profiling (42), can be used to identify the target protease(s). We have used these techniques to characterize the cleavage of RLQLKL ACPP and demonstrate cleavage by multiple serine proteases. We then generated an improved, acetylated substrate that was still efficiently cleaved by tumor extracts and elastases but had reduced cleavage by liver and kidney and was not cleaved by plasmin. Based on this and other presented data showing expression of elastase in PyMT tumors and inhibition of protease activity in tumor extracts by α1-antitrypsin and prior reports of elastase expression in breast cancers, we postulate that elastases are probably responsible for a major part of tumor uptake of RLQLKL ACPPs in PyMT tumors. Despite this claim, the identification of one specific protease that is responsible for 100% of cleavage and tumor uptake of any specific ACPP may not be possible because cleavage and tumor uptake probably results from a plurality of proteases present in the extracellular matrix of tumors. Additionally complicating the one enzyme-targeted mechanism is that peptide substrates and inhibitors that are exquisitely specific to a single protease remain a biochemical challenge. Interestingly, the identification of a protease substrate that is cleaved by multiple proteases could be expected based on the presented selection strategy that identified phage modified by extracellular enzymes expressed by tumors that probably contained more than one protease. Validation of this possibility for in vivo substrate cleavage by a multiplicity of proteases and utilizing this as a tool for synergistic multiprotease tumor targeting would represent a novel mechanism for targeting cancers and other tissues.
To highlight the technical improvements in the novel phage selection strategy presented here, it must be compared with prior in vivo phage selections. Previously published in vivo phage selection strategies relied almost entirely on tissue-specific binding, whereas the selection scheme presented here represents a novel use of phage display as an unbiased way to identify sites of in vivo enzymatic activity, in this case protease activity within tumors. Our strategy includes selection for covalent modification of the targeting sequence and not simply binding to the target tissue. Exclusive reliance on such tissue binding and localization for selection has the problem that phage particles have been shown to be nonspecifically retained in tissues for ≥24 h after in vivo injection (26). Future selections using our new strategy could become more elaborate by targeting specific intracellular compartments, either by including subcellular fractionation in the recovery step or by selecting for a compartment-specific biochemical modification, such as phosphorylation or biotinylation.
We thank Evangeline Mose, Holly Weld, Tim Salazar, Rachel Levin, and Perla Arcaira for technical assistance with SUVs and phage injections; Tao Jiang for discussions and training in peptide synthesis; Pfizer for a gift of prinomastat; and B. Ondek, S. Rodems, and Marco Gallio for critical review of and suggestions on the manuscript.
2The abbreviations used are: