Dissecting MMP P10′ and P11′ subsite sequence preferences, utilizing a positional scanning, combinatorial triple-helical peptide library

Matrix metalloproteinases (MMPs) are a family of zinc-dependent endopeptidases that remodel the extracellular matrix environment and mitigate outside-in signaling. Loss of regulation of MMP activity plays a role in numerous pathological states. In particular, aberrant collagenolysis affects tumor invasion and metastasis, osteoarthritis, and cardiovascular and neurodegenerative diseases. To evaluate the collagen sequence preferences of MMPs, a positional scanning synthetic combinatorial library was synthesized herein and was used to investigate the P10′ and P11′ substrate subsites. The scaffold for the library was a triple-helical peptide mimic of the MMP cleavage site in types I–III collagen. A FRET-based enzyme activity assay was used to evaluate the sequence preferences of eight MMPs. Deconvolution of the library data revealed distinct motifs for several MMPs and discrimination among closely related MMPs. On the basis of the screening results, several individual peptides were designed and evaluated. A triple-helical substrate incorporating Asp–Lys in the P10′–P11′ subsites offered selectivity between MMP-14 and MMP-15, whereas Asp–Lys or Trp–Lys in these subsites discriminated between MMP-2 and MMP-9. Future screening of additional subsite positions will enable the design of selective triple-helical MMP probes that could be used for monitoring in vivo enzyme activity and enzyme-facilitated drug delivery. Furthermore, selective substrates could serve as the basis for the design of specific triple-helical peptide inhibitors targeting only those MMPs that play a detrimental role in a disease of interest.

Matrix metalloproteinases (MMPs) 2 are a family of zinc-dependent endopeptidases that belong to the metzincin super-family. They have long been recognized for their roles in the degradation of extracellular matrix components (1). More recently, the indirect actions of MMPs on signaling pathways have been revealed (2)(3)(4)(5)(6)(7)(8). Numerous pathological states involve an imbalance between MMPs and their natural inhibitors, such as tumor growth and invasion, cartilage degradation, or atherosclerotic plaque formation and rupture (9 -13). Conversely, some MMPs have been shown to play a host of beneficial roles during certain pathological conditions (14).
Collagen is the major structural scaffold in the body and serves as barrier between tissues. Collagen catabolism is important in development and homeostasis, and during these processes collagen turnover is tightly regulated. When regulation is compromised, collagen catabolism can contribute to pathogenesis. Collagen triple-helical structure renders it resistant to general proteolysis. Several MMPs are capable of cleaving the triple-helical regions of collagen, including secreted MMPs (MMP-1, MMP-2, MMP-8, MMP-9, and MMP-13) and membrane-type MMPs (MT1-MMP/MMP-14 and MT2-MMP/ MMP-15) (15,16). MMP-mediated collagenolysis is associated with numerous diseases (16).
Triple-helical peptides (THPs) that model collagen have been developed in two ways to help better define the roles of collagenolytic MMPs in health and disease. First, THP substrates have (a) identified collagen-based sequence specificities of MMPs (17), (b) led to mechanistic proposals for the initial stages of collagenolysis (18 -21), and (c) been developed as in vivo imaging agents (22)(23)(24) and enablers of targeted drug delivery (25)(26)(27). Second, transition state analogs have been integrated into THP assembly to create triple-helical peptide inhibitors that are efficacious in in vivo models of disease (28 -32). THPs are promising systems for imaging, drug delivery, and biomaterial development, as triple-helical conformation is reasonably stable to general proteolysis, as observed in vitro in mouse, rat, and human serum and/or plasma and in vivo in rats (33)(34)(35)(36)(37). The stability of THPs has allowed for their administration orally (38).
A better understanding of the sequence preferences for individual MMPs will allow for discrimination between various family members and design of agents capable of targeting a selected enzyme. The need for highly selective MMP sequences is critical for the development of imaging agents and prodrugs (39). There have been numerous peptide combinatorial, phage display, and protease-generated peptide library studies designed to elucidate the sequence preferences of MMP family members (40 -51). However, none of these studies considered sequence preferences in the context of a collagen triple helix. We have found that MMP sequence preferences derived from single-stranded substrates do not translate to sequence preferences in a triple-helical context (32,52). Additionally, none of the prior library studies evaluated the impact of residues beyond the substrate P 5 -P 5 Ј subsites, whereas collagen triple helices are known to interact with MMPs both within and outside of these subsites (18,19,21,29,30,53,54).
To examine MMP sequence preferences in a triple-helical context, this study has generated a FRET THP (fTHP) library using positional scanning substrate combinatorial library (PSSCL) methodology. Such an approach (referred to as activity-based fingerprinting) has been described previously using single-stranded templates with several proteases (55,56), including the metalloproteases thermolysin and actinase E (57). This study utilized a PSSCL of fTHPs to identify collagenolytic MMP sequence preferences in the P 10 Ј and P 11 Ј subsites. The MMPs investigated here include the "classic" collagenases (MMP-1, MMP-8, and MMP-13), the two "gelatinases" (MMP-2 and MMP-9), and two membrane-type MMPs (MMP-14 and MMP-15), all of which have been shown to cleave at least one type of collagen, and MMP-3 serving as a noncollagenolytic MMP control.

Results
The template fTHP sequence ( Fig. 1) contained consensus residues 769 -783 from the MMP hydrolysis sites in types I-III collagen and was sandwiched by five Gly-Pro-Hyp triplets (30,53). The three chains were covalently linked by a C-terminal Lys branch (Fig. 1) to limit variation possibilities in PSSCL and to promote the formation of the triple helix. The template fTHP included subsites P 10 Ј and P 11 Ј, which are important based on collagen interactions with the HPX domains of MMPs (18,19,30,53). For positional scanning, one uses a defined single amino acid at one position and randomizes the other positions. This can be done either by a split-and-pool method (where the resin is split and each part is reacted with a single amino acid and then all parts are pooled together for the rest of the synthesis) (58 -61) or by competitive coupling of a mixture of Fmoc-pro-tected amino acids at isokinetic ratios to achieve equimolar incorporation (58,62). Using the latter approach (as illustrated by Ostresh et al. in Fig. 2 (63)) and our SPPS methods, one can create an fTHP with a defined amino acid in the P 10 Ј subsite but incorporate a mixture of 18 common amino acids (Cys and Pro are omitted) in the P 11 Ј subsite. The resulting fTHP sublibrary with the defined P 10 Ј subsite is then screened for relative activity by comparing initial velocities (RFU) of every sublibrary with MMPs. A second round of synthesis follows in which the P 11 Ј subsite contains the defined amino acid, and the P 10 Ј subsite incorporates the amino acid mixtures. Ultimately, following kinetic analyses of MMP activity, the best and/or most selective combinations of P 10 Ј and P 11 Ј are determined. The results of the positional scanning obtained for P 10 Ј and P 11 Ј subsites were examined herein.
The constructs had Mca as a fluorophore and Dnp as a quencher, each located at either side of the cleavage site, which allowed for monitoring of the proteolytic reaction by FRET. This permitted fast, convenient, and continuous monitoring of many samples at the same time. The incorporation of Lys(Mca) and Lys(Dnp) in the P 5 and P 5 Ј subsites, respectively, as well as larger fluorophore/quencher pairs such as fluorescein/4,4-dimethylaminoazobenzene-4Ј-carboxylic acid, have minimal effect on MMP triple-helical substrate specificity or selectivity (64,65).

Sublibrary synthesis
The isokinetic ratios for Fmoc-amino acids reported earlier (66) were initially examined here to determine whether they were effective for the microwave-assisted SPPS protocols utilized for assembling fTHPs. A short "test" peptide, Xaa-Lys(Dnp)-Ala-Arg-Gly-Lys (where Xaa stands for the mixture of 18 amino acids as listed under "Experimental procedures") was synthesized. The amount of each crude peptide was quantified by RP-HPLC using the absorption of the Dnp moiety (at ϭ 363 nm). Although separation of all peaks was not achieved, the area of the peaks that was separated was 5.55 Ϯ 0.67% of the total peak area, which would be expected for an equimolar amount of 18 peptides (Fig. S1). Additionally, the area of the peaks that was not completely separated added up to a number that would correspond to two or three peptides (Ϯ30% maximum deviation from equimolarity for peaks with the area corresponding to three peptides). MS analysis of each of the peaks revealed that the peaks not separated completely by HPLC contained more than one of the desired peptides, whereby the separated peaks contained only one compound. The MALDI-TOF mass spectrum of the crude product revealed the presence of all of the 18 desired peptides (Fig. S2). The present results were comparable with prior studies that evaluated the efficiency of isokinetic mixtures (63,67).
In the reference peptide, fTHP-15b, the P 10 Ј subsite was occupied by Pro and the P 11 Ј subsite by Hyp. In this study, two THP sublibraries were synthesized: P 10 Ј ϭ Z, P 11 Ј ϭ X, and P 10 Ј ϭ X, P 11 Ј ϭ Z. In the first case, the P 10 Ј position was determined ("Z" ϭ specific residue in each of the chains), and the neighboring P 11 Ј position was randomized ("X" ϭ any of the 18 amino acids present in the mixture, and not necessarily the same residue in each chain). In the second case, the P 10 Ј subsite , and the cleavage site is marked with "ϳ". The reference peptide (fTHP-15b) contained Pro in P 10 Ј and Hyp in P 11 Ј.

MMP THP combinatorial library
was randomized, and the P 11 Ј subsite was determined. In total, 38 mixtures from the sublibraries were synthesized, each containing 18 3 (5832) possible peptides.

Activity screening
Eight collagenolytic MMPs and one noncollagenolytic MMP (MMP-3) were chosen to test with the sublibraries in a FRETbased assay. To account for varying activity of different batches of enzymes, all data were normalized for each enzyme in relation to the results obtained for fTHP-15b.

P 10 ‫؍‬ Z and P 11 ‫؍‬ X sublibraries
In the reference peptide fTHP-15b, the P 10 Ј subsite was occupied by Pro. It was substituted by one of 18 amino acids to account for the specific residue plus the original Pro. The P 11 Ј subsite was randomized, and the rest of the sequence was identical to fTHP-15b.
MMP-3 proved to be the least active among the tested proteases, displaying very low to negligible hydrolysis rates toward the sublibrary (Fig. 2). This is not unexpected, as MMP-3 is noncollagenolytic and typically displays little or no activity toward THP substrates (52,68). When normalized to average activity compared with fTHP-15b (Fig. 3), some differences in MMP-3 activity could be observed based on the known residue in the P 10 Ј subsite. Overall, one must still consider that these activities are not very significant and almost certainly not physiologically relevant.
The highest average activity among the MMPs was observed for MMP-2 (Fig. 3). The most favorable mixtures contained charged residues in the P 10 Ј subsite: Lys (176.6%), Arg (145.7%), Glu (155.0%), and Asp (122.2%), with the exception of His (81.5%). Other mixtures that surpassed the activity observed with the reference peptide were Gly, Val, and Thr with 125.7, 109.2, and 116.5%, respectively. Results for Ala, Asn, and Ser were very close to those observed for fTHP-15b. The least pre-ferred mixtures were by far Leu and Ile, with only 34.6 and 46.0% activity, respectively. Very low MMP-2 activity was also observed for Met (52.2%) and Trp (56.7%). All remaining mixtures yielded results within a range of 70.4 to 82.2%.
MMP-1 was the second least active enzyme overall (Fig. 2). Contrary to both gelatinases, mixtures with charged residues in the P 10 Ј subsite were strongly disfavored by MMP-1. The mixtures that yielded the lowest activity were Lys (23.7%), Glu (24.8%), and Arg (28.3%) (Fig. 3). Asp was slightly better with 47.8% of the activity observed for fTHP-15b. The activity of the majority of the polar residue-containing substrates was in the same range as Asp (Asn ϭ 42.0%, Gln ϭ 49.9%, Ser ϭ 46.9%, and Thr ϭ 55.6%). His and Tyr in the P 10 Ј subsite yielded good relative activities for MMP-1, with 92.6% for His and 85.1% for Tyr. In general, mixtures containing nonpolar residues in the P 10 Ј subsite were more favored by MMP-1 then charged residues, and the highest activity was achieved with Ile (110.2%) and Pro (104.7%). The lowest activities observed for nonpolar residues were 59.4% for Ala, 64.6% for Phe, and 73.2% for Met.

MMP THP combinatorial library
The remaining mixtures yielded MMP-1 activity in the range of 78.9 to 93.0%.
MMP-13 was the second most active enzyme on average (Fig.  3). MMP-13 preferred mixtures containing negatively charged residues in the P 10 Ј subsite with Asp at 128.5% and Glu at 104.7% activity. The mixture with Gly was also a very good substrate with 112.9% activity compared with the reference peptide. Unlike Asp and Glu, positively charged residues induced decreased activities of 63.5% for Lys and 57.1% for Arg. His and Asn produced activities of 87.5 and 88.8%, respectively. The least preferred residues in the P 10 Ј subsite were Ile (19.1%), Met (27.8%), and Leu (29.5%). The remaining mixtures resulted in MMP-13 activity between 45.7% (Thr) to 72.5% (Ala).
MMP-8 showed the greatest preference for small, nonpolar amino acids in the P 10 Ј subsite, such as Gly (111.9%), Ala (91.5%), and Val (88.9%) (Fig. 3). The mixture containing Pro in the P 10 Ј subsite, as in the reference peptide, was also a reasonable substrate for MMP-8 (79.8% activity). Additionally, mixtures with basic residues yielded good MMP-8 activities (Arg ϭ 87.4%, Lys ϭ 80.0%, and His ϭ 77.1%). The remaining mixtures were weaker substrates yielding enzyme activity at around 50%. The least preferred residues in this position were Trp and Asp with 42.7 and 46.2% activity, respectively.
Both membrane-type MMPs (MMP-14 and MMP-15) showed very similar substrate preferences (Figs. 2 and 3). MMP-15 showed slightly higher tolerance of residues with aromatic side chains in the P 10 Ј subsite, whereas MMP-14 showed higher activity toward Ala (Fig. 3). The highest activity of these enzymes was observed toward mixtures with Gly, Val, Pro, and Ala in the P 10 Ј subsite (all in the same range as fTHP-15b). Mixtures with Lys and Arg also resulted in reasonable activity of these two enzymes (Fig. 3). Aside from the earlier mentioned aromatic residues, the lowest activity was observed toward Met, 54.7% for MMP-14 and 66.1% for MMP-15.

P 10 ‫؍‬ X, P 11 ‫؍‬ Z sublibraries
The second sublibrary contained a defined residue in the P 11 Ј subsite, whereas the neighboring P 10 Ј subsite was randomized. In the reference fTHP-15b, this position was occupied by Hyp, and thus a mixture containing this residue was included. Again, the lowest overall activity among the tested proteases was observed for MMP-3 ( Fig. 4).
MMP-1 preferred nonpolar aromatic amino acids in the P 11 Ј subsite, as the highest activity was observed for Phe (100.4%) and Trp (99.2%) (Fig. 5). The aromatic but polar Tyr in P 11 Ј provided a poorer substrate with 71.7% activity. Good activity was observed for nonpolar and bulky residues such as Met (97.0%) and Ile (89.1%). Leu in P 11 Ј, however, resulted in a lower MMP-1 activity (69.8%). Quite surprisingly, a significant difference was observed between mixtures containing Asp or Asn and Glu or Gln. MMP-1 favored the larger Glu/Gln (64.1 and 77.7% activity, respectively) over Asp/Asn (29.2 and 36.9% activity, respectively), and amide over acid. The only mixture that resulted in lower MMP-1 activity than Asp and Asn was His (29.0%). The remaining basic residues in this subsite provided activities of 56.7% for Lys and 68.2% for Arg.
Similarly, as in the P 10 Ј subsite sublibrary, significant variations in MMP-2 activity were observed for the P 11 Ј subsite sublibrary. The mixture most preferred by MMP-2 was Lys in P 11 Ј with an activity of 212.7% (Fig. 5). Another basic, charged residue (Arg) also resulted in quite high MMP-2 activity of 130.7%. However, Ile (192.0%), Asn (171.9%), and Gln (149.4%) were better substrates for this enzyme than Arg. Phe and Ala were also good substrates with activities of 124.6 and 111.4%, respectively. Unlike Asn and Gln, their charged counterparts yielded significantly lower activities (76.6% for Asp and 88.7% for Glu). The least preferred mixtures contained a hydroxyl residue in the P 11 Ј subsite (Ser ϭ 45.6%, Thr ϭ 45.1%, and Tyr ϭ 43.4%).

MMP THP combinatorial library
Trp in P 11 Ј also provided a rather weak substrate with 51.2% of the MMP-2 activity toward fTHP-15b.
In general, MMP-9 was less active than MMP-2 ( Fig. 4), but it showed similar preferences (Fig. 5). The best substrate turned out to be Ile with 191.1% activity, whereas Asn was the second best substrate with 163.9% activity. However, MMP-9 activity toward the Gln mixture was only 87.0%, much lower than MMP-2 activity toward this mixture (Fig. 5). The mixture containing Lys yielded 149.1% MMP-9 activity, one of the highest results for MMP-9, but 63.6% less than what was observed for MMP-2. A significant difference was observed between MMP-2 and MMP-9 activities for the Ala P 11 Ј mixture with a drop from 111.4% (MMP-2) to 64.9% (MMP-9). The only mixture that yielded higher MMP-9 activity compared with MMP-2 was His, 86.3% for MMP-9 and 70.7% for MMP-2. In similar fashion to MMP-2, the least favored MMP-9 mixtures were the ones with a hydroxyl residue, with Tyr yielding the lowest MMP-9 activity of 34.8%.
MMP-8 had better overall activities then MMP-1 (Fig. 4). Modifications in the P 11 Ј subsite had the most detrimental effect on MMP-8, as the average activity toward this sublibrary was only 52.3% of the activity observed for fTHP-15b (Fig. 5). Only the Arg mixture offered comparable activity (105.7%) to fTHP-15b. The second best substrate was Phe with 83.7% activity. The most interesting result was observed for Lys in the P 11 Ј subsite, as the MMP-8 activity observed toward this mixture  Relative activity displayed is based on normalizing the RFU increase using hydrolysis of fTHP-15b by a specific MMP as 100% activity.

MMP THP combinatorial library
was only 0.5% of that for fTHP-15b (Fig. 5). His also resulted in very low activity at 13.8%. The remaining mixtures resulted in MMP-8 activities in the range of 50 -60%.
fTHP-15b was cleaved the most efficiently by MMP-13 ( Fig.  4). None of the mixtures in this set turned out to be a better substrate for MMP-13 than fTHP-15b. The only mixtures that yielded comparable levels of MMP-13 activity were Lys (97.7%) and Asn (93.6%). Mixtures with Asp, Glu, and Gln turned out to be satisfactory substrates for MMP-13 with activities of 80.0, 75.2, and 70.0%, respectively ( For MMP-15, similar activity patterns to MMP-14 were observed. The most notable differences were the best substrates being Arg and Lys (100.7 and 100.5% activity, respectively) (Fig.  5). Phe and Ile provided good substrates for MMP-15, but not as good as for MMP-14 (87.1% for Phe and 74.1% for Ile). Also, the detrimental effect on MMP-14 activity observed for His was greater in the case of MMP-15 (17.6% compared with fTHP-15b). The second weakest substrate was the Hyp mixture, which resulted in 50.0% of MMP-15 activity, indicating a substantial gap between His and the second weakest substrate.

Individual peptides derived from sublibraries
Four peptides were designed to examine whether the findings from the sublibrary screening translated to the activity of single substrates. These peptides had the same sequence as the reference fTHP-15b with the exception of the P 10 Ј and P 11 Ј subsites. Peptides were designed to potentially discriminate within the collagenolytic MMP family based on results from the fTHP sublibraries (see below). The peptides were named according to the residues found in the P 10 Ј and P 11 Ј subsites. For example, Ile-Lys fTHPb contained Ile in the P 10 Ј and Lys in the P 11 Ј, and the rest of the sequence was identical to fTHP-15b ( Table 1).
The synthesis of the four individual peptides yielded crude products of good purity with ϳ50% yield, satisfactory for ϳ14.5 kDa fTHPs. Crude and purified individual peptides were examined for enzymatic activity and similar trends for peptides in both forms were observed (Fig. 6).
The Ile-Tyr fTHPb turned out to be a good MMP-1 substrate with 235% of the activity of the reference fTHP-15b ( Table 2). The P 10 Ј subsite Ile and P 11 Ј subsite Tyr mixtures were not favored by MMP-2 and MMP-9 (Figs. 3 and 5), and the substrate based on these sublibraries (Ile-Tyr fTHPb) resulted in reduced activity for MMP-2 (59.1% of that for fTHP-15b) and significantly lowered activity for MMP-9 (2.1% of that for fTHP-15b) ( Table 2). MMP-13 did not favor these mixtures (Figs. 3 and 5) but combining them provided similar MMP-13 activity of that for fTHP-15b ( In the second peptide Ile in P 10 Ј was paired with Lys in the P 11 Ј subsite. The mixture with Lys in P 11 Ј yielded different results than Tyr in this position (Fig. 5). Comparing Ile-Lys fTHPb with Ile-Tyr fTHPb, Lys was more favored by the gelatinases than Tyr, resulting in 1.35-fold higher MMP-2 activity and 4.1-fold higher MMP-9 activity ( Table 2). P 11 Ј Lys was less preferred by MMP-1 than Tyr in the mixtures (Fig. 5), and MMP-1 activity toward Ile-Lys fTHPb was lower than Ile-Tyr fTHPb ( Table 2). MMP-13 preferred Lys over Tyr in the P 11 Ј subsite mixtures (Fig. 5), whereas Ile-Lys fTHPb was a slightly worse substrate than the Ile-Tyr fTHPb for MMP-13 ( Table 2). The Lys P 11 Ј mixture produced only 0.5% MMP-8 activity ( The third peptide contained Lys in the P 11 Ј subsite and Asp in the P 10 Ј subsite. In the P 10 Ј sublibrary, charged residues had a detrimental effect on the activity of MMP-1 (Fig. 3). However, MMP-1 activity toward the Asp-Lys fTHPb was 214% compared with fTHP-15b (Table 2). Similarly, the P 10 Ј subsite Asp mixture resulted in one of the weakest substrates for MMP-8 (Fig. 3), whereas the Asp-Lys fTHPb yielded 92.8% relative MMP-8 activity. Moreover, Asp was one of the most preferred residues by MMP-2 in the P 10 Ј subsite sublibrary and was above average for MMP-9 (Fig. 3). Asp combined with Lys in the Asp-Lys fTHPb resulted in 59.6% MMP-2 activity and 10.4% MMP-9 activity compared with fTHP-15b (Table 2). Unlike the P 10 Ј subsite Ile, the P 10 Ј subsite Asp mixture was the most favored by MMP-13 (Fig. 3), but when paired with Lys in the P 11 Ј subsite, the Asp-Lys fTHPb was not as preferred as either of the mixtures and yielded only 22.8% MMP-13 activity ( Table   Table 1 Individual peptide sequences 2). In similar fashion, the Asp-Lys fTHPb was a good substrate for MMP-14 (81.3% activity compared with fTHP15b) and a poor substrate for MMP-15 (3.1% activity compared with fTHP15b) ( Table 2), which did not correlate with the results from the mixtures (Figs. 3 and 5). It is important to note that Gly-Asp-Lys-type THPs possess two lateral salt bridges that can be formed between residues in the i ϩ 1, i positions in the leading and middle chains and middle and lagging chains (Lys 3 Asp) (69,70). It is possible that hydrogen-bonding interactions that would occur between the various MMPs and Asp in the P 10 Ј subsite, either favorable or unfavorable for catalysis, are disrupted by salt bridges in Asp-Lys fTHPb.
The fourth peptide contained Lys in the P 11 Ј subsite and Trp in the P 10 Ј subsite. Trp in this position resulted in higher MMP-1 activity than the Ile-Lys and Asp-Lys fTHPbs ( Table  2). The Trp-Lys fTHPb was a very good substrate for MMP-2, providing 198.7% activity compared with fTHP-15b (Table 2). However, the other gelatinase showed much less preference for this substrate with only 22.6% MMP-9 activity. Trp-Lys fTHPb was the most preferred substrate for MMP-8 among the four individual fTHPbs (111.8% activity compared with the reference peptide) ( Table 2). The Trp-Lys fTHPb was also a good substrate for MMP-13 (117.6% activity) ( Table 2). Trp indicated a differentiation of the membrane-type MMPs substrate preferences, as MMP-15 had a higher tolerance for this residue in the P 10 Ј subsite (128.6% activity) compared with MMP-14 (53.5% activity). With the exception of MMP-8, the activities of the Trp-Lys fTHPb were consistent with the trends observed in the respective mixtures (Figs. 3 and 5).
After considering the results from the individual peptides, the P 10 Ј Ile and P 11 Ј Tyr substitutions were incorporated into the C-terminal region of fTHP-18. fTHP-18 has a longer C-terminal native collagen sequence compared with fTHP-15b (Table 1), which has been shown to be less favored by MMP-13 than fTHP-15 and preferred by MMP-1 (30,53). Additionally, the individual peptide strands in fTHP-18 were not covalently linked by the Lys branch. In the P 10 Ј subsite, fTHP-18 contains a Leu residue, which is thought to contribute to recognition of collagen by MMP-1 (19). Replacing this Leu residue and the neighboring Hyp with Ile and Tyr resulted in a significant increase of MMP-13 and MMP-15 activity, an increase of MMP-1 activity, a small decrease of MMP-8 activity, and a significant decrease of MMP-2 and MMP-14 activity (Table 2). MMP-9 preferred the longer C-terminal sequence of fTHP-18 over fTHP-15b, but the Ile-Tyr substitution reduced the activity (Table 2).

CD spectroscopic analysis
The thermal stabilities of the mixtures were evaluated using CD spectroscopy to determine whether (a) all peptides formed a triple helix and (b) whether differences in activity correlated to triple-helical stability. The CD spectra for each mixture showed characteristics consistent with the triple-helical structure, with a maximum at ϭ 225 nm and a minimum at ϭ 200 nm (data not shown) (71). Thermal transition curves were evaluated to determine T m values and thus the influence of each residue on the stability of the triple helix ( Table 3). The most stable triple helix in the P 11 Ј sublibrary was realized with Hyp, the residue that occupies this position in the template fTHP-15b. With a T m ϭ 49.8°C the mixture with Hyp in the P 11 Ј subsite was the most stable among all of the mixtures tested (Table 3). In the P 11 Ј sublibrary, four other mixtures reached a T m above 40°C: Val, Ala, Asp, and Leu. Four residues resulted in mixtures with T m below 30°C: Trp, Lys, Phe, and His.
Pro, which occupies the P 10 Ј subsite in the reference peptide, yielded among the most stable P 10 Ј mixtures with a T m ϭ 38.6°C (Table 3). However, mixtures containing Ser, Ile, Phe, and Asn were more stable with T m values of 41.6, 40.4, 39.2, and 38.8°C, respectively (Table 3). Gly was the residue that destabilized the triple helix the most in the P 10 Ј subsite (T m ϭ 25.2°C). This was not unexpected, as Gly in an Xaa position of the Gly-Xaa-Yaa collagen repeat (which is where the P 10 Ј subsite is) is very destabilizing for triple-helical structure (72).

MMP THP combinatorial library
Other P 10 Ј subsite mixtures that exhibited a T m lower than 30°C were Val, Asp, and His ( Table 3).
The CD spectra of individual peptides were subsequently analyzed (Fig. 7). All CD spectra showed distinct characteristics of triple-helical structure as described above (71). T m values were determined from the thermal transition curves (Table 3). For fTHP-15b, the P 9 Ј-P 10 Ј-P 11 Ј subsite Gly-Pro-Hyp triplet resulted in the most stable triple helix with a T m ϭ 44.5°C. The P 10 Ј Asp and P 11 Ј Lys mixtures both had a rather low T m value at around 28°C, and their combination in the Asp-Lys fTHPb resulted in the least stable triple helix with a T m ϭ 25.6°C. The P 10 Ј Ile was the most stable in its sublibrary with a T m ϭ 40.4°C, but when combined with a Lys in P 11 Ј, the resulting Ile-Lys fTHPb had a T m value of only 26.0°C. Likewise, the Ile-Tyr fTHPb had a significantly lower stability than for each of the respective mixtures. However, Trp-Lys fTHPb turned out to be more stable than either the P 10 Ј Trp or P 11 Ј Lys mixtures. Comparison of the thermal transition curves between crude and purified individual fTHPbs (Fig. 8) showed no significant differences. Thus, we can assume that differences between mixtures and individual peptides were not due to impurities found in the mixtures.

Discussion
The FRET-based positional scanning approach demonstrated here has the benefit of conveniently testing thousands of compounds in a relatively short amount of time. Moreover, it allows for the simultaneous evaluation of both the prime and nonprimed regions of the substrate, provided that there is no shift in the cleavage site, without the need of MS deconvolution. In positional scanning, small deviations in equimolarity are unavoidable even when using isokinetic ratios of reagents dur- ing sublibrary synthesis. However, reasonable deviations in equimolarity are not ultimately detrimental for screening. Even a 10-fold error in a mixture equimolarity resulted in less than a 2-fold error in the screening result (IC 50 value) (67), which does not eliminate the usefulness of a library for valid screening. Thus, we were satisfied with the maximum error in equimolarity (less than 12% for the separated peptides and less than 30% for peaks containing three peptides) of our test peptide obtained using the isokinetic ratios published previously (66).
Because of the complexity of the THP mixtures both in the number of compounds present and the triple-helical nature of collagen-model peptides, it is not possible to evaluate equimolarity in this system. However, based on the results of the test peptide and high efficiency of microwave-assisted coupling reactions, it is unlikely that there would be variation in the reaction rates significant enough to render any of the mixtures not suitable for meaningful screening. The activity obtained with individual peptides validated the approach as in general their results correlated to the corresponding mixtures (Table  2), and no significant differences were observed between crude and purified peptides (Fig. 6). Our approach allowed for a direct, quantitative evaluation of the different residues in two substrate positions on the activity of MMPs. This work revealed general trends in sequence preferences of several MMPs, including the preference of MMP-1 toward aromatic amino acids in the P 11 Ј subsite and the strong preference of MMP-2 for charged residues in the P 10 Ј subsite. The result that mixtures containing nonpolar residues in the P 10 Ј subsite were more favored by MMP-1 then charged residues was consistent with the known composition of the MMP-1 S 10 Ј pocket, which is delineated by Ile-271, Met-276, Phe-301, Trp-302, and the alkyl portion of Arg-272, all nonpolar entities (19). Moreover, we have been able to find motifs that differentiate within the same sub-group of the MMP family. Here, we report for the first time the differentiation between the two transmembrane MMPs (MMP-14 and MMP-15), which was observed for the Asp-Lys fTHPb, fTHP-18, and with the aromatic residues in the P 10 Ј sublibrary. Similarly, the Asp-Lys and Trp-Lys motifs introduced a distinction between the gelatinases (MMP-2 and MMP-9). Also of interest is the Ile-Lys motif, which yielded a substrate with a much lower MMP-1 activity compared with the other MMPs.
The results of this study show that the P 10 Ј and P 11 Ј positions of substrate play an important role in the collagenase-substrate interactions and that modifying these residues can drastically affect the affinity of an MMP toward its THP substrate. This effect can be explained by the binding of the primed region of the THP by the HPX domain as initially shown for MMP-1 (18,19,30,73). There are notable interactions of the THP P 10 Ј and P 11 Ј subsites with blades I and II of the MMP-1 HPX domain (19). Similar interactions of a triple-helical substrate with the HPX domains in other MMPs are very probable, as not only MMP-1 was affected by substitutions in the two subsites investigated here. It is important to note that the alignment of the THP can vary depending upon the MMP. THP alignment is shifted in MMP-14 compared with MMP-1 (20). Structural analysis of THP binding to the MMP-14 HPX domain suggested that the P 10 Ј and P 11 Ј subsites were at the very edge of interaction with blade II of the HPX domain (20).
In addition to THP alignment, one must consider the role of the MMP-2 and MMP-9 FN inserts in collagen binding. Molecular dynamics simulations yielded two modes of binding for fTHP-15 with MMP-2 (74). One mode indicated binding of MMP-2 HPX domain Arg-482 with Hyp in the P 11 Ј subsite (12% frequency), whereas the other mode indicated binding of MMP-2 FN III Cys-363 with Hyp in the P 11 Ј subsite (85% frequency), Met-373 with Pro in the P 10 Ј subsite (46% frequency), and Tyr-360 with Hyp in the P 11 Ј subsite (32% frequency) (74). Experimental data indicated that the MMP-2 FN III insert (particularly residue Arg-368) was primarily responsible for interaction with fTHP-15 (75). In the case of MMP-9, prior studies suggested that residues within the FN II insert (Arg-307-Asp-323) interact with the triple helix (76). The P 10 Ј and P 11 Ј resi-

MMP THP combinatorial library
dues from the present fTHP library would not quite reach this interaction site in MMP-9 (29). MMP-1 and MMP-8 have been proposed to have similar collagenolytic mechanisms (77). Data from the screening of P 10 Ј and P 11 Ј subsite sublibraries indicate that there are different interactions with these subsites for MMP-1 and MMP-8 (particularly when Met, Trp, Arg, or Lys occupies the P 11 Ј subsite and Lys, Arg, Trp, or Ile occupies the P 10 Ј subsite) (Figs. 2 and  4). Replacing the P 11 Ј Hyp with Met or Trp did not affect MMP-1 activity, but it had a negative effect on MMP-8 activity. Lys in the same position had a detrimental effect on MMP-1 activity, but MMP-8 displayed almost no activity at all toward this mixture. However, an Arg residue resulted in no loss of MMP-8 activity, but MMP-1 activity was significantly lower toward this mixture.
In the P 10 Ј subsite, both positively charged residues decreased MMP-1 activity but had no detrimental effect on MMP-8 activity. Ile was the residue most preferred by MMP-1 in the P 10 Ј subsite while being one of the least favored residue by MMP-8. Similarly, as in the case of the P 11 Ј subsite, a Trp residue in P 11 Ј did not affect MMP-1 activity drastically, but it had a detrimental effect on MMP-8.
Structures of MMP-1 interacting with triple-helical peptides have been obtained (18,19). Unfortunately, there is no X-ray crystallographic or NMR-derived structure of the MMP-8 HPX domain available for direct comparison with MMP-1. However, homology modeling using I-TASSER and MODELLER (78 -81) indicates high structural similarity between the HPX domains of MMP-1 and MMP-8 (Fig. 9). When the sequences of the two enzymes are aligned (82), there is only one difference in the residues that interact with the THP P 11 Ј subsite (Fig. 10). The MMP-1 Gln-354 residue that interacts with the THP leading strand is replaced by Tyr-355 in MMP-8. The MMP-1 residues interacting with the THP middle strand are Arg-291 and Glu-293, corresponding to Arg-292 and Glu-294 in MMP-8. The P 11 Ј Hyp residue in the trailing strand seems not to interact with any of the residues of MMP-1. However, the P 10 Ј Leu from the trailing chain exhibits hydrophobic contacts with Phe-308 and Tyr-309, which in MMP-8 are replaced with Gln-309 and Leu-310, respectively. Leu from the middle chain interacts with  Differences in residues interacting with the THP P 10 Ј and P 11 Ј subsites are indicated. The MMP-1 HPX domain (blue) is adapted from PDB entry 4AUO (19). The MMP-8 HPX domain structure was predicted by homology modeling using I-TASSER (yellow) or MODELLER (sea green). multiple residues, but only two are different between MMP-1 and MMP-8; Ile-290 and Met-295 are replaced by Leu-291 and Leu-296, respectively, in MMP-8 (Fig. 9). This suggests that differences in activities of MMP-1 and MMP-8 observed toward P 11 Ј mixtures were due to interactions with the leading strand (Fig. 10), and in the case of P 10 Ј, it was due to the interactions with the middle and trailing strands of the THP. One might hypothesize that lower MMP-8 activity observed toward Lys, Trp, or Met residues in the P 11 Ј subsite might be caused by the steric clashes considering the bulkier Tyr residue in MMP-8 compared with the Gln residue in the case of MMP-1 (Fig. 9). The MMP-1 Gln-354 may also play a role in the enzyme preference for amide side chains (Gln, Asn) versus acid side chains (Glu, Asp) in the P 11 Ј subsite (Fig. 4), based on forming a hydrogen bond analogous to that found when Hyp is in this subsite (Fig. 10). However, why Arg was preferred by MMP-8 in this subsite is not readily apparent and is subject to investigation in future studies.
MMP-2 and MMP-9 were initially recognized for their ability to process denatured collagen (gelatin) (1). The stability of the triple helix can influence the activity of MMP-2 and MMP-9, as well as other MMPs (68,83,84). MMP-2 favored the Trp-Lys fTHPb (Table 2), which was the most thermally stable individual peptide (Table 3). Thus, the relative stability of the triple helix was not a significant contributor to the MMP-2 activities. MMP-1 has been observed to favor less stable triple helices (68,83). MMP-1 favored one less stable substrate (Asp-Lys fTHPb), but this was not a general trend (Tables 2 and 3). Although the least stable mixture (Gly in the P 10 Ј subsite) showed some good activities (for MMP-8, MMP-13, and MMP-2), it did not exhibit significantly different activities compared with more stable fTHPs (Fig. 2). Overall, triple helix stability did not appear to direct MMP activities compared with individual amino acid preferences.
Data accumulated using the present combinatorial library approach has three potential weaknesses. First, nonadditive effects may occur between neighboring subsites, which can skew results obtained for amino acids in the fixed position.
Comparison of preferences obtained from analysis of the combinatorial libraries to substitutions in individual peptides indicated that some nonadditive effects may have occurred, but they did not significantly impact the overall trends observed. Second, the mixture subsite may contain two or three different amino acids (one in each strand), which may exacerbate nonadditive effects. Third, the preference for a specific amino acid in a specific strand of the triple helix (leading, middle, or trailing) cannot be evaluated using the present methodology. Prior structural studies of MMP⅐THP complexes have revealed that individual THP chains are differentiated by the enzyme during binding, and chain-specific interactions occur (see above) (18 -20, 85). In the future, individual chain interactions can be examined using heterotrimeric triple-helical constructs (31,86,87).
MMP-2 was found to have significant activity toward the triple-helical peptide library. The activity of MMP-2 toward native collagen has been controversial, and the role of this enzyme as a physiological collagenase has been questioned (16). This study, in concert with a prior study (84), further indicates that factors beyond local sequence can dictate whether or not an enzyme such as MMP-2 can catalyze the hydrolysis of collagen efficiently. This conclusion is similar to one obtained when comparing MMP-1 activity toward homotrimeric and heterotrimeric type I collagen (88,89).
This study allowed for differentiation between MMPs that have traditionally been challenging to target for selectively, such as MMP-14 versus MMP-15, MMP-14 versus MMP-8, and MMP-2 versus MMP-9. In the future, we plan to investigate the effects of other subsites, and non-native amino acid residues, in THP MMP substrates to further elucidate the sequence preferences of MMPs. This will provide more subsite data that could be used to rationally design a truly selective substrate or inhibitor of the targeted MMP. Such substrates could be used as selective probes for individual MMP activity (17) in the early diagnosis of pathological states and selective targets for MMPresponsive drug delivery vehicles (90). Selective motifs can facilitate the design of transition-state analog inhibitors (28 -32). These highly proteolytically stable molecules could act as therapeutic agents targeting only the MMPs involved in the disease, while sparing the ones with host beneficial functions.

Materials and enzymes
The reagents and solvents used for peptide synthesis were  The THP leading (L) strand is colored in cyan, the middle (M) strand in green, and the trailing (T) strand in red. P 11 Ј Hyp residues from each strand and the residues from the enzyme that are closest to them are represented as sticks. Enzyme surface that is within 4 Å from any of the THP strands is colored with respective to the individual strand. Adapted from PDB entry 4AUO (19,96).

Peptide synthesis
All peptides were synthesized by Fmoc-chemistry using TentaGel S RAM on a 0.025-mmol scale. The syntheses were carried out on a Liberty Blue automated microwave-assisted peptide synthesizer (CEM Corp., Matthews, NC) equipped with a Discover microwave module. Before the synthesis, the resin was solvated in 15 ml of DMF for 300 s. Coupling reactions were performed using 5 eq each of Fmoc-amino acid, DIC, and Oxyma (91 The branch component of all peptides was synthesized several times in larger batches. Lys(Dde)-Lys(Dde)-Gly was assembled at 0.25-mmol scale using standard microwave-assisted solid-phase protocols with Fmoc removal at the end. The resin was washed three times with DMF and three times with DCM. The Dde-protecting groups were removed using 2% hydrazine in DMF (three 5-min cycles), and the resin was washed three times with DMF and three times with DCM. At this stage, a small portion of the peptide-resin was cleaved with TFA and the product analyzed. MALDI-TOF MS analysis revealed the presence of the desired branched peptide. Fmoc-Aha was coupled to each of the amino groups available. Three times excess of Fmoc-Aha was used for each of the amino groups, accounting for a total of 2.25 mmol. The reaction was performed using HCTU (2.1 mmol) and DIPEA (4.5 mmol) for 1.5 h with the ninhydrin test used to confirm the completion of the coupling. If the ninhydrin test was not satisfactory, the coupling was repeated for an additional 1 h using fresh reagents. After the Fmoc-Aha coupling, the branch-resin was washed three times with DMF and three times with DCM, dried in vacuo, and then divided into 10 aliquots, which were later used for the synthesis of THP branched mixtures.
The amino acid mixture for a randomized position contained 18 common amino acids (Pro and Cys not included). The mixture was prepared so that the content of each amino acid was in agreement with the isokinetic ratios for competitive Fmocamino acid coupling reported by Kim et al. (66). The ratios used were 3.10% (Glu), 3.85% (Gly and Val), 4.60% (Ala, Leu, and Phe), 5.35% (Ile, Met, Gln, His, and Ser), 6.20% (Lys, Thr, and Tyr), 6.95% (Trp), and 7.70% (Asn, Arg, and Asp). The coupling reaction conditions for the synthesis of THP sublibraries were the same as described above with the exception of the competitive coupling of the mixture of amino acids, for which the initial coupling step was 15 s at 75°C, 225 W, followed by 220 s at 90°C, 45 W.
fTHP-18 and fTHP-18 Ile-Tyr did not require the use of the branching protocol. Incorporation of individual amino acids was as described above using DIC and Oxyma.
fTHPb mixtures and individual fTHPs were cleaved from the resin, and side chain protecting groups were removed using water/thioanisole/TFA (1:1:18). The reaction was carried out for 3 h under ambient gas atmosphere. fTHPs were precipitated with cold MTBE. The ether layer was removed, and the precipitate was dissolved in water and lyophilized.
Individual peptides were purified using RP-HPLC on an Agilent 1260 Infinity system equipped with a Vydac C 18 (15-20 m, 300 Å, 250 ϫ 22 mm) column and a multi wavelength detector. For Ile-Tyr, Ile-Lys, Asp-Lys, and Trp-Lys fTHPbs, the elution gradient was 30 -70% B in 60 min; A was 0.1% TFA in water; and B was 0.1% TFA in acetonitrile. fTHP-18 and fTHP-18 Ile-Tyr were purified using the elution gradient of 30 -50% B in 40 min. For all purifications, the flow rate was 10 ml/min, and the detection was set at ϭ 220 and 365 nm. Homogeneous fractions were combined and lyophilized.
Peptide purity (Figs. S3-S9) was evaluated using an Agilent 1260 Infinity series HPLC system with a Kinetex C 18 (2.6 m, 100 Å, 150 ϫ 4.6 mm) column and a diode array detector. The elution gradient was 2-98% B in 20 min; A was 0.1% TFA in water, and B was 0.1% TFA in acetonitrile. The flow rate was 1 ml/min, and detection was at ϭ 220 and 365 nm. The identity of synthesized peptides was evaluated by the MALDI-TOF MS analysis of tryptic fragments on a Bruker Microflex mass spectrometer (Bruker Daltonics, Billerica, MA). The digestion was performed using TPCK-trypsin immobilized on agarose beads according to the manufacturer's instructions. Briefly, the immobilized TPCK-trypsin was washed three times with the digestion buffer (0.1 M NH 4 HCO 3 , pH 8.0), and the beads were suspended in the buffer. The suspension was added to the peptide (previously dissolved in the digestion buffer), and the reaction proceeded at room temperature. Each sample was analyzed twice, first after 2 h of reaction and second after overnight incubation. For the branched peptides, fragments corresponding to the two trypsin cleavage sites (at the two Arg-Gly bonds) were observed (Table S1). fTHP15b Trp-Lys generated an additional fragment corresponding to Gly-Trp-Lys (Table S1). For the two nonbranched peptides, fTHP-18 and fTHP-18 Ile-Tyr, fragments corresponding to the entire sequences were identified (Table S1).

Enzyme activity assays
fTHPb mixtures or individual fTHPs were dissolved in Tris buffer (50 mM Tris, 100 mM NaCl, 10 mM CaCl 2 , 0.05% Brij-35, 0.02% NaN 3 , pH 7.5) at a concentration of 25 M and left overnight at 4°C to allow for triple helix assembly. The pro-MMP-1, pro-MMP-2, pro-MMP-8, pro-MMP-9, and pro-MMP-13 were activated by incubation with 1 mM APMA at 37°C for 2 h, with the exception of pro-MMP-9, which required 24 h of incubation. The pro-MMP-3 was activated by incubation with 5 g/ml chymotrypsin for 30 min at 37°C. The reaction was quenched using 2 mM PMSF. The pro-MMP-14 and pro-MMP-15 were activated by treating with 0.1 mg/ml trypsin-3 at 37°C for 2 and 1.5 h, respectively. Trypsin-3 was inactivated by addition of 1 mM AEBSF and incubation for 15 min at room temperature. After activation, all enzymes were immediately diluted to the desired concentration with Tris buffer. Relative MMP activity was initially evaluated by using fTHP-15b and compared with prior data (29,30). In this way, activity toward the substrate was used as an indicator of enzyme integrity (32), rather than TIMP titration, as performed previously (68). Enzyme activity was tested at 5 nM (MMP-2, MMP-8, MMP-13, and MMP-14) or 10 nM (MMP-1, MMP-3, MMP-9, and MMP-15) final concentration and 25 M substrate mixture concentration at room temperature (ϳ25°C). Based on our prior studies (68), the K m value for MMP hydrolysis of fTHP-15 is 7-25 M. Thus, using a 25 M concentration for substrate mixtures is appropriate, as individual substrates within a mixture are present at a concentration much lower than 25 M. Fluorescence was monitored continuously at excitation ϭ 324 nm and emission ϭ 394 nm for 1 h and measured once after 24 h. All measurements were done in triplicate on a Synergy H4 Hybrid Multi-Mode Microplate Reader (BioTek, Winooski, VT). All fluorescence data were normalized, assigning the result for the reference peptide (fTHP-15b) as 100% activity. Hydrolysis at the fTHP-15b Gly-Leu bond was confirmed by MALDI-TOF MS analysis (Table S2). Hydrolysis at the Gly-Leu bond in fTHP-branched mixtures was confirmed by MALDI-TOF MS analysis of randomly selected mixture-enzyme combinations ( Table S2). Determination of individual kinetic parameters was as described previously (68). Hydrolysis at the Gly-Leu bond in purified, individual fTHPs was confirmed by MALDI-TOF MS.

CD spectroscopy
Triple-helical structure was evaluated by near-UV ( ϭ 180 -250 nm) CD spectroscopy. Spectra were recorded on a Jasco J-810 spectropolarimeter (Easton, MD) using a 0.1-cm quartz cuvette. The fTHP concentration was evaluated using a Ther-moFisher Scientific NanoDrop 1000 (Waltham, MA) at ϭ 363 nm, using ⑀ Dnp ϭ 15,900 M Ϫ1 cm Ϫ1 (allowing for three Dnp moieties per triple helix) and was adjusted to 25 M in 0.5% aqueous acetic acid solution. Thermal transition curves were obtained by recording the molar ellipticity ([]) at ϭ 225 nm with the temperature increasing by 20°C/h from 5 to 80°C. Temperature was controlled using a Jasco PTC-348WI temperature control unit. The melting temperature (T m ) was determined as the inflection point in the transition region. The spectra were normalized by designating the highest [] 225 nm as 100% folded and the lowest [] 225 nm as 0% folded (92)(93)(94)(95).