Functionalized osteoarthritis targeting peptides for MRI, lubricant and regenerative medicine


 Despite the efforts made for osteoarthritis (OA) treatment, the results are limited and can be improved by enhancing the OA homing strategy. Here, we used a phage display system to identify OA-targeting peptides, and combined them with magnetic resonance imaging detection reagents to expand their application to early OA diagnosis in rat and swine models. OA-targeting peptides showed better static and kinetic friction characteristics than scrambled peptides, when conjugated with hyaluronic acid for rheological lubrication studies using human OA specimens. Furthermore, mesenchymal stem cells, through CD44 binding to hyaluronic acid conjugated with OA-targeting peptides, showed better capacity for OA homing and repair than those conjugated with scrambled peptides. Protein–peptide docking revealed WXPXW as the consensus binding motifs that bind to collagen XII, a protein exclusively expressed in human and animal OA models. These results suggest the potential of OA-targeting peptides to promote diagnosis, treatment, and regenerative medicine for OA.

Because articular cartilage lacks the capacity for self-repair, incidences of osteoarthritis (OA) are 1 increasing, especially for those older than 60 years of age 1 . Medication with anti-inflammatory 2 drugs, intra-articular injection with lubricating supplements, and surgeries including microfracture 3 and mosaicplasty remain the current modalities for OA treatment, only alleviating symptoms. 4 There are no disease-modifying agents for OA. Cell-based therapy using autologous chondrocyte 5 implantation was only effective in treating focal articular cartilage defects 2 . Transplantation of stem 6 cells or progenitor cells has now emerged as an alternative for chondrocytes in the treatment of OA 7 and osteochondral defects, especially for large lesions 3 . 8 Mesenchymal stem cells (MSCs), with capacities of self-renewal and multipotent 9 differentiation, are not only used to repair mesenchymal tissues but also used in tissue engineering 10 of cartilage and bone 4 . The long-term safety of intra-articular injection of MSCs has been 11 demonstrated in 41 patients with knee OA 3 . Furthermore, the clinical efficacy and safety of 12 MSC transplantation for OA treatment has been demonstrated in a meta-analysis with 11 eligible 13 trials and 582 knee OA patients 5 . 14 A two-year follow-up study regarding the efficacy of intra-articular injection of MSCs for 15 the treatment of knee OA revealed potential concerns about the durability of clinical and structural 16 outcomes in low-and intermediate-dose arms of treatment 6 , suggesting the need for further studies. 17 Intra-articular injection of MSCs with hyaluronic acid (HA) as a vehicle showed a superior effect 18 than injection of HA alone for the treatment of OA induced by anterior cruciate ligament (ACL) 19 transection 7 . This study and others 8 revealed that nonspecific binding of MSCs onto the synovium, 20 meniscus, and ligamentous tissues, highlighted the importance of the development of methods for 21 enhancing local delivery of cells to injured articular cartilage. However, there are few, if any, 22 studies focusing on this. Magnetically labeled MSCs have been applied for articular cartilage 23 repair 9 . Although MSCs labeled with magnetic particles exhibit no deterioration in chondrogenic 1 differentiation, there is concern about the uptake of iron by the tissues . 2 In the current study, we identified OA-targeting peptides through biopanning of a phage 3 display peptide library with the use of human OA specimens. The OA-targeting peptides were 4 further investigated for application in the delivery of diagnostic agents, lubrication supplements, 5 and MSCs to articular surfaces in an enzyme-induced OA rat model and in an ACL-transection OA 6 swine model. 7 8 Identification of OA-targeting peptides 9 Using a phage display peptide library, we probed the OA articular cartilage cut from the 10 subchondral bone of knee joints from patients who received total knee joint replacement. The OA 11 cartilage specimens were homogenized for acquiring tissue lysates or cut into square tissue pieces, 12 5 mm  5 mm in size. Through five rounds of selection of phage-displayed peptides (biopanning) 13 binding to tissue lysates and tissue pieces (Fig. S1), the titers of bound phages significantly 14 increased up to 388-fold (Fig. S2a), and 864-fold (Fig. S2b), respectively. Phage clones collected 15 from the fifth round of biopanning were further subjected to ELISA screening, and clones with 16 high affinity to tissue lysates or pieces were chosen, sequenced, and aligned ( Fig. S2c and S2d). 17 Finally, we identified five groups of targeting phages sharing distinctive consensus motifs (Table  18   S1). The binding abilities of selected phage clones were validated in the human chondrocyte cell 19 line, hPi-GL 10 , by immunocytofluorescent staining. All of the identified phage clones, labeled with 20 M13-PE (antibody conjugated to fluorescent dye), bound to hPi-GL in a dose-dependent manner 21 (Fig. S3a). Notably, C5-24 and C5-91 peptides showed specific and remarkable binding scenarios 22 in hPi-GL (Fig. S3b). To examine phage clones specifically binding to OA cartilage (Fig. S4) rather 23 than to other soft tissues, such as the synovium and meniscus (Fig. S5), the human OA tissue 1 sections were immunostained using horseradish peroxidase (HRP)-labeled phage clones, followed 2 by semi-quantification of the deposited 3, 3diaminobenzidine (DAB) intensity (-to +++) 11 . 3 Particularly, C5-24 and C5-91 peptides showed superior binding activity to cartilage, but no 4 binding activity to the meniscus and synovium (Table S2). Moreover, C5-24 peptide exhibited the 5 best specificity for targeting the territorial region of OA cartilage and was chosen for subsequent 6 studies. 7 8 Intravital imaging of OA targeting 9 To demonstrate the OA-specific targeting activity of C5-24 peptides, rhodamine-labeled C5-24 10 peptides and scrambled peptides were separately injected into OA joints in a rat model for two-11 photon microscopy observation of fluorescence and second harmonic generation (SHG) signals. A 12 scrambled peptide contains all the same amino acids as the original peptide but in new random 13 order. Surface-rendered 3D reconstructed images and transversal composite images of cartilage 14 showed sparse red dots randomly existing in the C5-24 peptide-injected control cartilage, 15 scrambled peptide-injected control cartilage, and scrambled peptide-injected OA cartilage. 16 Conversely, red dots were observed in the C5-24 peptide-injected OA cartilage. When probing type 17 II collagen with SHG, red dots were localized in the SHG signal-free area (Fig. 1a), corresponding 18 to the territorial region of OA cartilage. From the z-axial plane, it is determined that C5-24 peptides 19 reached at least 50 m in depth in the OA cartilage (Fig. 1a). In addition, the overall fluorescent 20 peptide binding area (Fig. 1b) and binding intensity (Fig. 1c) in all slices were further calculated, 21 showing a significant difference in C5-24 peptide targeting between OA and control cartilage.
These data demonstrate the distinguished recognition capability and specificity of C5-24 peptides 1 for targeting the territorial regions of OA cartilage. 2 OA 3 cartilage. The C5-24 peptides and scrambled peptides were fluorescently labeled and separately 4 injected into rat knee joints pre-established without (CTR) or with OA. The whole joint capsules 5 were removed at 24 h post-injection, and articular surfaces of the proximal tibiae were observed 6 intensity in each slice (upper panel) and all groups (lower panel). All images were reconstructed, 7 merged and analyzed using ImageJ Fiji software. **p<0.01 8 9

Fig. 1. Intravital imaging demonstrates the binding capability of C5-24 peptides to
Application in early OA diagnosis 10 To demonstrate the applicability of OA-targeting peptides in the delivery of diagnostic agents for 11 early diagnosis of OA, C5-24 and scrambled peptides were conjugated with superparamagnetic 12 iron oxide (SPIO) (Fig. 2a). Fourier-transform infrared spectroscopy (FTIR) revealed increased N-13 H band/C-O stretch ratios, indicating successful installation of SPIO into C5-24 and scrambled 14 peptides (Fig. 2b), which were intra-articularly injected into the OA joints of a rat model 15 established by enzyme digestion 12 . Magnetic resonance imaging (MRI) of the OA knee joints 16 without peptide-conjugated SPIO injection showed no difference compared to the sham control 17 knee joints, revealing the challenge of MRI for early OA diagnosis when the articular cartilage is 18 not severely denuded. Similarly, scrambled peptide-conjugated SPIO that did not bind to the OA 19 cartilage, and the MRI signal reduction also failed to differentiate early OA from sham controls. 20 Conversely, C5-24 peptide-conjugated SPIO bound to OA cartilage and caused MRI signal 21 reduction in OA cartilage but not in healthy cartilage (Fig. 2c). To get one step closer to the clinical 22 setting, the feasibility of C5-24 peptides-conjugated SPIO for early OA diagnosis was further 23 confirmed in a large animal OA model established by ACL-transection in Lanyu minipigs. After 2 24 months of ACL-transection, the sham control knee joints either with or without receiving C5-24 25 peptides-conjugated SPIO or the OA knee joints without receiving C5-24 peptide-conjugated SPIO 26 showed no difference in the T1-and T2-weighted MR images (Fig. 2d), indicating the difficulty in using MRI for early OA diagnosis. However, the OA knee that received C5-24 peptide-conjugated 1 SPIO showed enhanced signal reduction in T1-and T2-weighted MR images, demonstrating the 2 sensitivity of C5-24 peptide-conjugated SPIO for early OA diagnosis. Taken together, these data 3 suggest that imaging contrast agents, such as SPIO, when conjugated with C5-24 peptide, could be 4 applied in early OA diagnosis in combination with the MR imaging system. 5

Application in joint lubrication 4
To investigate the potential of C5-24 peptide to deliver HA into OA cartilage for lubrication, C5-5 exhibited better lubrication than non-modified HA and scrambled-HA in the rheological pre-1 condition stage (Fig. S7) and torque measurements (Fig. S8). Representative individual patient data 2 were synthesized and conjugated with HA through Michael-addition chemistry. (b) In the 1 H proton 5 NMR analysis, the methacrylate (MA) and peptide modified HA were dissolved in D2O and 6  examined by NMR, with the 2 HOH peak at 4.8 ppm used as the reference line. The proton-NMR 1 of the original HA and modified HA revealed distinct methylene (6.0 and 5.6 ppm) and methyl 2 resonances (1.8 ppm) (thin yellow circle indicates the corresponding resonance of the branched  3 group). (c) For rheological lubrication measurement of peptide-modified HA, the static (μs) and 4 kinetic (μk) friction coefficients were examined in paired human OA cartilage cylinder discs 5 collected from OA patients. The cartilage discs were immersed in 1% HA (non-modified) or 6 peptide modified HA solution (including modified C5-24 peptides or scrambled peptides) for 2 h, 7 washed, mounted onto specific plates, and immersed in phosphate-buffered saline (PBS). The 8 cartilage discs were subjected to a pre-conditioning stage for 3600 s and then to four stages of 9 relaxation period, which minimized the cartilage disc height change and reduced the factors 10 influencing the friction measurement in the rheometric program. 11 12 are placed in the supplementary information ( Fig. S9-Fig. S13), showing the same scenario with 13 the gradual loss of cartilage disc height in the 3600 s relaxation time in the pre-conditioning stage, 14 but returning to consistent cartilage disc height in the following four stages of the relaxation period, 15 which reduced the factors affecting the friction measurement. Together, these data suggest the 16 applicability of C5-24 peptides in the development of novel and effective joint lubricants for OA. 17 18 Application in OA regenerative medicine 19 C5-24-HA may be applied to MSC regenerative medicine by binding to CD44, the HA receptor, 20 which is extensively expressed on the MSC cell surface, and delivering MSCs to the OA cartilage 21 surface. Moreover, the chondrogenic activity of HA is likely to induce MSC chondrogenesis, as 22 demonstrated previously 14,15 . To demonstrate this, rat MSCs were fed with SPIO for subsequent 23 tracking and incubated with fluorescent-conjugated C5-24-HA or scrambled-HA (Fig. 4a). 24 Fluorescence microscopic observation demonstrated that MSCs were tightly surrounded by green 25 fluorescence (Fig. 4b). Furthermore, after incubation with C5-24-HA or scrambled-HA, MSCs 26 were immediately injected into OA joints in a rat model, and the joints were subjected to 27 histological examination 8 weeks post-transplantation. Histomorphometric analysis revealed the 28 successful induction of OA when compared OA group with sham control group (Fig. 4c, 4d).
Moreover, knee joints receiving MSCs delivered by C5-24-HA had apparent cartilage regeneration 1 and safranin-O staining (Fig. 4c), while those receiving MSCs delivered by scrambled-HA still 2 exhibited severe OA, showing multiple cracks on the surface of the cartilage with the loss of 3 safranin-O staining. Quantification of OA degree by modified Mankin score 7 also revealed that the 4 former had better improvement in OA than the latter (Fig. 4d). For cell tracking of SPIO-fed MSCs 5 To identify the putative target protein derived from human OA cartilage tissue that binds to C5-24 7 peptide, we used biotinylated C5-24 peptide combined with a chemical cross-linker, 3,3'-8 Dithiobis(sulfosuccinimidylpropionate) (DTSSP), and subsequent sodium dodecyl sulfate-9 polyacrylamide gel electrophoresis (SDS-PAGE) and liquid chromatography with tandem mass 10 spectrometry (LC-MS/MS) to identify the binding target (Fig. 5a). Silver staining revealed several 11 sharp bands, such as coimmunoprecipitated proteins COIP-1, COIP-3, and COIP-5 (Fig. 5b), which 12 were separately collected, digested with trypsin, and analyzed by LC-MS/MS. The fragments were 13 identified by searching the Swiss Protein Database through algorithms using MASCOT and 14 TurboSequest search engines (Table S3). We found several candidate proteins, including collagen 15 alpha-1 (XII) and collagen alpha-3 (VI) fragments with a probability score, indicating the 16 probability of that peptide belongs to a protein, of up to 850 and 372, respectively, higher than most 17 of the other identified peptides. To further confirm these protein fragments as the target protein of 18 the C5-24 peptide, we examined the mutual binding activities between the target protein and 19 biotinylated C5-24 peptide using ELISA. We first pre-coated the ELISA plate with a specific 20 collagen concentration, to identify the optimal collagen concentration for peptide binding (Fig. 5c), 21 and subsequently used the collagen alpha-1(XII) at 3.3 g/mL to examine peptide binding (Fig.  22 5d). We found that biotin-C5-24 peptides bound to collagen alpha-1(XII) and collagen alpha-3 (VI) 23 fragments, but biotin-scrambled peptides did not. However, a mutual dose-dependent binding was 1 observed only between collagen alpha-1(XII) and biotin-C5-24 peptide (Fig. 5c, 5d). In addition, 2 there was no difference in the binding of biotin-C5-24 peptide and biotin-scrambled peptide to 3 Bovine Serum Albumin (BSA). Together, these data suggest that collagen alpha-1(XII) is the target 4 protein of the C5-24 peptide. 5 To predict the structure of the protein-peptide complex, the protein-peptide docking was 6 approached through homology modeling (Fig. S15) and the establishment of several reliable 7 structure models targeting human collagen XII (Fig. S16), based on searching for sequence 8 similarities. These structural models were subsequently applied to calculate the possible molecular 9 docking poses with C5-24 and C5-91 peptide chains ( Fig. S17 showing the Collagen XII L1385 -10 S2295 template and Fig. S18 showing the Collagen XII C-terminus S2506 -P2724 domain), which 11 were both the most promising peptide chains and could be selected for further experiments in our 12 study. The protein-peptide docking models were mainly based on an algorithm in compliance with 13 the lowest Gibbs free energy and chemical thermodynamics, after peptide chain binding with the 14 target protein. Our data showed that both C5-24 and C5-91 peptide chains were targeted to the 15 pocket site of collagen XII in the region of L1385 -S2285 with pose 125 and pose 68, and targeted 16 to the C-terminus of S2506 -P2724 with pose 34 and 42, respectively, which share identical 17 docking sites for C5-24 and C5-91 with the highest pose frequencies (Fig. 5e). Furthermore, these 18 predicted poses share the important consensus binding motifs, WXPXW, which may dominate the 19 major docking affinity between peptide chains and collagen XII. In addition, the sequence 20 homology of collagen XII between humans, swine, rabbits, rats, and mice reached 90.3% similarity 21 and 83.7% identity (Fig. S19), and phylogenic analysis revealed a high genetic correlation of 22 Collagen XII between these five species. Fig. S20 demonstrates that C5-24 peptides were highly reliable in rodent, rabbit, and swine OA model examinations. Moreover, according to the consensus 1 domains labeled in colored letters in Table S1, the peptide sequences in the same groups shared 2 important and identical motifs, such as FVEW and DTH in groups 1 and 3, respectively. 3 Finally, we demonstrated the exclusive expression of collagen XII in OA articular cartilage. The 4 expression of collagen XII was only observed in rat OA cartilage, but not in normal articular 5 cartilage (Fig. S21). In addition, the expression of collagen XII was only observed in human OA 6 cartilage, but not in human non-OA cartilage ( Fig S22). Consistent with the area where C5-24 7 peptide binds (Fig. 1a), collagen XII is mainly expressed in the territorial regions of clustered 8 chondrocytes (Fig. S22). These data are also supported by a cohort study that included 161 OA 9 patients and 29 non-OA patients. The preliminary analysis revealed a significant increase in 10 COL12A1 mRNA levels in combined OA hip and OA knee cartilage relative to the non-OA 11 cartilage 16 . examination of the binding activity of C5-24 peptide with the target-protein by direct ELISA, the 5 ELISA plates were precoated with recombinant human collagen alpha-3 (VI) and alpha-1(XII) 6 proteins at the indicated concentration to determine the optimal collagen concentration for peptide 7 collagen alpha-1(XII) at 3.3 g/mL was selected to coat the ELISA plate, followed by incubation 2 with C5-24 peptides and scrambled peptides using the indicated concentration. ELISA plates 3 precoated with BSA were served as controls. The biotinylated DYLWQYPDITWH peptides, which 4 is not able to bind to OA cartilage, which was used as a scrambled peptide, and the sequence was 5 identical throughout the study. (e) In docking site prediction using homology modeling, human 6 Collagen XII models were established through homology modeling and served as templates for 7 docking pose prediction of candidate peptide sequences. The predicted poses were based on the 8 consensus region in Collagen XII for C5-24 and C5-91 and the highest pose frequencies. The 9 important consensus binding motifs, WXPXW shared between C5-24 and C5-91, were labeled in red 10 and X indicates non-specific amino acid. 11 12 Application in disease-modifying OA drugs (DMOADs) 13 No drugs are currently approved DMOADs. OA therefore can be a serious disease with an unmet 14 medical need for therapies that modify its underlying pathophysiology and translate to long-term, 15 clinically relevant benefits 17 . Currently, there are several drugs in phases II OR III or in the 16 preclinical stage, including fibroblast growth factor-18 (Sprifermin) targeting cartilage 17 regeneration 18 , and Kartogenin that promotes the dissociation and nucleus internalization of core 18 binding factor beta (CBF) and stimulates cascaded chondrogenesis 19 . All of these developing 19 DMOADs could be further assisted by OA-targeted peptides developed in this study to accelerate 20 delivery to OA tissues. Besides, most of these drugs are focused on an intra-articular route of 21 administration as opposed to systemic pharmacotherapy, aiming to enhance the local 22 bioavailability of drugs and bypass conventional barriers, and to minimize systemic toxicity and 23 enhance safety profile by reducing off-target effects. It is however, important to recognize the 24 marked placebo effect from local intra-articular administration, making the assessment of efficacy 25 more challenging. Improvements in the precision of technology applied to deliver therapeutic 26 agents to OA sites, such as the peptide discovered in this study may lead to the successful 27 development of effective therapies for OA. Future efforts should be directed at delivering disease-28 modifying drugs in a sophisticated carrier equipped with an OA-targeting peptide to enhance the development of DMOADs. 1 2 We have identified several phage-encoded peptide motifs (WXPXW and DTH) that home 3 selectively to an OA joint without any significant targeting to other articular soft tissues, including 4 synovial tissues, meniscus, and ligaments. Moreover, we identified C5-24 and C5-91 peptides that 5 specifically bind to the territorial region of chondrocytes in OA joints. C5-24 has been successfully 6 conjugated to SPIO and HA for OA diagnosis and lubrication purposes, respectively. Although the 7 C5-91 peptide has not been confirmed to deliver diagnostic agents or lubricants to the articular 8 surface in OA joints, C5-91 peptide was considered to have the same function since it has the same 9 size and shares the same motif as C5-24 peptides. 10 11 Although collagen II is the basis for hyaline cartilage, making up 85-90% of all protein in 12 articular cartilage, aging or OA leads to its damage, starting around chondrocytes (territorial region) 13 at the articular surface, and extending into the whole cartilage with progressive degeneration 20 . 14 Given that collagen II is not specifically expressed in OA, collagen II-targeting peptides may not 15 be applied in OA diagnostics, therapeutics, lubrication, and regenerative medicine 21   The authors declare no competing financial interests. 1

Preparation of cartilage specimens for biopanning and ELISA screening
To avoid interference from individual differences among patients, we used surgical articular cartilage specimens from the same OA patient for the five rounds of biopanning in a phagedisplay experiment. The following processes were used to ensure that the particle-size composition of the cartilage used for the five rounds of biopanning was consistent. A human surgical OA specimen weighed 3.2 g was added to two volumes of phosphate-buffered saline (PBS) and homogenized. The cartilage homogenate was centrifuged at 800 × g and 4°C for 10 min, and the precipitate was collected as "the large particle cartilage sample (C1)".
The supernatant was added to a new centrifuge tube, followed by centrifugation at 1,500 × g and 4°C for 10 min, and the pellet was collected as "the cartilage sample with medium particles (C2)". Following centrifugation of the supernatant again at 2,000 × g and 4°C for 10 min, the precipitate was collected as "the small-particle cartilage sample (C3)". The supernatant at this time was separately collected as the "cartilage tissue lysate" for another five rounds of biopanning, which was different from the biopanning performed on the "cartilage tissue pieces". For "cartilage tissue lysate" biopanning, C1, C2, and C3 were weighed and aliquoted into five equal parts, respectively. Each round of biopanning used a mixture of an aliquot of the C1, C2 and C3, for five rounds.
For "cartilage tissue pieces" bio-panning, the cartilage specimens were also cut into square pieces (5  5 mm in size) and adhered to a 96-well ELISA plate with nail polish, one piece per well, for the chondrocyte binding screening (Fig. S1a).

Bio-panning of phage clones targeting OA cartilage tissue lysate and pieces
For "cartilage tissue lysate" biopanning, the tissue lysate supernatant was diluted ten-fold with coating buffer [0.1 M NaHCO3, pH 8.6] and coated fresh on 10-cm Petri dishes for biopanning (and 96-well ELISA plates for screening) at 4°C for 24 h before use. The tissue lysate-coated plate was blocked with 1% BSA in PBS at 4°C overnight, 10 11 pfu of the Ph.D.-12 TM phage (New England BioLabs, Ipswich, MA, USA) display peptide library was added and incubated at 4°C for 1 h. After washing, the bound phages were eluted with 1 ml of the log-phase ER2738 culture at 37°C with 100 rpm shaking for 20 min. This eluted phage pool was amplified and titrated in an ER2738 overnight culture. The recovered phages were used as input for the next round of panning (Fig. S1b), and total 130 phage clones were randomly selected from the fifth round of biopanning to be cultured for ELISA screening.
The processed cartilage specimen was blocked with 1% bovine serum albumin (BSA) in PBS at 4°C for 1 h for each round of "cartilage tissue pieces" biopanning. The Ph.D.-12 TM (New England BioLabs, Ipswich, MA, USA) phage display peptide library, which initially contained 10 11 plaque-forming units (pfu), was added and incubated at 4°C for 1 h. After washing, the bound phages were eluted with 1 ml of log-phase Escherichia coli ER2738 culture (New England BioLabs) at 37°C with 100 rpm shaking for 30 min. This eluted phage pool was amplified and titrated in an ER2738 overnight culture. The recovered phages were used as input for the next round of panning (Fig. S1c), and total 95 phage clones were randomly selected from the fifth round of biopanning to be cultured for ELISA screening.

Identification of amino acid sequence motifs to target OA cartilage
The binding activity of the selected phage clones to cartilage tissue lysate (Fig. S2c) and cartilage tissue pieces ( Fig. S2d) was examined by ELISA. Phage clones with the highest binding affinity (A490 value > 0.15 for the cartilage tissue lysate and A490 value > 2.0 for the cartilage tissue piece) were selected and sequenced.
We identified five distinct groups with differing consensus motifs by amino acid sequence alignment (highlighted in Table 1).

Validation of peptides targeting OA cartilage using immunofluorescence of hPi-GL chondrocyte cell line
For examination of those phage clones, slide-cultured hPi-GL cells were fixed with 4% paraformaldehyde in PBS at room temperature for 15 min, washed with PBS and permeabilized with 0.1% Triton X-100 at room temperature for 30 min, blocked for nonspecific binding with 1% BSA/PBST. The slide-cultured hPi-GL cells were separately incubated with 4  10 8 pfu, 8  10 8 pfu, and 10 9 pfu selected phage clones at 4°C for 1 h.
After removing the unbound phages by washing, the cells were incubated with anti-M13 mouse mAb (GE Healthcare, Milwaukee, WI, USA) as the primary antibody and R-Phycoerythrin-AffiniPure F(ab')2 fragment goat anti-mouse IgG (Jackson ImmunoResearch Inc.) as the secondary antibody at room temperature for 1 h, respectively. Then, washed with PBST, and counterstained with Hoechst 33258 (1 μg/ml; Sigma-Aldrich) at room temperature for 10 min. The cells were analyzed for phage binding and localization by fluorescence using confocal microscopy (Zeiss LSM 700).

Selection of peptides targeting OA cartilage but not synovium and meniscus
To examine localization of the targeting phages bound to articular tissues, human OA cartilage specimens were used for examination. Paraffin-embedded human OA tissue, synovium, and meniscus sections were

Rat OA model establishment
The OA in rat model was established as previous described with slightly modification (1). In brief, the male SD rats ≅ 300 grams in weight were used in this study. All animal experiments were approved by the China Medical University Committee for the Use and Care of Animals. Rats were kept under standard laboratory conditions (temperature 24°, 12h light-dark cycle), fed standard diet and drank tap water. Rats were anesthetized with 2.5% isofluorane (Abott, USA) in 70 ml/min flow rate before every injection. Rat joint OA was induced in the right knees in each group by injecting 0.2 ml of 4% papain solution (Sigma-Aldrich, USA) with 0.1 ml of 0.03 M cysteine (Sigma-Aldrich, USA) as activator. Same amount of saline was injected into the left knees in each group. Injection was repeated on the fourth and seventh days, respectively, and two weeks after the last papain injection, rat knees were removed for histological analysis to confirm the formation of OA. The established OA model in rats was further used in the following experiments for intraarticular injection.

Preparation of rhodamine labeled C5-24 peptide and 2-phonton microscopic observation
To demonstrate the OA specific targeting activity of C5-24 peptide, the DYLWQYPDITWH peptides, which is not able to bind to OA cartilage, was used as scrambled peptides. The rhodamine-labeled C5-24 and scrambled peptides were separately injected into rat joints without (control) or with enzyme-induced OA (2). Aliquots of 1 μg rhodamine-labeled peptides in 40 μl PBS were used for intraarticular injection with using 30G syringes.
Rat Knees were removed at 1-day post-injection, and both femoral condyles and tibias were cleaned thoroughly, immersed in PBS and sophisticatedly attached on 3.5 cm dish for 2-phonton microscopic observation. The microscope system was operated using a near-infrared femtosecond laser (Mira 900, Coherent, USA) at the central wavelength of 810 nm, 76 MHz pulse repetition rate, and 200 fs pulse width for imaging. The laser power was controlled to 20 mW that is sufficient to produce SHG and TPEF, and also prevented photodamage during continuous illumination. Thus, the wavelength of SHG from collagen fibers is 405 nm, while the TPEF from collagen, elastin, FAD, and NADH is approximately ranging from 450 to 650 nm (3,4). All images were obtained by a laser scanning unit (Fluoview 300, Olympus, Japan), a pair of two objective lenses for both lasers focusing and collection of photons (UPlanSApo 20/0.75, Olympus, Japan), and two photomultiplier tubes respectively for SHG and TPEF detection (R3896, Hamamatsu, Japan). SHG and TPEF were filtered from the intense excitation laser background by a combination of band-pass filter (FF01-405/10, Semrock, USA) and color glass (BG39, Schott, Germany). And, then splitted by a dichroic mirror (FF435-Di01, Semrock, USA) and forward detected. Note that we used a cube polarizing beam-splitter (GT10-B, Thorlabs, USA) combined with a half (AHWP05M-980, Thorlabs, USA) and a quarter (AQWP05M-980, Thorlabs, USA) waveplates to demonstrate LP and CP imaging, respectively. Only the extinction ratio of linear polarization larger than 50:1 and the ellipticity of circular polarization (Imax/Imin) less than 1.1 after the focusing objective lens can then be used for the following two-photon imaging. The acquired images were mainly processed and analyzed with ImageJ/FiJi software (National Institutes of Health, Bethesda, MD, USA). The type II collagen structure reconstructed through the second harmonic generation images (5) (Fig. 1a) showed porous collagen fiber inter-connected structures (green color) surrounding nested chondrocytes (black area).

Preparation of C5-24 peptide-conjugated superparamagnetic iron oxide (SPIO) and IR spectroscopy
The C5-24 and scrambled peptides were chemically synthesized, meanwhile aminosilane modified SPIO particles in 50 nm in diameter (Chemicell GmBH, Germany) were firstly crosslinked with succinimidyl-[(Nmaleimidopropionamido)-tetraethyleneglycol] ester (Thermo Fisher Scientific, USA) in sodium bicarbonate buffer pH 8.5 to form amide bonds, and subsequently interacted with sulfhydryl group on cysteine of the peptides in pH 7.2 to form a stable thioether bond and subjected to dialysis in ddH2O in M.W. 10K cut-off to remove the free-forms of peptides, bridge linkers and salts, leaving groups and further concentrated in reduced pressure, resuspended in PBS and stored in 4 °C for experiments not longer than 2 weeks. To analyze the installation of peptides on SPIO, a part of prepared SPIOs were lyophilized, grounded with potassium bromide (KBr) thoroughly in 1:100 wt./wt. and compressed in 200 pound/inch 2 to form a thin pellet for further infrared radiation spectroscopic analysis (Perkin Elmer, USA). IR was scanned from 400-4000 1/cm frequency to record the characteristic-group and finger-print region molecular groups in transmission mode, respectively.

Magnetic resonance imaging (MRI) analysis of OA in rat model
Rats at the indicated time points as results shown were anesthetized by inhalation and subjected to MRI scanning. MRI scans were performed using a 4.7T MR scanning system (Bruker BioSpin, Germany) at the Institute of Biomedical Sciences, Academia Sinica in Taiwan. T1-weighted and T2-weighted sagittal sections were rendered using the following settings: fast spin echo sequence with a time to repetition of 2000 ms and time to echo of 72ms; slice thickness was 1 mm; interslice gap 1 mm; matrix 256; TE 60; TR 2000; field of view 60 mm; number of averages 2. A 60 mm volume resonator and a 2 cm diameter surface receive coil were used to maximize image resolution and quality. Tomographs DICOMs of MRI were analyzed by Osirix MD (Osirix Ltd., USA).

OA in mini-pig model, intraarticular injection of C5-24 peptide-conjugated SPIO and 3T-MRI analysis
For surgery of anterior crucial ligament (ACL) transection to establish OA, a Taiwan Lan-Yu minipig (9month-old, weight ≈ 50-60 kg) was anesthetized by combined intramuscular (i.m.) injection of Stresnil (20 mg/kg) and atropine sulfate (0.02 mg/kg), followed by i.m. injection of Zoletil® 50 (4 mg/kg, Virbac Animal Health, France) 15 min later. In order to get a more homogenous group of knee joints, only female pigs were included in the current study. During the surgery, the animals continued to be anesthetized with gas containing oxygen (flow rate of 1.5 L/min), nitrous oxide (flow rate of 1 L/min) and 1% isoflurane. The right rear limb was washed and covered sterilely. Following intravenous administration of cefazolin (2 g), an incision in the skin of approximately seven cm was made in the right knee from the patella to the tuberositas tibiae. The joint was then opened medial to the patellar ligament and the patella is partly luxated. The ACL was then fixed by a clamp and cut at the distal end using a scalpel. To avoid spontaneous healing of the ACL after this transection, a proximal resection was additionally carried out using an electrical arthrosector. Following successful rinsing with sterile 0.9% saline solution, the skin incision was closed in layers using 1-0 VICRYL® sutures (Ethicon, USA). The minipigs were able to walk and move normally after this procedure. MRI scans were performed at the indicated time points using a 3T MR scanning system (Achieva

Preparation of C5-24 and scramble peptide-conjugated hyaluronic acid (HA)
The peptide conjugated HA was synthesized as previous described with slight modification (6). In brief,

Lubricant performance analysis
Human articular cartilage samples collected from femoral condyles of cartilage were prepared for lubrication testing with a slight modification from previous publishes (7). Human osteoarthritic cartilage samples were sectioned from the patients who underwent total knee arthroplasty under stringent supervision by IRB committees from China Medical University Hospital (IRB number: CMUH108-REC1-046, and T-CMU-23728). Care was taken to avoid damaging the articular surface during dissection. The superficial layer of OA cartilage from individual patient was maintained intact, punch-cut to obtain a cylinder disc with diameter in 8.0 mm and 6.0 mm, respectively and only the deep layer of cartilage was cut to obtain a flat disc to glue to the metal counter-surface of the particularly designed testing modules while performing friction measurements in rheometer. Cartilage was used fresh without freezing or the addition of protease inhibitors so as not to change the surface lubrication properties. Samples were washed vigorously in PBS overnight to deplete the cartilage surface of any residual synovial fluid, after which they were separated into at least 3 groups. Cartilage discs were pre-incubated in 1 ml original HA or peptide modified HA (1% HA in PBS) for 2 h as indicated in Results for binding of the nonmodified HA or peptide modified HA with cartilage disc, followed by immersing them in 10 ml PBS in testing modules and mounting onto rheometer (HR-1, TA Instrument Ltd., USA) for friction measurements.
The rheometer was initially set to zero using standard protocol in compliance with manufacturer's instruction, and then we calculated the initial heights of the cartilage samples with an electronic caliper followed by loading the samples on the rheometer. The samples were glued with cyanoacrylate glue to the top and bottom rheometer fixtures in parallel plate configuration. Only a thin layer of glue bound to the cartilage and metal fixture surface. The 6.0 mm sample surface was positioned on top of the 8.0 mm surface. The top sample was lowered and pressed against the bottom sample until a load value of ~0.01 N to avoid insufficient contacts between the sample surfaces, load value fluctuations and minimize the errors in height measurements.
The corresponding recorded height, which was automatically sensed by rheometer, was taken for strain calculation. The instrument was programmed to record the total cartilage thickness and calculate the height for ≈14% compression. The total thickness of the human OA cartilage sample was in the range of ≈2.5 -3.5 mm, which were tested in a bath of HA/PBS fluid (10 mL) covered with protecting lid to prevent desiccation. ]. This preconditioning was repeated twice more, followed by a 3600 seconds stress-relaxation period to allow the pressurization of the fluid in the compressed cartilage to fully subside.
The equilibrium normal stress data recording and measurement was performed for each experimental group (Fig S8 ~ Fig. S20). Lubrication testing was performed in 14 stages. The first two stages were considered negligible and used as a clearing or pre-shear stage. Stages 3, 6, 9 and 12 were performed to analyze the effect of different durations of relaxation. Samples were allowed to relax between tests for 1200, 120, 12 and 1.2 seconds. Lubrication data were recorded during stages 4-5, 7-8, 10-11 and 13-14; each stage was in a different direction of rotation and at a constant shear rate. During each test, torque (τ) and axial force (N) were measured, and instantaneous measurements of μk, the kinetic friction coefficient, were determined from the following equation: μk = τ/(Reff × N). Instantaneous μk values were averaged over the second revolution in each direction to produce an average μk that was used for comparison. Static friction coefficients were calculated as the instantaneous μs = τmax/(Reff × N) at the maximal torque value found during the startup period of the test. After experimentation a central indentation due to ≈14% compression on the cartilage surface was confirmed.

Rat MSCs isolation and labeling with SPIO, and delivery via C5-24 peptide conjugated HA
Rat MSCs were isolated and expanded as previously described. Briefly, femora collected from 2 female Sprague-Dawley rats with 8 to 10 weeks of age (BioLASCO Taiwan Co Ltd, Taipei, Taiwan), and the soft tissues were detached aseptically. The bone marrow mononuclear cells were isolated by the density gradient centrifugation method and suspended in complete culture medium (CCM: α-MEM supplemented with 16.6% fetal bovine serum, 100 U/mL penicillin, 100 μg/mL streptomycin, and 2 mM L-glutamine), then seeded in culture dishes in the density of 1 × 10 5 /cm 2 . Nonadherent cells were removed by washing and changing medium at 24 hours later. When cells reached sub-confluence, the cells (passage 0) were harvested for further subcultures. Then, the cells were seeded at density of 100 cells/cm 2 and grown in CCM with medium change twice per week. The MSCs used in this study were passage 3-4.
For MSCs labelling with superparamagnetic iron oxide nanoparticles (SPIO), 50 µg/mL of SPIO (Chemicell GmbH, Gemany) was pre-mixed with 0.75 µg/mL poly-L-lysine (Sigma Aldrich, USA) in culture medium at room temperature for 1 h. For endocytosis of SPIO nanoparticles, MSCs were seeded in 6-well plate at density of 4 × 10 4 / well and grown for 24 h, followed by thoroughly washed with PBS. Then, the MSCs were collected to a microtube and incubated with 2% C5-24 peptide conjugated HA in serum-free medium in concentration of 1 × 10 6 cells/200 l at 37˚C for 30 min. For intra-articular injection, the volume of HA encapsulated MSCs was reduced to 25 l containing 1 × 10 6 cells, and were sophisticatedly injected into a OA rat knee joint synovium capsule.

Identifying the target protein of the C5-24 peptide by affinity trapping, liquid chromatography-tandem mass spectrometry (LC-MS/MS) and ELISA
To identify the binding target of C5-24 peptide chain, human OA cartilage specimens were homogenized for affinity trapping. The biotinylated C5-24 peptide in 1mg/ml in PBS was added to the cartilage homogenate and incubated at 4°C for 1 h. After washing, the DTSSP solution was added to a final concentration of 2 mM for peptide-target protein cross-linking. The reaction mixture was incubated and rotated at room temperature for 30 min. The reaction was stopped with 1 M Tris base. After lysing the chondrocytes with the first lysis buffer (1 M NaCl in 100 mM Tris acetate, pH 8.0) at 4°C for 24 h, the lysates were centrifuged, and the pellet was re-treated with the second lysis buffer (4 M guanidine HCl, 65 mM DTT, 10 mM EDTA in 50 mM sodium acetate, pH 5.8) at 4°C for another 24 h. Following centrifugation, the guanidine extracts were mixed with 100% ethanol (5:1 volume ratio) at -20°C for 16 h to ensure removal of the residual guanidine HCl. The target protein fraction was precipitated by centrifugation at 16,000 × g and 4°C for 45 min, the pellet was washed with 90% ethanol, dried, and re-dissolved with 100 mM acetic acid containing 100 μg/ml pepsin. MyOne thoroughly for 1 h. Immuno-magnetic separation was used to pull down the peptide-protein complexes. Finally, the purified proteins were separated by gradient sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) (Bio-Rad) and silver-stained with a SilverQuest Silver Staining Kit (Invitrogen).
The stained protein bands were cut into small pieces and washed with 10mM ammonium bicarbonate (ABC, Sigma, St Louis, MO) containing 50% ACN for 5 min three times. The gel pieces were dehydrated with 100% CAN and rehydrated with 25mM ABC (pH 8.2) solution containing 1 ng/l trypsin (Promega, Madison, WI) and then incubated at 37 C overnight. After digestion, the tryptic peptides were extracted from the gel using 1% FA in 50% ACN and dry using a centrifugal concentrator. The peptide fragments were Three different parts of human ColXII homology models were built using MODELER based on the templates (PDB code: 1FNF, 2B2X, 2UUR) from BLAST result. The length of first human ColXII model was from L1385 to S2285, with 30% identity to the template 1FNF, which indicated the fibronectin structure and could be used to model establishment due to the highly conserved structural topology. The second and third human ColXII models were from K2321 to L2513 and S2506 to P2724 with 31% and 36% sequence identity with template 2B2X and 2UUR, individually. All of the homology models were firstly checked by PDF total energy, DOPE (Discrete Optimized Protein Energy) and verify score, Ramachandran plot and refined the structure to obtain the reasonable backbone and sidechain conformation. The most representative protein templates were used to predict the binding sites and poses with C5-24 and C5-91 peptide chains due to the most promising results in IHC. Subsequently, Protein-peptide docking using ZDOCK was performed for searching the potential binding region. The Z_Dock score and E_R_Dock score were used to validate the docking capability and exactitude between peptides and target protein templates.

Statistical analysis
Data are presented as mean  SD, statistical comparisons were performed by Student's t-test or one-way analysis of variance (ANOVA) and p values <0.05 were considered significant. All calculations were performed using Statistics Analysis System (SAS) licensed to China Medical University. All in vivo data are representative of at least 3 independent experiments as indicated. Figures   Fig. S1. Biopanning of phage clones targeting OA cartilage. The OA articular cartilage specimens of knee joints from patients who received total knee joint replacement were separated into two parts, cell lysates and           were retrieved from Uniprot database for analysis and showed high sequence similarity (90.3%) and identity (83.7%). Background color intensity represents the 100% identical (deepest) to completely non-related (white).

References Supplementary
The common binding regions of C5-24 and C5-91 peptides at the indicated pose are labeled with dotted line.
The overlapped docking residues of C5-24 and C5-91 peptides to ColXII between these five species are showed in bold letters and underlined, which are highly conserved in five species.