An ex vivo tissue model of cartilage degradation suggests that cartilage state can be determined from secreted key protein patterns

The pathophysiology of osteoarthritis (OA) involves dysregulation of anabolic and catabolic processes associated with a broad panel of proteins that ultimately lead to cartilage degradation. An increased understanding about these protein interactions with systematic in vitro analyses may give new ideas regarding candidates for treatment of OA related cartilage degradation. Therefore, an ex vivo tissue model of cartilage degradation was established by culturing tissue explants with bacterial collagenase II. Responses of healthy and degrading cartilage were analyzed through protein abundance in tissue supernatant with a 26-multiplex protein profiling assay, after exposing the samples to a panel of 55 protein stimulations present in synovial joints of OA patients. Multivariate data analysis including exhaustive pairwise variable subset selection identified the most outstanding changes in measured protein secretions. MMP9 response to stimulation was outstandingly low in degrading cartilage and there were several protein pairs like IFNG and MMP9 that can be used for successful discrimination between degrading and healthy samples. The discovered changes in protein responses seem promising for accurate detection of degrading cartilage. The ex vivo model seems interesting for drug discovery projects related to cartilage degradation, for example when trying to uncover the unknown interactions between secreted proteins in healthy and degrading tissues.


Introduction 1
Osteoarthritis (OA) is a progressive disease involving mechanical, biochemical and ge-2 netic factors that disturb associations between chondrocytes and extracellular matrix 3 (ECM), alter cellular metabolic responses and result in degradation of articular carti-4 lage 1 . Prominent proteins associated with the pathophysiology of OA are pro-inflam-5 matory cytokines including the interleukins IL1a/b, IL6, IL8 and the tumor necrosis 6 factor TNFa 1 . Anti-inflammatory cytokines such as IL4, IL10 and IL13 are also ele-7 vated in OA tissues 2 . Moreover, aggrecanases and matrix metalloproteinases (MMPs) 8 that degrade the ECM as well as growth factor families of bone morphogenetic proteins 9 (BMPs), fibroblast growth factors (FGFs) and transforming growth factors (TGFs) are 10 all present in synovial joints of OA patients 3, 4 . The fact that proteins with opposing 11 effects are found in OA joints simultaneously (e.g. pro-inflammatory and anti-inflam-12 matory cytokines or matrix degrading enzymes and chondrogenic cytokines) suggests 13 nontrivial inherent interactions between these proteins. 14 Targeting these players separately in order to reverse or suppress OA has been tested 15 in many clinical studies in the past with very limited success. Monoclonal antibodies 16 such as Adalimumab (anti-TNFa) and Canakinumab (anti-ILb), MMP inhibitors, and 17 growth factor stimulators like Sprifermin (rhFGF18) have not been able to provide sig-18 nificant improvements until now or are still in clinical trials 5 . Such rather disappointing 19 results support the hypothesis that targeting a single protein is not sufficient for a suc-20 cessful therapy. Therefore, a deeper understanding of the cytokine interaction network 21 might be necessary to leverage drug discovery in OA. 22 One approach to achieve this aim is the use of antibody-based multiplexing assays 23 that simultaneously measure the abundance of a broad panel of proteins in a biological 24 sample. Applications of such assays related to OA and cartilage include re-construc-25 tions of chondrocyte cell signaling pathways based on phosphoproteomics and cytokine 26 release data from 2D chondrocyte cultures 6 , cytokine releases after anabolic stimula-27 tions of 3D chondrocyte scaffolds 7 and measurements of joint pathology dependent cy-28 tokine profiles in synovial fluids and cartilage tissues 8 . Notably, careful multivariate 29 analyses of such protein measurements also have great potential as a tool for discovery 30 of novel diagnostic biomarkers in the form of characteristic changes across subsets of 31 the proteins studied. However, in order to enable systematic large-scale measurements 32 of these protein secretion patterns, a sufficiently simple and cheap ex vivo tissue model 33 of cartilage degradation (CD) is needed. 34 stand how different protein stimuli affect the degraded state and to search for potential 23 diagnostic biomarkers that can discriminate between healthy and degraded cartilage 24 tissue. More specifically, we expanded the work by Grenier et al. 10 by looking at protein 25 secretion patterns after stimulation with major OA related cytokines/proteins, and eval-26 uated the possibility to use these response patterns for determination of the cartilage 27 state. As demonstrated below, this novel systematic approach revealed biomarkers with 28 potential to be used for accurate detection/diagnostics of degrading cartilage. More gen-29 erally, this approach was found to have potential to help uncovering the interactions of 30 CD related proteins, and thereby also help accelerating drug discovery and development 31 activities associated with OA. . The main idea of the ex vivo tissue model is to perturb healthy and degrading cartilage 6 tissue with a set of OA related stimuli followed by a measurement of the tissue re-7 sponses in terms of protein secretions. The resulting dataset is analyzed in order to 8 compare the two different tissue states and pinpoint individual stimuli yielding different 9 protein responses. Our hypothesis is that these protein responses depend on the tissue 10 state (healthy or degrading) and thus can be used to distinguish between them. 11

Explant isolation, state setting and washing 25
Cartilage tissue samples were obtained from the femoral heads of patients (n=5, age 72-26 94, 1 male and 4 females) undergoing total hip replacement due to fracture (n=3, P1-27 P3) or OA (n=2, P4 and P5) with patient's informed consent and protocol approved by The supernatant of an unstimulated disc (cultured in DMEM* during the perturbation 32 step) was taken as control. Blank measurements to evaluate the experimental noise for 33 each protein were included as well. All steps were conducted in a humidified incubator 7 at 37 Ο C and 5% CO2. The rather high number of stimuli reduced the possibility of 1 having biological replicates as 55 cartilage explants were needed for one run of single 2 measurements. As our main aim was to discover and compare response patterns of car-3 tilage explants and not to uncover new biological mechanisms, we decided to accept 4 the drawback of having a low number of replicates with simultaneously having a broad 5 panel of stimuli. Thus, single measurements were collected after the stimulations of 55 6 untreated cartilage discs of patient P2 and 55 collagenase II (2 mg/ml) treated cartilage 7 discs of patient P3.

Mechanical testing 27
In order to evaluate the change of mechanical properties of collagenase II (2 mg/ml) 28 treated samples, three DMEM* and three collagenase II (2mg/ml) treated tissue sam-29 ples were taken after the washing step and tested with the Bose Electroforce 3100 30 (Bose, Framingham, MA). Stress-relaxation tests and measurement of the equilibrium 31 stress were used to obtain information about the compression stiffness of the material 13 . 32 Initially, samples were pre-loaded with a force of F=0.1N. Then an instantaneous ramp 33 displacement of 5% of the initial height was applied and the relaxation of the force over 34 time was measured until a dynamical equilibrium was reached. The procedure was re-1 peated for a total of three loading steps. Equilibrium stress was calculated as the engi-2 neering stress σ=F/A0 with A0≈7.07 mm^2. 3 4 GAG release 5 The extracellular release of sulfated glycosaminoglycans was measured spectrophoto-6 metrically via a Dimethylmethylene Blue (DMMB) assay 14 using the Varioscan LUX 7 multimode microplate reader (Thermofisher Scientific Inc., USA). As s-GAGs belong 8 to the main constituents of the ECM 10 an increased presence in the explant supernatant 9 can be directly related to increased ECM destruction. The GAG release was quantified 10 for two groups. One group was treated with DMEM* and the other group was treated 11 with collagenase II (2 mg/ml). 50 µl of the supernatant was extracted at t=2h, 6h, and 12 24h. The measured s-GAG concentration was normalized to tissue wet weight. n=3 13 discs were chosen per group. 14 15

Collagen II content 16
Collagen II is the main constituent of the ECM 10 , thus reduction can be directly related 17 to ECM destruction. Tissue collagen II content was quantified as shown previouly 15 18 with a hydroxyproline assay kit (Abcam, Cambridge, UK). One group was treated with 19 DMEM* for 24h and the other group was treated with collagenase II (2 mg/ml) for 24h. 20 n=3 discs were chosen per group and tissue collagen content was normalized to tissue 21 wet weight. Meaningful multivariate data analyses required normalized MFIs that allow compari-1 sons between plates and between cytokines/proteins. The normalized difference D(i,j,p) 2 for stimulus i and cytokine/protein j present on plate p was determined as: The between-class separation sb and the within-class separation sw needed to calculate 27 the separation score Jseparation = sb/sw used to find the most discriminative pairs of meas-28 ured proteins are defined in equations 2 and 3. 29 In the analysis performed here we used K=2, corresponding to healthy and degraded 1 samples, respectively. xn denotes the n th sample (column vector) and Nk denotes the 2 number of samples belonging to class k. In order to reduce the risk of overfitting, the 3 actual score used to identify the most discriminating protein pair was calculated as the 4 maximum across 100 values of Jseparation obtained using a resampling approach where 5 each value was obtained by using a stimuli subset consisting of 80% of the stimuli in 6 the original dataset. 7 8

Statistical comparisons of individual treatments 9
Statistical comparisons between individual treatments were done with the unpaired 10 two-sided t-test at significance level p=0.05. 11 12 In order to investigate the individual protein secretions of collagenase treated tissue, 5 raw MFIs were compared between untreated (DMEM*) and treated (Col) samples. The 6 MFIs obtained from the three concentrations were pooled together. Cohen's d, defined 7 as d=(m1-m2)/s where mi are the class means and s is the pooled standard deviation of 8 the dataset, was taken as the measurement of the effect size. The identified cytokines 9 are presented in Table 2, sorted by decreasing effect size. Out of the 11 proteins with 10 significant differences between control and collagenase treated samples, 4 (bold, Table  11 2) were found in many OA related studies. These were TNFa and IFNG that are related 12 to inflammation and innate immunity responses as well as the two anti-inflammatory 13

21 22
In summary, these results suggest that cartilage discs treated with collagenase increase 1 the secretion of particular proteins that also have been observed to have increased levels 2 in clinical OA studies 3 4 Tissue state can be determined based on measured protein patterns 5 PCA of healthy and degrading tissue responses with subsequent k-means clustering 6 (k=2) based on the first 4 PCA dimensions (82% of variance covered) was performed. 7 The responses plotted in the resulting 2D space of the first two PCs are presented in 8

13
As the samples assigned to the gray color mainly belong to the healthy group, it has been given the label 14 "assigned as healthy" while the red samples has been given the label "assigned as degrading"

16
The shapes of the markers in Figure 2 represent the true tissue (sample) groups. There 17 are 55 squares (healthy perturbed discs) and 55 diamonds (degrading perturbed discs) 18 in the PC1-PC2 plane. The two colors gray and red represent the two categories identi-19 fied by the k-means algorithm (k=2). The subsequent manual assignment of them as 20 "assigned as healthy" and "assigned as degrading" protein secretion patterns was based 21 on the fact that the majority of the members of the gray cluster belongs to responses 22 from the healthy tissue and the majority of the red cluster belongs to responses the 23 degrading tissue. 24 25 Figure 2 also shows that protein secretions after perturbations lead to the formation of 1 two distinct clusters in the PC1-PC2 plane, indicating that tissue state can be determined 2 based on measured protein patterns. In particular, the coordinate along the PC1-axis 3 seems suitable for classification of the cytokine responses as "healthy" or "degrading". 4 This suggests that it is interesting to look at the elements (loadings) of the correspond-5 ing eigenvector in order to determine which of the measured protein changes are most 6 useful for separation between healthy and degrading tissue. The individual loadings of 7 each protein to PC1 are shown in Supplementary Material 4, Figure S1. Increases of 8 MMP9 and FGF2 as well as decreases of CXCL10, ZG16 and FST would cause a shift 9 along PC1 from degrading to healthy responses. 10 11

Levels of IFNG and MMP9 are promising for tissue state discrimination 12
PCA is an unsupervised method that disregards any class information. Therefore it is 13 not guaranteed that the resulting linear projection provides an optimal solution in terms 14 of discrimination between healthy and degrading tissues. Notably, 10 out of the 55 de-15 grading tissue responses get assigned to the wrong cluster (Figure 2, gray diamonds).

5
In summary stimulation of healthy and degrading tissue with a set of 55 stimuli lead to 6 changed secretion levels of various proteins, which were evaluated to discriminate be-7 tween the two tissue states using different approaches. An exhaustive search across all 8 pairwise combinations of protein levels identified MMP9 and IFNG, among several 9 other pairs, as promising for class discrimination. The main aim of this study was to measure and characterize cytokine/protein release 2 patterns in an ex vivo model of CD, created by exposing healthy cartilage to the ECM 3 degrading enzyme collagenase type II. First, it was shown that the induced ECM deg-4 radation resulted in secretions of proteins related to OA observed in various clinical 5 studies (see Table 2). Secondly, the in vitro responses of healthy and degrading tissue 6 to a broad set of OA related protein perturbations were investigated. Multiplexed 7 ELISA measurements of 26 secreted proteins were subject to exploratory (PCA, k-8 means) multivariate data analysis. These analyses showed that it is possible to success-9 fully distinguish between healthy and degrading samples for a majority of the samples 10 In many studies on in vitro models for OA related drug treatment, the typical experi-26 mental readouts are glycosaminoglycan (GAG) and collagen II content. In addition pro-27 totypic biomarkers such as matrix metalloproteinases (MMPs) or inflammatory factors 28 such as IL1a/b or TNFa are also often used 32 . Looking at such few readouts might be 29 too simplistic given the outstanding complexity of the human biological systems. In 30 particular, this might lead to misleading conclusions in general, for example overlook-31 ing truly efficient single drugs and drug combinations. Therefore, in our approach we 32 measured the changed release of 26 proteins simultaneously. Then we used the col-33 lected data in order to characterize the molecular processes associated with CD, includ-1 ing how they can be used for diagnosis and/or to accelerate drug discovery and devel- for 45-120 min, and claimed it to induce similarities with early stages of OA 10 . In our 12 work the collagenase II treatment was extended to cover 24h. The panel of released 13 proteins after such a treatment in our study (Table 2) indicates similarities with late 14 rather than early stages of OA, when comparing with clinical studies. However, this 15 difference in interpretation may simply be due to the fact that there is yet no clearly 16 defined distinction between early and late stages of OA. 17

Limitations 19
The systems biology approach introduced here based on protein profiling of CD used 20 55 stimulations per cartilage condition (healthy and degraded). Since a patient donation 21 usually results in less than 100 cartilage samples, the tissue responses recorded in this 22 study did not come from the same donor. More specifically, healthy perturbed (gray 23 squares in Figure 2) and degrading perturbed (red diamonds in Figure 2) came from 24 two different patients (P2 and P3). Therefore, one cannot exclude that there is a batch 25 effect that can explain some of the differences observed. However, as shown in Figure  26   Additionally, using duplicate measurements on the same experimental plate would also 31 be preferable. The relatively high number of 55 different perturbations was used to in-32 crease the probability to observe different protein releases from healthy and degrading 33 tissue, whilst accepting a statistically weaker significance of the biological differences 34 observed. Thus, the framework and results presented here should be considered as a 1 proof-of-concept that will be followed by more optimized experiments in the future, 2 which will be limited to a smaller set of stimulations. 3 4 A second limitation of the study reported here is that important OA related proteins 5 such as disintegrin and metalloproteinases with thrombospondin motifs ADAMTS 4 6 and 5 and the matrix metalloproteinase MMP13 were not included. In the current study 7 these proteins were not measured due to the lack of reliable assays and challenges in 8 terms of cross-talk. Additionally, it would be interesting to quantify to which extent 9 different stimuli affect the collagen II content of the explants. 10 11 Despite these limitations we believe that the novel systemic ex vivo and in silico ap-12 proach introduced here presents a viable way to investigate compound treatments for 13 CD in general, and related to OA in particular. Thus, we have shown that a more so-14 phisticated systemic analysis at the molecular level is feasible, and that it provides a 15 more detailed molecular picture of what happens during CD in terms of proteomic re-16 sponses to cytokine/protein stimulations.