SC I ENCE ADVANCES | R E S EARCH ART I C L E

DEVELOPMENTAL B IOLOGY

1Duke Molecular Physiology Institute, Duke University School of Medicine, Durham,NC, USA. 2Department of Clinical Sciences, Lund University, Lund, Sweden. 3Depart-ment of Orthopaedic Surgery, Duke University Medical Center, Durham, NC, USA.4Division of Rheumatology, Department of Medicine, Duke University School ofMedicine, Durham, NC, USA.*Corresponding author. Email: vbk@duke.edu

Hsueh et al., Sci. Adv. 2019;5 : eaax3203 9 October 2019

Copyright © 2019

The Authors, some

rights reserved;

exclusive licensee

American Association

for the Advancement

of Science. No claim to

originalU.S. Government

Works. Distributed

under a Creative

Commons Attribution

NonCommercial

License 4.0 (CC BY-NC).

Analysis of “old” proteins unmasks dynamic gradient ofcartilage turnover in human limbsMing-Feng Hsueh1,3, Patrik Önnerfjord2, Michael P. Bolognesi3, Mark E. Easley3, Virginia B. Kraus1,4*

Unlike highly regenerative animals, such as axolotls, humans are believed to be unable to counteractcumulative damage, such as repetitive joint use and injury that lead to the breakdown of cartilage and thedevelopment of osteoarthritis. Turnover of insoluble collagen has been suggested to be very limited in humanadult cartilage. The goal of this study was to explore protein turnover in articular cartilage from human lowerlimb joints. Analyzing molecular clocks in the form of nonenzymatically deamidated proteins, we unmasked aposition-dependent gradient (distal high, proximal low) of protein turnover, indicative of a gradient of tissueanabolism reflecting innate tissue repair capacity in human lower limb cartilages that is associated with expressionof limb-regenerativemicroRNAs. This association shows a potential link to a capacity, albeit limited, for regenerationthat might be exploited to enhance joint repair and establish a basis for human limb regeneration.

on June 14, 2020http://advances.sciencem

ag.org/D

ownloaded from

INTRODUCTIONSome animals, such as zebrafish, bichir, and axolotl, with a high regen-erative capacity, regulate limb regeneration by a circuit of microRNA(miRNA) conserved across species (1). Unlike these animals, humanscannot regenerate whole limbs. Humans are also believed to be unableto counteract the cumulative damage of repetitive joint use or one substan-tial, usually sports- or trauma-related, injury that leads to the breakdownof cartilage and the development of osteoarthritis (OA). Although humanshave a limited natural regenerative capacity, evidenced by the ability toregrow distal portions of amputated digits during childhood (2), by thecarbon-14 (14C) bomb pulse method, turnover of insoluble collagen(residual after hyaluronidase and trypsin digestion) was suggested to bevery limited in human adult cartilage and Achilles tendon (3, 4).

We have focused on endogenous molecular clocks based on spon-taneous posttranslational protein modification by deamidation—thenonenzymatic hydrolysis of the amide group on the side chains of as-paragine (Asn, N) and glutamine (Gln, Q) to yield aspartate (Asp, D)and glutamate (Glu, E) (5), respectively. Until a series of studies demon-strated the deamidation of cytochrome c in vivo (6–8), the deamidationof proteins was considered an in vitro artifact introduced by the proteinpurification process. Because both nondeamidated (“young”) and de-amidated (“old”) epitopes can be simultaneously quantified by massspectrometric analysis due to a mass shift increase of 0.984 Da with de-amidation (9–11), posttranslational modification due to deamidation isa particularly attractivemeans formonitoring protein turnover, definedas the balance between protein synthesis and protein degradation. De-amidation may be considered irreversible since there are no known ex-tracellular repair mechanisms for these spontaneous nonenzymaticmodifications (12), and this reversal has not been observed (5).

Because newly synthesized proteins eventually replace posttransla-tionallymodified proteins, monitoring the accumulation of deamidatedproteins by mass spectrometry (MS) can be a way of estimating proteinturnover and tissue repair. The turnover of cartilage proteins is partic-ularly suited to study using proteinmolecular clocks because many car-tilage proteins are long lived and therefore susceptible to accumulation

of many types of age-related posttranslational nonenzymatic proteinamino acid modifications (13–15). Deamidation rates of Asn and Glnvary from hours to decades. Each site of nonenzymatic amino acid de-amidation serves as itsown intrinsicmolecular clockdue to thedeamidationrate varying according to the two-dimensional (2D; pentapeptide se-quences) (5, 16, 17) and 3D (contextual) features (18) of a protein epitope.This method of determining biological half-life is theoretically applicableto any protein with Asn and/or Gln residues. However, to date, reliableprediction of deamidation rates based on 2D and 3D protein features isonly available for Asn. Moreover, median deamidation rates of Asn aremuch faster than those of Gln (19), allowing estimations of half-life foreven short-lived proteins. For all these reasons, we focused specificallyon Asn deamidation for our analyses.

For those animals with high limb-regenerative capacity, blastemaformation at the site of limb injury is critical (20, 21). Genetic defectsin blastema formation can block limb/appendage regeneration (22, 23).Most recently, an important blastemamiRNA regulatory circuit sharedby three highly regenerative animal systems has been shown to controllimb regeneration (1). We hypothesized that miRNA, critical for blas-tema formation during limb regeneration across species (20–23), wouldplay a role in articular cartilage protein turnover. To assess whetherhuman cartilage shares any of the evolutionarily conserved miRNAregulatory circuitry of the blastema, essential for limb regeneration inhighly regenerative animals (1), we extracted total RNA from humanankle, knee, and hip cartilages for quantification of blastema-relevantmiRNA (miR-21, miR-31, and miR-181c) expression in cartilage. Toevaluate protein turnover by joint site, disease state, and cartilage depth,we used proteomic tools to quantify the amounts of native and deami-dated forms of cartilage proteins. Last, we assessed the association ofthese cartilage proteins with regenerative miRNA. To our knowledge,this is the first study to comprehensively profile the regional patternsof protein turnover andmiRNAexpressions in human adult limb jointcartilage in health and disease. The results of this study can provideinsight into the potential regenerative capacity within human post-natal cartilages and lay the foundation for identifying the underlyingmechanisms determining and regulating this regenerative capacity.

RESULTSAdetermination of protein turnover based on spontaneous, nonenzymaticAsn deamidation of peptides requires crystal structure for the protein

1 of 9

SC I ENCE ADVANCES | R E S EARCH ART I C L E

on June 14, 2020http://advances.sciencem

ag.org/D

ownloaded from

domain containing the peptide of interest (16). Actual crystal structurewas available for cartilage oligomeric matrix protein (COMP) and fib-ronectin (FN1); in silico–predicted protein structure could be generatedusing the Iterative Threading Assembly Refinement server (I-TASSER)for COMP and FN1, and an additional five cartilage proteins, includingclusterin (CLU), prolargin (PRELP), collagen type III (COL3A1), col-lagen type II (COL2A1), and aggrecan core protein (G1 and G3) do-mains (24). The abundance of the deamidated (Asp containing)relative to the nondeamidated (Asn containing) forms of these pro-teins revealed distinct spatial and disease state–related differences inthe turnover of these proteins, with the highest protein turnover rates(lowest ratio of deamidated to nondeamidated protein) in ankle joints(compared with knee and hip; Fig. 1A), at the cartilage surface (com-pared with deep; Fig. 1B), and in OA cartilage (compared with healthy;fig. S1B). To evaluate the independent effects of joint site, disease state,age, and gender and account for the repeated measures from cartilagedepth, we used a multivariable mixed model. A significant gradient ofprotein turnover, indicative of tissue anabolism reflecting an innatetissue repair capacity by joint site (highest ankle, intermediate knee,and lowest hip), was evident for the majority of cartilage proteins, in-cluding the aggrecan G1 domain (aggrecan N terminus), COMP,PRELP, CLU, and COL2A1 (Fig. 1A). We also observed a uniformlyhigher turnover rate at the cartilage surface that was generalizable tomultiple proteins, including FN1, CLU, COL3A1, and aggrecan G1and G3 domains (Fig. 1B). We developed a high-sensitivity, sand-wich enzyme-linked immunosorbent assay (ELISA) specific to thedeamidated form of the N terminus of COMP (14). The abundanceof deamidated COMP (DCOMP) in cartilage extracts by sandwichELISA associated strongly with the abundance quantified by MS(Pearson r = 0.77, P = 0.01) (fig. S1A). These results suggested thatartifactual deamidation during tissue processing for MS was minimal.

In contrast to these proteins, the highest turnover of the aggrecanG3(C-terminal) domain was in the hip, as evidenced by the lowest abun-dance of deamidated aggrecan G3 in the hip compared with the kneeand ankle joints. This may indicate the retention of the G3 domain inthe ankle after cleavage by metalloproteinase and ADAMTS (A Disin-tegrin and Metalloproteinase with Thrombospondin motifs) enzymes(25–28). In fact, on the basis of our MS analyses in the same samples,the abundance of aggrecanG3was linked to the abundance of tenascin-C(Pearson r = 0.45, P = 0.01) (fig. S1B), a high-affinity ligand for aggrecan

Hsueh et al., Sci. Adv. 2019;5 : eaax3203 9 October 2019

G3 (29, 30). Tenascin-C was significantly enriched in the ankle com-pared with the knee and hip cartilage. This suggests that the cleavedG3 domain is preferentially retained in ankle cartilage due to its inter-action with extracellular matrix (ECM) proteins enriched in ankle car-tilage. A higher turnover with OA, reflected by less abundant meandeamidation ratio compared with healthy cartilage, was observed in allproteins, but only statistically significant for COMP (P = 0.012) aftercontrolling for all other factors, namely, joint site and cartilage depth.

Prior efforts to estimate the half-lives of cartilage proteins based onAsp racemization (31–33) required specific protein purification andwere restricted to very long-lived proteins, such as cartilage collagen,due to the very slow rate of Asp racemization (7.7 × 10−4 to 8.3 ×10−4 per year). In contrast, mass spectrometric analysis of deamidatedpeptides does not require specific protein purification. Because of thefact that MS can distinguish protein identity based on peptide sequenceand the rates of Asn deamidation are commensurate with humanbiology, ranging from hours to years (34), the estimation of proteinhalf-life can potentially be applied to any Asn-containing protein epitopein any tissue.We confirmed that predictivemodeling is feasible and validfor half-life determinationswhen crystal structures are not available basedon the strong association of half-lives derived from protein crystal struc-ture and in silico I-TASSER–predicted models for two proteins (Pearsonr = 0.93, P < 0.0001 for COMP; Pearson r = 0.76, P < 0.0001 for FN1).

As in silico–predicted structures were available for all proteins, weused these to compare half-lives among proteins. The estimated proteinhalf-lives based onAsn deamidation analyses ranged from days to years(Fig. 2A). Within healthy cartilage, the shortest mean half-life, 3.4 days[95% confidence interval (95% CI), 1.8 to 4.9], was observed for theC-terminal G3 domain of aggrecan. The N-terminal G1 domain had amuch longermean half-life, 87.9 days (95%CI, 63 to 112.7), than theG3domain (a mean 84.5-fold greater for matched samples, P < 0.0001)(Fig. 2B), consistent with the known retention of the N terminus ofaggrecan in the cartilagematrix aftermetalloproteinase andADAMTSproteolysis (25–28). In contrast to insoluble COL2A1—reported to haveno significant turnover in adults, evenwith the occurrence of disease (3),soluble COL2A1 had a mean half-life of only 2420 days. These datademonstrate that this is a powerful approach for obtaining data simul-taneously on multiple individual proteins and even epitopes within aprotein, as demonstrated here for the aggrecan G1 and G3 domains,without the need to purify the specific protein of interest. The major

A B

27

Fig. 1. Spatial and cartilage depth-related distributions of deamidated proteins in human adult cartilage. Protein turnover in human lower limb cartilage wasbased on Asn deamidation quantified by MS. Counting only proteins identified in at least two biological replicates of one joint site, a total of 469 proteins wereidentified in cartilage, of which 285 proteins (61%) had at least one deamidated residue. We observed significant independent differences in the deamidation proteinratio in articular cartilage (n = 54 samples in total) stratified by (A) joint site and (B) cartilage depth. A significant lower mean deamidated protein ratio, indicative ofhigher tissue anabolism reflecting a higher innate tissue repair capacity, was consistently observed (A) in ankle cartilage compared with knee and hip cartilage, and (B)at the surface zone of cartilage compared with the deep zone. Abundance of deamidated protein forms is represented as the mean standardized value of the de-amidation ratio (z score derived from the ratio of deamidated Asn/total Asn); the error bars represent the SD. P values were derived from multivariable mixed models.

2 of 9

SC I ENCE ADVANCES | R E S EARCH ART I C L E

on June 14, 2020http://advances.sciencem

ag.org/D

ownloaded from

determinant of half-life differences was protein identity; the next majordeterminant of protein half-life was joint site (cartilage protein half-livesof ankle < knee < hip; Fig. 2C). Shorter protein half-lives were observedin ankle cartilages for aggrecan G1, COMP, CLU, and PRELP.

To assess whether miRNA, critical for blastema formation duringlimb regeneration across species (20–23), would play a role in the dif-ferential cartilage protein turnover we observed in a gradient pattern ofthe human lower limb and in the up-regulation of cartilage proteinturnover with OA, we quantified the blastema-relevant miRNA(miR-21, miR-31, and miR-181c) expression in cartilage. miR-21, themost highly expressed miRNA of the three, was differentially expressedin the human limb cartilages in patterns matching the turnover of pro-teins based on deamidation.

Evaluated by a multivariable mixed model, including the indepen-dent effects of joint site, disease state, cartilage depth, age, and gender,we founda significant gradient ofmiRNAexpressionby joint site (highestin ankle) for miR-21 andmiR-181c (P = 0.0158 and 0.018, respectively;Fig. 3A). A significant gradient of miRNA expression (highest in thesuperficial layer of cartilage) was also observed for miR-21 and miR-31(P < 0.0001; Fig. 3B). We also observed that miR-181c was enriched inOA compared with healthy cartilage (P = 0.044; Fig. 3C). Abundance ofmiR-21 was inversely associated with proportions of deamidated carti-lage ECM proteins, including aggrecan, COMP, FN1, PRELP, CLU,COL3A1, and COL2A1 (Fig. 3D). These results are consistent with apositive association of miR-21 with cartilage tissue anabolism. Higher

Hsueh et al., Sci. Adv. 2019;5 : eaax3203 9 October 2019

miR-31 abundance was also linked to lower deamidation ratios ofCOMP, PRELP, and CLU (fig. S2). Stratifying cartilage by disease state,miR-21 was significantly associated with proportions of all seven dea-midated proteins in OA cartilage but only linked to PRELP and CLU inhealthy cartilage, suggesting a large effect of this miRNA on ECM pro-duction induced under the stress of OA (Fig. 3E).

We then investigated the association of thesemiRNAswith themostabundant 193 cartilage proteins that were identified by MS in at least50% of samples. On the basis of cluster analysis, all three miRNAsyielded similar overall patterns of association with the cartilage proteinsin OA cartilage (Fig. 4A; distance equal to 0.16 to 0.17). In contrast,the patterns of association with cartilage proteins were different for thethree miRNAs in healthy cartilage, especially miR-181c (Fig. 4A). Thepattern for miR-181c in healthy non-OA cartilage was inversely linkedto the pattern in OA cartilage (P = 0.0005). These findings suggest en-trainment or synchronization of cartilage protein expression by all threemiRNA in the context of the stress of OA. Only 19% of cartilage pro-teins that were significantly correlatedwithmiR-21were identified to beECMproteins (fig. S3). This indicates thatmiR-21 effects are not limitedto thematrix but rather thatmiR-21 and the other regenerativemiRNAorchestrate a generalized anabolic response in human OA cartilage.Nearly 50% of the cartilage proteins that associated significantly withmiR-21were identified exclusively inOA cartilage (Fig. 4B). This subsetof uniqueOAcartilage proteinswas highly enriched in collagenproteins(Fig. 4C). Associations of these collagen proteins and the threemiRNAs

A

C

B

10 1

100

101

102

103

104

105

ACAN-G1 COMPACAN-G3 FN1 PRELP CLU COL3A1Overall protein half-life in cartilage extracellular matrix

COL2A1

Est

imat

edp

rote

inh

alf-

life

(day

s)

Avg =2.308

Avg =82.92

Avg =563.3

Avg =687.9

Avg =69.07

Avg =2102

Avg =65.49

Avg =2420

ACAN-G1 ACAN-G30

50

100

150

200

Est

imat

ed p

rote

in h

alf-

life

(day

s)

P < 0.0001

Hip

Kn

ee

An

kle

10 1

100

101

102

ACAN-G3

Est

imat

edp

rote

inh

alf-

life

(day

s)

Hip

Kn

ee

An

kle

Hip

Kn

ee

An

kle

Hip

Kn

ee

An

kle

Hip

Kn

ee

An

kle

Hip

Kn

ee

An

kle

Hip

Kn

ee

An

kle

Hip

Kn

ee

An

kle

102

103

104 ACAN-G1 COMP FN1 PRELP CLU COL3A1 COL2A1

P = 0.673 *P *P P P*P *P P= 0.004 = 0.021 = 0.626 = 0.019 = 0.003 = 0.504 = 0.137

Fig. 2. Estimated protein half-lives based on protein deamidation. (A) The proof of concept for estimating protein half-lives by deamidation was demonstrated bythe analysis of abundant cartilage ECM proteins. The half-lives varied from days to years (range, 2.8 to 2420 days) depending on the protein. The symbols indicate the5th to 95th percentile. (B) Within healthy cartilage, the N-terminal G1 domain had a much longer mean half-life, 87.9 days (95% CI, 63 to 112.7) compared with the G3domain (3.4 days), representing a mean 84.5-fold greater half-life for G1 than G3 for matched samples (paired t test P < 0.0001). (C) Shorter protein half-lives wereconsistently observed in aggrecan G1 domain, COMP, PRELP, and CLU in ankle compared with knee and hip cartilage. Although the half-life of the COL2A1 protein wasshorter in ankle cartilage, it was not significant after accounting for all independent factors. The symbols indicate the mean half-lives, and the error bars indicate theSEM. P values were derived from multivariable mixed models.

3 of 9

SC I ENCE ADVANCES | R E S EARCH ART I C L E

on June 14, 2020http://advances.sciencem

ag.org/D

ownloaded from

miR

-21

miR

-31

miR

-181

C0.0

0.2

0.4

10

20

30

40HealthyOAP = 0.0947

*P = 0.044

P = 0.4529

Fo

ldch

ang

efr

om

ctrl

miR

-21

miR

-31

miR

-181

C0.00.20.40.6

10

20

30

40SupMidDeep

*P < 0.0001

P = 0.2919

*P < 0.0001

Fo

ldc h

ang

efr

om

ctrl

CB

miR

-21

miR

-31

miR

-181

C

0.0

0.2

0.4

0.6

10

20

30

40AnkleKneeHip

*P = 0.0158

P = 0.6231

*P = 0.018

Fo

ldch

ang

efr

om

ctrl

A

D

miR-21 miR-21

–7.0

5.0 3.0 1 .0

1.00

6 5 4 3

2 13 5.2 2 5. 1 1 5.0

0–

3 2 1 5. 0 5.05.1

1 5.0 0

- 7.0

5.0 3.0

1.0 1 .0

1 8.0

6.0 4.0

2.0 00 1x 5.1 1 5.0 0

10

-8 -6 -4 -2 0 2

P = 0.025

P = 0.003

P = 0.006

r = 0.323

r = –0.419

r = –0.388

Total

AggrecanG3

AggrecanG1

COMP

FN1

PRELP

CLU

COL3A1

COL2A1

–1

–0.

50

0.5

1–1

–

00.

50.

51

–0.5

0

0.5

1– 0

.5

0

0.5

1 – 1

– 0.5

0

0.5

1

– 1

–

–1

––1

–

00.

50.

51

00.

50.

51

00.

50.

51

1.5

2

- 8 -6 -4 -2 0 2

Deamidated

P = 0.001

P = 0.003

P = 0.022

P < 0.0001

P = 0.0002

P = 0.011

P = 0.028

r = 0.506

r = 0.368

r = 0.318

r = 0.416

r = 0.331

r = 0.591

r = 0.476

-1- 0

.8

- 0.6

- 0

.4

-0.2

0

0.2

0.

4- 1

-

0.5

0

0.5

1

- 0.5

0

0.5

1

- 0.5

0

0.5

1

-1

-0.

5

0

0

.5

1-1

-

0.5

0

0.5

1

1

.5

2-1

-0.

5

0

0.

5

1-1

- 0

.5

0

0

.5

1

AggrecanG3

AggrecanG1

COMP

FN1

PRELP

CLU

COL3A1

COL2A1

Healthy OA

P = 0.0002

P = 0.032

P = 0.035

P = 0.007

P = 0.028

P = 0.038

P = 0.006

P = 0.004

P = 0.050

P = 0.023

-4 -3 -2 -1 0 1 -8 -6 -4 -2 0

miR-21 miR-21

Deamidationratio

E

r = 0.435

r = 0.488

r = 0.723

r = 0.415

r = 0.504

r = 0.422

r = 0.401

r = 0.513

r = 0.534

r = 0.380

-0.5

0

0.5

1-

0.6

- 0.2

00.

20.

6- 0

.5

0

0.

5

1-0

.8

-0.

4

0

0.

4- 0

.5

0

0.5

1- 1

-0.

5

0

0

.5-0

.4

0

0.

4

0.

8-0

.5

0

0

.5

1

Protein abundance

(total or deamidated)

Lowanabolism

High anabolism

Lowanabolism

High anabolism

P = 0.062r = 0.278 P = 0.904

r = 0.029

P = 0.069r = 0.405

P = 0.156r = 0.321

P = 0.089r = 0.380

P = 0.615r = 0.120

P = 0.547r = –0.139

P = 0.824 r = –0.033

P = 0.117r = –0.229

P = 0.130r = –0.222

P = 0.709r = 0.056

P = 0.433r = –0.116

Fig. 3. miRNAs that regulate whole-limb regeneration in animals are enriched in the cartilage of ankle joints, OA-affected cartilage, and cartilage superficiallayer. (A to C) miRNA distribution in human postnatal articular cartilage stratified by (A) joint site, (B) disease state, and (C) cartilage depth (n = 48 in total). BlastemamiRNAs (miR-21, miR-31, and miR-181c) are presented as fold change from the mean expression of internal control (ctrl) miRNAs (miR-423, miR-191, and miR-U6), allquantified using miScript SYBR Green PCR Kit. P values were derived from multivariable mixed models. (D) Association of miR-21 with total protein and deamidatedprotein abundance. Higher miR-21 expression was consistently associated with lower deamidated protein ratio but had a less consistent association with total proteinabundance. (E) The association of miR-21 with deamidated protein ratio was observed consistently in OA cartilage but less consistently in healthy cartilage. P valueswere derived from Pearson correlations. r = correlation. All P and r values are reported, but regression (green) lines are only shown for correlations with P < 0.05.Sup, superficial cartilage; Mid, middle depth cartilage.

Hsueh et al., Sci. Adv. 2019;5 : eaax3203 9 October 2019 4 of 9

SC I ENCE ADVANCES | R E S EARCH ART I C L E

on June 14, 2020http://advances.sciencem

ag.org/D

ownloaded from

were highly variable in healthy cartilage but markedly similar in OAcartilage (Fig. 4D). We believe that collagens are indirect rather thandirect targets of miRNAs. Using a representative example of miR-21and collagen proteins, we now provide evidence in support of the abilityof miRNA to indirectly activate collagen proteins. In this study, miR-21gene expression was inversely associated with gene expression of its

Hsueh et al., Sci. Adv. 2019;5 : eaax3203 9 October 2019

known target, transforming growth factor-b receptor 2 (TGFBR2)(r = −0.66, P = 0.0002). In turn, TGFBR2 gene expression was inverselyassociated with the protein expression of numerous collagen proteins(COL1A1, COL1A2, COL2A1, COL5A2, COL9A1, COL9A3, andCOL16A1; range of P values, 0.042 to 0.0001; range of r values, −0.41to−0.73)quantifiedbyMS.Thenet effect is anup-regulationof expression

Fig. 4. miRNAs and cartilage proteins respond collectively to the stress of OA. (A) Heatmap representing the associations of miR-21, miR-31, and miR-181c withthe most abundant 193 cartilage proteins identified in at least 50% of samples in healthy (H) and OA cartilage. The distance (d) listed between each paired dataset wasderived from the cluster analysis; distance equal to 1 means two paired sets were unassociated. The associations of miRNAs and cartilage proteins were orderedaccording to the pattern produced by miR-21 in the healthy cartilage as the reference. The most similar association patterns were evident for the miRNA in OA cartilage(distance equal to 0.16 to 0.17), suggesting entrainment or synchronization of cartilage protein expression by all three miRNA in the context of the stress of OA. Theassociation of miR-181c with cartilage proteins changed the most from non-OA to OA cartilage (distance equal to 1 versus 0.16); in fact, the overall patterns for miR-181c innon-OA and OA were inversely associated (table below graphic). (B) The expression of miR-21 was significantly linked to 106 cartilage proteins: 50 of these proteins wereidentified exclusively in OA cartilage (among them 12 collagen components); 29 of these proteins were identified in both OA and healthy cartilage (included 3 collagencomponents); and 27 of these proteins were identified exclusively in healthy cartilage (no unique collagen components identified). (C) String analysis identified protein-protein interactions in OA cartilage that were significantly associated with miR-21 expression. Of note, a tightly integrated network of protein members of the collagenfamily was highly associated with miR-21 and exclusively identified in OA cartilage (B and C). (D) Heatmap illustrating the associations of miRNA and abundance of collagenproteins ordered according to the pattern produced by miR-21 in the healthy cartilage. Similar to results for the total measured proteome, the most similar associationpatterns were evident for the miRNA in OA cartilage (distance equal to 0.14 to 0.21), suggesting entrainment or synchronization of collagen protein expression by all threemiRNA in the context of the stress of OA.

5 of 9

SC I ENCE ADVANCES | R E S EARCH ART I C L E

of collagenproteinsbymiR-21.Together, these findings suggest thatmajorcartilage proteins, such as collagens, are coordinately regulated by evolu-tionarily conserved blastema miRNAs in response to the stress of OA.

on June 14, 2020http://advances.sciencem

ag.org/D

ownloaded from

DISCUSSIONIt is generally believed that human articular cartilage has limited regen-erative capacity. Still, studies have discovered progenitor cells residing inmature cartilage (35) and stem cells in Ranvier’s groove of cartilage andother intra-articular tissues (36). Intrinsic regenerative capacity of artic-ular cartilage has also been suggested by clinical studies observing anapparent increase in radiographic joint space width and cartilage thick-ness as a result of joint distraction (37). Together, these studies indicatethe potential of cartilage repair capability in postnatal (mature) articularcartilage.

Our study demonstrated a position-dependent gradient (distal high,proximal low) of protein turnover in human lower limb cartilage. Wealso demonstrated that the expression of blastema miRNA is notablyassociated with the protein turnover gradient. These findings reveal adynamic anabolic effect in human limbs, which reflect a potential in-nate, albeit limited, regenerative capacity in human cartilage. Theseresults are consistent with studies showing increased expression ofmatrix proteins inOAknee cartilage (38–40). The lack of repair capacityin proximal joints may explain, at least in part, the higher prevalence ofhip and knee OA compared with ankle OA, particularly the rarity ofsevere ankleOA (41–43), and the lower prevalence of surface fibrillationof ankle cartilage with age compared with knee cartilage (44). Together,these data suggest a role for regenerativemiRNA in cartilage homeosta-sis, turnover, and intrinsic repair capacity.

Successful regeneration in limb-regenerating animals is a position-dependent response (45). There are now several lines of evidenceemerging to suggest that this anabolic repair or limited regenerativecapacity in humans is primarily dictated by location rather than jointshape or loading. First, as mentioned above, in vivo evidence demon-strates a higher prevalence of hip and knee OA compared with ankleOA. Second, on the basis of principal component analysis andindependent of age and OA disease status, a recent study showed ama-jor and statistically significant difference in DNA methylation of kneecompared with hip joint cartilages (46). Several other recent studiesinvestigating genome-wide methylation profiles in cartilage have alsoidentified joint site differences (comparing knee and hip cartilages) inmethylation patterns that are independent of disease status (47–51).Third, a study comparing synovial fibroblasts from various joints, includ-ing upper and lower human limb joints, demonstrates a topographicaldifference in transcriptomic patterns (52).

There are several limitations of this study. ECM proteins wereextracted with guanidine-HCl, rather than by enzyme digestion orstrong acid solution, thereby reflecting a subset of cartilage proteinsand, in particular, only a subset of collagens and collagen-associatedproteins (53). The COL2A1 extracted here was likely only loosely asso-ciated with the collagen framework. This extractable COL2A1 likely re-presents a portion of the total collagen framework that has a muchhigher turnover rate than the insoluble fraction containing cross-linkedcollagens. There are also limitations based on the use of in silico–predictedmodels. Both COL2A1 and COL3A1 have triple helices, but the I-TAS-SER protein structure prediction program could not accommodate thisstructure. Therefore, for these proteins, the number of interstrand hy-drogen bondswas likely underestimated, whichwould have the effect ofunderestimating the deamidation rate and collagen half-life estimates.

Hsueh et al., Sci. Adv. 2019;5 : eaax3203 9 October 2019

Nevertheless, the half-life of COL2A1 estimated here was the longestamong the proteins we monitored.

Through novel use of in vivo protein deamidationmolecular clocks,we found a distal-proximal gradient of protein turnover, indicative ofinnate tissue anabolism in human postnatal lower limb cartilages that isnotably associatedwith the expression of evolutionarily conserved limb-regenerative (blastema) miRNAs. We also identified a coordinate reg-ulation of cartilage protein expression in associationwith the expressionof regenerative miRNAs in the context of the stress of OA. These find-ings indicate a hitherto unappreciated innate, albeit limited, regenerativecapacity of human cartilage that is linked to miRNAs and up-regulatedinOA. These data also suggest that anabolic treatmentsmay be requiredin addition to anticatabolic treatments, especially for hip joints, to pre-vent or slow the progression of OA. Future functional studies on theeffects of thesemiRNAswould be required to further elucidate their reg-ulatory role in cartilage repair.

Mechanisms associated with the mammalian regenerative responseinvolve the regulation of pathways that are intricately linked to a largenumber of human disease states (45) and as shown here, OA. Furtherinsights into the parallels of animal limb regeneration and human limbtissue repair and regeneration could inform future therapeuticapproaches to arthritis. Moreover, injection of key regenerativemiRNAin a joint, singly or in combination, could potentially enhance endoge-nous repair and resist degeneration of joint tissues in arthritis of all typesincluding after traumatic injury or insult. Findings such as these, relatedto innate repair, could also be applied to tissue-engineered constructs toenhance exogenous tissue generation prior to transplantation. Identifi-cation of key “missing” (expressed in limb-regenerating organisms butnot in mammals) factors, including miRNA, could lead to future use ofa “molecular cocktail” for attempted recapitulation of blastema-mediated limb regeneration in humans.

MATERIALS AND METHODSClinical cartilage specimen sourcesUnder the Duke Institutional Review Board approval, all articularcartilages were collected fromDukeUniversity Hospital as waste sur-gical specimens. Full-thickness cartilage was collected from perilesionalregions of the load-bearing area of the hip, knee, and ankle joints frompatients with end-stage OA who had total arthroplasty surgery. Full-thickness healthy non-OA cartilage was collected at the time of surgeryfor acute trauma; the absence ofOAwas determined by the surgeon andconfirmed bymacroscopic inspection upon acquisition of the sample inthe laboratory. A total of 18 individual specimens, including three typesof joints (hip, knee, and ankle), two types of disease state (healthy andOA), and three biological replicates of each type with matched agerange, were collected. The mean age of the healthy non-OA patientswas 58.8 years (range, 30 to 82); the mean age of the patients withOA was 59.8 years (range, 42 to 87). Within 2 hours of surgical acqui-sition, all specimenswere snap frozen in tubes on dry ice and stored in a−80°C freezer until extracted.

Tissue dissection and intracellular proteome isolationFrozen cartilage specimens without subchondral bone were embeddedin Tissue-Tek O.C.T. (Sakura, Alphen aan den Rijn, The Netherlands)for cryosectioning. Serial transverse frozen sections (12-mm thickness)were collected starting from the cartilage surface; 20 adjacent sectionswere collected at different distances from the surface to represent thesuperficial (0 to 240 mm), middle (480 to 720 mm), and deep (960 to

6 of 9

SC I ENCE ADVANCES | R E S EARCH ART I C L E

on June 14, 2020http://advances.sciencem

ag.org/D

ownloaded from

1200 mm) layers. Twenty sections between each layer were skipped toavoid cross-contamination. In all, a total of 54 samples were acquired,representing each joint site, disease state, and cartilage depth.

Sample preparation for MS analysisExtraction of the cartilage with guanidine-HCl, representing the ECM-enriched proteome, was performed as previously described (54). Briefly,formass spectrometric analyses, frozen cartilage sections were extractedusing 4Mguanidine-HCl extraction buffer with 0.2%RapiGest (WatersCorporation, Milford, MA) for 24 hours on an orbital shaker at 4°C.Protein extracts were reduced with 4 mM dithiothreitol, alkylated with16 mM iodoacetamide, and then precipitated with ethanol (9:1)overnight at 4°C. Samples were digested with 2 mg of trypsin, filteredthrough a 30-kDa filter (Pall Life Sciences, Port Washington, NY) andthen a reversed-phase C18 column (The Nest group, Southborough,MA) to remove residual polysaccharides and salts.

MS analysisDiscovery experiments (nontargetedMS) were performed on a quadru-pole Orbitrap benchtopmass spectrometer (Q Exactive) (Thermo FisherScientific, Waltham, WA) equipped with an EASY-nLC 1000 system(Thermo Scientific, Waltham, MA) as previously described (55). An MSscan [400 to 1200 mass/charge ratio (m/z)] was recorded in the Orbitrapmass analyzer set at a resolution of 70,000 at 200m/z, 1 × 106 automaticgain control target, and 100-ms maximum ion injection time.

Database searchingProtein identificationwas performedusing theHomo sapiens taxonomy(20,200 sequences) setting of the Swiss-Prot database (SwissProt_2015_06)with Proteome Discoverer 2.1 (version 2.1.1.21, Thermo Scientific) aspreviously described (53). Quantification of MS1 precursor ions wasanalyzed using the ProteomeDiscoverer 2.1 (version 2.1.1.21, ThermoScientific). A relative quantification approachwas used for deamidationmeasurement as follows: The ratio of a deamidated peptide/total pep-tide was used to represent the abundance of the deamidated peptide.The ratio of a specific deamidated peptide across the samples was stan-dardized (z score); themean standardized values of the peptides derivedfrom one protein were used to represent the level of deamidated proteinabundance.

miRNA quantification by real-time polymerasechain reactionMatched frozen cartilage tissue from patients used for proteomic analysiswas dissected transversely from the cartilage surface. Total RNA wasextracted using the miRNeasy Mini Kit (Qiagen) following the manufac-turer’s instructionwithmodifications. Reverse transcriptionwas conductedwith miScript II RT Kit (Qiagen) following kit instructions. Quantificationof blastema miRNAs (miR-21, miR-31, and miR-181c) and internal con-trol miRNAs (miR-423, miR-191, andmiR-U6) was assessed by quantita-tive real-time polymerase chain reaction (PCR) using the miScript SYBRGreenPCRKit (Qiagen). TheDCt values for blastemamiRNAs, represent-ing the miRNA normalized to the mean of the internal control miRNAs,were used to evaluate the correlation ofmiRNAand cartilage ECMproteinexpression; these DCt values were computed as the Ct of the blastemamiRNA minus the mean Ct of the internal control miRNAs.

Deamidation rate prediction and protein half-life estimationHalf-life determinations require an estimate of deamidation rate con-stants based on the local 2D and 3D protein structure. Protein crystal

Hsueh et al., Sci. Adv. 2019;5 : eaax3203 9 October 2019

structures were available through the Protein Data Bank (www.rcsb.org/pdb) for portions of some proteins (FN1 and COMP) or generatedusing the I-TASSER (http://zhanglab.ccmb.med.umich.edu/I-TASSER)(24, 56). Swiss-PdbViewer (v4.10) (57) was used to visualize thestructures and manually acquire the steric covariate values for estimat-ing deamidation rate (16, 34). Deamidation, like racemization, is a first-order process (17). Therefore, accumulation of deamidated residues canbe described by two linear processes: the rate constants of deamidation(Kid) and the protein turnover (KT). The algorithm can be described as

dCAsnðD=Dþ NÞÞ=dt ¼ KidCAsnN=ðDþ NÞ �KTCAsnN=D=ðDþ NÞ

where CAsn is the molar concentration of Asn in the protein, andD/(D + N) and N/(D + N) represent the fractions of the deamidatedandnondeamidatednative proteins. Themeasured amount of deamidatedresidues in the sample is the net result of accumulated deamidated residueaccumulation and their loss due to protein turnover. The original estima-tion of deamidation rates was reported in days (17). Therefore, the de-amidation rates derived from our algorithm also roughly correspond tounits of days.

Quantification of DCOMP by immunoassayProteins extracted, as described above from cryosections (but withoutreduction, alkylation, or trypsin digestion), were precipitated with ice-cold EtOH (95% ethanol with 50 mMNaAc). DCOMP was quantifiedusing a previously described sandwich ELISA assay (14).

Data analysisSince our experimental design contains repeatedmeasures fromthree typesof cartilage depth per individual specimen,multivariablemixedmodels forcontinuous outcome variables (e.g., deamidated protein ratio and miRNAabundance) includemultiple independent factors, joint site (hip/knee/ankle), disease state (healthy/OA), depth (superficial, middle, and deep),age, and gender of the cartilage samples. The first-order autoregressive[AR(1)]was used as a variance-covariancematrix to account for within-and between-specimen correlations. The primary predictors of interestwere joint site and cartilage depth. Similarity measurement of correlationsof miRNA and cartilage proteins was performed using Cluster 3.0 (58).Protein-protein interactions were visualized using STRING (59). Statisticalsignificance of each factor and the overall model was reported at thenominal significance level (P < 0.05). The multivariable mixed-modelanalyses were performed using JMP Pro 13 (SAS, Cary, NC). All thegraphswere prepared inMicrosoft Excel, PowerPoint 2013, orGraphPadPrism 8.

SUPPLEMENTARY MATERIALSSupplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/5/10/eaax3203/DC1Supplementary Materials and MethodsFig. S1. Disease-related distributions of deamidated proteins in human adult cartilage.Fig. S2. The correlation of miR-31 with total protein and deamidated cartilage proteinabundance.Fig. S3. Classification of cartilage proteins correlated with miR-21.

REFERENCES AND NOTES1. B. L. King, V. P. Yin, A conserved microRNA regulatory circuit is differentially controlled

during limb/appendage regeneration. PLOS ONE 11, e0157106 (2016).

7 of 9

SC I ENCE ADVANCES | R E S EARCH ART I C L E

on June 14, 2020http://advances.sciencem

ag.org/D

ownloaded from

2. C. M. Illingworth, Trapped fingers and amputated finger tips in children. J. Pediatr. Surg. 9,853–858 (1974).

3. K. M. Heinemeier, P. Schjerling, J. Heinemeier, M. B. Møller, M. R. Krogsgaard,T. Grum-Schwensen, M. M. Petersen, M. Kjaer, Radiocarbon dating reveals minimalcollagen turnover in both healthy and osteoarthritic human cartilage. Sci. Transl. Med. 8,346ra390 (2016).

4. K. M. Heinemeier, P. Schjerling, J. Heinemeier, S. P. Magnusson, M. Kjaer, Lack of tissuerenewal in human adult Achilles tendon is revealed by nuclear bomb 14C. FASEB J. 27,2074–2079 (2013).

5. N. E. Robinson, A. B. Robinson, Deamidation of human proteins. Proc. Natl. Acad. Sci. U.S.A.98, 12409–12413 (2001).

6. T. Flatmark, On the heterogeneity of beef heart cytochrome c. III. A kinetic study of thenon-enzymic deamidation of the main subfractions (Cy I-Cy III). Acta Chem. Scand. 20,1487–1496 (1966).

7. T. Flatmark, On the heterogeneity of beef heart cytochrome c. II. Some physico-chemicalproperties of the main subfractions (Cy I-Cy 3). Acta Chem. Scand. 20, 1476–1486 (1966).

8. T. Flatmark, K. Sletten, Multiple forms of cytochrome c in the rat. Precursor-productrelationship between the main component Cy I and the minor components Cy II and Cy 3in vivo. J. Biol. Chem. 243, 1623–1629 (1968).

9. H. Yang, R. A. Zubarev, Mass spectrometric analysis of asparagine deamidation andaspartate isomerization in polypeptides. Electrophoresis 31, 1764–1772 (2010).

10. N. E. Robinson, A. B. Robinson, R. B. Merrifield, Mass spectrometric evaluation of syntheticpeptides as primary structure models for peptide and protein deamidation. J. Pept. Res.57, 483–493 (2001).

11. P. P. Hurtado, P. B. O’Connor, Deamidation of collagen. Anal. Chem. 84, 3017–3025(2012).

12. P. A. Cloos, S. Christgau, Non-enzymatic covalent modifications of proteins: Mechanisms,physiological consequences and clinical applications. Matrix Biol. 21, 39–52 (2002).

13. J. B. Catterall, D. Barr, M. Bolognesi, R. D. Zura, V. B. Kraus, Post-translational aging ofproteins in osteoarthritic cartilage and synovial fluid as measured by isomerizedaspartate. Arthritis Res. Ther. 11, R55 (2009).

14. J. B. Catterall, M. F. Hsueh, T. V. Stabler, C. R. McCudden, M. Bolognesi, R. Zura, J. M. Jordan,J. B. Renner, S. Feng, V. B. Kraus, Protein modification by deamidation indicatesvariations in joint extracellular matrix turnover. J. Biol. Chem. 287, 4640–4651 (2012).

15. J. B. Catterall, R. D. Zura, M. P. Bolognesi, V. B. Kraus, Aspartic acid racemization reveals ahigh turnover state in knee compared with hip osteoarthritic cartilage. Osteoarthr. Cartil.24, 374–381 (2016).

16. N. E. Robinson, A. B. Robinson, Prediction of protein deamidation rates from primary andthree-dimensional structure. Proc. Natl. Acad. Sci. U.S.A. 98, 4367–4372 (2001).

17. N. E. Robinson, A. B. Robinson, Molecular clocks. Proc. Natl. Acad. Sci. U.S.A. 98, 944–949(2001).

18. N. E. Robinson, Z. W. Robinson, B. R. Robinson, A. L. Robinson, J. A. Robinson,M. L. Robinson, A. B. Robinson, Structure-dependent nonenzymatic deamidation ofglutaminyl and asparaginyl pentapeptides. J. Pept. Res. 63, 426–436 (2004).

19. R. A. Robinson NE (Althouse Press, 2004).20. D. C. Lo, F. Allen, J. P. Brockes, Reversal of muscle differentiation during urodele limb

regeneration. Proc. Natl. Acad. Sci. U.S.A. 90, 7230–7234 (1993).21. J. P. Brockes, A. Kumar, Plasticity and reprogramming of differentiated cells in amphibian

regeneration. Nat. Rev. Mol. Cell Biol. 3, 566–574 (2002).22. G. G. Whitehead, S. Makino, C. L. Lien, M. T. Keating, fgf20 is essential for initiating

zebrafish fin regeneration. Science 310, 1957–1960 (2005).23. S. Makino, G. G. Whitehead, C.-L. Lien, S. Kim, P. Jhawar, A. Kono, Y. Kawata, M. T. Keating,

Heat-shock protein 60 is required for blastema formation and maintenance duringregeneration. Proc. Natl. Acad. Sci. U.S.A. 102, 14599–14604 (2005).

24. A. Roy, A. Kucukural, Y. Zhang, I-TASSER: A unified platform for automated proteinstructure and function prediction. Nat. Protoc. 5, 725–738 (2010).

25. K. Huang, L. D. Wu, Aggrecanase and aggrecan degradation in osteoarthritis: A review.J. Int. Med. Res. 36, 1149–1160 (2008).

26. P. Roughley, J. Mort, The role of aggrecan in normal and osteoarthritic cartilage.J. Exp. Ortop. 1, 8 (2014).

27. J. Dudhia, C. M. Davidson, T. M. Wells, T. E. Hardingham, M. T. Bayliss, Studies on the G3domain of aggrecan from human Cartilagea. Ann. N.Y. Acad. Sci. 785, 245–247 (1996).

28. C. Kiani, L. Chen, Y. J. Wu, A. J. Yee, B. B. Yang, Structure and function of aggrecan. Cell Res.12, 19–32 (2002).

29. J. M. Day, A. I. Olin, A. D. Murdoch, A. Canfield, T. Sasaki, R. Timpl, T. E. Hardingham,A. Aspberg, Alternative splicing in the aggrecan G3 domain influences bindinginteractions with tenascin-C and other extracellular matrix proteins. J. Biol. Chem. 279,12511–12518 (2004).

30. A. Aspberg, The different roles of aggrecan interaction domains. J. Histochem. Cytochem.60, 987–996 (2012).

31. S.-S. Sivan, E. Wachtel, E. Tsitron, N. Sakkee, F. van der Ham, J. DeGroot, S. Roberts,A. Maroudas, Collagen turnover in normal and degenerate human intervertebral discs

Hsueh et al., Sci. Adv. 2019;5 : eaax3203 9 October 2019

as determined by the racemization of aspartic acid. J. Biol. Chem. 283, 8796–8801(2008).

32. S. S. Sivan, E. Tsitron, E. Wachtel, P. J. Roughley, N. Sakkee, F. van der Ham, J. DeGroot,S. Roberts, A. Maroudas, Aggrecan turnover in human intervertebral disc as determinedby the racemization of aspartic acid. J. Biol. Chem. 281, 13009–13014 (2006).

33. A. Maroudas, M. T. Bayliss, N. Uchitel-Kaushansky, R. Schneiderman, E. Gilav, Aggrecanturnover in human articular cartilage: Use of aspartic acid racemization as a marker ofmolecular age. Arch. Biochem. Biophys. 350, 61–71 (1998).

34. N. E. Robinson, A. B. Robinson, Prediction of primary structure deamidation rates ofasparaginyl and glutaminyl peptides through steric and catalytic effects. J. Pept. Res. 63,437–448 (2004).

35. Y. Jiang, R. S. Tuan, Origin and function of cartilage stem/progenitor cells in osteoarthritis.Nat. Rev. Rheumatol. 11, 206–212 (2015).

36. M. Iwamoto, Y. Ohta, C. Larmour, M. Enomoto-Iwamoto, Toward regeneration of articularcartilage. Birth Defects Res. C Embryo Today 99, 192–202 (2013).

37. F. Intema, P. M. Van Roermund, A. C. A. Marijnissen, S. Cotofana, F. Eckstein,R. M. Castelein, J. W. J. Bijlsma, S. C. Mastbergen, F. P. J. G. Lafeber, Tissue structuremodification in knee osteoarthritis by use of joint distraction: An open 1-year pilot study.Ann. Rheum. Dis. 70, 1441–1446 (2011).

38. M. B. Goldring, Chondrogenesis, chondrocyte differentiation, and articular cartilagemetabolism in health and osteoarthritis. Ther. Adv. Musculoskelet. Dis. 4, 269–285(2012).

39. M. B. Goldring, K. B. Marcu, Cartilage homeostasis in health and rheumatic diseases.Arthritis Res. Ther. 11, 224 (2009).

40. T. Videman, I. Eronen, T. Candolin, [3H]proline incorporation and hydroxyprolineconcentration in articular cartilage during the development of osteoarthritis caused byimmobilization. A study in vivo with rabbits. Biochem. J. 200, 435–440 (1981).

41. C. Muehleman, A. Margulis, W. C. Bae, K. Masuda, Relationship between knee and ankledegeneration in a population of organ donors. BMC Med. 8, 48 (2010).

42. V. B. Kraus, T. M. Kilfoil, T. W. Hash II, G. McDaniel, J. B. Renner, J. A. Carrino, S. Adams,Atlas of radiographic features of osteoarthritis of the ankle and hindfoot. Osteoarthr.Cartil. 23, 2059–2085 (2015).

43. S. Treppo, H. Koepp, E. C. Quan, A. A. Cole, K. E. Kuettner, A. J. Grodzinsky, Comparison ofbiomechanical and biochemical properties of cartilage from human knee and ankle pairs.J. Orthop. Res. 18, 739–748 (2000).

44. J. A. Martin, J. A. Buckwalter, Roles of articular cartilage aging and chondrocytesenescence in the pathogenesis of osteoarthritis. Iowa Orthop. J. 21, 1–7 (2001).

45. J. Simkin, M. C. Sammarco, L. A. Dawson, P. P. Schanes, L. Yu, K. Muneoka, The mammalianblastema: Regeneration at our fingertips. Regeneration 2, 93–105 (2015).

46. W. den Hollander, Y. F. M. Ramos, S. D. Bos, N. Bomer, R. van der Breggen, N. Lakenberg,W. J. de Dijcker, B. J. Duijnisveld, P. E. Slagboom, R. G. H. H. Nelissen, I. Meulenbelt,Knee and hip articular cartilage have distinct epigenomic landscapes: Implications forfuture cartilage regeneration approaches. Ann. Rheum. Dis. 73, 2208–2212 (2014).

47. E. Aref-Eshghi, Y. Zhang, M. Liu, P. E. Harper, G. Martin, A. Furey, R. Green, G. Sun,P. Rahman, G. Zhai, Genome-wide DNA methylation study of hip and knee cartilagereveals embryonic organ and skeletal system morphogenesis as major pathwaysinvolved in osteoarthritis. BMC Musculoskelet. Disord. 16, 287 (2015).

48. M. D. Rushton, L. N. Reynard, M. J. Barter, R. Refaie, K. S. Rankin, D. A. Young, J. Loughlin,Characterization of the cartilage DNA methylome in knee and hip osteoarthritis.Arthritis Rheumatol. 66, 2450–2460 (2014).

49. M. A. Jeffries, M. Donica, L. W. Baker, M. E. Stevenson, A. C. Annan, M. B. Humphrey,J. A. James, A. H. Sawalha, Genome-wide DNA methylation study identifies significantepigenomic changes in osteoarthritic cartilage. Arthritis Rheumatol. 66, 2804–2815(2014).

50. J. Fernández-Tajes, A. Soto-Hermida, M. E. Vázquez-Mosquera, E. Cortés-Pereira,A. Mosquera, M. Fernández-Moreno, N. Oreiro, C. Fernández-López, J. L. Fernández,I. Rego-Pérez, F. J. Blanco, Genome-wide DNA methylation analysis of articularchondrocytes reveals a cluster of osteoarthritic patients. Ann. Rheum. Dis. 73, 668–677(2013).

51. F. C. Moazedi-Fuerst, M. Hofner, G. Gruber, A. Weinhaeusel, M. H. Stradner, H. Angerer,D. Peischler, B. Lohberger, M. Glehr, A. Leithner, M. Sonntagbauer, W. B. Graninger,Epigenetic differences in human cartilage between mild and severe OA. J. Orthop. Res.32, 1636–1645 (2014).

52. M. Frank-Bertoncelj, M. Trenkmann, K. Klein, E. Karouzakis, H. Rehrauer, A. Bratus,C. Kolling, M. Armaka, A. Filer, B. A. Michel, R. E. Gay, C. D. Buckley, G. Kollias, S. Gay,C. Ospelt, Epigenetically-driven anatomical diversity of synovial fibroblasts guidesjoint-specific fibroblast functions. Nat. Commun. 8, 14852 (2017).

53. M.-F. Hsueh, A. Khabut, S. Kjellström, P. Önnerfjord, V. B. Kraus, Elucidating the molecularcomposition of cartilage by proteomics. J. Proteome Res. 15, 374–388 (2016).

54. C. Müller, A. Khabut, J. Dudhia, F. P. Reinholt, A. Aspberg, D. Heinegård, P. Önnerfjord,Quantitative proteomics at different depths in human articular cartilage reveals uniquepatterns of protein distribution. Matrix Biol. 40, 34–45 (2014).

8 of 9

SC I ENCE ADVANCES | R E S EARCH ART I C L E

55. Y. Wang, Y. Li, A. Khabut, S. Chubinskaya, A. J. Grodzinsky, P. Önnerfjord, Quantitativeproteomics analysis of cartilage response to mechanical injury and cytokine treatment.Matrix Biol. 63, 11–22 (2017).

56. Y. Zhang, I-TASSER server for protein 3D structure prediction. BMC Bioinformatics 9, 40 (2008).57. N. Guex, M. C. Peitsch, SWISS-MODEL and the Swiss-PdbViewer: An environment for

comparative protein modeling. Electrophoresis 18, 2714–2723 (1997).58. M. J. L. de Hoon, S. Imoto, J. Nolan, S. Miyano, Open source clustering software.

Bioinformatics 20, 1453–1454 (2004).59. D. Szklarczyk, J. H. Morris, H. Cook, M. Kuhn, S. Wyder, M. Simonovic, A. Santos,

N. T. Doncheva, A. Roth, P. Bork, L. J. Jensen, C. von Mering, The STRING database in 2017:Quality-controlled protein-protein association networks, made broadly accessible.Nucleic Acids Res. 45, D362–D368 (2017).

Acknowledgments: We thank B. Donald and S. Endo-Streeter for helping with the initialprotein structure modeling. We also thank Y.-J. Li, the Associate Director of the BiostatisticsCore of the Duke Translational Medicine Institute, for advice related to the statistical modelingfor this study. Funding: This study was supported by an OARSI Collaborative Scholarshipto M.-F. H., a Collaborative Exchange Award from the Orthopaedic Research Society toV.B.K., and NIH/NIA P30-AG-028716. Mass spectrometers were funded by the Crafoord

Hsueh et al., Sci. Adv. 2019;5 : eaax3203 9 October 2019

Foundation and the Inga-Britt and Arne Lundberg Foundation, and other financial supportwas obtained from the Swedish Research Council (2014-3303) to P.Ö. Author contributions:M.-F.H., P.Ö., and V.B.K. were involved in the conception and design of the study as wellas the interpretation of the data. M.E.E. and M.P.B. provided the cartilage specimens. M.-F.H.and P.Ö. conducted the MS experimental work. M.-F.H. conducted the analysis and draftedthe manuscript. All authors critically revised the article and approved the final version forsubmission. Competing interests: The authors declare that they have no competing interests.Data and materials availability: All data needed to evaluate the conclusions in the paperare present in the paper and/or the Supplementary Materials. Additional data related to thispaper may be requested from the authors.

Submitted 13 March 2019Accepted 10 September 2019Published 9 October 201910.1126/sciadv.aax3203

Citation: M.-F. Hsueh, P. Önnerfjord, M. P. Bolognesi, M. E. Easley, V. B. Kraus, Analysis of “old”proteins unmasks dynamic gradient of cartilage turnover in human limbs. Sci. Adv. 5, eaax3203(2019).

9 of 9

on June 14, 2020http://advances.sciencem

ag.org/D

ownloaded from

Analysis of ''old'' proteins unmasks dynamic gradient of cartilage turnover in human limbsMing-Feng Hsueh, Patrik Önnerfjord, Michael P. Bolognesi, Mark E. Easley and Virginia B. Kraus

DOI: 10.1126/sciadv.aax3203 (10), eaax3203.5Sci Adv 

ARTICLE TOOLS http://advances.sciencemag.org/content/5/10/eaax3203

MATERIALSSUPPLEMENTARY http://advances.sciencemag.org/content/suppl/2019/10/07/5.10.eaax3203.DC1

REFERENCES

http://advances.sciencemag.org/content/5/10/eaax3203#BIBLThis article cites 58 articles, 14 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.Science AdvancesYork Avenue NW, Washington, DC 20005. The title (ISSN 2375-2548) is published by the American Association for the Advancement of Science, 1200 NewScience Advances

License 4.0 (CC BY-NC).Science. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution NonCommercial Copyright © 2019 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of

on June 14, 2020http://advances.sciencem

ag.org/D

ownloaded from