Hyaluronic acid and tissue mechanics orchestrate mammalian digit tip regeneration

Fibrous tissue accumulates after nonregenerative amputations (3). We used second harmonic generation (SHG) microscopy to investigate the collagen matrix after digit amputations. Nonregenerative wounds contained dense, fibrosis-like collagen, which was largely absent from the blastema (Fig. 1A and fig. S1, A and B). We confirmed the collagen’s architectural features by transmission electron microscopy, which similarly showed more organized fiber bundles in nonregeneration (Fig. 1B and fig. S1C). To investigate what ECM factors underlie the difference in the collagen responses, we performed single-cell transcriptomic analysis of nonregenerative wounds and integrated these datasets with previously published blastema and nonregenerative datasets (5, 6) (Fig. 1C). To identify the primary cell type responsible for the ECM among the seven distinct populations detected (fig. S1, D and E), we assigned an ECM score to each cell based on its expression of core matrisome genes (29) (Fig. 1D). The Pdgfrα-expressing cluster showed the highest ECM activity, outpacing all other clusters (Fig. 1D). Gene set enrichment analysis revealed that this cluster up-regulated ECM-related pathways, including “Collagen Fibril Organization” and “Extracellular Matrix Organization” (fig. S1F). We deduced that fibroblasts constituted most of the Pdgfrα-expressing cluster and were the main cell type involved in establishing the extracellular milieu.

Because fibroblasts are heterogeneous, we hypothesized that nonregenerative and regenerative digit ECMs diverge owing to underlying differences in cellular composition. Subsetting Pdgfrα-expressing cells revealed eight transcriptional signatures of which Fibroblast 1 and 2 subtypes were preferentially enriched in nonregeneration (Fig. 1E and fig. S1, G to I). With no notable transcriptional distinction in fibroblast subtypes based on wound conditions (fig. S1J), we hypothesized that the predominance and activity of Fibroblast 1 and 2 cells accounted for the fibrogenic response. All Pdgfrα-expressing clusters expressed fibrillar collagens to a similar degree (fig. S1K). However, we identified elevated collagen modeling factors in the nonregenerative wound, such as tenascin X (TNX) expressed by PDGFRα⁺CD34⁺CD31⁻ Fibroblast 1 cells (Fig. 1, E and F, and fig. S2, A and B) and thrombospondin-4 (THBS4) expressed by SULF2⁺ Fibroblast 2 cells (Fig. 1, E and G, and fig. S2C). Of note, Fibroblast 3 cells, the major blastema subtype, highly expressed markers associated with limb regeneration, such as Msx1 and Mdk, but did not differentially express major collagen modeling genes (Fig. 1E). Taken together, Fibroblast 1 and 2 cells and their up-regulation of TNX and THBS4, respectively, were specific to nonregeneration, suggesting that collagen maintenance and/or modeling drive the fibrotic response.

Given that fibrosis is the antithesis of regeneration, we hypothesized that the blastema is enriched with cell subtypes responsible for synthesizing regeneration-specific ECM. Our cell subtype correlation and gene expression analysis demonstrated that Fibroblast 1 and 2 cells were most disparate from osteo-lineage (OL) cells (figs. S1J and S2D), which were approximately three times more prevalent in the blastema (fig. S1H). Furthermore, OL cells had the highest expression of collagens and proteoglycans (fig. S2E), which supports the role of OL cells as the blastema’s main ECM modulator. ECM-related gene ontology terms describing OL cells included “Glycosaminoglycan Binding,” “Proteoglycan Binding,” and “Hyaluronic Acid Binding,” suggesting an OL microenvironment in which noncollagenous matrix components, particularly hyaluronic acid (HA), appear prominently in the blastema (fig. S2F). OL cells highly expressed cartilage-associated genes Acan, Col2a1, and Hapln1 as well as related transcription factors Sox6 and Sox9 (fig. S2F). Indeed, at the tissue level, OL cells were more prevalent in regeneration and colocalized with accumulations of HA (fig. S2G), reminiscent of pericellular coats or a glycocalyx. We confirmed higher total HA in the blastema compared with that in the nonregenerative wound (Fig. 1I). Because HA binds proteoglycans through link proteins (30) (Fig. 1J), we assayed the distribution of hyaluronan and the proteoglycan link protein 1 (HAPLN1) and aggrecan (ACAN), finding a stronger presence of both in the blastema (Fig. 1, K and L). Thus, the blastema’s OL cells synthesize copious matrices composed of HA, HAPLN1, and ACAN in the absence of major collagen matrix, forming a distinct ECM niche in regeneration.

Given the importance of ECM composition, particularly collagen, in tissue mechanics (31), we used atomic force microscopy (AFM) and force-clamp force mapping to investigate the physical properties of wounded digits. We detected differences between nonregenerative wounds and the blastema, including greater stiffness and lower fluidity in the former (Fig. 1M and fig. S2H), suggesting that digits undergoing nonregenerative and regenerative responses exhibit distinct mechanical microenvironments. To explore whether the divergent niches are a product of preexisting differences between P2 and P3 cells or arise during wound healing, we performed single-cell analysis of uninjured P2 and P3 digits and integrated these datasets with previously published uninjured and injured datasets (fig. S3A). At baseline, anatomical location strongly segregated Pdgfrα-expressing cells (fig. S3, B and C). Upon injury, both nonregenerative and regenerative cells up-regulated matrisome genes and were transcriptionally more similar to each other than to their uninjured counterparts (fig. S3D). However, only regenerative digits up-regulated HA by 10 days postamputation (DPA), whereas HA remained scarce at earlier timepoints and in nonregenerative digits (fig. S3E). Thus, both intrinsic cellular properties and injury-induced ECM remodeling shape the distinct regenerative and nonregenerative niches.

Given the abundance of hyaluronic acid (HA) in the blastema, we asked whether digit regeneration depended upon HA. To answer this question, we degraded HA enzymatically by serial injections of hyaluronidase into the digit following amputations of the tip of the third phalanx (P3) (Fig. 2A). At 14 DPA, digits were markedly decreased in length, and the area was often smaller (Fig. 2A and fig. S4A). RUNX2⁺ OL cells and pericellular HA were also diminished after hyaluronidase treatment (Fig. 2B), corroborating the relationship between OL cells and HA. For continuous HA depletion, we incorporated 4-methylumbelliferone (4-MU) into the mouse diet to reduce levels of UDP-glucuronic acid, one of two substrates required for HA synthesis (32) (Fig. 2C). With reduction of extracellular HA (fig. S4, B and C), 4-MU digits at 14 DPA were reduced in length and area (fig. S4, D and E). By 28 DPA, control digits regenerated, but 4-MU digits remained significantly smaller (Fig. 2D and fig. S4E). We then examined the state of the blastema in 4-MU digits, hypothesizing that the digits’ defects were a result of compromised blastema formation. With 4-MU, digits globally down-regulated their expression of the blastema marker arylsulfatase I (ARSI) (6) (Fig. 2E). Another feature of the blastema is the prevalence of proliferative cells (33). The blastema contained high numbers of Ki67⁺ proliferating cells, particularly OL cells, which were substantially reduced with 4-MU treatment (Fig. 2F and fig. S4F). The fate of the blastema involves differentiation into mature tissue (7). Using microcomputed tomography, we showed that 14-DPA regenerating digits exhibited histolysis—the expulsion of the distal bone fragment resulting in bone shortening—and new bone formation (fig. S4, G and I). 4-MU delayed histolysis-induced bone shortening, and we detected no new bone (fig. S4, G and I). By 28 DPA, when regeneration was expected to be largely complete, 4-MU digit bone volume, surface area, and length remained diminished compared with those of controls (Fig. 2G and fig. S4, H and I). The regenerated digit contained an abundant population of SP7⁺ osteoblasts, whereas 4-MU reduced osteoblast numbers (Fig. 2H and fig. S4J). Additionally, fibrosis-like collagen was deposited in 4-MU digits, indicating a switch to a more nonregenerative ECM (Fig. 2H and fig. S4J). Thus, depletion of HA interfered with digit restoration after distal P3 amputation, likely owing to a failure of blastema formation and differentiation, which suggests an important role for HA in digit tip regeneration.

Given that HA and collagen were inversely correlated in wounded digits and that HA depletion elicited a fibrosis-like ECM, we asked whether HA matrices abrogate fibrotic collagen assembly. We first compared the collagen matrix of hyaluronidase-treated versus control digits. Hyaluronidase treatment increased collagen content as well as fibrotic architectural features (Fig. 3A and fig. S5A). We observed similar attributes after HA depletion using 4-MU (Fig. 3B and fig. S5B). Overall, disruption of HA produced fibrotic collagen matrix. Because collagen content impacts tissue mechanics, we hypothesized that the fibrotic collagen emerging with HA depletion would alter the tissue’s mechanical properties. To test this idea, we perturbed HA using 4-MU prior to distal third phalanx (P3) amputations and performed stiffness and fluidity measurements. 4-MU–treated digits were stiffer and less fluid compared with controls (Fig. 3C and fig. S5C). Thus, the viscoelasticity of 4-MU–treated digits closely resembled that of nonregenerating wounds (fig. S5C), highlighting how HA-collagen interactions impact the mechanical microenvironment.

To probe how tissue mechanics arising from HA depletion affected cell behavior, we performed single-cell transcriptomic analysis of distal P3–amputated digits after 4-MU treatment versus control. Unsupervised clustering and integration of the datasets resulted in nine major cell types (fig. S5D). Hyaluronic acid depletion demonstrably reduced the relative proportion of Pdgfrα-expressing cells (fig. S5D), indicating a suppression of their typical injury-induced expansion. Analyzing the Pdgfrα-expressing cells, we showed a diminished OL proportion after 4-MU treatment (Fig. 3D and fig. S5E). Comparing all Pdgfrα-expressing cells between conditions underscored the down-regulation of OL-related genes Ibsp, Alpl, and Sox6 and its target gene Acan (Fig. 3E). These results highlighted a disturbance in osteogenic differentiation and the down-regulation of cartilage-associated matrix elements. Notably, 4-MU–treated Pdgfra-expressing cells up-regulated their expression of S100a4 and Tnc (Fig. 3E), both of which are associated with increasing ECM stiffness (12, 34). These genes can also be expressed by tenocytes; however, the absence of a tenocyte-associated gene signature with 4-MU suggested that this transcriptional shift does not reflect tenogenic differentiation (fig. S5F). Indeed, the few Pdgfra-expressing cells that remained also up-regulated blastema-associated genes, but their limited abundance indicates a failure of expansion rather than loss of blastema transcriptional identity (fig. S5G).

Given that HA depletion resulted in skeletal defects and disruption of osteogenic differentiation following distal P3 amputations, we explored the relationship between HA, tissue mechanics, and the bone morphogenic protein (BMP) pathway. Unlike previous reports linking HA with Wnt signaling (18), we did not observe changes in Wnt activation in the blastema after HA depletion (fig. S5H). Consequently, we assayed the primary effector of BMP signaling, pSMAD1/5/8, within the tissue after HA depletion with 4-MU. pSMAD1/5/8 levels were elevated and increased as regeneration progressed in control digits (Fig. 3F). By contrast, 4-MU persistently suppressed pSMAD1/5/8 levels (Fig. 3F). Thus, OL cells not only help establish the ECM environment in regeneration but are also acutely sensitive to HA and changes in tissue mechanics.

To dissect the interplay of mechanical cues and soluble signaling, we created stiff and soft StemBond hydrogels (35) to model the mechanical microenvironment of the nonregenerative wound and blastema, respectively (Fig. 4A). Simultaneously, we administered blastema-associated growth factors to examine their combined effects on cellular responses (Fig. 4A). Back skin fibroblasts on soft substrates exhibited enhanced nuclear pSMAD1/5/8 signal after BMP-7 treatment (Fig. 4A and fig. S6A). Immunoblotting further corroborated these stiffness-dependent patterns in pSMAD1/5/8 levels (fig. S6B). We also profiled changes in the gene expression of Inhibitors of DNA Binding/Differentiation (Id), to assess early downstream effectors of pSMADs. Id1 expression responded to substrate stiffness, with fibroblasts on soft substrates up-regulating the highest levels of Id1 (Fig. 4B and fig. S6C). We found similar trends among freshly isolated cells from the second and third phalanx, which up-regulated their expression of Id1 most when stimulated under soft conditions (Fig. 4C). Given no baseline differences in the ECM in our hydrogel experiments (fig. S6, D and E), our data indicated that substrate stiffness tunes responsiveness to BMP ligands.

Next, we used platelet-derived growth factor BB (PDGF-BB), a blastema injury signal (36) with potent HA-synthesizing activity (37), to test how substrate stiffness influences HA (Fig. 4D). In addition to Hapln1 and Acan, we considered the three isoforms of hyaluronan synthases. Substrate stiffness had little effect on expression levels of Has1-3, and, under these experimental conditions, Acan was not responsive to either PDGF-BB or substrate stiffness (fig. S6, F to H). By contrast, fibroblasts cultured on soft hydrogels with PDGF-BB treatment significantly up-regulated Hapln1 (Fig. 4D). Reasoning that HAPLN1 confers structural integrity to the HA matrix, we asked whether fibroblasts assemble pericellular HA matrix more robustly under soft conditions. Following PDGF-BB treatment, fibroblasts produced more aggregates of HA and HAPLN1 at the cell surface on soft versus stiff substrates (Fig. 4, D and E, and fig. S6I). Because soft mechanical cues enhance proregenerative ECM synthesis, we investigated whether a stiff environment augments nonregenerative ECM and used transforming growth factor (TGF)–β1 and ascorbic acid to induce collagen synthesis (38) (Fig. 4F). After treatment, fibroblasts on stiff substrates produced fibrillar collagen matrix that largely colocalized with THBS4 (Fig. 4F and fig. S6J), both of which typified nonregenerative healing (Fig. 1, A, B, and G). Conversely, in soft conditions that mimic the blastema, collagen I–THBS4 network formation remained undetectable, irrespective of ascorbic acid and TGF-β1 addition (Fig. 4F and fig. S6J). Thus, the softness imbued by HA may initiate positive feedback mechanisms to enhance the production of HA matrix in contrast to stiff substrate-induced collagen fibrillogenesis (Fig. 4G).

Next, we investigated whether HAPLN1 promotes pericellular HA and impacts the collagen matrix. Not only was HAPLN1 abundant alongside HA in the blastema (Fig. 1, H and K), but blastema-like substrate softness also propagated the synthesis of both (Fig. 4, D and E, and fig. S6I). Analysis of uninjured digits showed that the third phalanx (P3) region contained wide swaths of HA- and HAPLN1-rich regions, which appeared only in specific areas of the second phalanx (P2), such as the periosteum (fig. S7A). We corroborated regional differences in HAPLN1 by quantitative polymerase chain reaction (qPCR) analysis of P2 and P3 cells (fig. S7B). These findings showcased strong associations between HAPLN1 and HA in injury and homeostasis, leading us to hypothesize that HAPLN1 is a key factor mediating the presence of HA. To test this hypothesis, we overexpressed Hapln1 (Hapln1^OE) or a scrambled sequence as the control (mCherry Control) in back skin fibroblasts (fig. S7, C to E). We cultured transduced cells on stiff hydrogels to simulate the fibrotic mechanical environment, with or without the presence of high–molecular weight HA to encourage pericellular HA accumulation (Fig. 5A and fig. S7, F and G). Despite HA supplementation, mCherry Control fibroblasts exhibited sparse and limited distribution of HA across their cell surface (Fig. 5A and fig. S7, F and G). Meanwhile, Hapln1^OE fibroblasts synthesized large quantities of HAPLN1 that corresponded with a significant increase in pericellular HA (Fig. 5A and fig. S7, F and G). Next, we tested whether stabilizing HA matrix using HAPLN1 could inhibit stiffness-induced collagen fibrillogenesis. To mimic the fibrogenic milieu, we cultured transduced fibroblasts on stiff hydrogels and used ascorbic acid to induce collagen fibrillogenesis (Fig. 5B). Hapln1^OE fibroblasts accumulated robust pericellular HA that coincided with fewer and shorter collagen fibrils compared with that of mCherry Control fibroblasts (Fig. 5B and fig. S7H). Thus, HAPLN1 increased the deposition of HA and restrained collagen fibrillogenesis, signifying HAPLN1’s potential to promote a regenerative ECM in a nonregenerative context.

Because regeneration requires HA, and HAPLN1 stabilizes the HA matrix, we hypothesized that overexpressing Hapln1 in nonregenerating wounds would initiate restorative repair. To test this hypothesis, we performed nonregenerative amputations on adult immunocompromised mice and transplanted mCherry Control or Hapln1^OE fibroblasts into the digit tip (Fig. 5C and fig. S8A). Notably, we observed one incidence of nail regrowth at 28 DPA with Hapln1^OE fibroblast transplantation (fig. S8A), which does not occur under nonregenerative conditions. Using microcomputed tomography, we showed enhanced bone repair in digits injected with Hapln1^OE fibroblasts, including greater bone elongation and growth (Fig. 5D and fig. S8B). Regenerating bone tissue in Hapln1^OE digits extended beyond the initial amputation plane, in contrast to minimal new tissue formation in controls (Fig. 5D and fig. S8B). To confirm that these regenerative effects were independent of cell transplantation, we injected Hapln1-overexpressing lentivirus into the digit, which similarly resulted in bone elongation and increased callus size (fig. S9A). Hapln1^OE fibroblasts decreased the presence of fibrosis-like collagen structures (Fig. 5E and fig. S8, C to E) and promoted HA and HAPLN1 accumulation (fig. S8, F and G). Because HA and collagen influence tissue mechanics, we hypothesized that HAPLN1-driven ECM remodeling promotes regeneration by altering the mechanical properties of the microenvironment. Using AFM, we found a decrease in stiffness and an increase in fluidity at the amputation site of Hapln1^OE digits compared with those of controls (Fig. 5F and fig. S8H).

Lastly, we explored how ECM modulation by HAPLN1 affects cellular activity. Because Sox9 was elevated in the blastema (fig. S2F), we assayed SOX9 distribution as a marker for early chondro-osteogenic commitment. Digits with mCherry Control fibroblasts exhibited low SOX9⁺ cell numbers, whereas clusters of SOX9⁺ cells localized distally from the plane of amputation in digits with Hapln1^OE fibroblasts (fig. S8I). We also assayed for more differentiated RUNX2⁺ OL cells. In digits transplanted with mCherry Control fibroblasts, we observed RUNX2⁺ OL cells only immediately adjacent to the preexisting cortical bone at 14 and 28 DPA (Fig. 5G and fig. S8, J and K). However, Hapln1^OE fibroblasts increased the presence and distal localization of RUNX2⁺ OL cells at 28 DPA, consistent with bone regrowth (Fig. 5G and fig. S8K). Furthermore, given that soft substrates enhanced responsiveness to BMP signaling in vitro, we assayed pSMAD/1/5/8 levels and found elevated signal with Hapln1^OE compared with that with controls (Fig. 5G and fig. S8L). Thus, HAPLN1 triggers restorative repair after nonregenerative amputations, likely by mediating the HA-collagen matrices and softening the tissue microenvironment.

To explore whether the HA-collagen dichotomy emerges in other regenerative settings, we examined transcriptomic datasets spanning three models of regeneration: ear pinna injury, myocardial infarction, and fracture healing (fig. S10, A to D). Although the specific HA-associated enzymes, cross-linkers, and receptors varied between models, the regenerative transcriptional programs converged on the up-regulation of HA-related genes and the down-regulation of fibrillar collagen and cross-linking genes that typify fibrogenesis. Thus, different organs may synthesize and remodel the ECM through tissue-specific mechanisms; however, an HA and collagen dichotomy is a shared feature of regenerative versus fibrotic healing.

Our understanding of fibrotic wound healing has grown rapidly in recent years (2). However, factors that orchestrate complex tissue regeneration in vertebrates remain elusive. Part of the challenge lies in the few instances of accessible, robust tissue regeneration models, particularly in adult mammals. By investigating wound healing after adult mouse digit amputations, we examined the relationship between the tissue’s physical environment and regenerative versus fibrotic outcomes. We identified the ECM and resultant tissue mechanics as key properties distinguishing nonregenerative from regenerative responses. Notably, we found large amounts of HA in the blastema, which was necessary for regeneration; its absence triggered a switch toward fibrotic ECM and tissue mechanics in the amputated digit. Substrate mechanics, in turn, impacted cellular responses to injury signals, which we propose reinforce wound healing trajectories through feedback loops. Our experiments also demonstrated partial digit restoration after nonregenerative amputations by targeting the ECM, supporting these conclusions.

Fibroblasts are the principal architects of the ECM and wound healing (20). For example, a distinct fibroblast arises in infarcted hearts, and blocking their activation by immune cells reduces scar formation (39). Similarly, mechanical activation of Engrailed-1 in specific fibroblasts drives scar formation in skin wounds, and inhibiting this activation promotes regeneration (28, 40). Our study builds on this body of work by examining both the cells and physical niche that distinguish nonregenerative wounds from the regenerative blastema. In the former condition, two distinct Pdgfrα-expressing fibroblasts predominated, promoting a stiff collagen matrix through the production of collagen-modeling factors. By contrast, OL cells contributed to the blastema’s soft, more fluid extracellular milieu by synthesizing HA and HA-associated components, such as ACAN and HAPLN1. These findings highlight the divergence of both cell and extracellular factors during regenerative and nonregenerative processes as well as the strong interdependence between the two. Moreover, these observations underscore the importance of considering the role of the physical environment as well as fibroblast diversity in determining wound healing outcomes.

By targeting HA through perturbation and rescue experiments, our study provides evidence that HA is essential for regeneration. We posit that HA facilitates wound repair by modulating collagen fibrillogenesis and tissue mechanics. Although this phenomenon has been documented in mammalian fetal wounds (14–16, 41) and in other species (25), our findings in adult mice show that HA-collagen interdependency is conserved in injuries across developmental stages and may extend to other organ systems. However, the exact mechanism by which HA mediates collagen assembly remains to be elucidated. One possibility is that bulky pericellular HA matrices regulate integrin accessibility to collagen (42). Although not directly examined in our study, integrin-mediated binding to collagen is a major way by which cells sense and transmit mechanical forces (43) and regulate the presentation of soluble ligands (44); together, these mechanisms govern collagen remodeling. Simultaneously, HA matrices may physically interfere with the self-assembly and organization of collagen polymers (45), possibly through steric hindrance or by limiting diffusion of procollagen.

Lastly, we showed that the ECM and substrate mechanics regulate both fibrosis and regeneration. Prior studies have also related extrinsic physical forces to cellular rejuvenation and regeneration. For example, brain stiffening with age impairs oligodendrocyte precursor cell function (46). Inhibiting the mechanosensitive ion channel Piezo1 mitigated age-related changes in these cells, enhancing their regenerative capacity after demyelinating injuries (46). Similarly, soft substrates enhanced regeneration-associated machinery. To modify the nonregenerative wound’s extracellular environment to resemble that of the blastema, we used HAPLN1 to promote HA deposition, reduce scarring, and enhance skeletal repair. Given that link proteins stabilize aggregates of HA and proteoglycans (30, 47), we propose that HAPLN1 protects HA against fragmentation, which is commonly associated with inflammation and fibrosis (48, 49). Thus, higher-order structural organization of HA matrix is an important player in tissue mechanics and biological function. Previous studies have emphasized biochemical cues or nutritional supplements (50–52) in tissue regeneration. Our data show that modulating the physical microenvironment through HA stability and tissue mechanics can also improve regenerative outcomes. Thus, combining physical and biochemical interventions may be an effective approach to promote tissue regeneration in mammals.

We thank the members of the Storer lab and K. To and L. Fei for comments and suggestions; the CRUK genomics core facility; D. Clements in the imaging core facilities at CSCI; J. Beck and R. McGinn for help with AFM analyses; and O. Ogundele and A. Raffaelli for helpful experimental and technical advice.