Multifaceted Collagen-DDR1 Signaling in Cancer

Abstract

In addition to immune cells and fibroblasts, the tumor microenvironment consists of extracellular matrix, which contains collagens, whose architecture and remodeling dictate cancer development and progression. Collagen receptors expressed by cancer cells sense signals generated by microenvironmental alterations in collagen state to regulate cell behavior and metabolism. Discoidin domain receptor (DDR)1 is a key sensor of collagen fiber state and composition that controls tumor cell metabolism and growth, response to therapy and patient survival. This review focuses on DDR1 to NRF2 signaling, its modulation of autophagy and macropinocytosis and role in cancer and other diseases. Elucidating the regulation DDR1 activity and expression under different pathophysiological conditions will facilitate the discovery of new therapeutics and treatments.

Collagen receptors: ECM sensing and intercellular communication

Cancer research has gone from being cancer cell centric to a more inclusive and holistic view in which the cancer is studied together with the tumor microenvironment (TME), which also contains immune cells, endothelial cells, fibroblasts and the extracellular matrix (ECM) they produce [1,2]. Collagens (COLs) are the most abundant ECM proteins, and they play important structural roles that affect tumor rigidity and TME composition. New results show that COLs, which are classified as fibrillar or lamellar, also have critical regulatory functions that affect tumor growth and metabolism through interactions with different COL receptors expressed on cancer cells and other TME constituents [3,4]. These regulatory functions have raised interest in the therapeutic targeting of COL receptors [3,5], which include integrins, the discoidin domain receptors DDR1 and 2, OSCAR, GPVI, G6b-B, leukocyte-associated immunoglobulin-like receptor-1 (LAIR-1), and the mannose family receptor uPARAP/Endo180 (MRC2). Amongst these molecules, DDR1 has received much recent attention due to its ability to control cancer cell metabolism in response to COL remodeling and ability to modulate ECM architecture and TME composition [4,6,7]. This review is focused on current progress in understanding COL-DDR1 interactions and downstream signaling.

The DDR family of COL-activated tyrosine kinases

The first DDR family members were the Dictyostelium discoideum lectin-like proteins or agglutinins discoidin I and II [8]. In addition to discoidins, DDR1 and 2 bear homology to transmembrane receptor tyrosine kinases (RTKs), a feature that facilitated their molecular cloning [9,10]. Whereas DDR1 has five isoforms (DDR1a-e) that are generated through alternative splicing of its kinase domain, DDR2 has only a single isoform [11–14] (Figure 1). DDR1a-c are kinase-active, whereas DDR1d-e are inactive kinases. DDR1 is mainly expressed on epithelial cells of different tissues, while DDR2 is found in mesenchymal cells including fibroblasts, myofibroblasts, smooth muscle cells, and chondrocytes [15]. Unlike typical RTKs, DDR1 and DDR2 are activated by microenvironmental COLs or their fragments rather than much less abundant growth factors. Using a series of overlapping triple-helical peptides, DDR1 and DDR2 were found to bind the GVMGFO (O=hydroxyproline) motif which is present only in fibrillar COLs, such as COL I-III [16,17]. Curiously, however, DDR1 can also be activated by non-fibrillar COLs, such as COL IV. Replacement of DDR2 amino acids with the corresponding DDR1 sequence at the periphery of the GVMGFO peptide-binding interface enables COL IV binding to DDR2 [17].

Figure 1. Schematic representation of the five DDR1 isoforms and the single DDR2 form.

DDR1a, 1b, 1c, and DDR2 are enzymatically active tyrosine kinase receptors. DDR1d and 1e are enzymatically inactive and their function is unknown. DS, discoidin domain; DS-like, discoidin-like domain; EJM, extracellular juxtamembrane region; TM, transmembrane segment; CJM, cytosolic juxtamembrane region; KD, kinase domain. AA, Amino Acid.

DDR1 and DDR2 share high degree of sequence conservation. Both are single-span transmembrane proteins, with 4 different domains: an extracellular region containing a N-terminal discoidin homology (DS) domain and a DS-like domain, a juxtamembrane domain (JM) that includes an extracellular JM domain and a cytosolic JM domain, and a C-terminal tyrosine kinase domain (Figure 1). Two surface-exposed loops (S52-T57 and R105-K112) and one amino acid (S175) within the DS domain are responsible for COL binding [18]. On binding of triple-helical COLs DDR1 undergoes trans-autophosphorylation between adjacent dimers, after it has formed densely packed clusters [19–21]. These findings also reveal that DDR1 is activated in a slow manner of hours rather than seconds as for other RTKs. How DDR1 kinase activity is regulated in vivo is an outstanding question. A new study performed in pancreatic ductal adenocarcinoma (PDAC) shows that DDR1 activity is regulated by the cleavage state of COL I. Whereas COL I that has been cleaved by matrix metalloproteinases (MMPs) promotes DDR1 autophosphorylation, full length COL I induces DDR1 proteasomal degradation [4]. This is consistent with previous studies showing that COL peptides have higher binding affinity to DDR1 than full length COLs [17] and DDR1 protein level is downregulated by COL fibrils [21]. Whether MMP-generated COL I fragments act similarly to small COL peptides and how full-length COL I promotes DDR1 degradation remains to be determined. The ability to differentially respond to full length COLs vs their MMP cleavage products sets DDR1 apart from other COL receptors and places it in a central position to control cancer metabolism in response to microenvironmental cues.

DDR1 and macroautophagy-macropinocytosis crosstalk

Macroautophagy (autophagy) is a conserved catabolic process in which denatured and aggregated proteins and organelles are encased within double membraned vesicles and are delivered to lysosomes for degradation [22–24]. Autophagy occurs via a series of well-defined steps: initiation, phagophore nucleation, elongation, autophagosome formation and autophagosome-lysosome fusion (Figure 2). Autophagy can be triggered by energy stress (nutrient deprivation) and AMPK activation or mTOR inhibition [22], playing important roles in health and disease. In cancer, autophagy is Janus-faced, being either tumor promoting or tumor inhibitory. A connection between autophagy and COL signaling is illustrated by glioblastoma stem cells whose homeostatic maintenance is regulated by COL I [25]. Moreover, DDR1 is the only COL receptor with kinase activity in glioblastoma and its inhibition induces therapy sensitizing autophagy [26]. 14–3-3 connects DDR1 to AKT which prevents Beclin-1 from initiating autophagy. Disruption of the DDR1:14–3-3:AKT:Beclin-1 complex impairs AKT and mTOR signaling and promotes formation of the Beclin-1:VPS34:ATG14 autophagy core complex (PI3K-III complex) (Figure 2 and Figure 3). DDR1 inhibition also alleviates osteoarthritis by blocking mTOR inhibition-induced autophagy.

Figure 2. A schematic overview of the (macro)autophagy pathway.

Under starvation conditions, autophagy is initiated by AMPK activation or mTOR inhibition, allowing ULK1 complex formation, and activation of the class III PtdIns3K (PI3K-III) complex which creates the PtdIns3P (PI3P)-rich regions (omegasome) on the surface of the ER membrane. The concerted action of ubiquitin-like conjugation system converts LC3-I to LC3-II, which leads to phagophore expansion or conversion of the isolation membrane into an autophagosome. Autophagosome and lysosome fusion is mediated by the SNARE complex result in formation autolysosomes within which the cargo is degraded [75]. The ULK1 complex, contains ULK1, ATG13, RB1CC1/FIP200 and ATG101; PI3K-III complex consists of ATG14, Beclin-1, PIK3R4/VPS15 and PIK3C3/VPS34; and the SNARE complex contains STX17, SNAP29, and VAMP8. Abbreviations: AMPK, AMP activated protein kinase; mTOR, mechanistic target of rapamycin kinase; ULK1, unc-51 like autophagy activating kinase 1; PtdIns3K, phosphatidylinositol 3-kinase; PI3P, phosphatidylinositol-3-phosphate; ER, endoplasmic reticulum; MAP1LC3/LC3, microtubule associated protein 1 light chain 3; RB1CC1/FIP200, retinoblastoma 1-inducible coiled-coil 1; PIK3R4/VPS15, phosphoinositide-3-kinase regulatory subunit 4; PIK3C3/VPS34, phosphatidylinositol 3-kinase catalytic subunit type 3; SNARE, soluble N-ethylmaleimide-sensitive factor (NSF) attachment protein receptor; STX17, syntaxin 17; SNAP29, soluble NSF attachment protein 29; VAMP8, vesicle-associated membrane protein 8; PE, phosphoethanolamine; ATG, autophagy-related protein.

Figure 3. Autophagy and macropinocytosis (MP) crosstalk.

On cCOL I binding, 14–3-3 connects activated DDR1 to AKT to prevent Beclin-1 from forming the Beclin-1:VPS34:VPS15:ATG14 autophagy core complex which promotes omegasome formation. Meanwhile, activated DDR1 promotes MP through the NF-κB-NRF2 module that induces transcription of MP-related genes. Loss of autophagy causes p62 accumulation which sequesters KEAP1 to activate NRF2-promoted MP.

In contrast to autophagy, macropinocytosis (MP) is an actin-dependent endocytic nutrient acquisition pathway in which soluble extracellular proteins are ingested and delivered to the lysosome in single membrane endocytic vesicles or macropinosomes [27]. Autophagy disruption causes accumulation of the autophagy chaperon SQSTM1/p62, which binds KEAP1 to induce transcription factor NRF2 activation [28], recently found to be the master activator of genes encoding MP-related proteins as well as MP activating growth factors and receptors [29]. DDR1 engagement by MMP-generated COL I fragments triggers NF-κB activation, SQSTM1/p62 induction and stimulation of NRF2-promoted MP [4], while suppressing autophagy through DDR1-Beclin-1 interaction [26]. Autophagy is normally inhibited when nutrients are high [22], but whether MP-produced ATP and amino acids inhibit autophagy needs to be determined. These results place DDR1 and NRF2 as key regulators of autophagy-MP crosstalk in those cancer cells that are surrounded by COL-rich stroma (Figure 3).

COL-DDR1 signaling in different cancers

PDAC

COL I-DDR1 signaling plays a critical role in PDAC metabolism and growth (Figure 4). DDR1 inhibition suppresses tumor growth and metastasis and improves the response to chemotherapy [4,30–32]. On activation, DDR1 directly interacts with focal adhesion kinase (FAK)-related protein tyrosine kinase 2 (PYK2), whose activation promotes cancer cell proliferation and metastasis through a p130 Crk-associated substrate scaffold (CAS)-RAP1-JNK1-c-Jun cascade [33]. DDR1-PYK2 activation also enhances expression of major ECM components [30]. Network-forming COL VIII promotes PDAC progression through DDR1 and integrin signaling, thereby activating pro-survival pathways, including PI3K-AKT and FAK-NF-κB [32]. MMP-mediated COL cleavage has been known to regulate tumor development [34] and a recent study showed that MMP-cleaved COL I (cCOL I) and intact COL I (iCOL I) differentially regulate PDAC bioenergetics, growth and metastatic spread via their effects on DDR1 activity and expression [4]. Whereas cCOL I engages DDR1 to activate an NF-κB-p62-NRF2-TFAM signaling cascade that stimulates MP, mitochondrial biogenesis, bioenergetics, and tumor growth, iCOL I triggers DDR1 proteasomal degradation, having opposite effects on these parameters. Patients with PDAC whose tumors are enriched for iCOL I and therefore are low DDR1 and NRF2 expressors exhibit greatly improved median survival compared to patients whose TME is high in MMP expression and enriched in cCOL I and their malignant cells are high in DDR1 and NRF2. These results justify development of therapeutics that target the DDR1-NF-κB-NRF2-TFAM metabolism simulating pathway. It’s also important to understand how DDR1, a rather weak RTK, exerts such profound effects on PDAC metabolism, although the answer may be somewhat trivial and related to the very high concentration of cCOL I in the TME, far surpassing the concentration of any growth factor by many orders of magnitude. cCOL I amounts in the PDAC TME correlate with high expression of inflammatory cytokines and MMPs [4]. cCOL I-DDR1 signaling may further augment TME inflammation by inducing cancer cell expression of CXCL5 via a PKCθ-SYK-NF-κB cascade, resulting in recruitment of tumor-associated neutrophils and formation of neutrophil extracellular traps [31]. In addition, tumor cell-derived α1(I) homotrimers resulting from epigenetic suppression of the COL1a2 gene promote PDAC development by activating DDR1 and α3β1 integrin on cancer cells and subsequent activation of FAK-PI3K-AKT signaling [35] (Figure 4). This was found to be associated with a microbiota enriched in anaerobic Bacteroidales in hypoxic and immunosuppressed tumors, and ablation of integrin β1 rather than DDR1 inhibits the autocrine response to COL I homotrimers. An earlier study showed that COL I homotrimers from lathyritic chick embryo tendons and calvaria have a greater solubility than COL I heterotrimers [36], providing a plausible explanation how the small amount of cancer cell derived COL I homotrimers may still lead to significant DDR1 activation [4].

Figure 4. COL-DDR1 signalings in cancer.

PDAC: (i) COL I activates DDR1-PYK2 and α2β1 integrin-FAK to stimulate cell proliferation and metastasis. Both receptor complexes rely on the p130 CAS-RAP1-JNK1-c-Jun cascade. (ii) COL VIII promotes PDAC progression through the DDR1-PI3K-AKT and integrin-FAK-NF-κB cascades. (iii) cCOL I engages DDR1 to activate an NF-κB-p62-NRF2-TFAM module that stimulates MP, mitochondrial biogenesis, bioenergetics, inflammation and tumor growth, iCOL I induces the opposite effects by causing the degradation of DDR1. (iv) cCOL I-DDR1 signaling modulates the TME by inducing CXCL5 expression via a PKCθ-SYK-NF-κB signaling cascade. (v) tumor cell-derived α1(I) homotrimers activate DDR1 and α3β1 integrin and inhibit tumor immunity via the FAK-PI3K-AKT pathway. TNBC: In a mouse model of TNBC, COL induced shedding of the DDR1 ectodomain (ECD) leads to alignment of wavy COL fibers resulting in prevention of T cell infiltration. However, in a human TNBC model (MDA-MB-231), COL III activated DDR1 maintains production of wavy COL III through activation of STAT1 to restrict cancer cell proliferation. In luminal BCa model, DDR1 inhibition promotes tumor growth through an unknown mechanism. HNSC: DDR1 promotes tumor growth in response to COL VIII and COL XI. Tumor-derived wavy COL III sustains tumor dormancy through DDR1-STAT1 signaling. HCC: cCOL I-activated DDR1 promotes tumor survival through the PI3K-AKT pathway and PSD4-ARF6-p38-ERK-c-Jun cascade. COL XV downregulates DDR1 expression through an unknown mechanism.

Breast cancer (BCa)

COL-DDR1 signaling supports BCa development by preventing effector T cell infiltration (Figure 4). BCa is the second most common cancer worldwide and is divided into several subtypes: luminal A, triple-negative BCa (TNBC), named after its lack of estrogen receptor alpha (ERα), progesterone receptor (PR) and epidermal growth factor receptor 2 (HER2), and HER2-positive BCa [37]. DDR1 is highly expressed in all subtypes compared to normal breast epithelium and DDR1 mutations are associated with poor prognosis in hormone receptor-positive patients [38]. The DDR1 kinase domain mutation R776W is supposed to be homologous to the EGFR hotspot mutation L858R [39]. In TNBC, DDR1 seems to promote tumor growth by inducing immune exclusion after cleavage of its COL binding ectodomain, which binds and aligns COL fibers [7]. However, how the DDR1 ectodomain aligns wavy COL fibers and how the aligned fibers prevent T cell invasion is unknown. Conversely, another study found that in an epithelial, human TNBC-derived MDA-MB-231 cell graft, DDR1 maintains a wavy form of tumor-derived COL III to restrict cancer cell proliferation through STAT1 activation [40] and that deletion of DDR1 in the MMTV-PyMT luminal BCa model converts the tumors to a more aggressive basal-type phenotype, resulting in increased lung metastasis through upregulation of fibrosis and hypoxia [41]. Although much more remains to be learned about DDR1 in BCa, its function seems to subtype and species specific.

Head and neck cancer (HNSCC)

DDR1 in HNSCC was also described as either stimulatory or inhibitory (Figure 4). Like BCa, DDR1 is overexpressed in HNSCC relative to normal tissue and shown to promote cancer cell growth and migration in response to COL VIII and XI [42]. However, tumor-derived COL III sustains tumor dormancy through DDR1-mediated STAT1 signaling, which tends to occur in lymph node-negative HNSCC that contain more COL III than lymph node-positive HNSCC [40]. The mechanism accounting for the differential signaling activities of COL VIII, XI and III is totally unknown, as DDR1 is still activated by other fibrillar COLs in the absence of COL I in PDAC [4].

Liver cancer

DDR1 is upregulated in mouse models of fibrosis-associated hepatocellular carcinoma (HCC) and corresponding patient specimens and high DDR1 is associated with poor prognosis [43,44] (Figure 4). As in PDAC, ECM containing cCOL I, but not iCOL I, activates DDR1 and downstream tumor-promoting effectors such as AKT [44]. DDR1 also interacts with ARF6 and activates it by recruiting PSD4, thereby stimulating MAPK signaling and metastasis [43]. Another interesting finding is that overexpression of COL XV, a basement membrane molecule that also interacts with DDR1, downregulates DDR1 mRNA and protein through an unknown mechanism [45–47]. Whether COL XV is a natural inhibitor of DDR1 that acts like iCOL I is an important question whose answer will provide new insights to DDR1 signaling. In contrast to its significant role in promoting HCC growth, DDR1 seems inessential in intrahepatic cholangiocarcinoma (ICC) that has been induced by hydrodynamic transduction of oncogenic drivers into hepatocytes [48]. Nonetheless, such results do not completely rule out DDR1’s role in ICC owing to the limitations of hydrodynamic transduction which mainly affects mature hepatocytes than hepatic stem cells or biliary epithelial cells which are also thought to give rise to ICC and can’t generate the desmoplastic microenvironment characteristic of human ICC [49,50].

COL-DDR signaling in neurodegenerative diseases

Although this review is focused on cancer, it should be mentioned that the ECM is also important for maintaining homeostasis in the central nervous system (CNS) and neuronal survival [51]. Structural, organizational, and functional ECM alterations due to genetic variation or environmental stress can result in neurodegenerative disease. As in cancer, different ECM components have opposing roles in neuropathology [52]. DDR1 and 2 are upregulated in Alzheimer’s and Parkinson’s diseases and their inhibition reduces α-synuclein, tau, and β-amyloid deposition and prevents cell loss, suggesting DDR1/2 inhibition as a potential target for clearance of neurotoxic proteins [53–55]. Differentially methylated regions (DMR) mapping in Parkinson’s patients and matched controls shows that eight CpG sites are hypomethylated in the DDR1 locus, suggesting that DDR1 as a molecular biomarker and a therapeutic target [56]. A tyrosine kinase inhibitor, nilotinib that potently inhibits DDR1, was found to induce autophagy and reduce inflammation and neurotoxicity [57], challenging the strategy of using COLs for induction of brain repair [58]. Another study reported that the cellular spread of alpha-synuclein fibrils depends on macropinocytic internalization and lysosomal degradation [59], implying an important role of autophagy-MP crosstalk in neurodegenerative diseases. Whether DDR1 promotes neurodegenerative disease through upregulation of MP and inhibition of autophagy and how DDR1 controls MP and autophagy in neurodegenerative diseases need to be further investigated. Anyhow, DDR1 is also an intriguing therapeutic target in neurodegenerative disease.

Concluding remarks

DDR1 is also tumor promoting in lung adenocarcinoma, retinoblastoma and colon carcinoma [60–62]. While the tumor promoting role of DDR1 can be easily explained through its effects on mitochondrial biogenesis, MP, and bioenergetics, how can a protein kinase that leads to NF-κB and NRF2 activation be tumor inhibitory in some cancers is not clear. This is certainly one of the major open questions regarding COL-DDR1 signaling (see Outstanding questions). A possible explanation could be divergent modes of COL remodeling and DDR1 gene mutations that alter substrate preference in different tumor contexts [63,64]. The targeting of DDR1 or DDR1 downstream effectors, including NF-κB, NRF2, mitochondrial biogenesis or MP may provide new opportunities for therapeutic development in those cancers whose metabolism and growth are DDR1 dependent (Table 1). DDR2 is also a promising target in PDAC [4,31,65], BCa [7,40,66–68], HNSCC [40,42,67], HCC [43,44,70,71] and ICC [48,72] (Table 1). Thus, dual DDR1 and DDR2 inhibitors may be quite effective. Indeed, several new DDR inhibitors found via high throughput screening for ATP competitive inhibitors or designed via in silico strategies have shown promising effects. KI-301690, a novel DDR1 inhibitor, potentiates the anticancer activity of gemcitabine in PDAC [30]. Another potent DDR1/2 inhibitor, VU6015929, with low cytotoxicity blocks COL-induced DDR1 activation and COL IV production, supporting the utility of DDR1/2 inhibition [73]. However, new and more selective and potent DDR inhibitors or combinations are needed to counteract drug resistance due to DDR1 and DDR2 mutations [64,74]. Inhibition of MP using tool compounds or blockade of mitochondrial protein synthesis with the FDA approved antibiotic tigecycline were also found to induce tumor regression in mouse PDAC models [4]. Whether and when any of these molecules or their derivatives will make it to the clinic is an open question.

Outstanding questions:

1. How does iCOL I induce DDR1 degradation?

2. How do non-fibrillar COLs, such as IV and XV bind DDR1 and how do they differentially control its activity or expression?

3. DDR1 regulates autophagy and MP through mTOR and NRF2, respectively. Can DDR1 directly phosphorylate autophagy- or MP-related proteins?

4. How does DDR1 activation in epithelial cancer cells affect tumor inflammation and immunity?

5. How can COL III, which is very similar to COL I, activate DDR1 and still have a diametrically opposing effect to COL I on tumor growth?

6. Can DDR2 substitute for DDR1? If yes, in what sense?

7. Which genes are specifically regulated by DDR1 or DDR2 and which genes can be regulated by both tyrosine kinases?

8. Are DDR1 and/or DDR2 involved in inside-out signaling and/or transactivation processes?

Table 1.

Overview of the functions of DDR1 and DDR2 in the development of distinct cancers.

Cancer types	DDR1 function	DDR2 function
PDAC	Proliferation, metastasis and poor survival [4, 31]	Proliferation and migration activated by COL X [63]
BCa	Immune exclusion via promoting COL fiber alignment, or tumor dormancy though maintaining a wavy form of tumor-derived COL III [7, 40]	Proliferation, migration, invasion and immune exclusion [64–66]
HNSCC	Cell growth and migration in response to COLs III and XI or tumor dormancy in response to COL III [40, 42]	Migration, invasion [67]
HCC	Proliferation, metastasis and poor survival [43, 44]	EMT, metastasis, sorafenib resistance [68, 69]
ICC	Not essential in ICC growth [48]	Chinese patients had at least 1 actionable genetic aberration, with a significantly higher frequency in DDR2 compared with US patients [70]

Highlights.

• DDR1 is an important signal transducer that mediates crosstalk between cancer cells and their surrounding stroma.

• Due to the very high concentrations of its ligand, cleaved collagen, DDR1 exerts a propound effect on tumor metabolism.

• DDR1 controls macropinocytosis-autophagy crosstalk by activating NRF2, a key transcriptional regulator of metabolic adaptation.

• Distinct collagen types and their cleavage products exert differential effects on DDR1 activity and expression.

• DDR1 is an interesting therapeutic target for those cancers whose metabolism and growth depend on its activation.