Collagen misfolding mutations: The contribution of the unfolded protein response to the molecular pathology

Abstract

Mutations in collagen genes cause a broad range of connective tissue pathologies. Structural mutations that impact procollagen assembly or triple helix formation and stability are a common and important mutation class. How misfolded procollagens engage with the cellular proteostasis machinery and whether they can elicit a cytotoxic unfolded protein response (UPR) is a topic of considerable research interest. This is well justified since modulating the UPR could offer a new approach to treat collagenopathies for which there are no current disease mechanism-targeting therapies. This review will scrutinize the evidence underpinning the view that endoplasmic reticulum stress and chronic UPR activation contributes significantly to the pathophysiology of the collagenopathies. While there is strong evidence that the UPR contributes to the pathology for collagen X misfolding mutations, the evidence that misfolding mutations in other collagen types induce a canonical, cytotoxic UPR is incomplete. To gain a more comprehensive understanding about how the UPR contributes to pathology, and thus what types of manipulations of the UPR might have therapeutic relevance, much more information is needed about how specific misfolding mutation types engage differentially with the UPR and downstream signaling responses. Most importantly, since the capacity of the proteostasis machinery to respond to collagen misfolding is likely to vary between cell types, reflecting their functional roles in collagen and extracellular matrix biosynthesis, detailed studies on the UPR should focus as much as possible on the actual target cells involved in the collagen pathologies.

Collagens form the structural backbones of the extracellular matrices (ECMs)^* that surround, support, and connect cells. Forty-four human collagen genes encode procollagen pro-α chains that assemble as either homotrimers or heterotrimers into the 28 recognized collagen protein types. These collagen types have tissue-specific expression patterns that are critical in generating the nuanced structural and functional characteristics of different ECMs during development, tissue homeostasis and repair. As a result, mutations in collagen genes cause a broad range of inherited pathologies. To date, mutations in twenty-seven collagen genes have been identified (Table 1). Many mutations, particularly in the more common fibril-forming collagens (collagens I, II and III) found in bone, cartilage, skin, and vasculature; the network forming hypertrophic cartilage collagen X; the microfibrillar collagen VI and the basement membrane collagen IV have contributed crucially to our understanding of collagen structure, intracellular proα-chain assembly, extracellular processing and fibrillogenesis. Our knowledge about collagen structure, biosynthesis and fibril formation is detailed and, as it has been described in many thorough reviews, will not be discussed here^1–4. Likewise, how collagen mutations affect these processes has been well-reviewed^1,5 and will not be discussed in detail in this article. This review will focus on what we know about the cellular molecular pathology of the collagenopathies caused by structural mutations in the collagen protein proα-chains. These may be missense mutations that introduce amino acid changes, or larger structural changes such as exon-skipping mutations in the large multi-exon collagen genes. These can impair the correct folding of the carboxyl-terminal trimerization domains, or more commonly, disrupt the triple helix domain. How these misfolded collagens engage the cellular protein quality control machinery is an increasing focus of research on collagenopathy mechanisms (Fig 1) and on therapeutic targets. In addition, while premature termination mutations most commonly lead to nonsense-mediated mRNA decay and collagen haploinsufficiency, unless the mutant mRNA breakdown is complete, some truncated misfolded collagens will be synthesized that could also activate the UPR (Fig 1). This review will particularly scrutinize the evidence underpinning the view that endoplasmic reticulum (ER) stress and chronic activation of the unfolded protein response (UPR) is a significant contributor to the pathophysiology of the collagenopathies.

Table 1.

Collagenopathies

Collagen Gene	Disease	OMIM
COL1A1	OSTEOGENESIS IMPERFECTA, TYPES I, II, III, IV; CAFFEY DISEASE; EHLERS-DANLOS SYNDROME, ARTHROCHALASIA TYPE 1	166200, 166210, 259420, 166220, 114000, 130060
COL1A2	OSTEOGENESIS IMPERFECTA, TYPES I, II, III, IV; EHLERS-DANLOS SYNDROME, ARTHROCHALASIA TYPE 2; EHLERS-DANLOS SYNDROME, CARDIAC VALVULAR TYPE	166200, 166210, 259420, 617821, 225320
COL2A1	STICKLER SYNDROME, TYPE I; STICKLER SYNDROME, TYPE I, NONSYNDROMIC OCULAR; SPONDYLOEPIPHYSEAL DYSPLASIA CONGENITA; KNIEST DYSPLASIA; AVASCULAR NECROSIS OF FEMORAL HEAD 1; ACHONDROGENESIS, TYPE II; CZECH DYSPLASIA. OSTEOARTHRITIS WITH MILD CHONDRODYSPLASIA; PLATYSPONDYLIC LETHAL SKELETAL DYSPLASIA, TORRANCE TYPE; SPONDYLOEPIPHYSEAL DYSPLASIA, STANESCU TYPE; SPONDYLOEPIMETAPHYSEAL DYSPLASIA, STRUDWICK TYPE; SPONDYLOPERIPHERAL DYSPLASIA; MULTIPLE EPIPHYSEAL DYSPLASIA WITH MYOPIA AND CONDUCTIVE DEAFNESS; LEGG-CALVE-PERTHES DISEASE	108300, 609508, 183900, 156550, 608805, 200610, 609162, 604864, 151210, 616583, 184250, 271700, 132450, 150600
COL3A1	EHLERS-DANLOS SYNDROME, VASCULAR TYPE	130050
COL4A1	BRAIN SMALL VESSEL DISEASE WITH OR WITHOUT OCULAR ANOMALIES; ANGIOPATHY, HEREDITARY, WITH NEPHROPATHY, ANEURYSMS, AND MUSCLE CRAMPS; PORENCEPHALY 1; HEMORRHAGE, INTRACEREBRAL (SUSCEPTIBILITY); SCHIZENCEPHALY	607595, 175780, 614519, 269160
COL4A2	PORENCEPHALY 2; HEMORRHAGE, INTRACEREBRAL (SUSCEPTIBILITY)	614483, 614519
COL4A3	ALPORT SYNDROME, AUTOSOMAL RECESSIVE AND AUTOSOMAL DOMINANT, HEMATURIA, BENIGN FAMILIAL	203780, 104200, 141200
COL4A4	ALPORT SYNDROME, AUTOSOMAL RECESSIVE AND AUTOSOMAL DOMINANT	203780, 104200
COL4A5	ALPORT SYNDROME, X-LINKED	301050
COL4A6	LEIOMYOMATOSIS, DIFFUSE, WITH ALPORT SYNDROME; X-LINKED DEAFNESS 6	308940, 300914
COL5A1	EHLERS-DANLOS SYNDROME, CLASSIC TYPE 1	130000
COL5A2	EHLERS-DANLOS SYNDROME, CLASSIC TYPE 2	130010
COL6A1	ULLRICH CONGENITAL MUSCULAR DYSTROPHY 1; BETHLEM MYOPATHY 1	254090, 158810
COL6A2	ULLRICH CONGENITAL MUSCULAR DYSTROPHY 1; BETHLEM MYOPATHY 1; MYOSCLEROSIS, AUTOSOMAL RECESSIVE	254090, 158810, 255600
COL6A3	ULLRICH CONGENITAL MUSCULAR DYSTROPHY 1; BETHLEM MYOPATHY 1; MYOSCLEROSIS, AUTOSOMAL RECESSIVE; DYSTONIA 27	254090, 158810, 255600, 616411
COL7A1	EPIDERMOLYSIS BULLOSA DYSTROPHICA, AUTOSOMAL RECESSIVE AND AUTOSOMAL DOMINANT; NONSYNDROMIC CONGENITAL NAIL DISORDER 8	226600, 131750, 607523
COL9A1	STICKLER SYNDROME, TYPE IV; MULTIPLE EPIPHYSEAL DYSPLASIA 6	614134, 614135
COL9A2	MULTIPLE EPIPHYSEAL DYSPLASIA 2; STICKLER SYNDROME, TYPE V	600204, 614284
COL9A3	MULTIPLE EPIPHYSEAL DYSPLASIA 3	600969
COL10A1	METAPHYSEAL CHONDRODYSPLASIA, SCHMID TYPE	156500
COL11A1	STICKLER SYNDROME, TYPE II; MARSHALL SYNDROME; FIBROCHONDROGENESIS 1	604841, 154780, 228520
COL11A2	OTOSPONDYLOMEGAEPIPHYSEAL DYSPLASIA, AUTOSOMAL DOMINANT AND AUTOSOMAL RECESSIVE; AUTOSOMAL RECESSIVE DEAFNESS 53; FIBROCHONDROGENESIS 2	184840, 215150, 609706, 614524
COL12A1	BETHLEM MYOPATHY 2; ULLRICH CONGENITAL MUSCULAR DYSTROPHY 2	616471, 616470
COL13A1	CONGENITAL MYASTHENIC SYNDROME 19	616720
COL15A1	KNOBLOCH SYNDROME 1	267750
COL17A1	EPIDERMOLYSIS BULLOSA, JUNCTIONAL, NON-HERLITZ TYPE; EPITHELIAL RECURRENT EROSION DYSTROPHY	226650, 122400
COL18A1	KNOBLOCH SYNDROME 1	267750
COL25A1	CONGENITAL FIBROSIS OF EXTRAOCULAR MUSCLES 5	616219
COL27A1	STEEL SYNDROME	615155

Figure 1. The pathophysiology of collagen mutations.

The possible fates and cellular impacts of the two most common collagen heterozygous disease-causing mutation categories, premature termination codon and missense mutations, are considered in this schematic. A, Premature termination mutations are commonly subject to nonsense-mediated mRNA decay (NMD) which can completely degrade the mutant allele mRNA. This results in collagen haploinsufficiency in the ECM. This collagen is structurally normal and haploinsufficiency most commonly causes milder disease phenotypes. If NMD is only partial some mutant truncated collagen chains may be produced, but they either cannot form trimers or form misfolded aggregates which may engage the endoplasmic reticulum (ER) proteostasis machinery. B, Missense mutations can induce structural changes that impair trimer assembly or helix formation and structure. Mutant proteins interact with the protein folding and quality control (QC) machinery (proteostasis) and potentially activate an unfolded protein response (UPR). The UPR mechanisms and possible downstream consequences are discussed in more detail in the body of the article, but broadly involve an initial response to assist mutant protein folding or reduce the mutant misfolded protein load by activating ER-associated degradation (ERAD) or autophagy pathways. If the UPR is unresolved, it could become pathological with major impacts on cellular functions, including altered cellular differentiation and ultimately, apoptosis. However, some mutant misfolded collagens may escape these quality control strategies and after secretion into the extracellular matrix (ECM) it can impact collagen fibrillogenesis, stability and interactions with other ECM components important for the functional competence of the ECM. Since collagen disease-causing mutations are heterozygous, some collagen trimers will contain only the normal and correctly folded collagen chains. For heterotrimeric collagen I, 25% of the trimers contain only normal chains and for homotrimeric collagens such as collagen II and III the proportion of normal timers is 12.5%. This small proportion of normal collagen, while folded and secreted normally, is deposited within the depleted and structurally and functionally abnormal mutant-containing collagen ECM.

The Unfolded Protein Response

Within the ER, the proteostasis network coordinates protein synthesis, folding, trafficking and degradation, ensuring that only correctly folded functional proteins exit the ER and are delivered to their functional destinations⁶. Within this broad proteostasis framework, the unfolded protein response (UPR) is a conserved pathway that is activated in a coordinated attempt to reduce the load of misfolded proteins⁷ (Fig 2). Unfolded and incorrectly folded proteins unable to exit the ER bind the molecular chaperone BiP, which recognizes exposed hydrophobic protein sequences normally buried within correctly folded protein structures. The three canonical ER stress sensors, IRE1, ATF6, and PERK, are normally held in an inactive form when bound by BiP. The accumulation of misfolded proteins titrates BiP off the ER luminal domains of the stress sensors, activating all three arms of the UPR via this mechanism. In addition to BiP release, there may also be other mechanisms to activate the ER stress sensors. Once activated, each sensor sets off a distinctive signaling cascade. IRE1 dimerizes or oligomerizes and trans-autophosphorylates via its kinase domain, triggering a conformational rearrangement that activates a site-selective RNase domain. The activated RNase splices an unconventional intron from the mRNA for an otherwise inactive transcription factor, XBP1. Spliced XBP1 mRNA translation produces an active transcription factor, termed XBP1_s. The IRE1-XBP1_s arm of the UPR is the most ancient and conserved UPR module, and XBP1_s correspondingly targets a tremendous number of secretory pathway-related genes via UPR-responsive promoter elements, particularly stimulating the synthesis of protein trafficking, quality control, and folding machineries. Long-term IRE1 activation can lead to JNK activation and ultimately induce apoptosis. PERK is structurally similar to IRE1 but lacks the RNase domain. PERK dimerizes upon activation, activating its kinase domain and leading to trans-autophosphorylation and downstream phosphorylation of the translational initiation factor eIF2α. Phosphorylated eIF2α prevents formation of the translation initiation complex and globally downregulates protein translation, thus reducing the protein folding load. Paradoxically, whereas phosphorylated eIF2α inhibits translation of most mRNAs, it promotes translation of a subset of stress response genes, such as the transcription factor ATF4. ATF4 upregulates transcription of genes involved in amino acid metabolism and transport, oxidation-reduction reactions and the apoptosis-inducing transcription factor CHOP. Thus, long-term PERK pathway activation is cytotoxic. The third UPR arm is mediated by ATF6, a member of the bZIP transcription factor family. Upon activation, ATF6 transits to the Golgi. There, it is cleaved by S1P and S2P to generate an active cytosolic transcription factor, ATF6p50. ATF6p50 translocates to the nucleus, binds to ER-stress responsive promoter elements and upregulates transcription of ER protein folding chaperones.

Figure 2. The canonical unfolded protein response (UPR).

Unfolded and incorrectly folded proteins in the endoplasmic reticulum (ER) bind the molecular chaperone BiP. This titrates BiP off the three canonical ER stress sensors (IRE1, ATF6, and PERK), activating them and triggering downstream signaling. Activation of IRE1 results in RNase splicing of an intron from the mRNA for XBP1, producing an active transcription factor, termed XBP1s. XBP1s upregulates UPR target genes, particularly stimulating the synthesis of protein trafficking, quality control, and folding machineries. Activated ATF6 transits to the Golgi where it is cleaved by membrane-bound proteases, S1P and S2P, to generate an active cytosolic transcription factor, ATF6p50. ATF6p50 upregulates transcription of ER protein folding chaperones. Activated PERK phosphorylates the translational initiation factor eIF2α. Phosphorylated eIF2α (eIF2α-P) prevents the formation of the translation initiation complex and downregulates protein translation, thus reducing the protein folding load. While eIF2α-P inhibits translation of most mRNAs, it promotes translation of the transcription factor ATF4 which upregulates additional UPR target genes, including the apoptosis-inducing transcription factor CHOP. The initial UPR activation is cytoprotective, upregulating target genes which improve folding or stimulate degradation by the ER-associated degradation (ERAD) proteasomal or autophagy pathways. However chronic UPR activation that cannot be addressed by a transitory UPR is cytotoxic and directs the cell to apoptosis via PERK-mediated induction of the pro-apoptotic transcription factor CHOP and IRE1-mediated induction of the JNK pathway.

Initial UPR activation is cytoprotective. Chaperones and quality control factors, including many chaperones known to engage collagen^8,9, are upregulated to address protein folding deficits, and translation is lowered to reduce the protein-folding burden. Together, these mechanisms provide an opportunity for cells to return to proteostasis (Fig 2). Chronic UPR activation mediated by a long-term unfolded protein load that cannot be addressed by a transitory UPR is actively cytotoxic. The balance in activities of the three UPR arms shifts away from inducing pro-survival folding and quality control activities, to directing the cell to apoptosis via PERK-mediated induction of the pro-apoptotic transcription factor CHOP and IRE1-mediated induction of the JNK pathway.

There has been a recent surge of interest in the possibility that a chronic UPR could be pathogenic in the collagenopathies and other disorders caused by misfolding ECM proteins. Several reviews canvass the role of the UPR in ECMopathies^5,10. This interest is well-justified, since if a chronic, apoptosis-inducing UPR is pathogenic, inhibiting that pathway may offer new therapeutic targets^11,12 for this broad range of disorders for which there are no current therapies that target disease mechanisms. It is fair to say at this juncture that, while there is compelling evidence that the UPR contributes to pathology in several specific instances, the evidence that a canonical, cytotoxic UPR plays a broad role in the collagenopathies remains scant. In fact, it may actually be that in the case that misfolded collagens, which have unique features due to their fibrillar rather than prototypical globular structure, failing to induce the pro-survival aspects of the UPR is a more significant contributor to pathophysiology than is chronic UPR activation¹³. The remainder of this review will explore the detailed evidence for a chronic, cytotoxic UPR in the collagenopathies.

Collagen misfolding mutations

The most prevalent disease-causing mutations in the fibrillar collagens disrupt the predominant structural and functional unit, the triple helix. These are typically missense mutations that result in replacement of a glycine residue with another amino acid. For example, 80–85% of all osteogenesis imperfecta (OI)-causing mutations in COL1A1 are glycine substitutions. The early finding that triple-helical domain glycine mutations underly the most severe OI forms could be readily understood from what was known about collagen triple helix structure. Triple helix folding and stability is critically dependent on having a glycine at every third position in the triplet repeat amino acid sequence (Gly-X-Y) that makes up the triple-helical domain. Replacing glycine with a bulkier amino acid breaks an essential inter-strand hydrogen bond and thereby disrupts triple helix folding and compromises the structural integrity of the triple helix. Delayed helix folding can lead to increased post-translational lysine hydroxylation and glycosylation. Many early studies demonstrated that triple-helical domain glycine mutations led to reduced triple helix thermal stability and over-modified mutant α-chains that migrated slowly on SDS-polyacrylamide gels. Furthermore, disrupted triple helix formation and/or stability caused by glycine substitutions often results in retention of the mutant trimers in the ER. Numerous disease-causing, collagen triple-helix disrupting glycine substitutions have been identified for OI (COL1A1 and COL1A2), chondrodysplasias (COL2A1), Ehlers-Danlos Syndrome Type III (COL3A1), and Ehlers-Danlos Syndrome Type I (COL5A1 and COL5A2). The cellular consequences of triple-helical domain glycine substitution mutations in collagens III and V are less characterized, but it is highly likely that the primary impact is similar to collagen I and II; triple-helix destabilization and ER retention. While glycine substitutions are the predominant cause of triple helix disruption in the collagenopathies, less common helical mutations that substitute arginine for cysteine in COL1A1 and COL2A1 can also impact triple helix structure, stability, and collagen secretion. These result in less severe clinical phenotypes than glycine substitutions. A second, less common, but still important group of collagen mutations disrupt the structure of the C-terminal propeptide domain affecting initial chain recognition and trimerization. Correct propeptide assembly in the lumen of the ER is driven by Ca2+-binding and the presence or absence of specific cysteine residues, as well as by short discontinuous amino acid sequences in the three-dimensional structures of the individual propeptides^14,15. Mutations in the C-propeptides that perturb inter- and/or intra-chain disulfide bonding, or introduce other structurally deleterious changes, can prevent or delay assembly and mutant chain secretion.

Mutations in the network-forming basement membrane collagen IV genes, COL4A3, COL4A4 and COL4A5 cause Alport syndrome with characteristic kidney glomerular basement membrane defects as well as deafness and eye pathologies^16,17. COL4A1 and COL4A2 mutations also cause eye, kidney and muscle pathologies and, in particular, porencephaly, intracranial hemorrhages and related cerebrovascular disease^16,17. For all collagen IV types, the most common mutations are glycine substitutions within the triple helix. For several mutations, there is strong evidence that they result in increased collagen IV trimer retention within the ER.

COL6A1, COL6A2 and COL6A3 mutations cause the muscle disorders Bethlem myopathy and Ullrich Congenital Muscular Dystrophy. Collagen VI has a more complex intracellular assembly pathway than the other collagen types. Triple-helical monomers align into antiparallel dimers and then form tetramers. Tetramers are secreted into the ECM where they interact end-to-end to form characteristic beaded collagen VI microfibrils. Mutations in the collagen VI triple-helical domain have complex consequences that reflect the multistep intracellular assembly process^16,18,19. Glycine substitutions seem to have a larger impact on higher order assembly than individual triple helix structure per se. Pathogenic heterozygous glycine mutations (the most common) and other structural mutations, such as in-frame deletions, are clustered at the N-terminal end of the triple helix in a region critical for tetramer and microfibril formation. As a result, most mutations have a dominant negative effect, causing intracellular accumulation, reduced secretion and impaired microfibril assembly. The pathogenic severity of dominant or recessive collagen VI mutations in regions other than the triple helix depends on whether they prevent incorporation into monomers and are then targeted for proteasomal degradation, or are incorporated into monomers, dimers and/or tetramers that disrupt downstream microfibril formation and/or organization.

In addition to the impact of structural mutations on procollagen assembly, triple-helix folding, structure, and secretion, it is important to also consider that structurally abnormal collagen can have direct, deleterious effects on the ECM. The mutant collagen can alter fibril or network formation and stability in the ECM, and interactions between the collagen structures and other ECM components. The ways in which mutant collagens can disrupt ECM structure and function will not be discussed further in this article, since we focus on the intracellular impacts of collagen misfolding. Readers are directed to several important studies that explore the ECM impacts of these mutations^5,20–22.

Collagen premature termination mutations

Premature termination codons (PTC), introduced by point mutations or insertions and deletions (indels) that cause a codon reading frameshift, account for approximately 1/3 of all disease-causing mutations²³, and they are also common in the collagenopathies. In response to PTC mutations, cells deploy a complex surveillance mechanism, nonsense-mediated mRNA decay (NMD), to rapidly degrade the mutant PTC-containing transcripts and thereby reduce the synthesis of truncated and dysfunctional proteins.

While the precise molecular mechanisms are not fully resolved, all NMD involves distinguishing the PTC from the normal stop codon by distance from a molecular landmark such as the 3′ UTR, or the most 3′ exon-exon junction²⁴. The most widely accepted model, the exon-junction-complex (EJC)-dependent model, proposes that this distinction between the normal stop codon and PTCs is dependent on the deposition of a protein complex (the EJC) upstream of each exon-exon junction during pre-mRNA splicing. During a pioneer round of translation, the ribosome traverses the mRNA displacing all the EJC complexes it encounters until reaching the normal stop codon in the last exon, triggering an interaction with several termination factors, and resulting in normal termination. In contrast, if a PTC occurs in any exon other than the most 3′, the ribosome will encounter the PTC while there is still at least one downstream EJC present on the transcript. Communication with any remaining EJCs recruits a range of NMD effectors and initiates NMD²⁴. The EJC-dependent model thus predicts that PTC mutations elicit NMD in multi-exon genes, except when they occur in the last exon where there are no downstream EJCs to identify these PTCs as premature.

Collagen genes are large multi-exon structures with the triple helix sequence of fibrillar collagens (collagens I, II, III, V and XI) are distributed over ~42 central exons. While there are numerous PTC mutations in collagen genes, in many cases with predicted NMD outcomes, relatively few have been functionally characterized to determine the outcome on mRNA decay. In COL1A1, PTC mutations in exons upstream of the most 3′ exon result in NMD²⁴ and, although haploinsufficiency is thus the implied functional outcome, it is possible that not all the PTC-containing mutant mRNA is degraded and some abnormal truncated proα1(I) is produced with implications for pathophysiology, as discussed later. However, these mutations consistently produce the milder osteogenesis imperfecta clinical phenotype (OI Type I), suggesting that haploinsufficiency, rather than the production of a dominant negative truncated protein, is the likely overwhelming functional consequence. In the case of an OI patient with an insertion inducing a frameshift and PTC in the most 3′ COL1A1 exon, NMD does not occur, which is consistent with the EJC-dependent NMD model. As a result, stable proα1(I) mRNA encoding an abnormal C-terminal sequence is produced, which interferes with procollagen assembly and secretion causing the severe lethal OI (OI type II) clinical phenotype²⁵.

PTC mutations in other collagens with comparable complex multi-exon gene structures (such as collagens II, III, IV, V, VI) where functional studies have been conducted are also broadly consistent with this EJC-dependent NMD model. These studies are discussed in more detail in a review by Fang et al²⁴.

Of the collagens functionally studied, COL10A1 is conspicuous in not following the EJC-dependent NMD rules. Unlike fibrillar collagen genes, COL10A1 has only three exons, with most of the protein-coding sequence in the large terminal exon, exon 3. All the PTC mutations causing the COL10A1 disorder, metaphyseal chondrodysplasia type Schmid (MCDS), are in exon 3. Functional analysis in vitro and in cartilage from patients showed that heterozygous nonsense mutations COL10A1 p.W611X, p.Y632X and p.Y663X all result in NMD. Thus, COL10A1 NMD falls outside the conventional EJC-dependent NMD rule that posits that NMD does not occur when PTC mutations are in the most 3′ exon. Furthermore, the NMD in cartilage was complete – all the detectable mRNA derived from the mutant allele was degraded. This result is unusual. Most evidence for a range of other genes suggests that NMD is rarely complete, with at least a small proportion of mutant mRNA avoiding NMD. Notably, another heterozygous nonsense mutation, COL10A1 p.Y663X, led to only partial NMD in mouse and human cartilage samples²⁶. Adding further complexity, in a Col10a1 p.Y632X mouse model²⁷ the nonsense-mutation containing mRNA, which was completely degraded by NMD in MCDS patient cartilage²⁸, did not undergo NMD in mouse cartilage, and the truncated misfolded collagen X induced endoplasmic reticulum (ER) stress (see Discussion below). These two functional studies suggest that we should be cautious in predicting that the only consequence of heterozygous COL10A1 PTC mutations is NMD and a 50% reduction in collagen X.

Does intracellular retention of mutant misfolded collagens cause ER stress?

Collagen X

The most compelling evidence for a chronic, toxic UPR causing pathology in the collagenopathies comes from studies on COL10A1 mutations that lead to metaphyseal chondrodysplasia, type Schmid (MCDS). The collagen X mutations are predominantly in the C-terminal trimerization domain (NC1)²⁹. The only exceptions are two mutations at the N-terminal signal peptide cleavage site; in transfected cells the mutant chains remain anchored to the ER membrane after translation, cannot form a triple helix, and are not secreted, resulting in collagen X deficiency³⁰. Detailed in vitro studies showed that a broad spectrum of NC1 MCDS structural mutations, including COL10A1 p.Y598D, p.N617K, and p.G618V, compromise trimer assembly, reduce secretion, and lead to intracellular degradation. The mutant α1(X) chains were also extensively bound by ER chaperones, suggesting that a UPR could be induced²⁹. UPR activation was shown most convincingly in several mouse studies^31,32.

A knock-in mouse with an MCDS patient mutation, Col10a1 p.N617K, recapitulated the clinical phenotype with limb shortening and an expanded hypertrophic zone in growth plate cartilage³¹. Mutant collagen X was retained intracellularly and a robust UPR was elicited. All three stress sensors were apparently activated, as indicated by XBP1 splicing (IRE1), ATF6 cleavage, and eIF2α phosphorylation (PERK). Global transcriptome profiling of hypertrophic cartilage revealed that key downstream targets of the UPR stress sensors were upregulated, including molecular chaperones, foldases, and the ER-associated degradation (ERAD) machinery³³. An important finding from the Col10a1 p.N617K mouse^31,33 and other mouse studies^32,34 was that the mutant collagen X-induced ER stress disrupted chondrocyte maturation, reflected in the persistence of proliferative chondrocyte-like gene expression in growth plate cartilage.

While the MCDS mouse skeletal phenotype and UPR signature was much stronger in the homozygous Col10a1 p.N617K mice than in heterozygous Col10a1 p.N617K mice³¹, which are analogous to the heterozygous human MCDS mutations, the data strongly suggest that the canonical UPR pathway offers new therapeutic targets for MCDS patients. The broad approaches being tested to target the UPR in protein misfolding diseases involve reducing mutant protein load by stimulating mutant protein folding or degradation and inhibiting downstream UPR signaling to reduce pathological impacts. In one approach, the antiseizure drug carbamazepine was able to stimulate intracellular mutant collagen X degradation in vitro via proteasomal and autophagy pathways. Critically, treating homozygous Col10a1 p.N617K mice with carbamazepine significantly reduced the MCDS pathology³⁵. These breakthrough studies have resulted in a clinical trial of carbamazepine in MCDS patients (https://mcds-therapy.eu). In another mouse MCDS model, the experimental drug ISRIB (integrated stress response inhibitor), which reverses the effects of eIF2α phosphorylation downstream of PERK activation, also alleviated the pathology³⁴. These exciting results provide proof-of-principle that inhibiting elements of the canonical UPR can be an effective treatment option in collagenopathies where a chronic UPR is central to the pathology. The evidence for a pathological UPR is somewhat patchy for other collagen types, and this is discussed below.

Collagen I

Many early studies clearly demonstrated the effects of osteogenesis imperfecta procollagen I (COL1A1 and COL1A2) structural mutations on initial heterotrimer chain assembly, trimerization and subsequent helix formation, stability and secretion. Despite this long research focus, whether different classes of collagen I mutations induce a canonical UPR, and its possible contribution to pathology, is unresolved.

Although relatively rare, mutations in the procollagen I C-propeptide that disrupt proα-chain trimerization offer some insight into how the proteostasis network can be engaged by misfolding collagens. Early studies on OI patient fibroblasts with C-propeptide domain mutations showed increased BiP and GRP94 protein synthesis^25,36, key UPR markers. Furthermore, mutant assembly-compromised collagen chains bound BiP, and were degraded by the ERAD proteasomal system²⁵, a result that has been confirmed in several subsequent related studies^37–39. Since the two early studies were conducted before an understanding of mammalian UPR mechanisms was fully developed, detailed biochemical analyses exploring the possible activation of a canonical UPR were not included.

Subsequent studies provided additional insights. For example, a procollagen I C-propeptide frameshift mutation that causes bone pathology in the Aga2/+ mouse OI model resulted in modest BiP and CHOP upregulation at the mRNA level⁴⁰. No other UPR markers were tested. The authors also observed an increase in Hsp47, a key molecular chaperone involved in collagen triple helix assembly and transport. While Hsp47 is critical for collagen secretion and Hsp47 deficiency results in OI, it is not known to be a UPR target and thus its upregulation does not itself reflect canonical UPR activation. More recently, the OI-causing C-propeptide domain mutations COL1A2 p.C1163R, p.G1176V, p.P1182R, and p.Y1263C were studied in transfected cells³⁸. None of these variants increased expression of UPR targets BiP, Erdj4, Grp94, CHOP, HYOU1, Sec24D or Gadd34, indicating these procollagen misfolding mutations did not activate a prototypical UPR. Proteomic analysis of the proα2(I) C-propeptide interactome revealed that partners included PDIs, Hsp40/70/90 chaperones, and other members of the ER proteostasis network. These folding and quality control network members interacted much more extensively with the COL1A2 p.C1163R C-propeptide folding mutant and the outcome was complete intracellular retention and degradation with no UPR³⁸. This result is consistent with earlier studies showing increased BiP binding to several OI proα1(I) C-propeptide mutants^25,36.

The OI-causing proα1(I) C-propeptide domain mutations, p.C1299W, p.G1272V, p.W1275R, p.A1286D and p.D1277H also did not induce the UPR in vitro, based on unaltered BiP, Grp94, and HYOU1 expression³⁹. Notably, the inability to induce a UPR depended critically on the conserved N-glycan in the C-propeptide domain providing access to the ER lectin chaperones calnexin and calreticulin. Removing the N-glycan resulted in strong mutant-dependent UPR activation. Thus, the N-glycan provides an extrinsic mechanism to buffer C-propeptide domain misfolding³⁹. In sum, the evidence that a chronic UPR is widely induced by disease-causing mutations in the collagen I C-propeptide domain is minimal. The notion that a chronic UPR underlies pathology in these cases should, therefore, be viewed with skepticism based on current data. However, more studies in proper in vivo contexts are needed, as the field still lacks any human-relevant C-propeptide domain mutant mouse model of OI or in vitro expression in human osteoblasts.

For the much more common disease-causing triple-helix glycine mutations, the picture is even less clear with current data showing considerable discrepancies. Early studies on 17 different OI patient fibroblasts with a range of triple-helix structural mutations in COL1A1 and COL1A2, including nine patients with confirmed glycine substitutions, found no increase in BiP levels compared to control fibroblasts³⁶. Moreover, BiP did not bind to mutant chains with helix mutations. In contrast, a recent study on OI patient fibroblasts, including five with COL1A1 and five with COL1A2 helix glycine mutations, reported that several UPR components were upregulated, including BiP, PDI, ATF4, phospho-PERK and XBP1 splicing with some, but not all mutations⁴¹. Three of the five COL1A1 or COL1A2 mutations tested had increased BiP and, surprisingly, in two patients with the same COL1A1 p.G845R (G667R)^† mutation, one had increased BiP protein while the other with a milder form of OI, did not. Interestingly, the two most C-terminal helix glycine substitutions, COL1A1 p.G1172D (G994D) and COL1A2 p.G949S (G859S), did not upregulate BiP, although they would be expected to be more disruptive mutations from a protein folding perspective owing to their C-terminal location. In most COL1A1 and COL1A2 cases phospho-PERK was increased as was its downstream target ATF4. Accurately quantifying phospho-PERK levels is notoriously difficult and ATF4 is also downstream of other pathways including the integrated stress response, and so is not itself a strong indicator that the PERK UPR arm is active. The IRE1 UPR arm was activated with several mutations, but the ATF6 arm was not with any COL1A1 and COL1A2 triple-helix domain glycine mutation studied. Downstream autophagy upregulation was seen in most cases⁴¹, consistent with earlier studies suggesting that helix mutations cause aggregation with degradation targeted to the autophagosome-lysosome system³⁷. While these data suggested a UPR can be activated by at least some α1(I) and α2(I) helix glycine substitutions, there is apparent mutation specificity in the nature and strength of the response. The discrepancy between these studies with the earlier studies showing that BiP and GRP94 were not upregulated in fibroblasts from other cases of OI with triple helix-disrupting mutations is currently unresolved and it will be critical to confirm that the UPR is activated in independent studies. This will provide the opportunity to drill down in more detail on the timing, mutation-dependence, and molecular signature of any chronic UPR. Furthermore, it will be important to study how osteoblasts respond to collagen I misfolding mutations to determine if/how the UPR occurs in the disease-relevant bone tissues.

In addition to glycine mutations, which are common, arginine substitutions in the Y-position of the Gly-X-Y amino acid sequence triplet in COL1A1 can also slow triple helix folding and reduce helix stability. In fibroblasts from clinically mild OI patients with a COL1A1 p.R958C (R780C) mutation, or osteopenia caused by a COL1A1 p.R958L (R780L) mutation, collagen I had reduced thermal stability, increased susceptibility to MMP cleavage, reduced secretion and increased post-translational modification⁴². In two other cases, COL1A1 p.R1066C (R888C) causing the OI/EDS phenotype⁴³ and p.R1093C (R915C) with an EDS phenotype⁴⁴ showed some helix destabilization but the UPR was not explored. Given the milder effect on helix structure by Y-position arginine substitutions, compared to glycine substitutions, it is highly unlikely that a toxic UPR would be elicited.

In vivo mouse models offer the opportunity to study how OI mutations affect osteoblasts and bone development and function. Early OI transgenic mouse models over-expressed mutant collagen in addition to the wild-type collagen background^45,46. These were transformative in directly demonstrating that OI helix mutations were pathogenic. However, because of this transgenic approach the mutant genes were not expressed in the correct developmental contexts or at the correct mutant:wt collagen stoichiometry. At the time, ER stress/UPR pathways were not known, so we will not discuss these studies further. Our discussion will focus on more recent knock-in and naturally occurring OI mouse models, where the impact of collagen I helix mutations have been studied in more detail and expression patterns are of more relevance to the heterozygous OI patients.

The first knockin OI mouse model was a Col1a1 p.G527C (G349C) mutation (Brtl) that resulted in a lethal to a moderately severe heterozygous OI clinical phenotype⁴⁷. These heterozygous Brtl^-/+ mice have been used in several studies on the OI phenotype and to explore disease mechanisms and therapies. In the Brtl^-/+ mice, a glycine residue is substituted with a cysteine residue in the triple-helical domain, which is normally devoid of cysteines. The introduced cysteine means that two mutant α1(I) chains could form disulfide-bonded dimers (~23–29% of the total α1(I) chains), which were apparently secreted normally and incorporated into the ECM with a similar efficiency to wild-type α1(I)⁴⁸. Detailed pulse-chase labelling studies indicated that collagen α1(I)₂α2(I) trimers containing a single mutant α1(I) chain were selectively retained and degraded within the cells⁴⁹. Thus, collagen I trimers with one mutant α1(I) chain were under-represented in skin, and fibroblast and osteoblast cell culture media. A proteasome inhibitor did not change the collagen composition, suggesting the mutant-containing collagen was not targeted for proteasomal degradation, consistent with earlier studies on collagen helix mutations³⁷.

Proteomic and transcriptomic analysis of wild-type and Brtl^-/+ bone showed Gadd153 (CHOP) was increased in the most severely affected (lethal) mice, and αB crystallin was increased in non-lethal mice⁵⁰. BiP was not upregulated, consistent with previous studies suggesting that collagen I helix mutations neither interact with nor upregulate BiP³⁶. Further, in a more recent study using a commercial UPR-targeted PCR array, only 8 out of 83 genes were significantly upregulated in Brtl^-/+ osteoblasts⁵¹, with none of the 8 being prototypical UPR targets like BiP, GRP94, Erdj4, HYOU1, CHOP, or similar. Eif2ak3 (PERK) was transcriptionally upregulated, although PERK transcriptional upregulation is not a typical UPR marker. ATF4 protein was increased, consistent with either induction of PERK or other integrated stress response components⁵². Most compelling, a modest increase in XBP1 splicing implicated IRE1 activation⁵¹. Several autophagy-related genes were upregulated and, along with a small increase in LC3 protein and increased autophagic flux, this was consistent with increased autophagy of the mutant collagen⁵¹, although there was no apparent increase in LC3I to LC3II conversion, often used to indicate autophagy. Taken together, these data suggest there is some form of stress in Brtl^-/+ osteoblasts, but the evidence for a chronic, canonical UPR is scant. Further characterization is needed to determine if chronic UPR signaling, per se, contributes to the pathology (see later discussion). It is worth bearing in mind that the unpaired cysteine residues in Brtl heterozygous collagen trimers⁴⁹ may present a specific challenge to the ER proteostasis network that is different from that created by other common glycine substitutions.

Irrespective, the Brtl mutation has significant downstream impacts on cellular outcomes. Osteoblast apoptosis was increased⁵¹, there was less cell proliferation, altered TGFβ signaling and reduced expression of the osteoblast transcription factor, Runx2⁵³. Mesenchymal stem cells from Brtl^-/+ mice showed reduced osteoblastogenesis, suggesting that the intracellular and/or extracellular consequences of the mutation could compromise osteoblast formation.

The Col1a2 p.G700C (G610C) mouse model with mild to moderate OI has an introduced cysteine in the α2(I) triple-helical domain which also does not normally have cysteines residues. The collagen I triple helix is disrupted, the mutant misfolded collagen I accumulates in the ER, and while the possible activation of a UPR has been explored, the results are again conflicting and unclear⁵⁴. Early studies on bone and osteoblast cultures showed that BiP was not upregulated at the mRNA or protein level. There was some CHOP upregulation although this is a general signature of cell stress rather than a specific UPR marker⁵². A more recent study using a commercial UPR PCR array provided a conflicting result, with BiP mRNA upregulated in osteoblasts from this mouse, but no CHOP upregulation⁵¹. The array-based analysis did not provide quantitative data on how much the BiP mRNA was increased, so its significance is difficult to evaluate^51,54. Notably, XBP1 splicing was not observed in either study. Mirigian et al⁵⁴ showed that CHOP, chaperones αβ crystallin and HSP47 were upregulated, and there was more eiF2α phosphorylation leading them to conclude that the mutation caused an unusual form of cell stress but did not activate the canonical UPR pathway. The UPR PCR array showed the Col1a2^+/G610C mutation upregulated many UPR genes, indeed many more than in Brtl^-/+ osteoblasts, with several of unknown significance upregulated in both; Foxo6, Htra1. As in the Brtl^-/+ osteoblasts, Col1a2^+/G610C osteoblasts had increased autophagy and apoptosis⁵¹. Although these conflicting data are yet to be reconciled to reveal the precise UPR response in the Col1a2^+/G610C osteoblasts, the consequence of the mutation also retarded osteoblast differentiation and caused abnormal responses in major signaling pathways⁵⁴.

While these mouse studies are important because they place the collagen mutations and downstream pathology in the in vivo bone context, unfortunately they have not provided clarity around the precise nature of the cellular stress response. Perhaps most importantly, none of these studies are on the prototypical OI-causing GlySer triple-helical mutations. To understand how collagen I mutations engage the proteostasis machinery and the UPR and produce OI pathology, new mouse models that will allow us to study glycine substitutions in different regions of the helix, reflecting the patient mutation spectrum, are needed. More information about how human osteoblasts handle collagen I with OI mutations is essential to better understand the molecular pathology and develop new therapeutics. The ability to rapidly introduce mutations into human induced pluripotent stem cells using gene editing technology and differentiate mutant and isogenic control iPSC into human osteoblasts and osteocytes offers us a powerful new approach to achieve this goal.

Collagen II

Missense mutations in COL2A1 causing glycine substitutions in the collagen II homotrimer are a major mutation group in the chondrodysplasias. Limited access to patient cartilage and chondrocytes has meant that studies on collagen II mutations have been less intense than those on collagen I, where primary fibroblasts are widely available. Several early studies on glycine variant collagen II patient cartilage samples and chondrocytes in vitro supported the suggestion that these substitutions impact helix stability and secretion. For example, extracellular collagen II protein could not be detected in cartilage from a patient with a heterozygous COL2A1 p.G969S (G769S) achondrogenesis II mutation, despite expression of the mutant at the mRNA level⁵⁵. The intracellular collagen II had increased post-translational modifications, indicating misfolding, and the collagen II was degraded intracellularly. Likewise COL2A1 p.G804A (G604A)⁵⁶, p.G1053E (G853E)⁵⁷ mutations resulted in dilated ERs and increased post-translational modification. Increased post-translational modification, indicative of abnormal folding and reduced secretion, was also seen with COL2A1 p.G909C (G709C), p.G504C (G304C) and p.G492V (G292V)⁵⁸. These early studies clearly support the idea that the ER is faced with a chronic collagen II folding challenge, but the possibility of UPR activation was not addressed. Support for a UPR in response to heterozygous collagen II helical glycine substitutions came from studies on patient fibroblasts and iPSCs differentiated into chondrocytes⁵⁹. These studies showed reduced collagen II secretion and upregulation of BiP, GRP94, and CHOP mRNA and ATF6 protein, eiF2α phosphorylation, XBP1 cleavage and apoptosis with p.G450R and an in-frame exon 41 skip mutation. A COL2A1 p. G1182A mutation also resulted in BiP, GRP94, and CHOP mRNA upregulation⁵⁹. A mouse chondrodysplasia model, Col2a1 p.G1170S⁶⁰, had increased UPR markers (XBP1s; BiP, ATF4 and AFT6) in the homozygous, but not heterozygous cartilage. CHOP was the only potential UPR marker upregulated in both heterozygous and homozygous mice. Apoptosis in the growth plate cartilage was only seen in the homozygous mice. Given that the patient mutations are heterozygous, it is unclear how these studies translate to human pathological mechanisms.

In addition to glycine substitutions, several arginine to cysteine substitutions at the X or Y position of the collagen II Gly-X-Y repeating amino acid triplet cause a range of mild to severe (non-lethal) chondrodysplasias, commonly spondyloepiphyseal dysplasia (SED). As is the case with collagen I, the collagen II triple helix does not normally contain any cysteines. A heterozygous patient COL2A1 p.R989C (R789C) mutation⁶¹ shows increased post-translational modification implying reduced folding, although the thermal stability of the helix was normal. Reduced mutant collagen II secretion was seen in patient chondrocytes. The most comprehensive study⁶² looked at arginine to cysteine mutations distributed along the helix from the N- to the C-terminal end using transfected cell models where only the mutant chain was expressed. The p.R275C (R75C), p.R334C (R134C) and p.R904C (R704C) mutations did not significantly affect the collagen II thermal stability whereas the more C-terminal p.R940C (R740C) and p.R989C (R789C) mutants had reduced thermal stability and were susceptible to trypsin digestion, indicating they had a compromised triple helical structure. Furthermore, COL2A1 p.R940C and p.R989C provoked a small increase in XBP1 splicing, increased expression of BiP protein, and increased apoptosis in vitro⁶². These data suggested that the more C-terminal arginine to cysteine mutations provoked a UPR. Other studies using cells recombinantly expressing another C-terminal mutation, COL2A1 p.R1192C (R992C), showed the collagen II had reduced thermal stability, atypical disulphide bond formation and reduced secretion associated with apoptosis⁶³.

It is important to recognize that while these transfected cell studies are useful in identifying fundamental structural/biochemical properties of the mutant, such as helix formation and stability, secretion, post-translational modifications and extracellular interactions, they may be less informative in alterations to signalling pathways such as the UPR, where mutant expression at the correct levels, at the correct wild-type to mutant ratio, and by the disease target cells, chondrocytes, is likely to be critical. In this regard, the Col2a1 p.R1192C mouse model⁶⁴ will be an important tool to determine if a UPR results from this C-terminal helix mutation in chondrocytes in vivo. To date, only some preliminary analysis has been conducted identifying BiP-positive cells and apoptosis in the mutant mouse growth plate cartilage⁶⁵. This tantalizing data may suggest a UPR, but much more analysis is required to understand the disease mechanism in this correct cellular context.

Several mouse models have collagen II C-propeptide mutations. The Col2a1 p.Y1391S⁶⁶ and p.D1469A⁶⁷ mouse cartilage showed procollagen II ER retention and increased GRP94 and CHOP expression, generally more pronounced in homozygous mutants. The p.Y1391S mouse, which corresponded to a platyspondyly lethal chondrodysplasia patient mutation, had increased chondrocyte apoptosis. Two other mouse chondrodysplasia models, Col2a1 p.S1386P (Lpk/+)⁶⁸ and a 3bp deletion in the C-propeptide sequence (Dmm/+)^69,70 have dilated ER and mutant procollagen II retention. The Dmm/+ mouse has a small increase in BiP and increased eIF2α phosphorylation⁷⁰. While these mouse models also provide some indication that the C-propeptide mutations might induce a UPR, given the expectation that this type of mutation should elicit a strong canonical UPR, further detailed analysis is required to understand the cellular response to the misfolded procollagen and how this relates to the pathology. A heterozygous frameshift mutation, COL2A1 p.K1447CInsfsX18, causes spondylo-peripheral dysplasia⁵⁹. The mutation is proposed to produce a truncated and misfolding C-propeptide. Chondrocytes produced by directed differentiation of fibroblasts and iPSCs from the patient showed no increase in BiP, GRP94, CHOP or apoptosis. No other procollagen II C-propeptide mutations have been functionally studied to shed light on how these interact with the proteostasis machinery.

Other fibrillar collagens, III and V

Many missense and other structural mutations in COL3A1 potentially affecting collagen III assembly and helix stability have been identified in the vascular form of Ehlers-Danlos Syndrome (EDS IV)^71,72. As with the other fibrillar collagens, these helix mutations reduced triple-helix thermal stability⁷³, impaired secretion, and the collagen was retained in the dilated ER^74,75. It is not known if misfolded collagen III can elicit an ER stress response. While the one study that considered the UPR showed increased ATF6, it did not test if any UPR signal transducers were actually activated⁷⁴. Transcriptomics using cultured patient skin fibroblasts with dominant negative triple-helical domain mutations⁷⁶ showed upregulation of several ER protein folding genes, such as PDIAs and members of the DNAJ chaperone family, along with down regulation of proteasome genes, but did not provide a clear gene signature that would indicate chronic UPR activation.

In classical EDS (cEDS), although the predominant disease mechanism appears to be collagen III haploinsufficiency, COL5A1 and COL5A2 mutations affecting procollagen V assembly and helix glycine and arginine to cysteine substitutions have been identified⁷². The UPR has not been evaluated in these disorders. With collagen V mutations, the main disease impact appears to be extracellular where reduced collagen V results in abnormal collagen I/V fibrillogenesis⁷².

Network forming collagens, IV and VI

Collagen IV, the principal basement membrane collagen, is encoded by six genes (COL4A1–6) which form α1α1α2, α3α4α5 and α5α5α6 heterotrimers. The collagen IV heterotrimer triple helix has approximately 20 interruptions to the Gly-X-Y repeat^16,17. Mutations in COL4A3, COL4A4, COL4A5 cause Alport syndrome where basement membrane defects predominantly affect the kidney glomerulus. COL4A1 and COL4A2 mutations result in cerebrovascular disease and kidney, eye and muscle defects^16,17. For COL4A1 and COL4A2 intracellularly retained misfolded collagen IV can upregulate UPR markers. COL4A2 p.G702D causes mutant retention in the ER and upregulation of BiP, calnexin, ATF4, ATF6, eIF2α phosphorylation and XBP1 splicing⁷⁷. A Col4a1^Δex40 transgenic mouse also showed increased BiP, XBP1 splicing, ATF6 cleavage, eIF2α phosphorylation, ATF4, CHOP and increased apoptosis⁷⁸. A heterozygous Col4a1 p.G1064D mouse had increased BiP and ATF4 protein⁷⁹. In contrast, fibroblasts from an allelic series of eight Col4a1 and Col4a2 glycine substitution mutations in mouse models of ocular dysgenesis⁸⁰ showed collagen IV ER retention but no induction of ER stress markers BiP, CHOP, ATF4 or ATF6. Thus, the role of the UPR in COL4A1 and COL4A2 pathophysiology is not fully resolved.

There is evidence for UPR activation with the Alport Syndrome knock in mouse Col4a3 p.G133E mutation⁸¹ where UPR markers BiP, p-PERK, and p-eIF2α are upregulated at the mRNA and protein level. In transfected human podocytes overexpressing the mutant chain, mutant collagen was retained in the ER. Microarray expression profiling and gene set enrichment analysis showed UPR pathway dysregulation in cells transfected with either wild-type or mutant protein chains, and this is likely to be a consequence of the massive protein folding demand produced by the extreme overexpression – ~30 fold at the mRNA level and ~10 fold at the protein level. While qPCR verified that BiP, CHOP, XBP1, calnexin, calreticulin, the autophagy modulator Gadd 34 and the ERAD component, EDEM1 were all upregulated more in mutant expressing cells than in wild-type expressing cells, it is unclear how biologically relevant this system is⁸¹. In fibroblasts from Alport Syndrome patients with COL4A4 p.G908R and p.G624D mutations, UPR markers ATF6, BiP and CHOP were upregulated. Several autophagy and apoptosis genes were also upregulated, predominantly with the COL4A4 p.G908R mutation⁸².

Mutations in the collagen VI genes, COL6A1, COL6A2 and COL6A3, cause the muscle diseases Bethlem myopathy and Ullrich congenital muscular dystrophy¹⁹. Glycine substitutions and exon skipping mutations towards the N-terminal end of the helix tend to interfere with intracellular dimer and tetramer formation and since tetramers are obligatory for collagen VI secretion, secretion is reduced. These mutations also have a dominant negative effect on microfibril formation. Despite intracellular retention, there is no evidence that a canonical UPR occurs in dominant negative mouse models⁸³, human muscle⁸⁴ or patient fibroblast cultures⁸⁵. Rather, common pathogenic mechanisms include mitochondrial dysfunction and impaired autophagy⁸⁶.

Is the UPR a critical part of the pathology of collagen misfolding disorders?

The above discussion highlights that, while it is tempting to implicate UPR activation in collagen misfolding disease cellular pathology, the evidence is mixed. It is likely that there is considerable mutation, collagen type, and target cell specificity in whether they respond to collagen misfolding with a canonical, chronic UPR and what the downstream consequences of that UPR may be.

Activating specific UPR modules and/or components could indicate that cells are responding in a beneficial, remedial manner to the misfolding protein rather than reflecting a pathological process. It is important to remember that the UPR and the whole cellular proteostasis machinery, is involved in protein folding. The UPR responds to misfolded proteins by regulating the ER chaperone and quality control networks and reducing translation to lower the load of mutant protein in the ER (Fig 1). Thus, the initial response at least, could improve mutant collagen folding and help with the cellular response to the proteostasis challenge. Indeed, as highlighted above, for many collagen variants (especially collagen I), there is little or no evidence for UPR activation. It could be that cells are not able to recognize that an atypical fibrous protein is misfolded, and this means a remedial UPR that could have alleviated pathology by improving proteostasis is not activated. On the other hand, long-term and unresolved ER stress leading to chronic UPR activation, especially of the PERK pathway, can lead to apoptosis and be deleterious. There are hints that this mechanism operates in the pathology of especially the collagen type II diseases. In all cases, a detailed analysis of the interrelated sensor networks and downstream signalling and how this interacts with cellular growth, development and function is required to understand the nature of the UPR and its consequences.

Importantly, a direct cause-effect between chronic UPR activation and pathology has been established only for the cartilage growth plate pathology in MCDS. As discussed earlier, MCDS is caused by collagen X misfolding mutations which induce a strong canonical UPR^31,33,87 and while all arms of the UPR are activated, the data suggest the PERK arm is responsible for the pathology. To test if the UPR and downstream signalling, or other impacts of the mutant collagen X, caused the delayed entry of the mutant chondrocytes into the terminal stages of hypertrophy, a transgenic mouse was produced where a known misfolding protein, the thyroglobulin Cog variant, was expressed in hypertrophic chondrocytes under the control of the Col10a1 promoter³¹. This targeted UPR induction by a protein whose only role in the cartilage is to induce ER stress caused the same growth plate pathology as the Col10a1 MCDS mutations. This finding shows that MCDS is caused by a chronic UPR, and not the presence of mutant collagen X per se. In another MCDS collagen X misfolding mouse model, treatment with ISRIB (Integrated Stress Response InhiBitor) which blocks downstream signalling in the PERK arm of the UPR, ameliorated the pathology³⁴, providing further support for the central role of the UPR in MCDS pathology. With the other collagen mutations, a direct connection between any UPR and pathology is yet to be established, and it is important to consider that mutant collagens also induce changes to ECM composition, structure, interactions, and cell-ECM communication that likely influence the pathology. With some collagen mutations, while there is some evidence that PERK signalling is activated, as well as downstream effectors such as CHOP, it remains possible that the altered ECM could also induce these changes. As just one example, with the Col2a1 p.G1170S mutation discussed earlier, the reduced ECM collagen II interacts poorly with its cellular receptor β1 integrin, influencing the BMP/SMAD1 pathway and accelerating chondrocyte hypertrophy⁸⁸.

The final pathological outcomes will reflect the balance between the cellular consequences (such as the UPR), intracellular degradation by the quality control pathways, and the dominant negative effects that secreted mutant collagen has on ECM architecture, stability, and function (Fig 1). In addition, the phenotype will be influenced by the functional role and expression level of the collagens, and importantly by the precise mutation, its location and destabilising effect on collagen folding. It is not sufficient to characterise a few key “canonical” UPR markers - detailed quantitative transcriptional, and importantly, post-transcriptional analyses, of well-validated UPR components are needed to document a comprehensive mutation-specific molecular signature.

As a final comment, it is crucially important to study collagens in the correct disease target tissues/cells involved in the pathologies using model systems where expression levels are biologically appropriate. The proteostasis machinery in different cell types will be tuned to their functional biosynthetic capacity, and thus their capacity to respond to collagen misfolding is likely to exhibit cell specificity. For example, chondrocytes are professional secretory cells and this is thought to contribute to their particular susceptibility to ER stress from ECM protein misfolding⁸⁹. Thus, using new mouse models, or in vitro cell models that accurately recapitulate cellular pathologies, will be vital to understand the molecular pathology caused by collagen misfolding, and will also form the basis for sophisticated therapeutic manipulation of the proteostasis network.