The Centennial of insulin’s discovery (1) coincides with renewed interest in cellular mechanisms of biosynthesis. The mature hormone is the post-translational product of a single-chain precursor, proinsulin (2, 3). Diverse dominant mutations in the human insulin gene (INS) have been identified associated with diabetes mellitus (DM) (4–10). Such mutations impair oxidative folding of nascent proinsulin in the endoplasmic reticulum (ER) of pancreatic β-cells (11, 12). Originally identified as a monogenic cause of permanent neonatal-onset DM (7, 13–15), this syndrome (designated mutant INS-gene-induced diabetes of youth; MIDY) can also present in childhood (16) or adolescence (17) (maturity-onset diabetes of the young; MODY). Such variation in onset is ascribed to mutation-specific differences in extent of perturbed folding (12, 18). The spectrum of phenotypes may also reflect polygenic differences in how the β-cell responds to chronic ER stress (19, 20).

MIDY patients are heterozygous. Although one wild-type (WT) insulin allele would ordinarily be sufficient to maintain metabolic homeostasis, studies of the Akita mouse [a corresponding mouse model (21–23)] first demonstrated biochemical dominance: misfolding of the variant proinsulin impairs wild-type (WT) biosynthesis (24, 25). Analogous biochemical interference occurs in β-cell lines (11, 12, 18). ER stress leads to distorted organelle architecture, impaired glucose-stimulated β-cell secretion and eventual cell death (26, 27). Discovery of the mutant proinsulin syndrome has stimulated renewed interest in structural mechanisms of disulfide pairing (28–33) as a critical step in the biosynthesis of insulin (2, 3, 34). The central importance of such mechanisms—both in β-cells and as a general model for oxidative protein folding—have motivated extensive cell-based and animal studies (35–39). Together, these efforts have deepened the biophysical understanding of classical structure-functional relationships in the insulin molecule (9, 10, 19, 40) in relation to cellular mechanisms of biosynthesis (4, 10, 41, 42).

The goal of this review is to provide a structural perspective on INS mutations in human proinsulin [for clinical background and history of discovery, see (43)]. A starting point is provided by a general biophysical paradigm: that key interactions in intermediate stages of protein folding often foreshadow spatial relationships in the native state (44–46). Accordingly and in the reverse direction, we will regard the classical crystal structure of insulin (47) as a framework for interpreting folding mechanisms. Given this context, we will restrict our attention to mutations in (or adjoining) the well-organized insulin moiety of proinsulin (48) ( Figure 1A and Table 1 ). Whereas traditional structure-activity relationships (SAR) pertain to receptor binding (9), contributions of the same residues to folding efficiency may be inapparent once the native structure is reached. The growing MIDY/MODY database of INS mutations ( Figure 1B and Table S1 ) may be exploited to decipher this hidden layer of meaning. As a seeming paradox in Darwinian medicine (49, 50), the biophysical non-robustness of proinsulin biosynthesis suggests that the hormone has evolved to the precarious edge of foldability (40, 51, 52). We envisage that foundational principles of protein folding, structure and stability will be found to rationalize the distribution of MIDY/MODY mutations and broad spectrum of clinical presentations.

The majority of INS mutations cause permanent neonatal-onset DM ( Figure 1 and Table S1 ) (14). Because impaired β-cell function develops prior to maturation of the immune system, the patients present with auto-antigen-negative DM. Similar phenotypes may be caused by mutations in other genes (151), most frequently a heterozygous activating mutation in the β-cell voltage-gated potassium channel (either KCNJ11 or ABCC8, respectively encoding its Kir6.2 and Sur1 subunits) (152, 153). The resulting diabetic phenotype in this genetic background may be transient or permanent. It is important to recognize this subset of neonates or toddlers as in favorable cases they can successfully be treated with oral agents that inhibit the channel (sulfonylureas) rather than by insulin injections (151).

Dominant INS mutations are the second most common genetic cause of permanent neonatal DM (7, 13, 14, 16). Such mutations occur in each region of preproinsulin: its signal peptide, B-, C- and A domains ( Table S1 ) (9, 10). The majority result in the addition or removal of a cysteine, leading in either case to an odd number of potential pairing sites ( Figure 1A ). Mutations have been found at each of insulin’s six canonical cysteines, generally associated with neonatal onset ( Figure 2B , Table 1A ). An additional cysteine may be introduced at various positions in the insulin moiety ( Figure 2C and Table 1B ). The resulting odd number of thiol groups leads in general to misfolding and aggregation (11, 12, 18). Even in this context structure may matter, as it is possible that some sites of Cys introduction lead more readily to aberrant intra- or intermolecular disulfide pairing than others, depending on the conformational properties of oxidative folding intermediates and their interactions surfaces. Such biophysical variability would be expected to be associated with differences in ER stress and hence age of DM onset.

Among human MIDY mutations is the same “Akita” substitution (Cys^A7→Tyr) as in the Ins2 gene of the Mody4 mouse (21–23); this dominant murine substitution has thus been characterized as a model of the human syndrome (25–27). The variant murine proinsulin in vitro undergoes partial unfolding with increased aggregation (154). Analogous perturbations were found in human insulin- and proinsulin analogues lacking cystine A7-B7 (66, 132). Heterozygous expression of related Ins2 allele Cys^A6→Ser in the mouse also causes DM (155).

Identification of identical human and murine mutations at position A7 suggests that the mechanisms of neonatal DM have shared pathogenetic features independent of species (21–23, 25–27). Although β-cell degeneration in the Akita mouse remains incompletely understood, early defects have been observed in the folding and trafficking of both wild-type and variant proinsulins. These defects are associated with elevated markers of ER stress, electron-dense deposits in abnormal ER and GA, and mitochondria swelling—together leading to a progressive decline in β-cell mass (25–27). Evidence for the clinical relevance of these findings has been obtained by the construction of innovative fluorescent proinsulin fusion proteins and their use in cell lines and transgenic mice to detect subcellular localization and aggregation (35–38).

INS mutations may also be associated with onset of DM in childhood or adolescence ( Table 1D ) (169–171); diagnoses may be carried as auto-antibody-negative presumed Type 1 DM or Type 2 DM. Substitution of Val^B18 ( Figure 3B , center) by Ala (172) was identified as a MODY allele (DM onset <25 years of age, autoantigen negative) in a three-generation Italian pedigree (three siblings, the parent and presumed grandfather) (172). Unlike MIDY patients with neonatal onset, birth weights were normal. The Ala mutation at position B18 would be expected to enhance segmental α-helical propensity (167, 168), but introduce a destabilizing cavity (163, 164) adjacent to the critical [B19-A20] disulfide bridge. Unlike the perturbations introduced by Gly^B18 (above), these effects would offset to yield, rationalizing a mild net impairment of initial disulfide pairing.

Four additional MODY mutations occur within the B20-B23 β-turn and its aromatic anchor at B24: Gly^B20→Arg, Arg^B22→Gln, Gly^B23→Asp and Phe^B24→Ser ( Figure 6 ). Although the mechanism by which Gln^B22 causes MODY is not apparent, L-amino-acid substitutions of Gly^B20 or Gly^B23 would be expected to alter their respective ϕ dihedral angles. It has previously been reported that Ala substitutions impair the expression of mini-proinsulin in S. cerevisiae and impede chain combination, whereas efficient disulfide pairing in vitro can be rescued by D-Ala substitutions (97). That B23 mutations may cause either neonatal onset (Val^B23) or delayed onset (Asp^B23) suggests that details of side-chain chemistry influence folding efficiency.

Ser^B24 (originally designated insulin Los Angeles) is associated with variable genetic penetrance with hyperinsulinemia. The latter finding indicates that Ser^B24-proinsulin can in fact fold in the β-cell ER, undergo proper trafficking and processing to mature Ser^B24-insulin (173). In cell culture the variant proinsulin nonetheless induces ER stress, albeit at a level below MIDY variants (12). In vivo mutational induction of mild or moderate ER stress can presumably cause (depending on other genetic risk alleles and environmental factors) slow but progressive loss of β-cell mass (174, 175) as in the Akita mouse (24, 176).

The final MODY-associated mutation occurs on the surface of the A domain: Glu^A4→Lys ( Table 1D and Figure 7A ). That this substitution should perturb the folding of proinsulin seems surprising given (a) the absence of structural constraints at this position in insulin and (b) the broad tolerance of insulin chain combination to substitutions within the N-terminal A-chain α-helix (95). We speculate that Lys^A4 introduces a subtle perturbation in proinsulin through electrostatic repulsion of the dibasic element at the CA junction (red box in Figure 7B ). In particular, nascent α-helical structure in the A1-A8 segment may be stabilized by a salt bridge between “Arg^A0” (i.e., the final residue of the C domain [position 89 of preproinsulin]; Figure 7A ) and WT Glu^A4 ( Figure 7C ). Such an interaction, together with Gly^A1, could in essence provide a favorable N-Cap (177), which could overcome the adverse helical propensities of the three β-branched residues in this segment (Ile^A2, Val^A3 and Thr^A8). This contribution would not pertain to insulin chain combination due to the absence of Arg^A0 (an analogous C-capping salt bridge from Glu^A4 to the A1 α-amino group would be blocked by its deprotonation at the reaction pH of 10.5).

For completeness, we note that mutations in the insulin gene that are not associated with impaired folding can nonetheless be associated with adult-onset DM phenotypes of variable penetrance (57) ( Table S1 ). Such heterogeneity is in accord with “Murphy’s Law of genetics”: in a complex pathway or set of mechanisms, what can go wrong will go wrong. For example, insulin variants Wakayama and Chicago (i.e., classical insulinopathies Val^A3→Leu and Phe^B25→Leu respectively) markedly impede receptor binding (173) in association with mutant hyperinsulinemia (178). These mutations directly perturb the hormone-receptor interface (160). A complementary example is provided by diabetes-associated mutation His^B10→Asp, which enhances receptor binding (179). Although Asp^B10 would introduce a favorable electrostatic interaction at the hormone-receptor interface, in β-cells Asp^B10-proinsulin exhibits inappropriate sorting to a constitutive granule (180, 181). Unlike glucose-regulated secretory granules, constitutive granules lack prohormone convertases, and so the patients exhibit mutant hyperproinsulinemia. Yet another syndrome is characterized by impaired prohormone processing leading to circulation of a split proinsulin with reduced activity (182).

Protein evolution is generally enjoined by overlapping biological constraints, including biosynthesis, structure, and function ( Figure 8A ). Particular residues in insulin may thus contribute to one or more critical mechanisms, including nascent foldability in the ER, protection from intra- or extracellular toxic misfolding, trafficking from the ER through the GA to glucose-regulated secretory granules, self-assembly within these granules, disassembly of Zn²⁺-insulin hexamer in the portal circulation and in turn receptor binding. The stringency of these overlapping constraints rationalizes the limited sequence variation among vertebrate insulins (47).

Evolutionary constraints may be coincident or opposing at a given position. An example of a coincident constraint is the concurrent contributions of invariant Phe^B24 to core packing, dimerization and receptor binding. Opposing constraints call for compromise. An example is provided by Gly^B8, invariant as an achiral amino acid free to roam in the Ramachandran plane. Systematic studies of L- or D substitutions have suggested that at B8 kinetic determinants of foldability are at odds with conformational requirements of receptor binding (96, 98). Whereas a positive ϕ dihedral angle (enforced by a D-substitution) facilitates disulfide pairing, a D side chain impedes receptor binding. Conversely, negative dihedral angle (like that of an L-amino acid) impair folding efficiency but may be compatible with receptor binding (96, 98). These opposing requirements presumably underlie the invariance of glycine – the only achiral amino acid – at a site of conformational change. A switch in conformation of Gly^B8 between the right side of the Ramachandran plot and the left (respectively corresponding to positive or negative ϕ angles) was anticipated by the classical TR transition among zinc insulin hexamers (187). Although such allostery may pertain only to hexamers (99), the TR transition exemplified the long-range transmission of conformational change (188) —a theme central to the transmembrane propagation of an insulin signal via receptor reorganization (189–191). The examples posed by clinical mutations at B24 and B8 ( Table 1C ) suggest that premature adoption of the hormone’s receptor-engaged conformation within β-cells (either by proinsulin in the ER or GA or by insulin in secretory granules) may trigger toxic misfolding.

Recent co-crystal and cryo-EM-derived structures of insulin bound to receptor fragments have demonstrated the function of a protective hinge in B chain (160, 189, 190, 192). Mechanisms of hormone-receptor recognition [for review, see (191)], extend to IGF-I as visualized in a landmark series of homologous cryo-EM-derived structures of respective receptor ectodomain complexes (189, 190, 193–195). As predicted based on studies of “anomalous” insulin analogues (157, 196, 197), detachment of the C-terminal β-strand (residues B24-B28) enables both its own binding in a groove between receptor elements L1 and αCT (respectively at the N- and C-terminal ends of the IR α-subunit); the latter element also packs against the N-terminal A-chain α-helix.

Insulin’s B-chain hinge—opened on receptor binding—may represent an evolutionary response to the danger of proteotoxicity. This danger, aggravated by exposure of non-polar surfaces, is intrinsic to the coupled folding/misfolding landscapes wherein the true ground state is defined by β-sheet-rich amyloid ( Figure 8B ). Models of insulin amyloid as superhelices of protofilaments have been derived at low resolution by cryo-EM ( Figure 8C ). Studies of insulin fibrils by infrared and Raman spectroscopy have demonstrated a predominance of β-sheet (198–200) in accordance with fibril X-ray diffraction (201–203). Despite the striking biophysical features of fibrils as a universal thermodynamic ground state of polypeptides ( Figure 8B ) (85), oligomeric intermediates in the pathway of fibrillation pose the greater cytopathic danger ( Figure 8D ) (204).

Recent evolutionary studies of insulin have highlighted the importance of Phe^B24, whose conserved aromatic ring plays multiple roles: anchoring the native B-chain β-strand, stabilizing the α-helical core, and contributing to both self-assembly (47) and hinge opening on receptor binding (197). In the open state the aromatic ring binds within a classical nonpolar pocket at the hormone-receptor interface (189, 190, 195). On substitution of Phe^B24 by Gly, native function is paradoxically retained (157). Comparative studies of “register shift” analogues indicate that an alternative mode of receptor binding supervenes in which Phe^B25 takes the place of the missing Phe^B24 (52); residues B20-B24 form a flexible pentaloop rather than an aromatic-anchored β-turn (205). This alternative binding mode is apparently disallowed in evolution due to toxic misfolding of Gly^B24-proinsulin (as evidenced impaired folding efficiency, induction of ER stress and impaired secretability in transfected cell models) and possibly by the heightened susceptibility of Gly^B24-insulin analogues to fibrillation (52). Evidence for the paradoxical evolution of vertebrate insulins to the edge of foldability has been provided by biophysical studies of a native-like variant, Tyr^B24-insulin (40). Although providing the C-terminal B-chain β-turn and β-strand with an homologous “aromatic anchor,” Tyr^B24 is also disallowed due its perturbation of biosynthesis and induction of ER stress. Indeed, of the 20 natural amino acids, only Phe at position B24 enables the efficient biosynthesis of proinsulin (40). We speculate that such marked sensitivity to mutation—signifying the paradoxical non-robustness of an adaptive landscape (40)—will be found at many or most sites associated with neonatal-onset DM ( Table 1C ).

Because, to our knowledge, clinical mutations that selectively perturb insulin’s hexameric structure and storage in secretory granules have not been described, this review has not focused on these processes. Any such perturbations would be downstream from the major sites of perturbation in the MIDY syndrome: misfolding in the ER and impaired trafficking through the GA. It is possible, however, that processes in the secretory granule are affected by Ser^B24 and Asp^B10 in concert with other perturbations. (i) Phe^B24→Ser. The invariant aromatic ring of Phe^B24 packs at the dimer interface. Its substitution by Ser^B24 impairs self-assembly (as monitored by gel-filtration) and leads to accelerated disassembly of the R₆ hexamer once formed (40). Receptor binding and biological activity are low. (ii) His^B10→Asp. The conserved imidazole ring of His^B10 coordinates the axial zinc ions at the trimer interface of insulin hexamers (47). Genetic variant Asp^B10 causes a diabetes syndrome characterized by baseline mutant proinsulinemia due to constitutive secretion (180) as the mutation perturbs specific trafficking to glucose-regulated secretory granules (34, 57). The corresponding substitution in insulin blocks both zinc binding and trimer formation in vitro (206, 207). Asp^B10-insulin exhibits increased affinity for both IR and IGF-1R with prolonged residence times in association with augmented mitogenic signaling (179, 208–211).

Given the ancestral history of metazoan insulin-like proteins over the past 540 million years and its broad radiation among diverse body plans (212–214), why might vertebrate proinsulins be susceptible to misfolding and lacking in mutational robustness? A possible answer is given by the history of the INS gene as traced by the late D.F. Steiner and coworkers (215–218). This seminal study characterized an insulin-like gene encoding an insulin-like protein (ILP) in an extant protochordate (amphioxus; Branchiostoma californiensis) ( Figure 9A ). The predicted polypeptide precursor pro-ILP contains a C-terminal peptide resembling the D and E domains of vertebrate IGFs, suggesting an intermediate form linking the ancestral proto-insulin gene with modern IGF genes. In accordance with this perspective, ILP is the only INS-like gene in amphioxus; its genome also contains a single gene encoding a putative insulin-IGF receptor (216) and a single gene encoding a putative IGF-binding protein (IGFBP) (220, 221).

ILP was thus proposed to combine the functions of insulin, IGF-I and IGF-II prior to the duplication the proto-insulin gene and specialization of distinct factors (215). Evolutionary changes in intron-exon structures are shown in greater detail in Supplemental Figure S1A . In this scheme conversion of a metazoan proto-insulin gene to ILP would have been effected by a nonsense-to-sense mutation at the end of the A-domain-encoding sequence in the putative proto-insulin gene; conversion of ILP to proto-lGF would have been effected by an upstream shift in the intron donor site into the B-domain-encoding exon. IGF-I and -II genes subsequently evolved from the posited proto-IGF by insertion of an intron into the E-encoding domain followed by gene duplication.

Thus, predating the divergence of insulin and IGFs as distinct gene products, ILP retains framework residues conserved among vertebrate insulins and IGFs, including the six canonical cysteines, Leu^B6, Leu^B11, Leu^B15, Val^B18, Leu^A16 and Tyr^A19—hotspots for MIDY mutations ( Table 1C ). Whereas mammalian insulins contain Leu^B17, however, residue B17 in ILP is Phe as in IGFs. Similarly, ILP residue A8 is Tyr, resembling the homologous Phe in IGF-I and IGF-II but unlike Ala^A8 or Thr^A8 in mammalian insulins (215, 218). ILP would not be expected to undergo insulin-like self-assembly: (a) it lacks a His at position B10 and so would not be expected to coordinate zinc ions; and (b) dimerization would be predicted to be impaired by ILP residues Ala^B12 and Ser^B26 (in place of Val^B12 and Tyr^B26) (183, 222). Representative vertebrate insulin B-chain sequences and IGF-I B-domain sequences are shown in Supplemental Figure S1B .

Given the evolutionary framework established by Steiner and coworkers (215–217), we hypothesize that the primordial insulin/IGF precursor protein folded as a heterodimer in partnership with a proto-IGFBP. Such heterodimeric folding occurs in vertebrate IGF-IGFBP systems (221, 223–226) and appears to compensate for the ambiguous refolding properties of IGFs in vitro (28, 29, 123, 130). We envisage that in heterodimeric folding requirements of foldability are relaxed in each partner (when considered in isolation). Crystal structures of human IGF-I/IGFBP complexes [illustrated in a representative case in Figure 9C (219)] exhibit extensive engagement of IGF surfaces adjoining disulfide bridges and sites of MIDY mutations. This model predicts that the foldability of pro-IGF variants in mammalian cells would be more robust to MIDY-like mutations than is proinsulin—but only in cells co-expressing one or more IGFBPs.

Proinsulin by contrast folds in the ER as an autonomous monomer, aided by chaperonins and oxidoreductases but not, to our knowledge, by specific proinsulin-binding proteins. Native zinc-mediated self-assembly of insulin ( Figure 9B ) can include proinsulin [which can form corresponding hexamers (69)], but such self-assembly occurs in secretory granules and not in the zinc-poor environment of the ER. Although IGFBPs do not bind to insulin or proinsulin, this model predicts that engineered proinsulin-binding proteins may be designed to enhance the foldability of WT and variant proinsulins. Although such artificial proteins are unlikely to find therapeutic application, they may be of interest as reagents to probe the mechanism of proinsulin folding in vivo, including steps susceptible to misfolding.

Arvan and colleagues have studied the ER folding of proinsulin in β-cell lines using a systematic set of variants that can form only one or two disulfide bonds; to this end, specific disulfide bridges were removed by pairwise mutagenesis (227). These constructs differed in biosynthetic properties and so provided probes of quality-control determinants. Their results demonstrated that cystines A20-B19 and A7-B7 (but not cystine A6-A11) are critical to enable native folding and ER exit. Prior biophysical studies of an insulin analogue lacking cystine A7-B7 (due to pairwise Ser substitution) demonstrated a more marked decrease in stability and chain-combination efficiency relative to analogous analogues lacking A6-A11 (60, 132). Further studies on single-chain insulin analogues (67) and IGF-1-related peptides and peptide fragments (29, 31, 60) provided evidence for a kinetic pathway in which pairing of cystine A20-B19 provides a required first step to stabilize a native-like molten mini-core (31). Native-like NOEs were observed in such a one-disulfide peptide model even in the absence of stable secondary structure (31). Together, these cell-based and in vitro studies suggest possible structural features that might be sensed by ER quality control: as a general principle, the more destabilizing the disulfide intermediate or isomer destabilizing, the greater the degree of exposed non-polar surfaces and in turn the intervention of detection and degradation by the quality-control machinery.

Whereas variant proinsulin polypeptides without interdomain disulfide bridges cannot be secreted (227), non-native disulfide isomers can accumulate and evade ER quality control (138, 140). Early work from the Arvan laboratory demonstrated secretion of mispaired disulfide isomers in cells using various single-chain insulin constructs (138, 140). Indeed, prior studies of IGF-I revealed that its oxidative refolding in vitro yielded two isoenergetic products (28, 123). Although these had similar α-helical propensities and thermodynamic stabilities, 2D ¹H-NMR spectra were remarkable for distinct well-dispersed patterns of chemical shifts, indicative of different three-dimensional structures (28). Unlike IGF-I and its disulfide “swapped” isomer, insulin disulfide isomers are less stable and less well-ordered than is native insulin (75, 76). The respective N-terminal segments of proinsulin and IGF-I contribute to such salient differences in the fidelity of disulfide specification and relative stability (228, 229). The chain asymmetry of non-Cys-related MIDY mutations—more in the B domain than in the A domain ( Figure 1 )—is consistent with a hierarchical disulfide pathway in which nascent structure in the B domain provides a structural template for folding of the A domain (95).

Mutations that impair the foldability of proinsulin (or efficiency of insulin chain combination (95) can nonetheless be compatible with high activity (108). This lack of correlation suggests that determinants of quality control in the ER differ from determinants of receptor binding. A prominent example is provided by substitution of invariant Leu^A16 by Val (156). This cavity-associated mutation (not [yet] seen among MIDY patients) markedly impairs both cellular folding of Val^A16-proinsulin and chain combination, and yet substantial biological activity is retained once the native state is reached (94, 156). Similarly, folding of Ser^B8-proinsulin is significantly reduced in vitro, yet IR affinity is similar to WT insulin (109). A recent study reported that substitution of Phe^B24 by Tyr (also not [yet] seen among MIDY patients) blocks cellular folding (40) whereas the corresponding two-chain insulin analogue retains substantial activity in a rat model of DM.

We imagine that insulin’s conserved side chains, as exemplified by Phe^B24, play different roles in the course of a complex conformational “life cycle.” If so, this would represent a marked compression of structural information within a short protein sequence. The present cryo-EM revolution promises to provide snapshots of structures through this life cycle, likely to be extended by solid-state NMR-based models of non-native insulin aggregates and fibrils. As we celebrate the Centennial of insulin’s discovery in Toronto in 1921 (1, 230)—and coincidentally the gold anniversary of its high-resolution crystal structure at Oxford in 1971 (161)—it is remarkable to appreciate how much remains to be discovered in relation to biosynthesis, folding, function and evolution. Further, in a related review article, J. S. Flier and C. R. Kahn have discussed how the discovery of insulin has defined a milestone in the history of molecular medicine that extends beyond the insulin molecule itself (231).

This review has focused on the structural lessons of the mutant proinsulin syndrome (4–6). Patient-derived experiments of nature are providing an opportunity to investigate biophysical principles at the intersection of cell biology and human genetics. As envisioned by classical diffusion-collision and framework models (232), folding of globular proteins (such as proinsulin) represent the coalescence of discrete subdomains (233, 234). Even as funnel-like energy landscapes make possible parallel events in folding (126), the existence of preferred trajectories (235) is implied by disulfide trapping studies of insulin-related polypeptides. Biophysical studies of these trajectories and equilibrium models promise to deepen a structural understanding of MIDY/MODY mutations. Sites of mutation reflect mechanisms of folding or misfolding that may not be apparent in the native state (9, 10, 19, 51). Many of the principles discussed here were foreshadowed in pioneering efforts toward the total chemical synthesis of insulin wherein specific disulfide pairing posed a key challenge to chain combination (89).

Foldability is an evolved property (236), highlighting the general threat of toxic misfolding as a hidden constraint in protein evolution. In the genetics of proteotoxic diseases these principles connect bench to bedside. Critical questions for continuing investigation include: can over-expression of the WT INS gene in response to peripheral insulin resistance likewise tax the folding capacity of the β-cell and induce ER stress analogous to that of the mutant proinsulin syndrome? Might structural mechanisms of misfolding due to MIDY mutations broadly inform a hidden landscape of toxic aggregation awaiting WT biosynthesis? A key frontier in molecular metabolism is thus defined by the role of the UPR and chronic ER stress in the progression of non-canonical Type 2 DM (9, 24, 25). Molecular dissection of how β-cells respond to the challenge of proinsulin overexpression (237) is of compelling translational interest as a strategy to arrest the progression of prediabetes to frank diabetes (9, 38, 238). Structural lessons of the mutant proinsulin syndrome may thus inform UPR-based approaches to mitigate the growing pandemic of diabesity.

MW oversaw preparation of the manuscript and wrote the first draft. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.