From structural biology to designing therapy for inborn errors of metabolism

Inborn errors of metabolism are model subjects to be studied by structural biology, due to two distinctive features. First, the predominant majority of disease-causing mutations are within the exonic region of the affected gene, rather than in the non-coding introns. Hence they are expected to affect the integrity (e.g. stability) and function (e.g. catalysis) of the encoded enzyme (Yue et al 2014), a largely globular conformation well suited for protein structure determination. Second, a large proportion of these mutations (e.g. ~65 % of all genetic variants for the metabolic disorders phenylketonuria, classical homocystinuria and ornithine transcarbamylase deficiency) are of the missense type that result in non-synonymous changes of a single amino acid, as opposed to insertion or deletion of larger peptide fragments. Missense changes are therefore more ‘subtle’ in character, context-dependent in function, and dictated by the physico-chemical properties (e.g. size and hydrophobicity) of the mutated amino acid. Structures of the wild-type and mutant metabolic enzymes determined at atomic resolution (i.e. with the details to resolve individual amino acid residues) prove useful in ascertaining the structural and molecular principles for the amino acid substitution in the local environment, as well as detecting mutation trends and hotspot regions on the protein polypeptide as a whole (Froese et al 2013; Shafqat et al 2013; Riemersma et al 2015).

Nevertheless, structural studies of disease-associated mutant proteins are often more intractable than their wild-type counterparts (Kang and Stevens 2009). This is because the mutation defects that impact on the native integrity and function of the enzyme often precludes its recombinant production in the quantity (milligrams) and quality (purity, homogeneity) that are required for the in vitro biophysical, biochemical or structural experiments. To complement this gap, a plethora of in silico prediction methods have been developed (e.g. FoldX, PolyPhen-2, SIFT and SNPs3d), adopting different algorithms of sequence, physico-chemical and structural information, to interpret the molecular effect and pathogenicity of amino acid substitutions (For reviews, see Stefl et al 2013; Studer et al 2013 and references therein). For the modern-day high-throughput pipeline in characterizing sequence variants, in silico approaches are more preferable to experimental approaches, as the latter are time-consuming and require significant resource for the design and setup.

In silico methods also prove particularly useful in characterizing those metabolic disorders with broad genotypes (i.e. large numbers of disease-causing alleles), such as phenylketonuria (MIM 261600), the most common inborn errors of amino acid metabolism, where more than 400 out of 834 disease alleles annotated in the locus-specific database PAHvdb (http://​www.​biopku.​org/​home/​pah.​asp) are due to missense changes. A general lesson from surveying such a large catalogue of mutations (Wettstein et al 2015) is that disease-causing mutations, as opposed to the harmless variants, are often accompanied by drastic changes in the physico-chemical properties (polarity, hydrophobicity, charge, side-chain geometry) and interaction networks (hydrogen bonds, salt bridges, hydrophobic interactions) of the amino acid involved. The mutated residues are also more commonly localized within the protein core and conserved functional sites, as reflected by their strong amino acid sequence conservation. The utility of in silico prediction tools is also exemplified in its application to triage a subset of all known disease mutations as a starting point for experimental characterization of mutant proteins. This is illustrated in the example of methylmalonyl CoA mutase (MUT), one of the two destination enzymes requiring the vitamin B12 cofactor for catalysis, where to date >130 missense mutations are found on this 750-amino acid polypeptide to cause methylmalonic aciduria (MIM 251000). From a mapping of all known MUT mutations onto the protein crystal structure (Froese et al 2010), we selected 23 of them representative of diverse exonic regions, clinical phenotypes and ethnic populations, and performed a series of biochemical and cellular studies that characterize the protein stability (recombinant expression level, differential scanning fluorimetry) and enzyme catalysis (activity assay in patient cells) (Forny et al 2014) (Fig. 1). Among the different molecular defects catalogued from these mutations, we found that a subset of mutants are thermally less stable than wild-type in vitro, and can be partly rescued in vitro and in vivo by exogenous supplementation of chemical chaperones (CC) such as glycerol, proline, betaine and trimethylamine N-oxide. These CCs are low-molecular-weight osmolytes that stabilise proteins, without directly interacting with them, by preferential exclusion at the protein surface, thereby altering the thermodynamic free energy of the protein in the solvent environment (Arakawa et al 2006). Together, this study provides a proof of concept that stabilizing mutant protein could be one therapeutic strategy to rescue its defective function.

The availability of high-resolution protein structures has since facilitated greatly the process of small molecule drug development in human medicine (Yue et al 2014). Even before its high-throughput advances, structural biology already played a role in the lead optimization step of drug development, in which structures of an established therapeutic target in complex with a drug candidate can direct the chemical alterations of the drug compounds to improve their affinity, potency and selectivity. Nowadays, the timeframe of protein structure determination has become sufficiently short, hence allowing its application in other stages of the conventional drug discovery pipeline. For example, protein structures can reveal distinguishing features across different members of an enzyme/domain family to explore specificity, detect binding pockets and evaluate their ‘druggability’ (i.e. how amenable a small molecule can bind to these pockets), guide compound screening campaign using biophysical/in silico assays, and develop structure-activity relationship for a chemical series of compounds.

When considering therapeutic possibilities, inborn errors of metabolism, like other genetic diseases, can be seen as a dysregulation of dosage for the associated protein (Beaulieu et al 2012). Depending on the particular disease, or even the disease allele, a genetic lesion can theoretically lead to a higher- or lower-than-physiological level of the associated protein that signify a gain-of-function (GOF) or loss-of-function (LOF) phenotype, respectively. Our systematic study of PAH and MUT mutations, as well as of other enzymes (McCorvie and Timson 2013; Balmer et al 2014; Burda et al 2015), all conform to the general concept that IEMs are by and large LOF diseases due to pathogenic mechanisms that effect the structure (e.g. misfolding, aggregation) and function (loss of catalysis, loss of interactions) of the enzyme. This therefore poses a conceptual challenge for drug discovery, since the most intuitive therapeutic target for an IEM (i.e. the metabolic enzyme harbouring a LOF mutation itself) implies an imperative to develop an activator of the deficient or defective enzyme (Segalat 2007). While the drug development industry has more traction in GOF diseases, by means of small molecule inhibitors aimed at reducing the mRNA, protein or activity levels, the road is much less travelled for developing a therapy to activate or upregulate the levels of mRNA, protein or activity as treatment for LOF diseases. With the explosion of genomic data and disease linkage from the advent of next-generation sequencing, novel therapy design and principles are urgently required for LOF diseases (Boycott et al 2013).

One emerging therapeutic approach for LOF diseases involves the use of small molecule ligands known as pharmacological chaperones (PCs) to stabilize and activate the mutant enzyme (Muntau et al 2014), with the rationale that a moderate increase in the mutant enzyme activity beyond a certain threshold level (e.g. 10 % of many lysosomal storage enzymes) could suffice to delay disease onset and ameliorate phenotypes (Suzuki et al 2009). To date, PCs have promising potentials for several IEMs, including the Fabry, Gaucher and Pompe Diseases, which have reached early-stage clinical trials (Boyd et al 2013; Parenti et al 2015); while others are gaining proof of concept (Santos-Sierra et al 2012; Jorge-Finnigan et al 2013; Makley et al 2015). These first-generation PC molecules often originate as substrate-mimetics or cofactor-mimetics of the target enzymes, and the field of structural biology has been useful in characterizing how the PC molecule interacts with the enzyme active site (Bateman et al 2011; Guce et al 2011; Torreblanca et al 2012; Suzuki et al 2014). However, active-site-directed PCs create a conundrum in which by binding to the active site, these ligands could potentially compete with the native substrate or cofactor of the target enzyme. Hence the counter-intuitive mode of action for these PCs could restrict their application to certain sub-inhibitory concentration and dosage window.

We propose that the next generation of PC molecules should target an enzyme’s domains or pockets that are distant from its active site, and therefore would not compete with the enzyme’s native ligands. These allosteric, non-catalytic binding sites are often specific to an enzyme’s unique structure and function, and necessitate structural and chemical biology approaches to help identification and validation. As a first lesson to design allosteric PCs, my group has the vested interest in understanding the molecular mechanism of naturally-occurring enzymes built with a modality for ligand-binding allosteric regulation. An example for such enzyme is cystathionine-β-synthase (CBS) (Fig. 2a), which catalyzes the conversion of homocysteine to cystathionine, a reaction up-regulated by the native ligand of CBS, S-adenosyl-L-methionine (SAM). Inherited mutations of the cbs gene lead to classical homocystinuria (MIM 236200), in which protein misfolding is the major cause of malfunction among the predominant alleles. In addition to a core catalytic domain harbouring the enzyme active site, CBS contains a regulator domain at its C-terminus which binds SAM in order to activate the catalytic domain (Pey et al 2013). We have applied crystallography, biochemical and biophysical methods to explain how CBS is allosterically activated by SAM, by showing that binding of SAM to the regulatory domain results in large conformational change that relieves its steric inhibition of the enzyme catalytic domain (McCorvie et al 2014). We therefore posit that targeting the CBS regulatory domain with a small molecule could be a strategy to activate mutant CBS enzyme using the same native mechanism, thereby rescuing disease alleles with defective enzyme activity.

CBS is not a lone example of allosteric activation via a multi-domain protein architecture (Jaffe and Lawrence 2012; Jaffe et al 2013). Similarly, the enzyme phenylalanine hydroxylase (PAH), of which genetic mutations lead to PKU, is activated by its own substrate phenylalanine (Phe), postulated to bind not only to the enzyme active site in the catalytic domain, but also to a separate regulatory domain (Zhang et al 2014) (Fig. 2b). We recently provided the first structural evidence that binding of Phe to the PAH regulatory domain causes this domain to dimerize and, like CBS, relieves its steric inhibition on the catalytic domain (Patel et al 2016). We therefore believe that screening for small molecule binders in allosteric sites such as the regulatory domains of CBS and PAH could be a common principle for drug discovery of LOF inborn errors of metabolism, at least in those enzymes with allosteric sites. There are emerging studies identifying non-inhibitory ligands in allosteric sites and cryptic pockets of metabolic enzymes, either by means of high throughput screening of large compound libraries (Marugan et al 2010; Patnaik et al 2012; Porto et al 2012) or by the more chemistry-efficient small molecule fragment approach (Landon et al 2009).

When determining the appropriate therapeutic targets for a metabolic disorder, it is often beneficial to look at the different reaction steps around the affected enzyme of a metabolic pathway, in order to maximize intervention possibilities (Fig. 3). As an example, normal glycogen synthesis is catalysed by the concerted action of three enzymes in humans: glycogenin (GYG), glycogen synthase (GYS) and branching enzyme (GBE). Andersen disease (MIM 232500) and adult polyglucosan body disease (APBD; MIM 263570) caused by GBE mutations, as well as Lafora disease due to mutations in malin or laforin proteins (MIM 254780), are all neurological diseases sharing a common neuropathology of malformed glycogen (‘polyglucosan’) accumulation, although their genetic defects affect different aspects of glycogen metabolism. We have recently applied structural biology to characterize GBE disease-causing mutations, in particular the prevalent allele causing APBD, p.Y329S. A homology model of the GBE mutant, based on the experimental wild-type crystal structure (Froese et al 2015), shows that the Tyr329-to-Ser substitution creates a cavity at the protein surface away from the active site, and results in lower protein expression in vitro and in vivo, and reduced protein stability. We reasoned that a small molecule specifically targeting this surface-accessible cavity could function as a ‘molecular strut’ to stabilize the local defect of this mutant enzyme. As proof of therapeutic principle, a tetra-peptide (Leu-Thr-Lys-Glu) designed computationally to fill this cavity can be taken up in patient cells, binds specifically to mutant GBE protein and rescues mutant enzyme activity to a moderate level (Fig. 3a).

The peptide approach above can be considered an example of pharmacological chaperone that is targeted directly at the site of mutation defect, distant from the enzyme active site. An alternative therapeutic avenue, however, can be based on substrate inhibition, with the rationale that the same biosynthesis enzymes for normal glycogen production are also responsible for polyglucosan formation. Recently, a gene knockdown of muscle glycogen synthase (GYS1) eliminated polyglucosan formation and restored neurological functions in a mouse model of Lafora disease (Pederson et al 2013), as well as a neuronal model for APBD (Kakhlon et al 2013). These clinical findings are further supported by the natural existence of very rare GYS mutations which lead to its enzyme deficiency (MIM 611556) with mild phenotypes in humans. Together, they provide proof of principle that downregulation of glycogen synthesis eliminates neurological abnormalities due to polyglucosan formation, and lend support to the inhibition of GYS1 enzyme activity, or its interaction with the functional partner GYG necessary for glycogen synthesis, as potential molecular therapy for Lafora and Andersen diseases, as well as for the lysosomal storage disorder Pombe (GAA mutations) whose hallmark is muscle glycogen accumulation. The inhibition approach is particularly attractive to the drug development industry, and can be considered as an example of ‘substrate reduction therapy’ (Fig. 3b) where there is precedence in IEMs (e.g. Miglustat for the treatment of Gaucher disease (Platt and Jeyakumar 2008)) to reduce the consequence of toxic metabolite accumulation due to defective enzymes (Schiffmann 2015). We anticipate that progress in the structural biology of GYS and GYG (Chaikuad et al 2011; Zeqiraj et al 2014) should pave the next step forward in developing novel inhibitors for these enzymes. This example therefore illustrates the principle of targeting different components of a metabolic pathway for the treatment of a genetic disease. As we enter the post-genomic era from the advent of next generation sequencing, with emerging new diseases (Ebrahimi-Fakhari et al 2016) and unmet medical need, there is an urgent call for novel, ‘imaginative’ routes of therapeutic intervention to tackle IEMs, and rare diseases as a whole. To this end, a recent example of functionally bypassing a deficient protein by a suppressor mutation (Yoon et al 2014) may represent one of many future therapeutic approaches aimed at activating an alternative gene, isoform or enzymatic mechanism (Fig. 3c) to functionally compensate for the defective gene and product.