Identification of pathogenic MMAB variants
This study presents information obtained from 97 individuals with confirmed cblB-type MMA, all of which were found to carry two disease-causing variants in the MMAB gene. A summary of the affected individuals and the identified variants are presented in Table 1, 2, respectively. In total we identified 33 different variants, of which 16 were novel, including the predicted missense variants c.380C > A (p.(Ala127Asp)), c.462G > T (p.(Glu154Asp)) and c.650G > T (p.(Ser217Ile)). Variants resulting in single amino acid substitutions accounted for the majority of identified changes (61.9%), while splicing and truncating changes were the next most frequent, constituting 18.6% and 18% of variants, respectively (Fig. 1a). Some changes occurred at a high frequency in the cohort (Table 2). These included the missense alleles p.(Arg186Trp) and p.(Arg191Trp) (57 and 19 alleles, respectively), and the splicing variant c.197-1G > T (22 alleles) (Fig. 1b). The most frequently identified truncating variant was c.700C > T (14 alleles), which results in a predicted premature termination of translation at peptide position 234 (p.(Gln234*)) (Fig. 1b). The majority of disease-causing variants identified resulted in predicted changes to the C-terminal half of the polypeptide chain (Fig. 1c). A particular hotspot was identified in exon 7, corresponding to amino acids 173–195, which contribute important residues to the binding sites of cobalamin and ATP (Fig. 1c and inset). These changes are likely to result in a direct impact on catalytic activity.
Clinical and biochemical cohort characterization
In addition to molecular genetic information, we assessed enzymatic function, in the form of propionate incorporation (PI) activity, from fibroblasts of 76 individuals (Table 1). PI activity was performed in the absence and presence of hydroxocobalamin (OHCbl) supplementation to the assay media. On average, PI activity in media supplemented with OHCbl was higher than in its absence, but for each individual both measures correlated strongly (Table 1, Suppl. Figure 1a).
The ratio of PI activity with and without OHCbl supplementation (PI ratio; Fig. 2a), typically using a cut-off of 1.5, has been used to discriminate between cells with residual activity that are responsive to OHCbl supplementation and those without (Fowler et al. 2008). Here, we found that individuals who were clinically responsive to vitamin B12 supplementation provided fibroblasts with a much higher PI ratio than those who were not, and the majority of clinically responsive individuals indeed had a PI ratio higher than 1.5 (Fig. 2b). Thus, in the context of MMAB deficiency, a PI ratio cut-off of 1.5 may be used to discriminate between cobalamin responsive and non-responsive fibroblasts. Accordingly, individuals whose fibroblasts were cobalamin responsive tended to have a lower maximal plasma ammonia at presentation (Fig. 2c). Finally, there was a clear positive correlation between PI ratio and the age at onset of presentation (Fig. 2d), which was further supported by the fact that individuals who presented with a later onset (> 30 days) had fibroblasts with a higher PI activity following OHCbl supplementation (Fig. 2e).
From a clinical perspective, there is an important interrelation between an early age at onset of presentation and the maximal measured concentrations of plasma ammonia and urinary methylmalonic acid, as all three parameters have been implicated in a worsened disease progression in MMA (Hörster et al. 2007; Kölker et al. 2015). For the individuals with available data in our cohort, we found that the latter two biochemical variables correlated positively with each other, while they were negatively correlated with age at onset (Fig. 2f), which showed two density peaks in our cohort at 2.5 and 300 days, respectively (Suppl. Figure 1b). These findings reinforce the predictive value of PI activity and the PI ratio, which are strongly correlated to these same clinical and biochemical parameters.
Functional impact of variant classes
Since PI activity and PI ratio correlate with important clinical parameters, we sought to use them to investigate the genotype–phenotype correlation of variant types and the presence of specific (frequent) variants found in our cohort.
Examination of variant classes as a function of PI activity with and without OHCbl supplementation reveals missense and truncating variants to distribute across the entire range of very low to very high PI activity and present as both OHCbl-responsive and unresponsive (Fig. 3a). Since missense changes have the potential impact range of very disruptive to mild alterations, depending on the nature and location of the substitution, for them such a finding might be expected. However, finding very high PI activity and OHCbl-responsiveness as a consequence of premature truncation is more surprising. This latter, however, appears to be driven by a single variant, p.(Gln234*), as discussed further below. In contrast to missense and truncating variants, splicing variants almost exclusively resulted in low to moderate PI activity.
We further determined the contribution of each variant class to OHCbl responsiveness by plotting the proportion of responsive and non-responsive cells according to variant type (Fig. 3b). Consistent with their low PI activity in the presence and absence of OHCbl, cells harboring splicing variants on both alleles (splicing/splicing) were found only in the non-responsive category. By contrast, a larger proportion of cells with the combination of missense/truncating alleles were found to be responsive than non-responsive. Nevertheless, cells containing two missense variants (missense/missense) or two truncating variants (truncating/truncating) were found to be responsive with a relative chance of about 50%.
Functional and clinical impact of frequent variants
We investigated the contribution of the most frequently identified alleles to PI activity (Fig. 3c) and responsiveness (Fig. 3d). For p.(Gln234*), the most frequently identified truncating variant (N = 14), cells that contained this variant in one or two alleles had medium to high residual activity and all were responsive to the presence of OHCbl. Correspondingly, 6/7 individuals harboring this variant, including all 3 in the homozygous state, from which we had clinical information, were described as clinically responsive to vitamin B12 (Table 1). Consistent with this dual responsiveness, two homozygous individuals showed a later onset (150 and 180 days) and one had onset at 2 days of age, while 4/5 individuals heterozygous for this variant showed a later onset (3, 90, 300, 820 days; one asymptomatic) (Table 1). The only other truncating variant associated with in vitro OHCbl responsiveness was p.(Tyr219Serfs*4), identified once in the homozygous state (Suppl. Figure 1c); all others were associated with non-responsiveness. Therefore, given its high frequency amongst our cohort, p.(Gln234*) was the source of the generally high PI activity and OHCbl responsiveness found in all cells with truncating variants and may be considered a cobalamin responsive variant.
By contrast, no clear dose-dependent allelic response was identified for the missense variant p.(Arg186Trp) (Fig. 3c, d), the most frequently identified allele in our cohort. It was detected in a homozygous state in fibroblasts from 18 individuals, which showed low PI activity and a median PI ratio of 1.4, corresponding to a weak or no response to OHCbl addition (Suppl. Figure 1d). Correspondingly, cell lines harboring this variant in one or two alleles were identified to be responsive approximately 50% of the time. Consistently, 7 of the 14 individuals containing this variant in one (3/7) or two (4/7) alleles were described to be clinically responsive to vitamin B12, while individuals carrying this variant in a homozygous state and of which information was available had mostly (4/5) early-onset of disease (Table 1). A different amino acid exchange at the same position (p.(Arg186Gln)) also resulted in non-responsiveness (Suppl. Figure 1e).
Similarly, fibroblasts from individuals homozygous for p.(Arg191Trp), the second most frequently identified missense variant, showed moderate activity with no clear indication of cobalamin responsiveness (Fig. 3c, d). Individuals homozygous for this variant had mostly (3/4) early-onset of disease, while a heterozygous individual (with p.(Arg190Cys)) and an individual homozygous for p.(Arg191Gln) also had early onset disease (Table 1).
In regards to the novel missense variants identified in this study, p.(Ala127Asp) was identified once each in the homozygous and heterozygous state, showing in vitro and in vivo response to cobalamin in the homozygous state (Suppl. Figure 1e, Table 1). p.(Glu154Asp) was identified twice in the homozygous state, and was not cobalamin responsive (Suppl. Figure 1e), while we unfortunately received no information about their clinical responsiveness (Table 1). Finally, p.(Ser217Ile) was identified in one homozygous individual, whose cells displayed low PI activity without cobalamin responsiveness (Suppl. Figure 1e), and was clinically not responsive to cobalamin administration (Table 1).
Molecular consequences of missense variants
To get a clearer indication of the potential molecular impact of the novel and frequent missense variants identified in our cohort, we mapped them onto the human MMAB protein structure (Fig. 4). MMAB assembles as a homotrimer with cobalamin and ATP binding sites at the subunit interfaces.
A top down view of MMAB (Fig. 4a), illustrates the positioning of Arg191 in the trimeric core, whereby the frequently identified substitution to tryptophan (p.(Arg191Trp)) would result in a loss of charge and change in side-chain size, which might be expected to result in protein instability. By contrast, Glu154 is a fully solute accessible outer edge residue. The novel identified substitution p.(Glu154Asp) results in a one carbon unit side-chain shortening, but conservation of negative charge. Structurally this appears to be quite a conservative change with no clear consequences, which belies its clinical severity and lack of cobalamin responsiveness (Table 1).
A side view of the MMAB protein, focusing on the AdoCbl (Fig. 4b) and ATP (Fig. 4c) binding pocket at the subunit interfaces, highlights the role of catalytic residues. Just outside of the active site are Ala127 and Val209, which are close enough (4.4 Å) to engage in hydrophobic (van der Waals) interactions (Fig. 4d). We found deleterious variants at both residues in our cohort, including the novel p.(Ala127Asp). These appeared to have a very similar effect on PI activity, and both were responsive to cobalamin supplementation (Table 1). Within the subunit interfaces, residues Arg186 and Arg190, the sites of the frequent deleterious substitutions p.(Arg186Trp/Gln) and p.(Arg190His), along with Glu193 and Arg194, substituted as p.(Glu193Lys) and p.(Arg194Ser) in our cohort, coordinate both AdoCbl and ATP. Ser217, site of the novel variant p.(Ser217Ile), supports this coordination, likely through polar interactions with the side-chain of Arg186 and cobalamin in the presence of AdoCbl (Fig. 4e), and polar interactions with Arg186 and Arg190 in the presence of ATP (Fig. 4f). Substitution to isoleucine at residue 217 would be expected to break these polar interactions, resulting in disruption of the active site. Such an explanation is consistent with the cobalamin non-responsive loss of function identified from an individual homozygous for (p.(Ser217Ile)).
To better understand how catalytic variants may affect enzymatic function, we performed a biochemical characterization of human MMAB.
Recombinant human MMAB was expressed as an N-terminally truncated protein, consisting of amino acids 56–250, which omits the mitochondrial leader sequence and incorporates the same residues represented in the recently solved human MMAB structure (Campanello et al. 2018). Final protein purity was > 95% as visualized by SDS-PAGE (Suppl. Figure 2a) with a native molecular weight of 78 kDa (Suppl. Figure 2b), as determined by size exclusion chromatography, corresponding to its biological assembly as a homotrimer (Schubert and Hill 2006; Campanello et al. 2018).
As a first characterization, we examined binding to AdoCbl using spectrophotometry. Titration of increasing concentrations of MMAB to AdoCbl revealed a transition from an absorption maximum at 525 nm, consistent with the “base-on” six-coordinate state of free AdoCbl, to an absorption maximum at 458 nm (Fig. 5a, b top), corresponding to the “base-off” five-coordinate state of MMAB-bound AdoCbl (Padovani et al. 2008). Curve fitting of the change in absorbance at 525 nm to a one-site binding model suggests a provisional dissociation constant (Kd) = 27.6 ± 2.0 µM (Fig. 5b top inset). However, Scatchard analysis indicates the presence of two non-equivalent binding sites, with Kd1 = 0.55 μM and Kd2 = 8.4 μM (Fig. 5b bottom). This latter analysis is more consistent with values published for the bacterial homologue ATR (Kd1 = 0.14 ± 0.02, Kd2 = 2.1 ± 0.5 μM, (Padovani and Banerjee 2009) and the recently published Kd = 0.96 ± 0.31 for MMAB (Campanello et al. 2018).
We next examined binding of ATP to apo-MMAB in two ways. First, we used spectroscopy of the fluorescent nucleotide analogue 2',3'-O-(N-Methyl-anthraniloyl)-adenosine-5'-triphosphate (MANT-ATP). Using titration of MMAB to MANT-ATP and monitoring the resulting change in fluorescence yielded a Kd = 21.1 ± 3.4 µM (Fig. 5c top). Binding specificity was confirmed by competitive exchange of MANT-ATP with unlabelled ATP (Suppl. Figure 3a). Second, as an orthogonal assay, we determined ATP binding by isothermal titration calorimetry. Following injection of ATP to MMAB we identified a Kd = 14.2 ± 0.6 µM (Fig. 5c bottom), comparable to that of MANT-ATP. The isothermal calorimetry measurement further identified an N = 0.968 ± 0.0032, indicating that the three potential ATP binding sites have equal affinity, while a negative ΔH suggested that binding is enthalpically driven.
Release of AdoCbl from MMAB can be initiated by binding to ATP (Padovani and Banerjee 2009). We monitored this release spectrophotometrically, by measuring the dose-dependent increase in absorption at 525 nm and decrease at 458 nm following addition of ATP to AdoCbl-bound MMAB (holo-MMAB) (Fig. 5d top). Curve fitting of the change in absorbance at 525 nm indicates ATP-induced released of AdoCbl has an activation constant (Ka) = 23.9 ± 4.8 µM (Fig. 5d top inset). This Ka is in line with the Kd of apo-MMAB for ATP (Fig. 5c), suggesting the affinity of MMAB for ATP is not significantly impacted by the presence of bound AdoCbl. Since the absorbance spectrum generated after AdoCbl release by ATP is comparable to the absorbance spectrum generated by free AdoCbl, we expect that the cofactor was fully released into solution.
Altered MMAB biochemistry in the presence of MMUT
We previously found that release of AdoCbl from MMAB upon ATP binding is favored in the presence of MMUT (Plessl et al. 2017). Here, we pre-incubated holo-MMAB with purified recombinant human MMUT (Suppl. Figure 2a), which retains its biological assembly as a dimer (Suppl. Figure 2b). Consistent with our previous findings, addition of ATP in the presence of MMUT resulted in a Ka of 13.3 ± 1.3 µM (Fig. 5d bottom inset), which is reduced compared to MMAB alone. Here, released AdoCbl was not free in solution, but instead bound by MMUT, as indicated by the additional absorbance peak at 565 nm (Fig. 5d bottom, Suppl. Figure 3b). We did not identify cofactor transfer from MMUT pre-loaded with AdoCbl (holo-MMUT) to MMAB (Suppl. Figure 3c). The direct binding of AdoCbl by MMUT, as well as the reduced Ka for ATP by MMAB in the presence of MMUT, suggests a direct transfer of AdoCbl from MMAB to MMUT, in line with findings from bacterial homologs (Padovani et al. 2008).