Structural and functional defects of mitochondrial serine hydroxymethyltransferase genetic variants responsible for a novel neurodevelopmental syndrome

1 Introduction

In 2020, a new clinical syndrome was described in five pediatric patients from four different families. These patients present a similar phenotype, consisting of several clinical manifestations: intellectual disability, microcephaly, corpus callosum anomalies, motor dysfunction in the form of spastic paraparesis, ataxia, and/or peripheral neuropathy, cardiac defects (mainly hypertrophic cardiomyopathy) and dysmorphic features (García-Cazorla et al., 2020). All patients carry biallelic serine hydroxymethyltransferase 2 (SHMT2) variants. Six out of the seven variants are single nucleotide polymorphisms (SNPs) whereas one is a deletion variant. The seven mutations are: P157S, T186R, G375del, N379D, G423S, Q435P and P499A (Figure 1A).

SHMT is a pyridoxal-5′-phosphate (PLP) dependent enzyme (EC: 2.1.2.1) that catalyzes the reversible transfer of the Cβ carbon atom from serine to tetrahydrofolate (THF), generating glycine and methylene-THF (CH₂THF) that acts as a one carbon (1C) unit carrier (Figure 1B). SHMTs play a key role in 1C metabolism, a series of interlinked metabolic pathways fueling crucial cellular functions, such as purine and thymidine synthesis, amino acid and neurotransmitter metabolism, redox homeostasis. 1C metabolism can generate and consume redox equivalents, and as such it is linked to the respiratory state of both the cell and the mitochondria. These pathways, while connected, are highly compartmentalized; with cytosolic and mitochondrial pools differing in the distribution of enzymes and folate species (Cutruzzolà et al., 2021; Ducker et al., 2017; Monti et al., 2021). Human SHMT1 and SHMT2 are isozymes with 66% sequence identity. SHMT1 is localized in the cytosol, while SHMT2 is localized in the mitochondria. SHMT2 also presents a splicing variant (SHMT2α) that is localized in the cytosol. SHMT2 and SHMT2α share the same sequence, with the only difference being the absence of the 21 N-terminus residues representing the mitochondrial import sequence in SHMT2 (Girgis et al., 1998). Both SHMT1 and SHMT2α can undergo SUMO-dependent translocation into the nucleus, where they bind dihydrofolate reductase (DHFR) and thymidylate synthase (TYMS) in a ternary complex that provides thymidylate for DNA repair and synthesis (Anderson et al., 2012a; Anderson et al., 2012b; Giardina et al., 2015; 2018; MacFarlane et al., 2011; Spizzichino et al., 2022a).

Unlike bacterial SHMTs, which are dimeric and in which the dimer represents the minimal catalytic unit (Florio et al., 2009), eukaryotic and plant SHMTs are tetrameric enzymes organized as dimers of dimers (Figure 1C). The PLP cofactor is bound to the active site through a Schiff base linkage with a catalytic lysine and interacts with specific loops from both subunits of the dimer. These same loops plus an additional one are also involved in folate binding. The higher oligomeric state has recently been associated with the ability of eukaryotic SHMTs to perform moonlighting functions as RNA-binding proteins and, at the same time, to act as riboregulated metabolic enzymes (Guiducci et al., 2019; Liberati et al., 2025; Monti et al., 2021; Spizzichino et al., 2024).

The scheme in Figure 1C illustrates the complex conformational and oligomerization equilibrium of SHMT2 that is strongly dependent on PLP binding. In the absence of PLP, the active site region is largely unstructured, and the dimers can adopt an “open” or a “closed” conformation. The open dimer is not catalytically active, and the tetramer can only be formed by two closed dimers. PLP binding promotes a disorder-to-order transition stabilizing the closed dimer and, consequently, the tetrameric state. The tetrameric assembly is stabilized by an extensive network of intersubunit interactions, with His158 acting as a central interaction hub at the dimer–dimer interface (zoom in Figure 1C). On the other hand, mutations that destabilize either the tetrameric or the dimeric interfaces are supposed to shift the equilibrium towards the open dimer, thereby lowering PLP affinity (Giardina et al., 2015). Similar conformational transitions have also been observed in other dimeric PLP-dependent enzymes, such as aromatic amino acid decarboxylase (AADC) and glutamate decarboxylase (GAD) (Giardina et al., 2011; Kass et al., 2014). However, in the case of SHMT2 the oligomerization equilibrium further complicates the conformational landscape.

This wide range of allowed conformational states endows SHMT2 with a plethora of non-canonical (or non-catalytic) moonlighting functions (Cao et al., 2019; MacFarlane et al., 2011; Paone et al., 2014; Walden et al., 2019; Zheng et al., 2013). Although the main clinical features of this novel developmental syndrome have been associated with defects in SHMT2 catalytic activity (mitochondrial dysfunction and/or altered 1C metabolism (García-Cazorla et al., 2020), these non-canonical functions may also play an important role that has been overlooked so far. Besides the already mentioned RNA binding and riboregulation, it is now clear that SHMT2, in the open dimer conformation, participates in the immune response. It promotes inflammatory cytokine signaling by interacting with the deubiquitinating BRCC36 isopeptidase complex (BRISC) (Walden et al., 2019; Zheng et al., 2013). This interaction can be promoted by fatty-acylation of K245 and regulated by HDAC11 deacylase (Cao et al., 2019). The open conformation can also undergo acetylation of the PLP binding lysine, that has been linked to colorectal carcinogenesis (Wei et al., 2018). Finally, SHMT2 has been identified, through a yeast two-hybrid screen, as potential interactor of the intrinsically disordered intracellular domain of KIRREL3, a synaptic cell-adhesion molecule belonging to the Immunoglobulin family (Liu et al., 2015). Defects of KIRREL3 have been linked to several neurological and cognitive disorders (Bhalla et al., 2008; Taylor et al., 2020).

Here we present a functional and structural analysis of the seven SHMT2 pathogenic variants described in the newly identified pediatric syndrome. For six of them (Q435P proved to be poorly soluble) we determined: i) the residual catalytic activity compared to wild-type; ii) the oligomerization state in solution; iii) the PLP affinity and iv) the thermal and/or chemical stability. We then solved the crystal structure of the three variants that showed both reduced activity and structural defects. Our results shed light on the structural determinants of the functional impairment of SHMT2 variants and uncover three new conformational states indicating that the structural flexibility of the enzyme is even higher than expected.

2 Results and discussion

2.1 All recombinant SHMT2 variants are soluble and folded except for Q435P

All seven mutants were expressed and purified following the same protocol used for the wild-type protein (Giardina et al., 2015). Since the oligomerization state of SHMT2 depends on an equilibrium between dimeric and tetrameric species, with tetramer formation stabilized by PLP binding (Giardina et al., 2015), an excess of PLP was included in the buffer throughout purification and protein storage to promote protein stabilization in solution. Under these conditions, Q435P was the only variant that was poorly soluble and unstable, showing a marked propensity to aggregate and precipitate (data not shown), and was therefore excluded from further biochemical analyses. P157S and T186R formed a white precipitate during the desalting steps of purification, indicating that loss of PLP promotes denaturation and aggregation for these variants. N379D formed a yellow precipitate, but this was likely due to high protein concentration and not to an intrinsic instability. By contrast, the variants G375del, G423S and P499A were fully soluble and stable. The results and yields of protein purification are summarized in Supplementary Table S2. All six variants showed the spectroscopic features typical of PLP-containing proteins, corresponding to the internal aldimine form of PLP, with two peaks at approximately 330 and 420 nm. However, the PLP signal was consistently lower than that observed for the wild-type protein, indicating an increased tendency to dissociate the cofactor (Supplementary Figure S1).

Differential scanning fluorimetry (DSF) was used to compare the thermal stability of the six variants with that of the wild-type protein (Figure 2). In a DSF assay the temperature is gradually increased, leading to protein denaturation; as this happens, a dye (SYPRO orange) binds to the newly exposed hydrophobic side chains, emitting a strong fluorescence signal that can be monitored in the visible region. Three mutants, P157S, T186R and G375del, displayed anomalous fluorescence profiles, with no evident transition from the folded to the unfolded state, and could not be reliably assessed using this technique (Figures 2A,B). This behaviour is not uncommon for DSF assays and may depend on favourable interactions between the dye and exposed disordered regions present in the folded state. Among the three variants displaying a typical DSF profile, N379D and P499A showed alterations in stability that, however, remain broadly comparable to the wild-type (ΔT_m: N379D = - 2.5 °C; P499A = + 1.3 °C). G423S showed a slightly different denaturation profile and was marginally more stable than the wild-type protein (ΔT_m = + 3.2 °C) (Figures 2A–C). We then performed far-UV circular dichroism (CD) thermal denaturation experiments to assess the stability of the mutants that could not be measured by DSF (P157S, T186R, G375del) and to confirm the higher stability of G423S with respect to the wild-type. CD thermal denaturation yielded apparent melting temperatures (T_m-app) of approximately 66 °C for SHMT2-WT, whereas all three mutants showed a marked decrease in thermal stability, with T_m-app values approximately 12–14 °C lower than that of the wild-type protein for P157S and T186R (54 °C and 52 °C, respectively) and 6 °C lower for G375del (60 °C) (Figure 2D). On the other hand, the T_m-app of G423S (70 °C) was 4 °C higher than that of SHMT2-WT, confirming the DSF result.

2.2 Catalytic activity assay shows that the six variants are either active, mildly impaired, or inactive

SHMT2 catalyzes the PLP-dependent reversible interconversion of L-serine to glycine, coupled to the conversion of tetrahydrofolate (THF) into methylene-THF (CH₂-THF) (Figure 1). We compared the serine-to-glycine conversion activity of the wild-type enzyme with that of the pathogenic variants. Catalytic activity was determined by measuring the initial velocity of the serine to glycine conversion using a coupled enzymatic assay based on NADP⁺-dependent methylenetetrahydrofolate dehydrogenase from Escherichia coli (FolD). This is a bifunctional enzyme that oxidizes the CH₂-THF produced by SHMT2, yielding 10-formyltetrahydrofolate (CHO-THF), with the concomitant reduction of NADP⁺ to NADPH. The formation of NADPH can be monitored spectroscopically at 340 nm (Figure 3A). FolD was used in excess to ensure that the SHMT2-catalyzed reaction remained rate-limiting; serine was used at saturating concentration whereas THF could not be used in excess because of substrate inhibition. Therefore, it was added at the concentration that allows the enzyme to work at the apparent V_max as already reported (Tramonti et al., 2018). Using the initial velocity of SHMT2-WT as reference, the pathogenic variants can be classified in three groups: i) active variant, such as P499A that exhibits an enzymatic activity comparable to that of the wild-type; ii) mildly impaired variants, G375del, N379D and P157S which retain more than 45% of wild-type activity; iii) highly impaired or inactive variants, G423S that retains only 20% activity with respect to the wild-type and lastly, T186R for which no enzymatic activity could be measured, even when tested at ten-fold higher concentration than the other mutants (data not shown; Figure 3B).

2.3 Impairment of the oligomeric state and loss of PLP affinity in the variants P157S and T186R are responsible for partial or total inactivity

As previously detailed, P157S has been characterized as mildly impaired variant while T186R is an inactive variant. The two residues are located at key oligomeric interfaces: P157 at the tetrameric interface and T186 at the dimeric interface (Figure 1A). Their substitution was therefore expected to affect the oligomeric equilibrium and/or the ability of the enzyme to bind the PLP cofactor. To qualitatively assess both PLP retention and the oligomeric state of the two variants in solution, we performed semi-analytical size-exclusion chromatography (SEC) using a PLP-free running buffer. Elution was monitored at 280 nm to detect the protein and at 420 nm to detect protein-bound PLP, allowing a qualitative evaluation of cofactor association with the eluting species. P157S eluted predominantly as a PLP-bound tetramer, although the broad 280-nm peak and its tailing toward higher elution volumes suggest the presence of a minor fraction of dimeric protein (Figure 4A). The 280/420-nm signal ratio further indicates that P157S retains most of its bound PLP, despite a modest reduction compared to the wild-type protein. To evaluate whether the oligomerization equilibrium of this mutant is concentration-dependent, we performed the same analysis starting from a four-fold diluted sample. In these conditions (5 µM protein) the peak corresponding to the dimer was the most abundant species (Figure 4B). The 280/420-nm signal ratio of the dimeric species was nearly double that of the tetramer (7.6 vs. 4.0), indicating that the dimeric assembly has a lower PLP saturation. Indeed, when we analyzed by analytical ultracentrifugation (AUC) the oligomeric state of a 15 µM P157S sample stored at −20 °C for 3 months only the dimeric form was observed (Supplementary Figure S2). This result may be explained by a tendency of this mutant to lose PLP favoring an irreversible dissociation of the tetramer. Consistently, PLP-binding measurements showed that P157S has a reduced affinity for the cofactor, with a K_d of 1.36 ± 0.05 µM, approximately 10.5-fold higher than that of SHMT2-WT (K_d = 0.13 ± 0.05 µM) (Figure 4C). In contrast, T186R eluted mainly as a dimer, also at higher protein concentration, and largely as an apo-protein, retaining only about 35% of the PLP-associated signal observed for the wild-type protein (Figure 4A). This oligomeric state was further confirmed by AUC (Supplementary Figure S2). In agreement with this behavior, T186R displayed a much stronger impairment in PLP binding affinity, with a K_d of 5.50 ± 0.04 µM, approximately 42-fold higher than that of SHMT2-WT (Figure 4C). The altered oligomeric equilibrium and reduced PLP affinity of P157S and T186R are likely associated with the observed decreased thermal stability (Figure 2D). Collectively, these data indicate that the conformational equilibrium of both P157S and T186R is shifted toward less stable dimeric species, with T186R showing the strongest tendency to adopt the dimeric state in solution and to lose the PLP cofactor.

The crystal structure of T186R and P157S confirmed and integrated the biochemical data indicating that an impaired oligomeric assembly is the main defect of these two variants. SHMT2-WT crystallizes as a tetramer in the hexagonal P6522 space group. However, in the absence of PLP (apo form) it can crystallize as an open dimer in the monoclinic C121 space group (Giardina et al., 2015). The distinctive features of the latter conformation are the loss of structure of the regions interacting with PLP, which become disordered in the absence of the cofactor (Figure 1C). Therefore, to favor the tetrameric oligomeric state, all the mutants were crystallized in the presence of PLP in the protein buffer. Even though the SEC and AUC analysis showed that in solution T186R is mainly dimeric and P157S can populate both the tetrameric and dimeric conformations, the two mutants crystallized in both the space groups. This indicates that these variants display an oligomeric equilibrium in solution between the tetrameric and dimeric forms, although with very different relative populations. Interestingly, for T186R, both crystal forms could be observed in the same crystallization drop (Supplementary Figure S3A). These results can be explained by the crystallization process acting as a conformational selector. During the phase transition the different crystal packing requirements thermodynamically favor either the open dimer or the tetramer incorporation into the monoclinic C121 or the hexagonal P6522 crystal lattice, regardless of the relative abundance of these species in solution.

The structures of T186R, in the open dimeric form and in the tetrameric form, were solved at 2.2 Å and 3.1 Å resolution, respectively. The open dimer displays a conformation similar to the open dimer of SHMT2-WT in the apo form shown in Figure 1C (PDB 6DK3; unpublished and Giardina et al., 2015) and is not very informative as the mutation falls within a disordered region (Supplementary Figure S3). On the contrary, the tetrameric structure reveals the conformational effects of the mutation. The replacement of threonine with the large arginine sidechain creates a steric hindrance that causes the unfolding of the PLP interacting loop (residues 165 to 186, yellow in Figure 1C). As a consequence, the cofactor is not visible and the regions surrounding the active site are highly disordered (Figures 5A,B). The dimeric interface is reduced, and only half of the tetrameric interface is retained. The tetramer is obtained by the interaction of two semi-open dimers (the distance between the H158 of adjacent subunit is 12 Å, whereas in the open dimer structure of T186R variant is 27 Å). Since the closed dimer conformation cannot be achieved, the tetramer is formed by the interaction of two subunits (A and C) belonging to different dimers (Figure 5C). The other two subunits, B and D, are not in contact and this increases the degree of disorder, extending the regions that are not visible in the electron density. Additionally, subunits B and D appear to have a more dynamic C-terminus domain, and no electron density is observed for the folate binding loop (residues 414 to 420, green in Figure 1C), whereas this region is visible and structured in the partner subunits A and C.

The structures of P157S, both in the dimeric and in the tetrameric form were solved at 2.7Å and 2.6Å resolution, respectively. As for T186R the open dimer is not very informative, being similarly disordered as the apo-wild-type enzyme (Supplementary Figure S4). In the tetramer the mutation affects the local conformation of the protein backbone. The carboxyl group of H158 is flipped changing the position of the sidechain that is no longer able to engage in the cluster of aromatic stacking and H-bond interactions central to the tetrameric interface (Figure 6A). The mutation also affects the correct assembly of the dimeric interface and results in an unstructured PLP binding loop (no density for residues 167–181). This structural effect is due to the change in the backbone conformation in the region of the mutation that shifts the position of an adjacent α-helix (residues 184–189) of about 1.2 Å (Figure 6A). Although modest, this shift is incompatible with the native conformation of the PLP-binding loop that is forced to unfold. Interestingly, the cofactor position is shifted in a relaxed conformation. The pyridine ring of PLP is tilted by 90° and the phosphate group flips to the opposite side and interacts with R425 (Figure 6B). In the native conformation, this arginine residue provides a crucial interaction to stabilize the external aldimine formed when the enzyme binds to the amino acid substrate. Therefore, it is likely that this represents an auto-inhibited conformation. To the best of our knowledge, this PLP tilt was never observed by X-ray crystallography and was only very recently reported in the cryo-EM structure of a mutated SHMT isoform from soybean (Samarakoon et al., 2026). Finally, as observed for T186R, for the P157S variant as well only half of the tetrameric interface is conserved with the two subunits B and D failing to properly interact and the C-terminus folate binding loop (residues 413–419) is also unstructured (Figures 6C, D).

As P157S retained almost 50% of residual activity (Figure 3B), it must be assumed that the inhibited and partially disordered conformation observed in the crystals is not the only one present in solution and that this variant can also be present in an ordered and catalytically active form.

2.4 The paradox of G423S: High stability and low activity

G423S is located approximately 20 Å from the PLP-binding region. The data obtained for this mutant were particularly intriguing since it retains only about 20% of wild-type activity (Figure 3), yet it did not show any evident structural defect: it eluted as a tetramer (Figure 7A), and its spectroscopic features (UV-vis and CD) were comparable to SHMT2-WT (Supplementary Figure S1). Surprisingly, for an inactive variant, G423S showed increased thermal stability in DSF and CD experiments (Figure 2). To further explore this unexpected behavior, we performed chemical denaturation experiments by CD spectroscopy. As shown in Figure 7B this mutant was more stable than SHMT2-WT, with a decrease in the extrapolated Gibbs folding free energy of ΔΔG°_D-N = − 1.71 ± 0.13 kcal/mol (SHMT2-WT ΔG°_D-N = - 6.01 ± 0.09 kcal/mol; G423S ΔG°_D-N = - 7.72 ± 0.09 kcal/mol). These observations become even more puzzling when we determined the affinity of G423S for PLP. The mutant showed a K_d for the cofactor one order of magnitude higher than the wild-type (1.48 vs. 0.13 µM) (Figure 7C). AUC analysis was thus performed to check whether this mutant was also undergoing an oligomerization change over time as observed for P157S, but G423S resulted stably tetrameric (Supplementary Figure S2). The reduced PLP affinity may provide an explanation for the impaired catalytic activity of G423S, although the kinetics were performed in the presence of a 1:2 excess of PLP to stabilize the holo-enzymes. This result is unexpected also because the mutation increases the overall stability of the protein and is located far from the active site. Moreover, typically in SHMT a reduced PLP binding affinity correlates with a higher degree of structural disorder and a reduced stability (as observed for T186R and P157S) and not vice versa. The behavior of this mutant seems to unveil the constant trade-off between function and structural stability that is typical of protein evolution (Dindo et al., 2022).

To decipher these contrasting results, we solved the crystal structure of G423S mutant at 1.8 Å resolution. While the tetrameric core remains intact, the mutated side chain introduces a steric effect on C412 that shifts the position of its Cα (Figure 8A). In SHMT2-WT the side chain of this cysteine residue points towards G423. In the mutant, the serine side chain would clash with the native conformation of C412. This apparently insignificant conformational change results in a “domino” effect that alters the whole conformation of an important allosteric loop involved in folate binding (see Figure 1C). In the mutant, the residues from 413 to 420 are not visible in the electron density indicating that they are disordered (Figures 8A–C). From a structural point of view, since G423S is more stable than the wild-type, this may indicate that in the native state the folate binding loop is in a frustrated conformation (i.e. a sub-optimal energetic conformation). Conformational frustration is often observed in protein regions that are involved in substrate binding, catalysis or allosteric regulation (Gianni et al., 2014; 2021; Giardina et al., 2015). The increased stability of the mutant may be a result of the lower frustration of the folate binding loop that becomes disordered. On the other hand, since this loop is positioned at the entrance of the active site, the disorder of the loop also results in a more solvent-accessible active site, which may explain the loss in PLP affinity. Finally, the impaired folate binding is also likely responsible for the reduced activity of G423S. It is likely that THF, lacking some of the native interactions, binds in a suboptimal conformation. However, the same interactions also serve to bind CH₂-THF in the reverse reaction (i.e. the conversion of glycine to serine), which should therefore also be impaired. To support this hypothesis, we measured the reverse reaction for this mutant, which was drastically reduced, similar to the forward reaction (from serine to glycine) (Figure 8D).

2.5 Characterization of the most active variants: P499A, N379D, G375del

The location of these residues is shown in Figure 1A: P499 is located at the C-terminus interacting with the N-terminus of the other subunit of the dimer; N379 and G375 are also positioned in the C-terminus domain (residues 343–501) and at the interface with the PLP binding domain (or large domain: residues 75–342). All these three variants eluted in the tetrameric state in the SEC analysis (Figure 9A). Two of them - P499A, N379D - displayed a thermal stability comparable to the wild-type, whereas G375del was slightly less stable (Figure 2). Of these three variants N379D and G375del both displayed 65% of residual catalytic activity, while P499A was fully active (95%) (Figure 3). To better investigate if the P499A mutation may affect catalytic efficiency, substrate binding, or cofactor dynamics, we calculated the kinetic parameters for the forward reaction (serine to glycine) (Figure 9B). No significant difference was found between the K_M (both for serine or THF) and the k_cat values of P499A (0.53 mM; 33 μM; 790 min^-1, respectively) with respect to the known wild-type parameters (0.40 mM; 53 μM; 850 min^-1, respectively) (Tramonti et al., 2018). Surprisingly enough, P499A, that is active and stable as the wild-type, displayed an affinity for the cofactor 8-fold lower than SHMT2-WT (K_d P499A = 1.05 ± 0.07 μM), whereas G375del and N379D slightly lower than the wild-type (K_d G375del = 0.34 ± 0.08 μM; K_d N379D = 0.44 ± 0.03 μM) (Figure 9C). Notably, for these three mutants the loss in cofactor affinity inversely correlates with their thermal stability and activity.

We could not determine the crystal structure for these three mutants; therefore, the mechanistic details of their structural defects cannot be easily explained. Indeed, it must be noted that, apart from P157S and T186R, all the other mutations are in the C-terminus domain of SHMT2. It was demonstrated that the dynamics of this domain affect the allosteric regulation of the fold-type I PLP-dependent enzymes (Paiardini et al., 2017; Spizzichino et al., 2022b; Tran and Brown, 2022). In particular, this domain is crucial in nucleic acid binding by SHMT that was shown to determine its moonlighting and riboregulation properties (Fu et al., 2003; Owuocha et al., 2024; Spizzichino et al., 2024). Moreover, as observed for the variant G423S, small changes in the structure or dynamics of the C-terminus domain of SHMT can have a great influence on the active site accessibility or the PLP environment, and consequently the catalytic activity. Previous molecular dynamics (MD) calculations, indicate that these three mutations have the following effects: G375del, altered active site shape, indirectly affects the ligand interaction pocket; N379D, altered active site making it more spacious; P499A, altered active site shape, local instability and dimer interface alteration (García-Cazorla et al., 2020).

Although highly speculative, based on our data and on the previous MD calculations, the following hypothesis may be formulated for each mutation:

N379D: Asparagine 379 side chain is buried in a neutral/hydrophobic pocket in close contact with two His residues (H380 and H254) that are part of a specific hydrogen-bonding network that stabilizes local secondary structures (helices/loops) near the active site (Figure 9D). Introducing a negative charge into a packed environment likely affects the histidine residues position and breaks the native interaction network. The mutated side chain is at H-bond distance from H380 and may shift its position or it may flip towards R359 making a salt bridge. However, the overall structural perturbation is mild. The mutant retains an oligomeric state and thermal stability similar to the wild-type, 65% of the activity and has a PLP affinity that is only 3.3 times lower than SHMT2-WT.

G375del: Glycine 375 is located in the solvent exposed region also harboring N379 (Figure 9D). Deleting a residue shortens the polypeptide backbone loop and pulls neighboring residues out of their native position. This mutation certainly disrupts the native interaction between the carbonyl oxygen of S374 and R230. Therefore, this mutation is expected to have an effect similar to that of N379D, disrupting the geometry of the nearby active site cavity, altering the precise positioning required to accommodate either the PLP cofactor or the substrate. Indeed, apart from the thermal stability that is lower for G375del, all the other measured parameters or properties (oligomeric state, residual activity, K_d for PLP) are similar for these two mutants.

P499A: is involved in a domain-swapping interaction between the C- and N-terminus region of opposite subunits that is critical for the dimeric interface (Figure 9E). Since neither the tetrameric assembly nor the thermal stability are affected by the mutation, it is likely that the substituting alanine has a minor impact on the dimeric interface but probably introduces an unexpected conformational flexibility that alters the dynamic properties of the whole C-terminus domain. Given that the catalytic activity is also unaffected while PLP affinity is strongly reduced, the final effect of this mutation must be an altered dynamic of active site access. Since the mutant retains the ability to efficiently convert the substrates, the structural environment of the PLP pocket is likely conserved. However, the high K_d for PLP binding may be explained by an increased dissociation rate (k_off) due to a widening of the active site access or by decreased association rate (k_on) due to a restriction of the access. Given the activity and stability values of this mutant, the second hypothesis is more likely. It must also be considered that this mutant does not dissociate the cofactor during the SEC elution as observed for P157S, despite the two mutants having a similar K_d for PLP, further supporting the hypothesis that for P499A the high K_d may be due to a reduced k_on rather than an increased k_off. Further experiments will be required to validate this hypothesis.

3 Concluding remarks

In this systematic study, we sought to rationalize the effects of seven SHMT2 mutations on the structure and function of the enzyme. A summary of the key findings is reported in Table 1. The impact of two variants, T186R and P157S, could be readily attributed to alterations at the dimeric or tetrameric interface. These findings reinforce the notion that the minimal catalytically active assembly of SHMT is the closed dimer, with the inability of T186R to adopt this conformation rendering the enzyme completely inactive. In contrast, structural analysis of the P157S tetramer revealed that the PLP cofactor readily adopts a relaxed conformation when increased flexibility within the active-site pocket permits structural rearrangement. This observation suggests that the cofactor of SHMT is constrained in a frustrated conformation that is essential for catalysis, as previously observed for human alanine:glyoxylate aminotransferase (Giardina et al., 2017). Although the concept of conformational frustration is now well established in protein science, directly visualizing this phenomenon, as observed for P157S, provides compelling structural evidence for its functional importance.

TABLE 1. Summary of the key properties of the SHMT2 variants.

MD, results are reported from (García-Cazorla et al., 2020)

Apart from these two variants, all of the remaining mutations cluster within the C-terminal domain. Our results demonstrate that this region plays a pivotal role in shaping the conformational landscape of the active-site cavity and regulating substrate access. Even subtle perturbations of domain dynamics or local structure are sufficient to compromise protein stability and catalytic activity. The intriguing cases of G423S, which exhibits increased thermal stability despite a marked loss of catalytic activity, and P499A, which retains native-like stability and activity despite an approximately eightfold reduction in PLP affinity, indicate that additional studies will be required to fully elucidate the gatekeeping role of the C-terminal domain in SHMT2 allosteric regulation. In this respect, it is noteworthy that recent studies have shown that SHMT2 RNA binding is mediated by both the C-terminal domain and the tetrameric assembly, further highlighting the multifunctional role of this region beyond catalysis. The Q435P variant deserves particular attention. This was the only mutant that could not be fully characterized because of its extremely low expression in the soluble fraction and pronounced tendency to aggregate and precipitate during the various purification steps. This behavior may indicate that the mutation compromises the overall stability and/or proper folding of the protein, thereby exposing hydrophobic or disordered regions that promote aggregation. Residue Q435 is located within a short loop connecting two alpha-helices, in close proximity to P499 (Figure 9E). In the native enzyme, Q435 forms an important hydrogen bond with the backbone carbonyl group of W43 in the N-terminus region of the partner subunit within the dimer. Replacement of this residue with proline is expected to disrupt this interaction while simultaneously altering the local backbone conformation, leading to local unfolding of this critical region of the dimeric interface. However, further experimental data will be necessary to clarify the structural perturbations that are responsible for the reduced solubility and instability of this variant.

Beyond providing mechanistic insight into the biochemical properties of individual variants, our study establishes a structural framework for understanding the molecular basis of SHMT2-associated disease. Pathogenic variants in SHMT2 have recently been identified as the cause of a severe developmental syndrome characterized by brain and heart abnormalities, underscoring the essential role of mitochondrial one-carbon metabolism during embryonic development. Our findings extend these clinical observations by demonstrating that disease-associated SHMT2 variants impair enzyme function through multiple, distinct molecular mechanisms, including defective oligomerization, altered active-site conformational dynamics, impaired PLP binding, and compromised protein folding and stability. Rather than converging on a single structural defect, these mutations perturb different elements of enzyme architecture that ultimately lead to reduced catalytic competence. Overall, this work highlights the intricate interplay between quaternary structure, cofactor organization, protein dynamics, and conformational stability in governing SHMT2 function. The identification of distinct structural mechanisms underlying the loss of enzymatic activity provides a molecular rationale for the pathogenicity of SHMT2 missense variants and offers a framework for interpreting newly identified variants emerging from clinical sequencing studies. More broadly, these findings emphasize how subtle perturbations of protein conformational landscapes can translate into profound functional defects, contributing to the molecular pathology of disorders associated with mitochondrial one-carbon metabolism.

4 Materials and methods

4.1 Mutagenesis

The seven mutants P157S, T186R, G375del, N379D, G423S, Q435P, P499A were obtained by QuikChange Site-Directed Mutagenesis Kit (Agilent), with proper primers produced by Eurofins Genomics (Germany) (Supplementary Table S1) from wild-type SHMT2α (isoform 3, UniProt entry P34897, NCBI Reference Sequence: NP_001159831.1) cloned into a pET28b(+) vector, with a N-terminus His-tag, as previously described (Giardina et al., 2015). Successful insertion of each mutation was verified by sequencing (Eurofins Genomics).

4.2 Proteins expression and purification

The same expression and purification protocol used for the wild-type SHMT2 was applied to all the mutants regardless of their yield, solubility and/or stability. Briefly, E. coli BL21 (DE3) cells were transformed, grown in LB medium supplemented with kanamycin at 37 °C at 180 rpm to OD₆₀₀ = 0.6 and then induced with 0.1 mM Isopropyl-β-D-thiogalactopyranoside (IPTG). After induction of protein expression, cells were grown at 22 °C overnight. Cells were harvested and the pellet was frozen at −20 °C until use. The pellets from 2L of culture were resuspended in 100 mL of lysis buffer: 20 mM Hepes pH 7.2, 250 mM NaCl, 10% glycerol, 40 μM PLP (hereafter identified as Buffer A) supplemented with 1 mM phenylmethylsulfonyl fluoride (PMSF), on ice for 30 min, then they were lysed by sonication. After centrifugation and clarification by a Millex filter (Millipore), the soluble fraction was added with 10 mM imidazole prior to being loaded on a 5 mL HiTrap Chelating column (Cytiva) previously loaded with NiSO₄. After extensive washing with purification buffer (Buffer A plus 10 mM imidazole), imidazole concentration was increased stepwise up to 500 mM and fractions were collected. All the wild-type and mutant proteins eluted at 300 mM imidazole. SHMT2 containing fractions were concentrated by ultrafiltration (Amicon Ultra, Millipore) and loaded on a G25 desalting column (Cytiva) equilibrated with Buffer A, to remove imidazole prior of the incubation with thrombin to remove His-tag. 1 U bovine thrombin (Sigma) was applied per mg of protein, with o/n incubation at 16 °C. A HiTrap Chelating column (Cytiva) was used to remove the protease and the tags as SHMTs without His-tag elutes at 100 mM imidazole. Fractions were concentrated to 5 mL and loaded on a SEC column Superdex 200 16/600 (Cytiva) and eluted in Buffer A. Protein concentration was assessed by UV-VIS spectrum, with the extinction coefficient ε₂₈₀ = 43,900 cm^-1M^-1 (Malerba et al., 2007). Proteins were aliquoted, flash frozen in liquid nitrogen and stored in Buffer A at −20 °C until use. A summary of the purification yields and outcomes is reported in Supplementary Table S2.

4.3 Differential scanning fluorimetry

Samples for DSF were prepared as follows: 2.5 μL of 20 μM protein solution; 97 μL of 20 mM potassium phosphate buffer (pH 7.2); 0.5 μL of SYPRO orange. Samples were mixed by flicking the tubes by hand and then centrifuged. 25 μL of sample were loaded into each well, and each sample was loaded in triplicate. DSF assays were performed on a Real Time PCR Instrument (CFX Connect Real Time PCR system, Bio-Rad). Temperature was increased by 0.4 °C every 30 s, with fluorescence (excitation 450–490 nm; detection 560–580 nm) recorded for each increase. Starting temperature was 25 °C and final temperature was 95 °C. Denaturation profiles were analyzed using Graph Pad software after removal of points representing quenching of the fluorescent signal due to post-peak aggregation of protein:dye complexes. All curves were normalized and fitted to sigmoidal equation to obtain the melting temperatures (T_m).

4.4 Enzymatic activity

The catalytic activity of the mutants was measured with respect to the wild-type, in 20 mM potassium phosphate buffer, pH 7.2, at 30 °C in a Hewlett-Packard 8453 diode-array spectrophotometer (Agilent Technologies, Santa Clara, CA). For the measurements of the initial velocity of the SHMT forward reaction a coupled assay was employed, in which SHMT catalyzes the reaction Serine + THF → Glycine + 5,10-CH _*2* -THF, and CH₂-THF is in turn oxidized by the NADP⁺-dependent E. coli FolD. The initial velocity of the reaction was recorded by following the increase of NADPH absorbance at 340 nm as described in (Tramonti et al., 2018). Activity assays were performed at 30 °C in 20 mM potassium phosphate buffer (pH 7.2). The absorbance at 340 nm of a solution containing 0.1 mM Serine, 50 μM THF, 0.25 mM NADP⁺; 0.67 μM FolD from E. coli, was acquired as a baseline before starting the reaction by adding the different proteins to a final concentration of 0.2 μM. The SHMT reverse reaction Glycine + CH _*2* -THF → Serine + THF was monitored by measuring the initial velocity in a coupled spectrophotometric assay in which the NAD⁺-dependent L-serine dehydrogenase from P. aeruginosa was used to oxidize L-serine produced by SHMT into 2-aminomethylmalonate semialdehyde (Tramonti et al., 2018). Assays were carried out at 30 °C in 50 mM TRIS/HCl pH 8.8 using 2 μM final enzyme concentration. Both the forward and the reverse reactions were performed in the presence of a 1:2 excess of PLP with respect to the protein concentration. The kinetic parameters of the reaction (L-serine + THF → glycine + CH _*2* -THF) catalyzed by the P499A variant were obtained by measuring the initial velocity using 0.1 μM enzyme and L-serine and THF as substrates by means of the spectrophotometric coupled assay already described (Tramonti et al., 2018). All assays were carried out in 20 mM potassium phosphate buffer, pH 7.2, 0.2 μM PLP, at 30 °C. Saturation curves were obtained varying the concentration of one substrate while keeping the other at a fixed concentration. Kinetic parameters were determined by fitting the data to Equation 1 when L-serine was the fixed substrate or to the Michaelis–Menten equation when THF was the fixed substrate.In Equation 1, which accounts for substrate inhibition acted by THF, V_i stands for the initial velocity, V_max is the maximum velocity, [THF] is the concentration of the variable THF substrate, K_mTHF is the Michaelis–Menten constant for THF, and K_i is the substrate inhibition constant.

V i = V max T H F T H F + K m T H F 1 + T H F K i (1)

4.5 Size Exclusion Chromatography (SEC)

To assess the oligomerization state of the mutants, they were analyzed by SEC in the absence of an excess of PLP in the buffer. 500 μL of 20µM (or 5 µM) protein samples were applied on a Superdex 200 10/300 (GE Healthcare) column equilibrated with 20 mM Hepes pH 7.2, 250 mM NaCl, 10% glycerol and eluting peaks were detected at 280 and 420 nm.

4.6 Analytical Ultracentrifugation

Sedimentation velocity experiments were performed on a Beckman Coulter Proteomelab XLI analytical ultracentrifuge equipped with absorbance optics. The experiments were conducted at 25,000 rpm at 20 °C. Each sample (400 µL) was diluted to 0,8 AU_280nm in a 1.2 cm optical path cell in 20 mM Hepes pH 7.2, 250 mM NaCl, 10% glycerol, 40 μM PLP. Radial absorbance scans were collected at 280 nm at a spacing of 0.003 cm with three averages in continuous scan mode (50 scans/cell). Sedimentation coefficients were determined using the program Sedfit (Schuck, 2000) (provided by Dr P. Schuck, National Institutes of Health) and were reduced to water (S20,W) using standard procedures.

4.7 Apo enzymes preparation

PLP removal from the holo proteins was performed by hydroxylamine, according to Giardina et al. 2015: 500 µL of 20 µM holo SHMT2 in purification buffer A were mixed with 500 µL of buffer 50mM Na Hepes pH 7.2, 200 mM NaCl, 0.1mM EDTA, 0.1mM DTT, 40 mM hydroxylamine, for 1 h at room temperature and loaded on a PD10 desalting column (GE Healthcare) and eluted with the same buffer without hydroxylamine. PLP removal was checked by UV-VIS spectrum, following the decrease of the visible peak at 430 nm. For some mutants, apo form was not obtained with this protocol, and further cycles of ultrafiltration, incubation with hydroxylamine and desalting were applied.

4.8 Measurement of K_d for PLP binding

Equilibria of PLP binding to apo-SHMT2 mutants were analyzed by recording the quenching of the protein intrinsic fluorescence upon PLP binding (Florio et al., 2009). In the fluorometric measurements, PLP from 0.05 to 30 µM (in the case of mutant T186R, the cofactor from 0.25 to 75 µM) was added to 1 µM apo-SHMT2 samples at 20 °C in 50 mM sodium HEPES, pH 7.2, containing 0.2 mM dithiothreitol and 0.1 mM EDTA. Fluorescence emission spectra (300–420 nm) were recorded in a 1 cm path length quartz cell 5 min after mixing PLP and apo-enzyme, using a FluoroMax-3 Jobin Yvon Horiba (Kyoto, Japan) spectrofluorometer with the excitation wavelength set at 280 nm. Fluorescence change at 336 nm as a function of PLP concentration was normalized between 0 and 1 as fraction of protein:ligand complex formation (Y) and K_d was determined by fitting these points to the equation: Y = B_max·PLP/(K_d + PLP), where B_max is maximum bound to unbound protein ratio and PLP is the cofactor concentration (µM), in Prism GraphPad.

4.9 Thermal denaturation

Protein samples were diluted to 5 µM in the purification buffer containing 40 µM PLP and placed in a quartz cuvette with a 1 mm optical path. CD spectra were acquired at 20 °C using a Jasco J-815 Spectropolarimeter. Thermal unfolding profile was monitored heating the sample from 30 to 90 °C at a constant rate of 1 °C/min while recording the CD signal at 222 nm every 0.5 °C T increment. A complete CD spectrum (from 270 to 200 nm) was also acquired every 5° up 70 °C. Thermal denaturation data were normalized between 0 and 1 as a fraction of unfolded protein. Data were not fitted as the denaturation transition was not reversible.

4.10 Chemical denaturation

For each denaturation experiments 2 protein solutions were prepared: (A) containing 2 µM protein and 4 µM PLP in 20 mM Hepes pH 7.2, 250 mM NaCl, 10% glycerol; and (B) containing the same components plus 7.6 M urea. Different volumes from solution A were exchanged with equal volume of solution B to obtain protein solutions at different urea concentration (ranging from 0 to 7.6 M) with a stepwise increase of 0.4 M urea. Each solution was allowed to equilibrate for 10 min at RT before being transferred into a quartz cuvette (2 mm path length) to acquire complete CD spectra from 250 to 200 nm at 20 °C on a Jasco J-815 Spectropolarimeter. The denaturation curve was then obtained by plotting the fraction of unfolded protein (obtained from the normalized CD signal at 222 nm) as a function of urea concentration. Data were then fitted according to a two state model in Prism GraphPad as reported in (Catalano et al., 2026; Troilo et al., 2026).

4.11 Crystallization, data collection, model building and refinement

Protein crystals were grown by the sitting drop vapor diffusion method, by mixing 0.4 μL of protein solution (T186R 55μM; P157S 60 μM; G423S 25 μM) and 0.4 μL of the MORPHEUS screen (Molecular Dimensions) at 21 °C. All the crystallization conditions are reported in Supplementary Table S3. Crystals grew in about 1 week to the final dimensions and were flash frozen in liquid nitrogen before X-ray exposure. Data collections were performed either at ESRF (Grenoble, FR) on beamline ID30B or at Elettra (Trieste, IT) on beamline XRD2 synchrotrons. Data collection parameters and diffraction statistics are reported in Supplementary Table S4. Auto processing of diffraction images was performed using the automated software pipeline available at the two synchrotrons. Phasing was performed by Molecular Replacement using different templates depending on the oligomeric state of the protein (SHMT2-WT apo PDB 6DK3 dimer - unpublished; SHMT2-WT holo PDB 4PVF tetramer - Giardina et al., 2015) with MOLREP (Vagin and Teplyakov, 2010). Model building and refinement were performed with Coot (Emsley et al., 2010) and Refmac5 (Murshudov et al., 2011; Nicholls et al., 2013) respectively. All statistics related to this step are reported in Supplementary Table S4.

The atomic coordinates and structure factors for the reported structures have been deposited in the Protein Data Bank (PDB) under accession code: T186R dimer - 32JA; T186R tetramer - 32JC; P157S dimer – 32JN; P157S tetramer 32JD; G423S tetramer – 32JY.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. To proofread the revised manuscript, figures and Supplementary Material for typographical errors, inconsistencies in nomenclature, and non-native English expressions. To create a draft of the summary table containing the key properties of all SHMT2 variants analyzed in the present study. The table was then inspected and revised for errors.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

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.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fchbi.2026.1941005/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Massimiliano Perduca, University of Verona, Italy

Reviewed by: Matteo Ardini, University of L’Aquila, Italy

Ales Hnizda, Charles University, Czechia